388.4042 
 An 13 
 
 PB80-192412 
 
 An Analysis of User Cost and Service 
 Trade-Offs in Transit and Paratransit Services 
 
 Cambridge Systematics, Inc., MA 
 
 Prepared for 
 
 Transportation Systems Center 
 Cambridge, MA 
 
 Aug 79 
 
 U.S. DEPARTMENT OF COMMERCE 
 National Technical Information Service 
 
► 
 
 > 
 
 l'\’IVERSITY OF 
 _ ILLINOIS LIBRARY 
 AT URSANA-CHA.MPAinN 
 
 
 
 
 
 
UMTA-MA-GS-3043TS-10 
 
 UMTA/TSC Evaluation Series 
 
 An Analysis of User Cost and 
 Service Trade-Offs in 
 Transit and Paratransit Services 
 
 Final Report 
 August 1979 
 
 Service and Methods Demonstration Program 
 
 U S. DEPARTMENT OF TRANSPORTATION 
 Urban Mass Transportation Administration 
 Research and Special Programs Administration 
 Transportation Systems Center 
 
 REPRODUCED 8Y 
 
 NATIONAL TECHNICAL 
 INFORMATION SERVICE 
 
 U£. DEPARTMENT OF COMMERCE 
 SPRINGFIELD. VA 22161 
 
THIS DOCUMENT HAS BEEN REPRODUCED 
 
 FROM THE BEST COPY FURNISHED US BY 
 
 THE SPONSORING AGENCY. ALTHOUGH IT 
 
 IS RECOGNIZED THAT CERTAIN PORTIONS 
 
 ARE ILLEGIBLE, IT IS BEING RELEASED 
 
 
 
 . •••-.' 1 
 
 . ‘ 
 
 • ‘ • • ' 
 
 . ' ’ ' ... ’ ‘ ; ' s ' • ' ' ’ 
 
 
 
 
 
 
 
 
 
 
 . .. 
 
 
 . . 
 
 
 
 
 
 
 
fasru 13 
 
 NOTICE 
 
 This document is disseminated under the sponsorship 
 of the Department of Transportation in the interest 
 of information exchange. The United States Govern¬ 
 ment assumes no liability for its contents or use 
 thereof. 
 
 NOTICE 
 
 The United States Government does not endorse pro¬ 
 ducts of manufacturers. Trade or manufacturers' 
 names appear herein solely because they are con¬ 
 sidered essential to the object of this report. 
 
Appiorimatt Convmiorts to Mr.Jfic Measures r "" - « Appro* mate Conversions Iron-. Metric 
 
 „ •*> *-• 
 
 : 4 p a; 
 
 19^ O 
 
 I 
 
 vt -A , 
 
 a. cs jfi . 
 
 u u a ^ r: 
 
 - > £ ' 
 
 ® © « 
 
 _• j. /}• 
 
 
 s - 
 
 
 VJ *0 >' » 
 
 « | £ 
 
 „ * “ B » Si 
 
 5 ? 1 T f ■§ 
 
 c. a a a» C> <j 
 
 i 9 
 
 u 2 
 
 
 C 
 
 ,£ *-'■ 
 
 >• - 
 
 i P 
 
 a 
 
 < 
 
 3 vr m r- sc 
 
 s»* 
 
 JS 
 
 CXI 
 
 5T 
 
 r- rw «tr \f> > 
 
 “ __ 
 t5 tv w* 72 
 
 W 
 
 U4 
 
 <n'. a 
 
 
 '£ - Z 
 
 t? 04 
 
 O C- >ri ^ O 
 
 < 
 
 o’ -- d' tv 
 
 w 
 
 o ^ -• 
 
 . ® 
 
 t cv '»- d W' - 
 
 < 
 
 • " : 
 
 
 
 < 
 
 . s ] 
 
 > 
 
 NOTICE 
 
 cc 
 
 UJ 
 
 , 
 
 * 
 
 V ] 
 
 , m o 
 
 V° 
 r c- 
 
 
 THIS DOCUMENT HAS BEEN REPRODUCED 
 FROM THE BEST COPY FURNISHED US BY 
 THE SPONSORING AGENCY. ALTHOUGH IT 
 IS RECOGNIZED THAT CERTAIN PORTIONS 
 
 x 0& 161 ' 81 it ! 91 I si - I H; I El . 12V |li j 01 |6. i'8 l '9 IS i* |C [l )l 
 
 ARE ILLEGIBLE, IT IS BEING RELEASED 
 IN THE INTEREST OF MAKING AVAILABLE 
 
 .. 
 
 O I 
 
 ! 1 1 ' . 1 : l 1 1 T- 
 
 ;'!TiYITmTr 
 
 mv\ 
 
 : , i. , | , {TfTf , | 
 
 
 rmm 
 
 AS MUCH INFORMATION 
 
 If i 7 ■ U V U 1 4 
 
 AS 
 
 POSSIBLE. 
 
 ] ‘ 
 
 1 1 inches 1 
 
 . £ ~ A 
 
 z z S 
 
 > g 
 
 I 5 v ' f 
 
 i 2 S i 
 
 s st- 
 
 C C C OC 
 
 4> S> ; a> • 
 
 I °’ 
 
 if. 
 
 2 i : 
 
 © « • r ■ " 
 
 ? 2 ‘. 2 g 
 
 l I 
 
 I j 
 
 111 
 ^ jc £ 
 
 “ ■ </i v. tr u t> 
 
 — * 5 5 © £ x 3 
 
 e c r /: .r .r s 3 
 
 h e ZZJZ-JZ w o 
 
 It 
 
 ■E 
 
 ' sfc 
 
 '5 
 
 ■|fr «£> 
 
 n 6 o' t* 
 
 : ttV: 
 
 ' tv'-© 
 
 o> 
 
 '«> 
 
 5 
 
 w j 
 
 <1 
 
 t 
 
 WU 
 
 4^3 
 
 < 
 
 .c —. v ® 
 
 ■;« 
 
 5 
 
 a 1 -’ wn o. © © o 
 
 tt> o f*> 
 
 <n © © 
 
 < 
 
 CP 
 
 O. 
 
 S' 
 
 4U 
 
 P u 
 
 I : i 
 
 
 Vi 
 
 e © 
 
 t © 
 1 * 
 I ;f 
 
 i i i 
 
 5 P 
 
 n»s 
 
 o ~ 
 
 * 5 $ 
 
 •tt -JS >-. 
 
 g V u. 
 
 r a .a 
 
 o cr d j u 
 
 ir* > £ 
 
 :S. -- >■ c 
 
 a u -ci - 
 
 ; §. 
 
 * JT 
 
 . 
 
 •« • r ■ 
 
 ' ' ' ' 
 
 if 
 
Technical Report Documentation Page 
 
 1. Report No. 
 
 UMTA-MA-06-0049-79-10 
 
 2. Government Accession No. 
 
 4. Title and Subtitle 
 
 An Analysis of User Cost and Service Trade-Offs 
 in Transit and Paratransit Services. 
 
 7. Author's) 
 
 J. Louviere, and G. Kocur 
 
 3. Recipient's Catalog No. 
 
 P'S io- j<fOL<-+iz 
 
 S. Report Oate 
 
 August 1979 
 
 6. Performing Organisation Code 
 
 8. Performing Organization Report No. 
 
 UM927-R9742 
 
 9. Performing Organisation Name and Address 
 
 Cambridge Systenatics, Inc.* 
 
 238 Main Street 
 
 Cambridge, Massachusetts 02142 
 
 10. Work Unit No. (TRAIS) 
 
 MA-06-0049 
 
 11. Contract ar Grant No. 
 
 DOT-TSC-1405 
 
 12. Sponsoring Agency Nome and Address 
 
 U.S. Department of Transportation 
 Urban Mass Transportation Administration 
 400 Seventh Street, S. W. 
 
 Washington, D. C. 20590 
 
 13. Type of Report and Period Covered 
 
 Final Report 
 November 1977-December 
 1978 
 
 14. Sponsoring Agency Code 
 
 UPM-30 
 
 15. Supp I ament or y Notes 
 
 *Under contract to: 
 
 Transportation Systems Center 
 
 Research and Special Programs Administration 
 
 Cambridge, Massachusetts 02142 
 
 16. Abstract 
 
 The Xenia Model Transit Service served as a test of several alternative transit 
 services operated in a small city setting. This project was undertaken as an 
 aid to the evaluation of transit and paratransit systems in Xenia, Ohio, and 
 as a test of a new technique for assessing travel demand. The technique, direct 
 utility assessment, was designed to test a new method for assessing user trade¬ 
 offs in costs and service based on attitudinal methods. A tradeoff survey was 
 administered as part of a home interview survey. Data from the tradeoff survey 
 were used to develop separate equations for each sample respondent to explain 
 and describe their tradeoffs over transit fare, travel time, walk distance, 
 type of service, and headway. An aggregate equation was also developed, 
 assuming that all respondents shared common tradeoffs. 
 
 These equations were employed to retrospectively predict changes in transit 
 system patronage (since 1974). Both sets of models performed well, producing 
 forecasts that were in the same direction and range of experience, although 
 magnitudes were somewhat different. Coefficients of the individual tradeoff 
 equations were then analyzed to see if they could be predicted on the basis of 
 interpersonal characteristics of the respondents. Results indicated that 
 differences in coefficients could be attributed to some differences in individuals 
 such as income and auto ownership. Overall results were promising for policy 
 evaluation and forecasting. This report contains a bibliography. 
 
 17. Kay Words 
 
 Transit Service 
 Travel Demand 
 Demand-Responsive 
 Attitudinal Models 
 
 Surveys 
 
 Forecasting 
 Paratransit 
 Small City 
 Fares 
 
 Assessment Technique Travel Time 
 
 18. Distribution Sfotement 
 
 Document is available to the public 
 through the National Technical 
 Information Service, Springfield, 
 Virginia 22161. 
 
 19. Security Clossif. (of this report) 
 
 Unclassified 
 
 20. Security Clossif. (of this page) 
 
 Unclassified 
 
 21* No. of Poges 
 
 210 
 
 22. Price 
 
 Form DOT F 1700.7 (8-72) 
 
 Reproduction of completed poge authorized 
 
 l 
 
METRIC CONVERSION FACTORS 
 
 
 «> 
 
 2 
 
 E 
 
 o 
 
 © 
 
 *5» 
 
 © 
 
 CJ 
 
 9 
 
 E 
 
 ? 5 : > i 
 
 XOKS for c? 
 
 O — « 
 
 usitS 
 
 Relative 
 •in Xenia. 
 
 t > E 
 
 in 2 
 
 s 
 
 C 
 
 sis 
 
 » a a o w u 
 
 I | 
 
 | I 
 
 O 
 
 S -* 
 
 © r> 
 
 odn 
 
 N v in 
 d ^ cd c>J 
 
 to 
 
 < 
 
 « S 
 
 •S </> C 
 
 t. a> 
 
 lie 
 
 V5 © 
 
 ^ © i 
 
 Ewe 
 
 li 
 
 Is 
 
 I s 5 • 
 
 •; « i 2 
 
 © i r - 
 
 u C -* *> 
 
 k (t n ~ 
 
 3 3 3 W 
 
 r cr 9 © 
 
 *• </> «fl i 
 
 ,V ■<■ : 
 
 Horn*: 
 
 sir 
 
 s | 
 
 e n ^ 3 
 
 ON*" 
 
 O 
 
 > 
 
 r of Alte 
 
 f 
 
 E •—■ 
 
 • *> 
 
 m <£> 
 © «- © 
 © fvJ «— 
 
 for 
 
 tl&t 
 
 ibut 
 
 S n 
 © to 
 
 r> 
 
 Att 
 
 s 
 
 c 
 
 © _ 
 
 € ™ 
 
 C. f*> 
 
 <5 ? 
 
 uate< 
 
 i»i 
 
 E g 
 
 C3> JZ i- 
 
 T *■ tfl 
 
 ^ © © 
 
 e-H 
 
 I I 
 
 *> u o 
 
 © «o d 
 
 «- 3 3 
 
 — U O 
 
 m «> 
 
 e 
 
 jo 
 
 E 
 
 *01 
 
 E E I 
 
 ot 
 
 c E ic 
 v E x 
 
 for Dir 
 na‘ 
 
 Assessment 
 
 of- 
 
 V06 
 
 E-E E 
 
 ana 
 
 u 
 
 8 - 
 
 _o 
 
 , _o 
 
 VO 
 
 CD 
 
 CD 
 
 O - 
 
 .§o 
 
 — o 
 
 '5 11 
 
 6~1 
 
 -o 
 
 » O 
 
 o X o o 
 
 V VO 
 
 A-l 
 
 EI 
 
 If!! 
 
 zz 
 
 !!(!]):• 
 
 xz oz 
 
 :«»"!«1J i • ; I ' ( *|t 
 
 ’V* ; 1 
 
 HI' 
 
 61 
 
 81 
 
 i! 
 
 ttt! ill! !!li:!li;:!i#f|!i! 
 
 91 I SI 
 
 !!!' f|!f!!W | !ll! 
 
 •thod of Construction of Orthogonal 
 
 *T 
 
 EX 
 
 b 
 
 ilflf!!!!!!!!! 
 
 zx 
 
 XI 
 
 It! 1 
 
 01 
 
 ’illt'll! 
 
 mii'iii 
 
 !!!li!!i! 
 
 M»t; !'■ i M i: it ■! 
 t: 11 i x i •«> i t i • 1111 
 
 
 t!!'l|!'t 
 
 
 
 i 11 t I i i (' t u.; 11111 >«i i i! 11 i I i i, 1111 > i H It 111 • t i 111 
 
 TIT 
 
 TIT 
 
 TIT 
 
 TIT 
 
 TIT 
 
 TJT 
 
 TIT 
 
 TIT 
 
 TIT 
 
 T|T 
 
 T|T 
 
 TIT 
 
 T(T 
 
 T|T 
 
 TIT 
 
 T 
 
 T 
 
 T 
 
 T 
 
 T|T 
 
 i 
 
 i 
 
 l 
 
 1 
 
 :dlV; 
 
 
 ' 
 
 1 
 
 1 
 
 1 
 
 
 :.cie 
 
 5 
 
 
 Wei; 
 
 
 
 
 
 
 
 9 
 
 8 
 
 
 7 
 
 
 6 
 
 
 6 
 
 
 4 
 
 
 3 
 
 
 2 
 
 i 
 
 inches 
 
 LIOGRAPHY WITH SELECTED. ANNOTATIONS 
 
 E , E 
 v E jc 
 
 © * - 
 
 C £ «> 
 
 3 s r 3 
 
 s s 
 
 II 
 
 19 
 
 *• < 
 
 ^ L1J 
 
 5 J E 
 
 E 'V E © 
 u E E jc x 
 
 S 5 
 V •- 
 
 © © © © z 
 
 3 3 3 3 U 
 ^ O’ O’ O' © 
 
 IT. © 
 
 <£> d 
 
 = 5 > E 
 © © © © 
 
 t -r > e 
 
 £ 5-. © 
 e c = 
 £ “ 5 
 
 T> JC Z 
 
 E E £-EE 
 
 c © 
 
 T ~ ~1 *) if. in in y w 
 
 5© © 0 ) £ is 
 
 EEEti^^XyS 
 
 V) 
 
 00 
 
 < 
 
 v n ifl r*» © 
 N V « © O N 
 
 40 to o © o d ro o © 
 
 5 - 
 
 JC 
 
 <* ° © c ,, 
 
 ®-o««£roS.S 
 
 fi. 
 
 s 
 
 t ” S3 
 
 © •- 
 
 ~ U 
 
 ffi £ - 
 v 3 N 
 is w n 
 
 2 ® 
 
 3 O 
 
 1 z 
 £ ? 
 
 " 3 
 
 > 
 
 D '/? 
 % a 
 J X 
 
 5 • 
 
 I I. 
 
TABLE OF CONTENTS 
 
 ■?age 
 
 PREFACE 
 
 1. EXECUTIVE SUMMARY 1-1 
 
 1.1 Methodology • 1-1 
 
 1.2 Conclusions on User Service and Cost Tradeoffs 1-10 
 
 2. SUMMARY OF DEMONSTRATION 2-1 
 
 2.1 Overview and Objectives 2-1 
 
 2.2 Project Setting 2-2 
 
 2.3 Project History 2-3 
 
 2.4 Conclusions 2-8 
 
 2.5 Implications of Demonstration Results for Other Cities 2-15 
 
 3. DESCRIPTION OF HOME INTERVIEW SURVEY 3-1 
 
 3.1 Telephone Screening 3-1 
 
 3.2 Home Interview Survey 3-3 
 
 4. METHODOLOGY 4-1 
 
 4.1 Overview of Research Objectives 4-1 
 
 4.2 Direct Value Assessment: An Overview 4-10 
 
 4.3 Factorial Sampling Plans 4-22 
 
 4.3.1 Fractional Factorial Sampling Designs 4-25 
 
 4.3.2 The Xenia Sampling Plan 4-35 
 
 4.4 Analytical Techniques for Model Specification 4-47 
 
 4.5 Forecasting From Individual Utility Models 4-57 
 
 5. RESEARCH RESULTS 5-1 
 
 5.1 Individual Level Response Specifications 5-1 
 
 5.1.1 Analysis of Individual Differences in 
 
 Coefficients 5-4 
 
 5.1.2 Individual Forecasting System Results 5-16 
 
 Aggregate Level Response Specifications 
 
 5.2 
 
 5-23 
 
TABLE OF CONTENTS (Continued) 
 
 6 . 
 
 IMPLICATIONS FOR DEMONSTRATION EVALUATION 
 
 6.1 
 
 6.2 
 
 6.3 
 
 6 . 4 
 
 Relative Importance of Cost and Service Tradeoffs 
 in Xenia 
 
 General Applicability of the Technique to Other 
 Demonstration Projects 
 
 Issues to be Addressed in Future Applications 
 
 '' <” T| fT . i T * 1 I" P ■ • ■ • . I 
 
 - / 
 
 6.3.1 Alternative Survey Approaches 
 
 6.3.2 Interface and Comparison with Other Disaggregate 
 Modelling Strategies 
 
 6.3.3 Finding Appropriate Measures for Attributes 
 
 6.3.4 Expanding the Number of Alternatives Evaluated 
 
 6.3.5 Expanding the Number of Attribute Dimensions 
 
 6.3.6 Alternatives to Home Interviews 
 
 6.3.7 Problems of Missing Data 
 
 Summary of Conclusions > - ^ - 
 
 1 '• ; ;: Service, Holding Fare Constant 
 
 iteraction Graphs of Time x Cost 
 
 6-1 
 
 6-1 
 
 . i 
 
 6-5 
 
 6-8 
 
 6-8 
 
 6-11 
 
 3-2 
 
 6-12 
 
 6-13 
 
 6-14 
 
 6-15 
 
 6-16 
 
 4-1 
 
 APPENDIX A: 
 
 5.2 Obsi 
 
 C O 
 
 APPENDIX B: 
 
 Algebraic Theory for Direct Value Assessment and 
 
 Overview of Functional Measures 
 
 rved- ver ted Work Mode Spl. 
 
 AT Mrtfl Tin « r« 
 
 Purpose and Method of Construction of Orthogonal 
 Polynomial Contrasts 
 
 APPENDIX C: Xenia Home Interview Survey 
 
 APPENDIX D: Results of Multiple Analyses of Variance for 
 
 Individual Regression Coefficients and Weights 
 
 . ~ 16 
 ‘-48 
 -49 
 
 6-17 
 
 4-50 
 
 4— 54 
 
 A-l 
 
 5- 21 
 5-26 
 
 B-l 
 
 BIBLIOGRAPHY WITH SELECTED ANNOTATIONS 
 
 I 
 
LIST OF TABLES 
 
 2.1 Operating Statistics for Demonstration Services 
 
 4.1 Factorial Combinations in Bus Service Example 
 
 4.2 Full Factorial Coding Example 
 
 4.3 Orthogonal Coding of the Design of Table 4.2 
 
 4.4 1/3 Fractional Plan, Polynomially Coded 
 
 4.5 Correlation Matrix of Effects in Xenia Design 
 
 4.6 Combinations within Types of Service for the Xenia Design 
 
 4.7 Time x Cost Interaction Means 
 
 5.1 Descriptive Statistics for All Individual Coefficients 
 
 5.2 Results of Individual Forecasting System 
 
 5.3 Aggregate Sketch Planning Results 
 
 5.4 Comparison of Model Forecasted Elasticities with 
 Observed Xenia Values 
 
 • • 
 
 6.1 Relative Importance Comparisons 
 
 D.1 Results of Multiple Analyses of Variance for Individual 
 Regression Coefficients and Weights 
 
 Page 
 
 2-6 
 
 4-24 
 
 4-28 
 
 4-33 
 
 4-34 
 
 4-38 
 
 4-46 
 
 4- 53 
 
 5- 2 
 5-19 
 5-29 
 
 5- 31 
 
 6- 3 
 
 D-l 
 
 V 
 
LIST OF FIGURES 
 
 Page 
 
 3.1 Results of Phone Screening and Interviewing, 1977 3-2 
 
 4.1 Alternative Possibilities for the Behavior of Marginal 
 
 Preferences for Levels of Fare 4-12 
 
 4.2 Additive and Multiplicative Joint Function 4-16 
 
 4.3a Average Sample Response to Combinations of Frequency of 
 
 Service and Fare, Holding Walking Constant . 4-48 
 
 4.3b Average Sample Response to Combinations of Fare and Walking 
 
 Distance, Holding Service Constant 4-49 
 
 4.3c Average Sample Response to Combinations of Walking 
 
 Distance and Frequency of Service, Holding Fare Constant 4-50 
 
 4.4 Interaction Graphs of Time x Cost 4-54 
 
 5.1 Results of Analysis of Interpersonal Effects 5-7 
 
 5.2 Observed Versus Predicted Work Mode Split 5-21 
 
 5.3 Summary of Aggregate Model Responses 5-26 
 
PREFACE 
 
 As part of the Urban Mass Transportation Administration Service 
 and Methods Demonstration (SMD) Program, Cambridge Systematics, Inc., 
 under contract to the US Department of Transportation, Transportation 
 Systems Center, has prepared the following final report on An Analysis 
 of User Cost and Service Tradeoffs in Transit and Paratransit Systems. 
 
 This report is based on analyses of data collected with the assis¬ 
 tance of the Xenia Department of Transportation, the Transportation 
 Coordinating Committee and the Montgomery - Greene County Transporta¬ 
 tion and Development Program. Carla Heaton of the Transportation 
 Systems Center was the Project Monitor for this study and provided 
 guidance for the effort. 
 
 The authors would also like to thank the Cambridge Systematics, 
 Inc. staff who assisted on the project. Melissa Laube assisted in the 
 writing of Chapters 2 and 3; James Wojno assisted in data preparation 
 and graphics. Several University of Iowa faculty provided valuable 
 advice and assistance with data analysis: George Woodworth in 
 Statistics, Jerry Eskin in Marketing, and Jim Stoner in Civil 
 
 Engineering. 
 
very difficult. This situation arose because the fixed-route transit, 
 system in Xenia operated at uniform.headways, fares, and speeds on 
 ail routes, and auto travel also was characterized by relatively -k-.y/';' 
 uniform speeds and no parking costs. The paratransit system had only 
 system-level data on average wait times and.travel times, and very 
 limited variation in fares. In all cases, there was insufficient 
 variation in most variables of interest to adequately assess their 1 
 impact on travel choice. Thus, the evaluation was extremely limited 
 
 between transit and paratransit systems, a key objective of. the 
 evaluation.. The direct assessment technique was introduced in 
 
 response to this need. 
 
 The lack of variability in the explanatory ievel-of-service 
 
 r, ; v ;. • 
 
 variables found in Xenia is. likely to be encountered in many 
 
 ■ 
 
 
 small -if 
 
 ii >an areas, with s Lmpl* transit and highway networks,. Thus, this 
 
 methodology, if successful, can have applications in many other 
 demonstration projects. Other cases in which direct utility assess¬ 
 ment offers the promise of improved analysis capabilities are the 
 introduction of new modes (often done in SMD projects), the effect 
 of new settings on travel behavior (for example, gas rationing, or> 
 which no data currently exists)., or for variables affecting travel 
 demand which are unmeasurably invariant or too expensive to measure 
 (for example, the effect of taxis versus buses in a service in which 
 only one vehicle type has been used). All of these areas may be of 
 interest in future projects. 
 
 1-2 
 
1. EXECUTIVE SUMMARY 
 
 1.1 Methodology 
 
 This project was undertaken as an aid to the evaluation of tran¬ 
 sit and paratransit system in the Xenia, Ohio Service and Methods 
 Demonstration (SMD) project, and as a test of a new technique for 
 assessing travel demand. The technique is called direct utility 
 assessment , and consists of an experimental design (described below) 
 containing a set of situations in which individuals are asked to rate 
 their likelihood of making a certain choice, in this case of using 
 transit versus driving. These responses can then be analyzed to 
 produce a utility function with respect to the variables contained in 
 the situations, for each individual or for groups of individuals. 
 
 The utilities can then be related to the actual decision-making of 
 the sample. 
 
 The heart of the technique is the experimental design. In it, a 
 set of variables are defined, each with several discrete levels, in 
 
 t 
 
 # » 
 
 such a way that each variable is completely uncorrelated with every 
 other variable. This independent, or orthogonal, design allows one 
 to analyze the response to each variable totally unconfounded with 
 any other variable. This propensity of the direct utility assess¬ 
 ment technique is what motivated its use on the data from Xenia, as 
 the actual data describing travel choices in Xenia was very highly 
 correlated, and made the calibration of a more typical demand model 
 
very difficult. This situation arose because the fixed-route transit 
 
 XU 
 
 system in Xenia operated at uniform headways, fares, and speeds on 
 
 all routes, and auto travel also was characterized by relatively 
 
 ■ 
 
 uniform speeds and no parking costs. The paratransit system had only 
 
 Because there are 18 numerical response, observations for 
 
 system-level data on average wait times and travel times, and very 
 
 U-*- v x,uu5.i iroin tn6 sufvsy^ it Weis possibls to dsv^lou 
 
 limited variation in fares. In all cases, there was insufficient 
 
 a in 3,nd describe 
 
 variation in most variables of interest to adequately assess their 
 
 tradeoffs. The equations are of the following general form* 
 
 impact on travel choice. Thus, the evaluation was extremely limited 
 
 in its ability to address the cost and service level tradeoffs 
 
 l i = S 0 + B^TSp + (BK ) + 6 (TK.) + 0. (F ) + 8 t (F.) 
 
 between transit and paratransit systems, a key objective of the 
 
 evaluation. The direct assessment technique was introduced in 
 
 1 . 0 ) 
 
 response to this need. 
 
 - ct^d numerical response to each of the 
 
 i (—18) alternatives for trio Purnose i (work or shooV 
 
 The lack of variability in the explanatory level-of-service 
 
 variables found in Xenia is likely to be encountered in many small 
 
 urban areas with simple transit and highway networks. Thus, this 
 
 methodology, if successful, can have applications in many other 
 
 Zs the rieaaway for paratrMsit (taxi) system (call two 
 
 j . ead or rt on demand) . , . , ,. ...... 
 
 demonstration projects. Other cases m which direct utility assess- 
 
 15 
 
 ment offers the promise of improved analysis capabilities are th 
 
 level squared, used to detect-nonlinearity In the re- 
 
 spoils C- tQ f £1X0 
 
 introduction of new modes (often done in SMD projects), the effect 
 
 W i is the number of blocks to walk to the nearest transit 
 
 of new settings on travel behavior (for example, gas rationing, on 
 
 the 
 
 which no data currently exists), or for variables affecting travel 
 
 •T. is the difference in travel time between auto and transit; 
 
 demand which are unmeasurably invariant or too expensive to measure 
 
 minutes longer than auto) 
 
 (for example, the effect of taxis versus buses in a service in which 
 
 B 0 - 0 are regression coefficients 
 
 only one vehicle type has been used). All of these areas may be of 
 
 e. is the error term , : 
 
 interest in future projects. 
 
 1-2 
 
The analysis of user cost and service tradeoffs in transit and 
 paratransit systems in Xenia, Ohio involved four major components 
 which were designed to assess user tradeoffs among cost and level-of- 
 service attributes through the application of direct utility assess¬ 
 ments. These research components were: 
 
 1. To develop a survey instrument which could be used to measure 
 respondent tradeoffs among the attributes of cost, type of transit 
 service, headway, walk distance, and travel time differences of 
 transit compared to auto. This instrument was developed for use with 
 
 the Xenia, Ohio Home Interview Survey conducted as part of the 
 
 • • t # » 
 
 evaluation of a model transit service demonstration program in opera¬ 
 tion in Xenia since 1974. The instrument consisted of 18 different 
 hypothetical transit systems, each with a different combination of 
 levels of the four attributes.^ For example, one alternative was a 
 fixed-route bus system which passed by a bus stop six blocks from the 
 respondent's residence on a 15-minute headway, charged a 50-cent 
 fare and took the same amount of time to reach the respondent’s 
 destination as an automobile; a second alternative was a paratransit 
 (taxi) system that could be called on demand, charged a 50-cent 
 fare and took 20 minutes longer to reach the respondent's destination 
 than a private automobile. The respondent was asked to estimate, by 
 responding on a numerical scale to each alternative, "how frequently 
 
 ^ Headway and type of transit service were considered a single com¬ 
 posite attribute; cost, walk time, and travel time were the others. 
 
 1-3 
 
he or she might use” each of these two services and an additional 16 
 alternatives to make a typical work or shopping trip. In addition 
 to these data, each respondent supplied current and historical 
 personal data about themselves and their household. 
 
 2. Because there are 18 numerical response observations for 
 each individual from the survey, it was possible to develop unique 
 regression equations for each respondent to explain and describe 
 their tradeoffs. The equations are of the following general form: 
 
 D€ 
 
 L i “ 6 0 + 6 i (TS i ) + 6 2 ( BH i> + B 3 (TH i ) + W + e 5 (Fp 
 + B 6 (W.) + 6 7 (W. 2 ) + 6 g (T i ) + 6 9 (T. 2 ) + £i> 
 
 ( 1 . 0 ) 
 
 where L 
 
 j 
 
 ipproaches were developed. First, a "totally individual" 
 
 is the predicted numerical response to each of the 
 i (=18) alternatives for trip purpose j (work or shop) 
 
 TS^ is the type of transit system (bus or paratransit) 
 
 BH. is the headway for fixed-route bus (every 15 or every 
 
 30 minutes) 
 
 interview survey for each of the five system changes of interest 
 
 TEL is the "headway" for paratransit (taxi) system (call two 
 
 hours ahead or on demand) 
 
 2 
 
 is the fare charged (free, 50 cents, $1); F^ is fare 
 level squared, used to detect nonlinearity m the re¬ 
 sponse to fare 
 
 4 
 
 W. is the number of blocks to walk to the nearest transit 
 
 stop (0, 3 or 6 blocks); W.^ is the square of this 
 walking distance 
 
 choice it . was assumed that the respondent 
 
 alte 
 
 A. IJ C JL V 
 
 ai< 
 
 T. is the difference in trayel time between auto and transit; 
 
 T.^ is the square of travel time (transit 0, 10 or 20 
 minutes longer than auto) 
 
 transit syst^ e time/ which his or her equation pre- 
 
 8n - Bn are regression coefficients 
 0 9 
 
 e. 
 i 
 
 is the error term 
 
 1-4 
 
 70 
 
Note that the presence of squared terms allows one to test for non¬ 
 linear and threshold effects in individuals’ responses. Out of 670 
 respondents surveyed, 451 supplied sufficient data to estimate a 
 unique equation. 
 
 3. Once the individual equations were estimated, the third 
 research task involved testing whether the personal and household 
 measures obtained during the home interview could account for 
 (explain) differences in the individual coefficients; for example, as 
 household income rises, one might hypothesize that individual coeffi¬ 
 cients for cost might become less negative and/or cost might have 
 less impact or explain less variance relative to other factors such 
 as time. A number of similar hypotheses regarding interpersonal 
 measures were tested by means of a multiple analysis of variance in 
 which the individual regression coefficients are dependent variables 
 and the interpersonal measures are explanatory variables. Results 
 indicate that there are a number of highly significant relationships 
 between individual coefficients and interpersonal measures. These 
 results can be used to assess differences in the responses of differ¬ 
 ent traveller groups to proposed level-of-service and cost changes. 
 Hence, this type of analysis can guide design and implementation of 
 demonstration projects as well as evaluate the impacts and results 
 
 of such projects. 
 
 4. The next task was to develop forecasting systems that could 
 be used to retrospectively predict the transit mode shares back 
 through time for each of the major stages in the demonstration in 
 
 1-5 
 
order to provide a preliminary assessment of validity and usefulness. 
 There were a number of significant changes in fares and levels of 
 service between 1974 and the present in Xenia beginning with free 
 
 fixed-route bus service with a 30-minute headway in 1974. There were 
 two route structure changes and one fare increase to 25 cents in 
 1975; then the system was converted to a paratransit mode with a 
 50-cent fare for call-ahead or immediate request service in 1976, 
 followed by a fare differentiation in 1977 of 50 cents for off-peak 
 call-ahead, 75 cents for peak call-ahead and $1 for peak immediate 
 request service. 
 
 In order to forecast the transit mode shares for these systems, two 
 modelling approaches were developed. First, a "totally individual" 
 system using all 451 regression equations described above was created. 
 Values for the four attributes were measured for each individual 
 intra-Xenia trip observed in the travel survey portion of the home 
 interview survey for each of the five system changes of interest 
 since 1974. These attribute measures for each trip (type of mode, 
 headway, cost, walking distance and travel time difference) were 
 substituted in the equation of the respondent who made the particular 
 trip to generate predicted "likelihood to use" values for each mode 
 alternative. To predict choice it was assumed that the respondent 
 would be most likely to select that mode alternative (auto or the 
 transit system available at the time) which his or her equation pre¬ 
 dicted them to be most "likely to use"; i.e., the highest predicted 
 
 1-6 
 
value. Mode shares were generated by counting the total number of 
 
 respondents who were "most likely" to choose each alternative mode. 
 
 / 
 
 Results of this exercise were reasonably consistent with the 
 pattern of ups and downs in the observed Xenia patronage, particularly 
 for work trips. Also, the magnitude of the estimates was within the 
 range of experience in Xenia, although too low in 1974 and too high 
 in 1977. Reasons for discrepancies cannot be discerned from the 
 data but may be due to replacement of autos damaged in the 1974 
 tornado, influences of variables such as reliability which were not 
 included in the model, or error in the measurement of the attribute 
 values. Nonetheless, a strong rank-order relationship exists between 
 observed experience and mode forecasts at an aggregate level. 
 
 A second alternative analysis approach was developed that focused 
 on the proportion of "regular" users of any of the systems. The pro¬ 
 portion of responses which indicated a greater than 50 percent "likeli¬ 
 hood of use" was calculated for each of the 18 alternatives. These pro¬ 
 portions were used as the dependent observations and Equation 1.0 was 
 fit to these data using regression analysis. This is an ad hoc approach 
 based on the observation from previous applications of the technique 
 that respondents may overstate their intent to use transit. By using 
 
 I 
 
 only responses of seven or more on an 11-point scale (greater than 50 
 percent likelihood of transit use), this bias is partially compensated 
 for. If this second model were to fit the data significantly better than 
 the first, it would be indicative of this bias. More formal techniques 
 should be developed to deal with it, however. 
 
 1-7 
 

 • ions on use 1 
 
 bervi( 
 
 and Cos t- 
 
 .radeoj 
 
 
 The average system-wide values of the four attributes were calcula- 
 
 The results of the research permit - a number of conclusions about 
 
 ted for each of the five system changes of interest from 1974-1977. 
 
 'cost ar.'i service tradeoffs' by type of user. 1? • 
 
 (This is a reasonable approximation for reasons mentioned earlier; 
 
 1. Consistent with a considerable amount of previous research, the 
 
 namely, the lack of variation in level-of-service measures.) These 
 
 results of the consumer tradeoff study demonstrated that levels of cost 
 
 average values were substituted into the equation described above to 
 
 and levels of services do not Compensate for one another. Contrary to 
 
 predict the proportion of "regular" users for each system. 
 
 current models oi traveller choice., the results of the tradeoff 
 
 Results again were in the same rank order as experienced in Xenia, 
 
 cleany. show that low cost cann ot make up for poor service levels or 
 
 and the magnitudes of proportions were also within the range of observa- 
 
 that good service levels cannot make up for high costs. Rather, results 
 
 tion. Once more, 1974 estimates were too high and 1977 too low. 
 
 suggest that if service levels are poor (or costs are high) it hardly 
 
 Elasticity values for the fare changes were calculated from the equation 
 
 matters at all what level the cost (or service) attributes are because 
 
 for "regular" users, and these values were compared with values observed 
 
 the service will have very low patronage. Section 4.2, and particularly 
 
 in Xenia during the demonstration. Again, the rank order was correct, 
 
 
 
 
 HA 1 VA ft 
 
 
 
 .cure 
 
 illustrate 
 
 :he 
 
 -Oil 
 
 jmei 
 
 :rom the analysis. 
 
 but the magnitudes were not. However, the consistency of the rank-order 
 
 For example, the utility curves associated with several levels of fare 
 
 relationships suggests the possibility that a simple transformation can 
 
 would all. cpnverge:..a's .travel, time .(or anv other variable) increased to 
 
 be found that would permit one to derive the "correct" estimates from 
 
 an unattractive level. If fare and travel time were traded off by 
 
 the survey data. Also, there is some indication that bus and taxi values 
 
 individuals, then the utility curves for several levels of' fare would be 
 
 are different, but there is insufficient data to draw a stronger 
 
 pc . all el, and at any leve.L of fare, utility would be higher if travel 
 
 conclusion. 
 
 time were lower. 
 
 5. The final task was to develop a traditional logit choice model 
 
 2, The utility of transit compared to auto based on a single model 
 
 for the Xenia data and compare the results. This could not be completed 
 
 over the entire sample may be expressed as the following linear 
 
 because there was insufficient variation in the attribute data in 1977 
 
 approximation: ' 
 
 to permit model estimation. 
 
 Conclusions are that results provide positive support for continued 
 
 research into the application of this approach to assessing service and 
 
 1-10 
 
 1-8 
 
 
cost tradeoffs and more general -travel demand issues. Results show 
 that the approach can be implemented easily and that it has the poten¬ 
 tial to generate data useful for policy formulation and evaluation. 
 
 1-9 
 
1.2 Conclusions on User Service and Cost Tradeoffs 
 
 The results of the research permit a number of conclusions about 
 
 user cost and service tradeoffs by type of user. 
 
 « Walk time is the next most significant factor for bus service; 
 
 1. Consistent with a considerable amount of previous research, the 
 
 results of the consumer tradeoff study demonstrated that levels of cost 
 
 ® In-*vehicle travel time has very little effect on mode choice; 
 
 and levels of services do not compensate for one another. Contrary to 
 
 current models of traveller choice, the results of the tradeoff analysis 
 
 Fxn&j-.xy, tn© errect 01 dvis ti^cidw&ys is vsry sii£~riu in nic r&xig6 
 
 clearly show that low cost cannot make up for poor service levels or 
 
 that good service levels cannot make up for high costs. Rather, results 
 
 fTy. - ... 1 4- J 3 j C £ — r* & . .l i i ir\c *7 t* r\\TG> v q 11 f* n *1 g, A U At 
 
 suggest that if service levels are poor (or costs are high) it hardly 
 
 last influence of the attributes studied, but their influence 
 
 matters at all what level the cost (or service) attributes are because 
 
 aiffers by trip purpose: headway and transit mode are more influentia' 
 
 the service will have very low patronage. Section 4.2, and particularly 
 
 tor work than flopping trips; travel time difference of automobile ove: 
 
 Figure 4.2, illustrate the situation that emerges from the analysis. 
 
 
 i L- idx zoj snopping cnsn woxtk. wi ips * wst 
 
 For example, the utility curves associated with several levels of fare 
 
 to have./equal influence for work and shopping, trips. . 
 
 would all converge as travel time (or any other variable) increased to 
 
 3. Results indicate that there are significant differences in the 
 
 an unattractive level. If fare and travel time were traded off by 
 
 slopes of the cost and level~of-service attributes by demographic and 
 
 individuals, then the utility curves for several levels of fare would be 
 
 other interpersonal or household characteristics. Some of the differ- 
 
 parallel, and at any level of fare, utility would be higher if travel 
 
 ences include the following: 
 
 time were lower. 
 
 a. Regardless of the type of transit system, its cost or 
 
 2. The utility of transit compared to auto based on a single model 
 
 service components, some persons are more disposed to use transit than 
 
 over the entire sample may be expressed as the following linear 
 
 others. For example, likelihood of using transit decreases slightly 
 
 approximation: 
 
 io simple analytic means to derive this elasticity; it emerge; 
 .ndividual level simulations done for the forecasting phase 
 :udv. 
 
 1-10 
 
 1-12 
 
U t = 0.860 + 0.456TS' + 0.78BW - 1.298TH' - .0159F 
 
 o (1.1) 
 
 -.000193F - .077WT - .0047T 
 
 where: 
 
 TS' = type of service: 1 if bus, 0 if taxi 
 
 BW = bus wait (assumed half the headway), minutes 
 
 TH* = cab "headway": 1 if call-ahead, 0 if on-demand 
 
 F ■ fare, cents 
 
 2 2 
 F = fare squared, cents 
 
 WT = walk time (2 minutes per block), minutes 
 T * travel time, minutes. 
 
 2 2 
 
 Two terms with insignificant, small coefficients (WT and T ) are omitted. 
 As discussed in Chapter 4, a linear model can be used as an approximation 
 to a multiplicative model, which is the most likely "true" model for the 
 Zenia data. 
 
 This model shows that: 
 
 • Bus service is preferred to even on-demand taxi service, all 
 other variables being equal, by a weight equivalent to a 29-cent 
 fare difference.! 
 
 • Within the taxi options, the on-demand option is very strongly 
 preferred to the call-ahead option, by a weight equivalent to an 
 81-cent fare difference. 
 
 This equivalent weight is found by dividing the coefficient of TS' 
 by the coefficient of F (ignoring the F^ effect). 
 
 1-11 
 
• Fare has the greatest effect on mode choice of all the level-of- 
 service variables; the elasticity implied by the model’s coeffi¬ 
 cient is close to the -.4 to -.7 range observed in the 
 
 demons tra tion.^ 
 
 here appear to be few reliable differences in traveller 
 
 • Walk time is the next most significant factor for bus service; 
 its "disutility” is 50 percent greater than that of in-vehicle 
 travel time (per minute). 
 
 For example, there are no income differences in reactions to travel time 
 
 « In-vehicle travel time has very little effect on mode choice; 
 the implied value of travel time based on the model coefficients 
 is only 18 cents per hour. 
 
 e Finally, the effect of bus headways is very slight in the range 
 studied; in fact, 30-minute headways were slightly preferred to 
 15-minute headways. 
 
 plainable on the basis of sociodemographic and interpersonal character- 
 
 Travel time difference of public transit over automobile and headway 
 
 istics, The following findings can be highlighted: 
 
 have the least influence of the attributes studied, but their influence 
 
 individuals 
 
 differs by trip purpose: headway and transit mode are more influential 
 
 relative to the other attributes examined. Licensed drivers without 
 
 for work than shopping trips; travel time difference of automobile over 
 
 previous transit experience place more weight on .cost than those who. 
 
 transit is more influential for shopping than work trips. Cost appears 
 
 to have equal influence for work and shopping trips. 
 
 b . - If ••'an individual has previous- transaC experience-, the .weight 
 
 3. Results indicate that there are significant differences in the 
 
 on walking distance decreases with increasing household size. If an 
 
 slopes of the cost and level-of-service attributes by demographic and 
 
 individual has no previous transit use, the weight on walking distance 
 
 other interpersonal or household characteristics. Some of the differ- 
 
 increases with increasing household’size. 
 
 ences include the following: 
 
 c. There seems to be a reliable sex difference in weighting 
 
 a. Regardless of the type of transit system, its cost or 
 
 of travel time: females place a higher weight on time than males. 
 
 service components, some persons are more disposed to use transit than 
 
 . /. ror ■ vVTii.'!. a p of one .to two autos and 
 
 others. For example, likelihood of using transit decreases slightly 
 
 t-hpn Harlinp.R.- The sex difference is accentuated when auto, ownership 
 
 There is no simple analytic means to derive this elasticity; it emerges 
 from the individual level simulations done for the forecasting phase 
 of this study. 
 
with age. When income is also taken into account, the analysis shows 
 that likelihood of using transit decreases with age in lower-income 
 categories, but increases with age at higher-income levels. Young, low- 
 income individuals are the most likely transit users, followed by older, 
 higher-income individuals. Also, probability of use decreases with 
 increasing auto ownership and is a U-shaped function of adult household 
 size. 
 
 b. The influence service type and headway is related to adult 
 household size in a U-shaped manner, indicating that very small and very 
 large households favor bus to paratransit service more strongly than 
 average-size households, though all prefer bus to paratransit service. 
 
 A person’s having previous experience with public transit results in more 
 preference for 15-minute headways compared to 30-minute headways as auto 
 ownership increases. However, persons without previous public transit 
 experience showed no preference for 15- or 30-minute headways regardless 
 of auto ownership. Also, results indicate that older persons are more 
 sensitive to headways than younger persons. Finally, males are consider¬ 
 ably more sensitive to paratransit headways than females, strongly 
 preferring cn-demand compared to call-ahead service. 
 
 c. There are interesting differences in responses to costs of 
 transit service associated with age and sex: older males are less sensi¬ 
 tive to cost than females, but younger males are more sensitive to cost 
 than females on the average. There also appears to be an association 
 between low or no auto ownership and increased sensitivity to transit 
 cost. 
 
 1-13 
 
d. It appears that females are less sensitive to walking 
 distance at all levels of auto ownership compared to males. 
 
 e. There appear to be few reliable differences in traveller 
 sensitivity to travel time differences over the ranges studied in Xenia. 
 For example, there are no income differences in reactions to travel time 
 differences. 
 
 A. Results indicate that there are differences in how much weight'*' 
 individuals accord cost and level-of-service attributes which are ex¬ 
 plainable on the basis of sociodemographic and interpersonal character¬ 
 istics. The following findings can be highlighted: 
 
 a. Younger and older individuals place more weight on cost 
 relative to the other attributes examined. Licensed drivers without 
 previous transit experience place more weight on cost than those who 
 have used transit. 
 
 b. If an individual has previous transit experience, the weight 
 on walking distance decreases with increasing household size. If an 
 individual has no previous transit use, the weight on walking distance 
 increases with increasing household size. 
 
 c. There seems to be a reliable sex difference in weighting 
 of travel time: females place a higher weight on time than males. 
 
 Weight on travel time increases from ownership of one .to two autos and 
 then declines. The sex difference is accentuated when auto ownership 
 
 Weight is defined as variance explained by one attribute relative to 
 total variance explained by all attributes. 
 
 1-14 
 
is also examined: females place more weight on travel time than males 
 with increasing auto ownership up to three autos. The higher weight for 
 females is possibly due to differences in trip purpose, time of day, or 
 destinations, which were not examined. 
 
 The resylts of analyzing user cost and service tradeoffs suggest 
 that these types of results could be used to guide design and implemen¬ 
 tation of transit demonstration projects as well as to guide or supple¬ 
 ment the evaluation of their impacts. The areas in which this technique 
 can be used include: 
 
 • Assessment of demand response to parameters with insufficient 
 variability in the demonstration to allow the use of other 
 techniques. 
 
 • Improving demonstration design by using the technique, either 
 before implementation or at interim points, to assess the prefer¬ 
 ences of potential users, and alter the system fare or operating 
 policy to match. 
 
 a Providing guidance to marketing efforts by identifying market 
 segments with different responses to system characteristics; 
 this information can be used to decide which aspects of the 
 service should be emphasized to which groups, and what biases 
 may need to be overcome. 
 
 • Overall evaluation of a demonstration, by using the technique to 
 allow people to assess their overall satisfaction (more broadly 
 defined than just willingness to use) with the service. 
 
 Results of the forecasting and validity tests provide encouraging support 
 
 for the method and suggest its potential for future application. 
 
 The Xenia application provided additional evidence to support the 
 indications of the project results that bus service was preferred to 
 taxi service, and on-demand taxi service was preferred to call-ahead 
 
 1-15 
 
service. These findings, plus the relatively strong role of fare levels 
 in determining use of transit, were suggested by the project data, but 
 could not be confirmed. This effort, when combined with a thorough 
 validation exercise, is able to give strong answers to these issues. 
 
 comprised largely of single family homes. 
 
 Xenia has close economic ties with Dayton; approximately 50 percent 
 'of workers residing in Xenia are employed in the Dayton vicinity,; and 
 residents- travel frequently to destinations throughout the Dayton metro¬ 
 politan area for shopping and recreational purposes. 
 
 The core of the city is small and was-severely damaged in the 1974 
 tornado, when approximately one-quarter of the structures throughout the 
 entire city were destroyed. Recovery from the tornado, in terms of the 
 quantity of rebuilding which occurred, is complete, although the pattern 
 oi redevelopment, has deviated substantially from plans adopted by: the. 
 City Commission in June, 1974 . Growth has occurred principally in areas 
 
 at the city’s periphery, while a large part of the, downtown has not been 
 
 * 
 
 rebuilt, However, the City has recently approved plans for a shopping 
 mall to be constructed in the vacant section downtown. 
 
 1-16 
 
 2-2 
 
2. SUMMARY OF DEMONSTRATION 
 
 2.1 Overview and Objectives 
 
 The Xenia Model Transit Service Demonstration Project was funded by 
 the UMTA Service and Methods Demonstration (SMD) Program. The demonstra¬ 
 tion served as a test of several alternative transit services operated in 
 small city setting. Over the course of the demonstration, the system 
 evolved in stages from the original free-fare fixed-route bus system, to 
 a flat-fare fixed-route system, a jitney system, dial-a-ride services 
 operated first in off-peak periods only and then all day, culminating in 
 a mix of differentially-priced paratransit services operated with taxi 
 vehicles. 
 
 The system was operated by a public agency during the first half of 
 the project and by a private contractor during the second half. Total 
 funding from all sources for the Xenia project totalled $1,399,739; SMD 
 funding was $655,000. 
 
 During the first phase of the demonstration, from July, 1974 through 
 December, 1975 the major demonstration objective was to test the feasi¬ 
 bility of providing fixed-route transit service in Xenia. During the 
 demonstration's second phase, from January, 1976 through December, 1977 
 there were two major demonstration objectives: 
 
 1. Demonstrate the technical, operational, and economic 
 feasibility of providing paratransit service as the 
 sole transit service in a small city. 
 
 2. Determine the appropriate role for the private sector 
 in the provision of public transportation service. 
 
. 
 
 2.2 Project Setting 
 
 Xenia is a city of 8.47 square miles, with a population of approxi¬ 
 mately 28,600 people, located 13 miles to the east of Dayton, Ohio. 
 
 . £ X* ^ i v 1 A W O prn- . f 4 4- >. ^ 
 
 The population is predominantly middle-income and the housing stock is 
 comprised largely of single family homes. 
 
 on all services wer 
 
 Xenia has close economic ties with Dayton; approximately 30 percent 
 of workers residing in Xenia are employed in the Dayton vicinity, and 
 
 residents travel frequently to destinations throughout the Dayton metro¬ 
 politan area for shopping and recreational purposes. 
 
 t-i r'hftar and extended dialra-riofi. service hours until 10 PM A 
 
 The core of the city is small and was severely damaged in the 1974 
 
 tornado, when approximately one-quarter of the structures throughout the 
 entire city were destroyed. Recovery from the tornado, in terms of the 
 quantity of rebuilding which occurred, is complete, although the pattern 
 of redevelopment has deviated substantially from plans adopted by the 
 City Commission in June, 1974. Growth has occurred principally in areas 
 at the city’s periphery, while a large part of the downtown has not been 
 
 -TV | • ^ ^ V"i^S3 i V • V f V 1 Q y ^ ^ ^ • 
 
 rebuilt. However, the City has recently approved plans for a shopping 
 
 mall to be constructed in the vacant section downtown. 
 
 service levels. Under the revised fare structure, immediate-request . 
 
 shared-ride service was differentiated from advanced-request service, 
 and the latter priced differently during peak and off-peak hours. The 
 fare for immediate-request, shared-ride service was increased from 50 
 
 rent 
 
 nts to $1. Passengers who requested service at least two hours 
 
 2-2 
 
 
2.3 Project History 
 
 The initial free-fare fixed-route service was started as a relief 
 measure on April 6, 1974, three days after the city was struck by a 
 tornado. The service operated on half-hour headways on four routes 
 from 6:30 AM to 6:30 PM, six days per week, and on one-hour headways on 
 Sunday. The routes on this service were designed to minimize walk times, 
 
 and it is estimated that over 90 percent of Xenia residences were 
 located within one-quarter mile of a bus route. The Miami Valley 
 Regional Transit agency operated the service under contract. 
 
 SMD funding of the demonstration-began on July 22, 1974 though 
 the service remained unchanged until September 1, 1974 when the City's 
 Transportation Department, established with demonstration funding, took 
 over operation of the transit service with Flxette minibuses acquired 
 for the system. A fare of 25 cents was imposed at this time. The 
 route structure in effect during the period of emergency service was 
 retained until November, 1974 when routes were revised to provide more 
 direct access to the downtown area. Routes were subsequently revised 
 again in July, 1975 to accommodate observed travel patterns. Additional 
 system changes in the fixed route phase of the demonstration included 
 the following programs and service innovations: a 24-hour advanced re¬ 
 quest dial-a-ride service for the handicapped; a pre-paid pass program; 
 a half-fare program for the elderly and handicapped; and the replacement 
 of fixed-route service with dial-a-ride on Sundays and holidays, begin- . 
 ning on July 4, 1975. 
 
 2-3 
 
The paratransit phase ..of the demonstration began on January 1, 1976 
 
 when the regular fixed-route service was replaced with an unscheduled 
 
 jitney service during peak hours (7-10 AM and 2-7 PM) and dial-a-ride 
 
 1 . « . ! B £ ! 
 
 re o ft j i rsi i-" <r cm j *h re 
 
 service during off-peak hours (10 AM - 2 PM). The former route struc- 
 
 a.8 u i 1 ft re u 
 
 On!'-' « > a 
 
 ture was retained on the jitney service. Fares on all services were 
 
 |\ . . . j W-.. »*;•.- • - - . ■ 5 ■*. : : *1 « ' - ( 
 
 raised to 50 cents at this time, with senior citizens retaining half- 
 
 : 
 
 fare privileges on the jitney service. On March 1, 1976, the Xenia 
 
 i. X-< W j * w* - —< CM . .04 | ft- re 
 
 Taxi Company entered into contract with the City to operate the service 
 
 re jz 
 
 >> ft 
 
 The private operator increased the base jitney headway from 25 minutes 
 
 .3 re x 
 
 CO MS 
 
 
 to 1 hour, and extended' dial-a-ride service hours until 10 PM. A 
 
 y 
 
 o 
 
 ft 
 
 charter service was also started at this time. 
 
 —• - > 
 
 — I ft ft © 
 
 ill , | > o a 6) | .c ft o 
 
 1 < x C Q l ( xj re © 
 
 The next major service change occurred on May 9, 1976 when jitney 
 
 .....-*—— - ——---- ! o .o 
 
 re 
 
 service was eliminated, and shared-ride taxi was instituted throughout 
 
 c >•» r-' j , ’ *p 
 
 the service day. During the week of July A, 1976*seven-passenger Check- 
 
 r— < re M ft •“ ; PO tj-i iff,' r- c-- . , .c 
 
 Or i-'-re C ft) K 1 Ih ‘ - cm "■ - • —< ' ■ O ft —( 
 
 er cabs were phased into service, replacing the minibuses and the single 
 
 r —---4—--—----J re u 
 
 ■ t- j , j Ci ci re 
 
 taxicab which had previously been in use on the service. 
 
 ; C 
 
 GC i w TO Q) 
 
 The final service change, which occurred in March,1977 was a revi¬ 
 se r-' W | O rsl o <■ r ■ "O -o | • 
 
 -h : i cc jz. V- | ^ co o> r**^ 
 
 sion of the fare structure, intended to set fares as a function of 
 
 ; O --H | <3- cm o i ■*-» O 
 
 £•' i** PP r- , V 
 
 service levels. Under the revised fare structure, immediate-request 
 shared-ride service was differentiated from advanced-request service, 
 
 C \C ft rv 
 
 .. r 
 
 and the latter priced differently during peak and off-peak hours. The 
 
 -re ft 00 c ^ ft »H s *0 .© 
 
 r- re rep- d \o o re o \c 
 
 fare for immediate-request, shared-ride service was increased from 50 
 
 I O 2 3 ■ —(■ . ft) I OW-<C c* 
 
 . L <y y re G U 
 
 cents to $1. Passengers who requested service at least two hours 
 
 QO 
 
 ^ f—* 
 
 r^ 
 
 tH {s. 
 
 > 
 
 
 O 
 
 
 
 
 
 
 •n Q. 
 
 XJ 1 
 
 c 
 
 re 
 
 <; 
 
 . ’ 
 
 
 3 1 
 
 1- < 
 
 •H 
 
 
 c 
 
 
 o cy 
 
 
 
 
 m o 
 
 JC 
 
 c. 
 
 U 1 
 
 O 
 
 
 
 ' 1 
 
 C I s - 
 
 00 
 
 ft 
 
 
 p,y c. 
 
 V 
 
 V. 
 
 
 re cn 
 
 3 
 
 3 
 
 01 i—l 
 
 
 2-A 
 
 >n re 
 
 K. 'r* 
 
 ft —< 
 
 O 
 
 Cl 
 
 
 *"w ^ 
 
 
 re ~ 
 
 ft ft 
 
 XJ 
 
 r* 
 
 
 
 CL 
 
 
 C d ' 
 
 ft ft. 
 
 re >> 
 
 xj 
 
 ft 
 
 a,' c- 
 
 X & 
 
 (j 
 
 « re 
 
 J re 
 
 ft : re 
 
 rft 
 
 O 
 
 E < 
 
 •H CC 
 
 o 
 
 —: T 
 
 •H 2 
 
 
 < 
 
 
 
 'w' 
 
 c 
 
 s-r 
 
 r>' 
 
 Cft -'w> 
 
 
 
 0) 
 
 * 
 
 * 
 
 2-6 
 
in advance would now pay a lower fare than riders using the immediate 
 request service—75 cents during peak hours (7-9:30 AM; 1:30-7 PM), 
 and 50 cents during off-peak hours. 
 
 Table 2.1 summarizes data on monthly ridership, operating cost, 
 operating deficits, cost per passenger, and operating ratios for each 
 of the major service types operated over the course of the demonstration. 
 As the table shows, average monthly ridership declined from approxi¬ 
 mately 40,000 trips for the fare-free emergency service to 6,000 for 
 the mix of paratransit services operated during the last 10 months of 
 the demonstration. Average monthly operating costs ranged from $54,000 
 during the fare-free period to $17,000 for the jitney service under 
 private management and $19,000 for the paratransit services. The oper¬ 
 ating deficit ranged from $54,000 (the total operating cost) for the 
 fare-free service, to $12,000 for the .paratransit services. The cost 
 per passenger was lowest for the emergency service, at $1.23, and equal 
 to or greater than $2.45 for jitney and paratransit. The operating 
 ratio ranged from 12.5 during the fixed-route period to 1 2.6 during the 
 
 I 
 
 I 
 
 paratransit period. \ 
 
 The demonstration ended on December 31, 1977 when the UMTA opera¬ 
 ting subsidy was discontinued. However, UMTA continued to fund the 
 Xenia Transportation Department from January through June of 1978, 
 while the Greene County Transit Board attempted to develop a plan for 
 continuation of transit service in Xenia. The fare structure was 
 revised on January 1, 1978 in an attempt to cover operating costs in 
 
 2-5 
 
CM 
 
 w 
 
 -2 
 
 O 
 
 < 
 
 H 
 
 CO 
 
 3 
 
 o 
 
 •H 
 
 2m 
 
 0) 
 
 to 
 
 c 
 
 o 
 
 C3 
 
 u 
 
 Ml 
 
 CO 
 
 c 
 
 o 
 
 6 
 
 3 
 
 o 
 
 u 
 
 o 
 
 w 
 
 •H 
 
 Ml 
 
 CO 
 
 u 
 
 CO 
 
 to 
 c 
 
 •H 
 Ml 
 CO 
 
 >M 
 0) 
 CM 
 
 o 
 
 CO 
 
 0) 
 
 3 
 
 C 
 
 02 
 
 > 
 
 setting is 
 of this ' f6 
 door-to-door 
 portion of 
 downtown. 
 times, • an 
 
 
 M 3 
 
 •H 0- 
 
 
 
 
 tn 
 
 
 4 
 
 n- 
 
 m 
 
 CD 
 
 taut is 
 
 
 Ml ^ 
 
 
 1 
 
 
 • 
 
 
 
 • 
 
 • 
 
 • 
 
 
 
 CO O HI 
 
 
 1 
 
 
 CM 
 
 
 
 
 <r 
 
 CM 
 
 
 
 H rt CO 
 0) Ml O 
 
 
 
 
 
 
 
 
 »se iss; 
 
 jes 
 
 are di 
 
 
 CM TO O 
 
 c os 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 tior 
 
 
 not begin 
 
 
 
 60 H 
 
 
 
 
 
 
 ocus 
 
 o c 
 
 n the issues rais 
 
 
 C H 3 
 •H 02 60 
 
 
 TO 
 
 
 
 
 
 CM 
 
 CO 
 
 m 
 
 
 
 mi a c 
 
 CO 02 
 
 
 CM 
 
 • 
 
 
 r*» 
 
 • 
 
 
 
 • 
 
 as from 
 
 <r 
 
 • 
 
 e fixer. 
 
 
 Ml Ml CO 
 
 
 *H 
 
 
 rH 
 
 
 
 CM 
 
 CM 
 
 CM 
 
 
 
 0 ) co to 
 
 
 <!> 
 
 
 
 
 
 
 
 
 
 
 Cm O TO 
 OOP- 
 
 
 
 
 
 
 
 in 
 
 p ar t i c*ul ar > man* 1 
 
 
 
 
 
 
 
 
 
 ip i. 
 
 isues d 
 
 
 nc pbe 
 
 0! 
 
 60 
 
 s 
 
 PH *H Ml 
 
 
 r- 
 
 CH 
 
 
 r>- 
 
 IT' 
 
 
 
 n- 
 
 vD 
 
 (HI 
 
 cn 
 
 cn 
 
 CM 
 
 ndirect 
 
 60 rH Ml *H 
 
 
 CD 
 
 
 H-* 
 
 
 
 fH. 
 
 in 
 
 n- 
 
 
 CO 
 
 d to o 
 
 
 #> 
 
 
 
 
 
 n 
 
 
 •c 
 
 
 H 
 
 Ml 2-1 -H 
 
 
 m 
 
 
 CHI 
 
 
 
 rH 
 
 cn 
 
 rH 
 
 
 02 
 
 e 02 m 
 
 
 LT) 
 
 
 cn 
 
 
 rs.c l 
 
 CM 
 
 rH 
 
 
 
 > 
 
 O CM 02 
 
 
 <r> 
 
 
 
 
 
 
 
 
 
 < 
 
 ZOO 
 
 
 
 W^L 
 
 
 
 cna 
 
 
 IT i S • a* C. 
 
 
 
 
 emen 
 
 tl 0 
 
 
 
 enia 
 
 
 
 
 
 
 60 
 
 
 
 
 C. Ill--' ^ 1 
 
 S.H 
 
 V n H •{ 
 
 
 M i M. - ♦ V 
 
 0 S 
 
 was to 
 
 
 c 
 
 
 f-» 
 
 
 cn 
 
 
 
 rH 
 
 m 
 
 HP 
 
 
 02 
 
 pH *H 
 
 
 ON 
 
 
 CH 
 
 
 
 cn 
 
 r*» 
 
 cn 
 
 
 60 rH Ml 
 
 
 cD 
 
 
 in 
 
 
 
 cn 
 
 cn 
 
 o 
 
 small. < 
 
 CO 
 
 JS CO 
 
 
 m 
 
 TV Hi 
 Md 
 
 »■ 
 
 m 
 
 an 
 
 sa w 
 
 r 
 
 
 0k 
 
 
 u 
 
 iJ M Ml 
 
 
 
 
 
 <r 
 
 
 
 
 0) 
 
 C 02 CO 
 
 
 UO 
 
 
 C-> 
 
 
 
 CM 
 
 rH 
 
 rH 
 
 
 p> 
 
 < 
 
 o o o 
 
 2 0 0 
 
 •to- 
 
 e an 
 
 u 
 
 
 
 ansi 
 
 t services 
 
 were oper. 
 
 
 
 
 
 
 
 T\V 
 
 ov~> r 
 
 ed t 
 
 he oppo 
 
 rtt 
 
 Ph 
 
 U 
 
 09 
 
 
 
 
 
 
 
 
 
 
 
 
 m 
 
 
 a 
 
 
 
 
 
 
 
 
 
 
 0 . 
 
 v s 
 
 •H 
 
 3 Ph .3 
 
 
 o 
 
 
 O 
 
 
 
 o 
 
 o 
 
 hes 
 
 g'dif 
 
 
 60 rH CO 
 
 
 o 
 
 
 (HI 
 
 
 
 o 
 
 
 
 i 
 
 
 CO 3= H 
 
 
 
 
 CO 
 
 
 
 cn 
 
 n- 
 
 
 
 
 2-i mi 02 
 
 
 
 
 #c 
 
 
 
 
 r» 
 
 
 o 
 
 op 
 
 02 C *3 
 
 
 ro 
 
 
 o 
 
 
 CH 
 
 m .in 
 
 a s 
 
 o 
 
 
 > O -H 
 
 
 MT 
 
 
 (HI 
 
 
 
 
 
 
 o 
 
 
 < 2 PS 
 
 
 
 
 
 
 
 
 
 
 r> 
 
 
 
 
 
 
 
 
 
 
 
 
 cD 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 «* 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 /■-s 
 
 Ml 
 
 
 
 
 st it 
 
 
 
 
 
 
 
 Ml /—N 
 
 c-sed 
 
 
 n- 
 
 
 
 
 
 
 
 
 
 C CD 
 
 3 
 
 
 fH. 
 
 
 
 
 
 
 
 
 
 3 
 
 E 
 
 
 cn 
 
 
 
 
 
 •k 
 
 
 
 
 E cn 
 
 3 
 
 
 fH 
 
 rm 
 
 of dev 
 
 
 Q p Q 6 
 
 /^N 
 
 0) 
 
 
 
 Qj rH 
 
 60 
 
 
 
 
 
 
 
 HP 
 
 H 
 
 
 
 60 
 
 TO -—' 
 
 
 u 
 
 
 
 
 
 r- 
 
 CO 
 
 
 
 TO Ph 
 
 C iD 
 
 
 3 
 
 
 
 
 t3 ^ 
 
 DO 
 
 CH 
 
 (*- 
 
 
 
 C H 
 
 TO n- 
 
 
 -D 
 
 r>H 
 
 servic 
 
 
 
 
 i 
 
 vt T'yp 
 
 TO TO 
 
 2 cn 
 
 
 E 
 
 
 
 
 o 
 
 
 
 2 3 
 
 ^H 
 
 
 3 
 
 
 
 
 
 Ml 
 
 m 
 
 HT 
 
 
 2-i 
 
 3 
 
 
 3 
 
 
 
 
 
 CO 
 
 CM 
 
 
 
 •Sfi 
 
 Ml rH 
 
 
 3 
 
 
 1"Vi f y 
 
 ; > f 
 
 
 3 
 
 
 CH 
 
 m 
 
 TO -H 
 
 s i 
 
 0 
 
 
 
 
 
 00 
 
 
 rH 
 
 n» 
 
 *H [m 
 
 > H 
 
 
 
 
 
 
 
 3 
 
 02 
 
 
 CH 
 
 X) 
 
 •H Cl 
 
 Ml 
 
 1 
 
 
 
 
 
 < 
 
 Ml 
 
 H 
 
 •H 
 
 3 1 
 
 H < 
 
 •H 
 
 
 F n 
 
 rtherm 
 
 
 Ph 
 
 
 3 
 
 3 
 
 
 Cm 
 
 CM 
 
 CO 
 
 cD 
 
 
 
 o 
 
 1 
 
 O 
 
 .10 
 
 u 
 
 h-' Ph 
 
 H-- | 
 
 c 
 
 
 
 
 
 c 
 
 
 co 
 
 E 
 
 3 
 
 Sm 
 
 
 TO 
 
 cn 
 
 
 
 
 02 
 
 rH 
 
 
 3 
 
 -3 
 
 Ph TO 
 
 Ph J3 
 
 2m 
 
 
 wa 
 
 lk tim 
 
 
 60 -H 
 
 *3 
 
 MJ 
 
 E 
 
 3 3 
 
 3 U 
 
 Mi 
 
 
 
 
 
 H 
 
 2-i 
 
 3 
 
 C- 
 
 3 
 
 C C 
 
 3 H 
 
 TO 
 
 Ph 
 
 
 
 
 0) 
 
 rv 
 
 X 
 
 3 
 
 U 
 
 Ml f8 
 
 Ml TO 
 
 2-i 
 
 TO 
 
 
 
 
 E 
 
 < 
 
 •H 
 
 CO 
 
 3 
 
 •H ^ 
 
 H 2 
 
 TO 
 
 2 
 
 
 
 
 W 
 
 
 U- 
 
 >w' 
 
 a 
 
 r-, w 
 
 i-> w 
 
 CM 
 
 
 « ec 
 
 02 
 
 *3 00 
 0) CO 
 TOO 0) 
 3 rH 
 rH *H 
 
 O E 
 
 C 
 
 •H 0) 
 
 t—1 
 03 JD 
 •H CO 
 
 i—I 
 
 0) *H 
 O CO 
 •H > 
 
 6 * 
 E 
 O 
 u 
 
 02 
 
 to 
 
 fre. 
 
 ly 
 
 i <u 
 w 
 
 to *o 
 cm 02 
 f mi 
 02 CO 
 0) E 
 
 2m -H 
 CM Ml 
 
 CO 
 
 ?H 02 
 
 C B 
 02 CO 
 60 5 
 W 
 
 02 02 
 
 £ v 
 
 02 -H 
 
 > 
 
 02 Sm 
 
 .c oj 
 
 Ml co 
 
 n- 
 
 e'¬ 
 
 en 
 
 >m 
 
 02 
 
 *§ 
 
 0) 
 
 o 
 
 0) 
 
 a 
 
 i 
 
 
 c 
 
 o 
 
 CO 
 
 Ml 
 
 CO 
 
 *o 
 
 c 
 
 o 
 
 *r-< 
 
 Ml 
 
 CO 
 
 2m 
 
 Ml 
 
 CO 
 
 c 
 
 o 
 
 E 
 
 PH 
 
 o 
 
 c 
 
 0) 
 
 60 
 
 M 
 
 3 
 
 E 
 
 02 
 
 02 
 
 -U 
 
 CO 
 
 u 
 
 o 
 
 CM 
 
 CO 
 
 XJ 
 
 co 
 
 TOO 
 
 r- 
 
 e'¬ 
 
 en 
 
 o 
 
 Ml 
 
 fO 
 
 2 
 
 E 
 
 O 
 
 u 
 
 CM 
 
 o 
 
 o 
 
 o 
 
 0 
 
 cD 
 
 *o 
 
 TO 
 
 r— 
 
 r- 
 
 Cn 
 
 PH 
 
 CO 
 
 c 
 
 0) 
 
 (M 
 
 I • 
 
 02 Ml cD 
 
 JD CO ft. 
 
 Ml O C?c 
 
 O — 
 
 ■U c 
 
 M. O 
 
 co tru 
 
 CM *H • 
 U CO 
 rH to Ml 
 rH D, ^TO 
 
 o 
 
 O J-> 
 CO 
 
 O cm O 
 CO O V 
 
 CO 
 
 3 
 
 O 
 
 c 
 
 CO 02 
 02 iH 
 CO *H 
 
 o E 
 
 J= Cw i 
 60 >m U 
 
 3 3 0) 
 O O. CM 
 /- 
 
 Ml M *0 
 rH O C 
 < CM CO 
 * 
 
 ?H 
 
 CO 
 
 z 
 
 E 
 
 O 
 
 Jm 
 
 O 
 
 cD 
 
 CO 
 
 — 
 
 O' 
 
 CO 
 
 CO 
 
 p 
 
 CO 
 
 M 
 
 GJ 
 
 ■u 
 
 •H 
 
 OC 
 
 * 
 
 •* 
 
 2-6 
 

 the absence of the subsidy. The city was divided into 12 zones, and 
 the fare for travel between zones was set at a minimum of $1, with 
 $1 increments in several steps for longer trips. During this six- 
 month period, average monthly ridership was 2,423, average monthly' 
 '’operating cost was $9,302, and the average monthly operating deficit 
 was $3,537. • 
 
 Operation of the service was further extended under private 
 operation until September, 1978 when the insurance coverage on the 
 vehicles expired and service was suspended. In January,'1979 Greene 
 County begin providing county-wide transportation services to the elderly 
 using five vans with the Flexicabs as back-up, taking over all transpor¬ 
 tation functions from the City of Xenia. In April, 1979 county-wide 
 fixed-route and paratransit services for the general public are also 
 slated to begin using the Xenia cab and bus vehicles. 
 
 2-7 
 
2.4 Conclusions 
 
 The service changes implemented over the course of the demonstration 
 provided data for the analysis of a number of important issues. The eval 
 uation findings with regard to each of these issues are discussed below. 
 It should be noted that the evaluation did not begin until the midpoint 
 of the demonstration and thus it focuses on the issues raised in the para- 
 transit portion of the project. Key issues from the fixed-route period 
 are also covered, but an less detail. In particular, many of the inter¬ 
 esting travel demand and auto ownership issues during the free-fare and 
 later fixed-route periods could, only be addressed indirectly. 
 
 1. How do the operating characteristics of the paratransit 
 
 options compare with the characteristics of fixed-route 
 
 systems implemented in Xenia ? 
 
 One of the principal demonstration objectives was to evaluate the 
 feasibility of operating paratransit services in a small city setting. 
 Since both fixed-route and paratransit services were operated during 
 the demonstration, the project provided the opportunity to compare data 
 on level of service, patronage and finances for these different types 
 of service operated at the same location'within a short span of time. 
 
 • A potential advantage of demand-responsive service*in a low density 
 setting is that it can accommodate the dispersed trip patterns typical 
 of this form of development while providing users with the convenience of 
 door-to-door service. However, survey results indicate that a high pro¬ 
 portion of both fixed-route and paratransit trips in Xenia began or ended 
 downtown. Furthermore, the service levels in terms of travel times, wait 
 times, and walk times were quite similar for the two transit options. 
 
 2-8 
 
Spatial coverage and reliability were excellent on the fixed-route 
 service as well, again making its service level very comparable to the 
 paratransit system. 
 
 In summary, both systems provided average travel times near 15 min¬ 
 utes, wait times near 10 minutes, and short or no walks for the great 
 majority of trips made on the transit system. The average trip length 
 was three miles. This service level was quite good for a small-city 
 setting such as Xenia. 
 
 Comparison of ridership data during periods of fixed-route and 
 paratransit service shows that paratransit ridership was only 33 percent 
 of fixed-route levels. Although this decline is partially due to the 
 higher fare charged for the paratransit service, the magnitude of the 
 
 i 
 
 difference suggests a preference for the fixed-route service. For fare 
 alone to account for the difference between fixed-route and paratransit, 
 the implied fare elasticity would have to be -1.5, a value well in 
 excess of any observed for other Xenia fare changes. Other service 
 level variables, as mentioned previously, were quite similar in the 
 two systems. 
 
 Cost per passenger was substantially lower on the fixed-route 
 service than the paratransit service in two eight-month periods of 
 comparison: $1.28 versus $2.12, due both to lower productivity on the 
 paratransit service in terms of passengers per mile, and reduced total 
 ridership. However, the total monthly operating deficit was lower for 
 the paratransit service than the fixed-route service, due to higher fares 
 
 2-9 
 
and a curtailment of operations in terms of vehicle miles, vehicle 
 hours, and ridership, on the paratransit service. Operating losses 
 were therefore brought closer to a politically acceptable level, 
 although a monthly deficit near $10,000 persisted. 
 
 In summary, the evaluation of each system’s economic performance 
 depends upon the criteria by which the system is judged. The fixed- • 
 route system was operated at a lower cost and deficit per passenger, and 
 
 with greater ridership. Considering the budget constraint, however, 
 paratransit may be more viable, since higher fares are typically 
 charged for this service, resulting in lower total deficits, and also 
 in lower patronage. ot&l trips throughout the demonstration, lhe intra* 
 
 2. What impacts do fare-and service changes have on patronage ? 
 
 Patronage declined throughout the course of the demonstration, due 
 in part to increases in fare. Elasticities were calculated for each of 
 these fare increases; the elasticities ranged from -.51 for the change 
 from free-fare to 25-cent fare, to -1.2 for the change in fare from 25 
 cents to 50 cents, implemented at the same time the schedule fixed-route 
 
 bus service was replaced with unscheduled jitney service. In the cases 
 
 transit services ? 
 
 where the fare alone was changed, demand proved to be inelastic. In the 
 
 Comparison of cost data for the taxi vehicles and the Flxette mini- 
 
 single instance where demand proved to be elastic, the apparent explana- 
 
 buses shows that the taxis were more economical to operate on para-' 
 
 tion is that the simultaneous service change was responsible for some 
 
 transit services. -The lower passenger capacity of the taxis had.no 
 
 of the decrease in ridership. 
 
 effect on the number of passengers carried per mile or per hour, and 
 
 operating cost per mile was lower for the taxis than the Flxettes. 
 
 Fuel and oil cost averaged 7 cents per mile for the taxis versus 10 
 
 2-10 
 
A single headway elasticity of -1.5 was computed when jitney headways 
 were changed. A ridership increase of 35 percent occurred when the hourly 
 jitney service was replaced by shared-ride taxi service. Also, during the 
 period when the differentially priced paratransit service was operated, 
 approximately two thirds of the users chose the higher priced immediate- 
 request service ($1) over the less expensive advance-request service (50 
 cents or 75 cents) even though the advance request travel times were lower 
 as well. These results all indicate that users are more sensitive to 
 service than fare in a small urban area. 
 
 A final inference can be made from the fact that paratransit rider- 
 ship was one third of fixed-route ridership over comparable periods. For 
 fare alone to account for this observed change, the implied fare elasticity 
 would have to be -1.5, a value well in excess of any observed in other 
 Xenia fare changes. Since the measured service attributes of the transit 
 and paratransit systems were essentially the same (travel time, wait time, 
 walk time, and reliability), this comparison also suggests a user preference 
 for fixed-route service over paratransit. 
 
 Passengers' perceptions of the tradeoffs between cost and level of * 
 service in Xenia are also analyzed in another report, Analysis of User Cost 
 and Service Tradeoffs in Transit and Paratransit fystems . 
 
 3. Does the initiation of a new transit service impact auto 
 
 ownership and travel behavior ? 
 
 The Xenia demonstration provided the opportunity to assess the 
 degree to which transit availability could influence the course of 
 community development through modification of travel behavior. 
 
 2-11 
 
One aspect cf this analysis is evaluation of the impact of transit 
 availability on auto ownership. Estimates of the percentage of auto¬ 
 mobiles destroyed in the tornado vary between 16 and 29 percent. 
 
 Survey results indicate that approximately 90 percent of the households 
 who lost automobiles in the tornado replaced them within one year, and 
 that in subsequent years, auto ownership returned to pre-tornado levels. 
 
 Thus the demonstration results indicate that transit availability had 
 
 private taxi operation ? 
 
 virtually no effect on auto ownership. 
 
 Data provided by the owner of the Xenia Cab Company indicate that 
 
 The impact on travel patterns and community redevelopment was 
 
 private taxi revenues declined dramatically during the first two years 
 
 also slight, due to the fact that transit accounted for only a very 
 
 of the demonstration, when the fixed-route service, was-in operation. 
 
 small percentage of total trips throughout the demonstration. The intra- 
 
 During the subsequent, two years of the demonstration, the Xenia Cab 
 
 Xenia transit modal share ranged from an estimated 4.7 percent in the 
 
 Company, operated the public transit system under contract. Thus, it 
 
 summer of 1974 to 0.4 percent in the fall of 1977. Mode shares for 
 
 appears, that'competition. from a relatively high quality, low-fare 
 
 CBD-bound trips ranged from 7.4 percent to 0.7 percent. Survey results 
 
 .public, transit..', service can. have:., a : maj-pr: impact ..^on private taxi 
 
 indicate that transit use was proportionately greater among women, persons 
 
 operations. 
 
 without driver's licenses, lower income groups, and very large and very 
 small households. 
 
 4. How do taxicab vehicles perform on shared-ride para- 
 
 transit services ? 
 
 Comparison of cost data for the taxi vehicles and the Flxette mini¬ 
 buses shows that the taxis were more economical to operate on para- 
 transit services. -The lower passenger capacity of the taxis had no 
 effect on the number of passengers carried per mile or per hour, and 
 operating cost per mile was lower for the taxis than the Flxettes. 
 
 Fuel and oil cost averaged 7 cents per mile for the taxis versus 10 
 
cents per mile for the Flxettes. Vehicle reliability, as reflected in 
 maintenance cost per mile, was approximately the same for the two 
 vehicle types: maintenance cost per mile averaged 8 cents for both 
 vehicle types, though maintenance cost would decrease for-the taxis 
 relative to the Flxettes if the data were adjusted to account for 
 inflation. 
 
 According to a survey of passenger attitudes administered shortly 
 after the introduction of the taxis on the paratransit service, 88 per¬ 
 cent of users preferred the taxis to the buses. Respondents to this 
 survey were also asked to rate the ease of entry, noise, smoothness of 
 ride, seating comfort, overall comfort, and privacy of both vehicles; 
 a substantially higher percentage gave the taxis more favorable ratings 
 than the buses. 
 
 Hence, it appears that the taxis' performance with respect to 
 both passenger preference and operating economy was superior to that 
 of the buses in paratransit service. 
 
 5. How effectively did the mechanics of the contract 
 
 between the City and the private operator function ? 
 
 The private operator had some success in reducing direct operating 
 costs, though he/she accomplished this primarily through reductions in 
 service. The operator's flexibility in setting wage rates was another 
 important factor, however. Under private management, drivers' w^ges 
 were calculated as a percentage of receipts, where previously drivers 
 had received a flat hourly wage. 
 
 2-13 
 
Administrative cost increased, however, as a consequence of the 
 change in administrative structure resulting from the contract agree¬ 
 
 ment, though this situation was due in part to the project’s demonstra¬ 
 
 tion status and would not occur under normal system operating 
 
 conditions. In total, however, the direct operating savings outweighed 
 the added administration costs. 
 
 6. How does the introduction of transit service affect a 
 
 private taxi operation ? , 
 
 e ■ or - . : cos :s 
 
 Data provided by the owner of the Xenia Cab Company indicate that 
 
 i-usi'v } demonstrate .. s&lue of contracts between private 
 
 private taxi revenues declined dramatically during the first two years 
 
 operators and the public sector'because much of the cost‘reduction was 
 
 of the demonstration, when the fixed-route service was in operation. 
 
 due to a reduction in service. However, it does suggest that in,cases 
 
 During the subsequent two years of the demonstration, the Xenia Cab 
 
 Jhere municipal personnel or administrative policies restrict manage - 
 
 Company operated the public transit system under contract. Thus, it 
 
 raent cotio.ns, transit service might be provided more effectively by. 
 
 appears that competition from a relatively high quality, low-fare 
 
 •raters under contract. In Xenia, this was illustrated by, 
 
 public transit.service can have a major impact on private taxi 
 
 changes in the methods: of- computing driver wages introduced by the 
 
 operations. 
 
 
 private operator which resulted in a substantial cost decreases. 
 
 
 2-16 
 
 2-14 
 
2.5 Implications of Demonstration Results for Other Cities 
 
 The major demonstration results are potentially transferable to 
 other small cities with similar demographic and spatial characteristics. 
 
 The analysis of data for fixed-route and paratransit services 
 suggests that a fixed-route system can provide service of comparable 
 quality to that of paratransit systems, at lower cost per passenger, 
 
 depending on the degree to which users’ trip destinations are centralize 
 Since the percentage of trips with either a downtown origin or 
 
 destination was high for both fixed-route and demand-responsive service 
 in Xenia (near 70 percent, with one fixed-route survey showing 85 
 percent), improved access to decentralized locations was not important 
 to most system users. This finding might also apply to other cities 
 where travel purposes and characteristics of transit users are similar 
 to- those in Xenia. 
 
 The fare elasticities calculated for Xenia transit services tended 
 to be relatively high compared to larger cities, which is probably due 
 to the fact that most transit users either had access to automobiles 
 or could walk to their destinations. 
 
 Since a large percentage of Xenia trips had either origin or 
 destination outside the city limits or a reasonably defined transit 
 service area, it is not surprising to find that transit availability 
 had little long-term impact on auto ownership. In other small cities 
 located in metropolitan areas, where the incentive for auto ownership 
 is similarly strong, similar results can be expected. Furthermore, with 
 
 2-15 
 
high levels of auto ownership, adequate downtown parking, and the 
 
 absence of significant traffic congestion, the transit modal share is 
 
 jniunbers dialed 
 
 likely to be low, and the impact of the transit system on travel behavior 
 and development patterns minimal. The primary function which transit 
 
 should reasonably be expected to serve in a city of this type is to 
 
 Business or I 1 Residences Busy or no 
 
 increase mobility among those who lack convenient access to automobiles, 
 
 2730 (53%) 1526 (27%) 
 
 as well as the elderly and handicapped. 
 
 The private operator’s success in reducing direct operating costs 
 does not conclusively demonstrate the value of contracts between private 
 
 I used screeningj J J up interview i lor refused to be 1 
 
 operators and the public sector because much of the cost reduction was 
 due to a reduction in service. However, it does suggest that in cases 
 where municipal personnel or administrative policies restrict manage¬ 
 ment options, transit service might be provided more effectively by 
 
 private operators under contract. In Xenia, this was illustrated by 
 
 ' 
 
 changes in the methods of computing driver wages introduced by the 
 
 ■ - ■ 
 
 private operator which resulted in a substantial cost decreases. 
 
 (1.3 per HH) 
 
 FIGURE 3.1 
 
 Results of Phone Screening and Interviewing, 1977 
 
 between private 
 
 lor refused to bej 
 
 reduction was 
 
 |i-OJZ so * *) 
 
 that in cases 
 
 2-16 
 
3. DESCRIPTION OF HOME INTERVIEW SURVEY 
 
 3.1 Telephone Screening 
 
 A home interview survey was administered to 504 households in Xenia 
 by the City of Xenia in November and December, 1977. There were two 
 
 i 
 
 phases to the survey—a telephone screening and the home interview. The 
 telephone screening was used to ensure that the sample met pre-established 
 quotas for three criteria: at least 80 percent of the households must 
 have lived in Xenia over the entire 1974-1977 period of the demonstration; 
 at least 20 percent must have used Flexicab; and at least 40 percent must 
 have no more than one auto. These restrictions were imposed to ensure 
 that a sufficient sample was collected of users and low auto-owning 
 households who are potential users: a purely random sample would not 
 have been adequate to cover these groups. 
 
 A household found to fit into one of the needed categories was asked 
 whether an interviewer could come to the residence to administer a survey. 
 Figure 3.1 shows the results of the screening process. About half the 
 households in the "outside Xenia or refused to be interviewed" category 
 lived outside the city and were not in the transit service area. 
 
 I 
 
 3-1 
 
assistance from the interviewer : , who is administering the survey to the 
 
 second member of the household . In 30 perce nt of the households, two 
 
 Total' number of 
 numbers dialed 
 
 5681 (100%) 
 
 » __ 
 
 .interviewe 
 previous e 
 
 
 ind was administered by 
 
 
 ihe interviewei 
 
 had no 
 
 Business or 
 not in service 
 
 d iOCd- 
 
 1425 (25Z) 
 
 IUS, 
 
 Residences 
 
 reached 
 
 ana superv; 
 
 2730 (53%) 
 
 Busy or no 
 answer (up to 
 3 tries) 
 
 1526 (27Z) 
 
 i from 
 l simple 
 
 ' 
 
 
 
 . completed the dir 
 
 1 ) 
 
 1 
 
 • 
 
 manner all 
 
 rowing a un 
 
 radeoff informati 
 
 )di 
 
 combinations which is ne 
 No major problems w 
 
 Household ref- 
 
 
 Adult not home 
 
 
 Household set 
 
 
 HH outside Xenia 
 
 used screening 
 
 
 An addi‘ 
 
 3.1 
 
 up interview 
 
 
 or refused to be 
 
 questions 
 
 
 
 
 
 
 interviewed 
 
 83 (3%) 
 
 
 38 (1Z) 
 
 
 777 (28%) 
 
 
 1832 (67Z) 
 
 
 
 - 
 
 Hot home, can- 
 
 
 Survey adminis- 
 
 celled, refused 
 
 
 tered to HH 
 
 at door 
 
 
 p tne afreet 
 
 273 (34%) 
 
 
 504 (66%) 
 
 — 
 
 complete all 18 
 
 encountered b 
 xpressed 
 
 about people's ability to rate 18 combine 
 the sample had difficulties with the - 
 
 —i 
 
 670 persons 
 surveyed 
 (1.3 per HH) 
 
 small portion of 
 jurlty found it easy 
 
 and were able to complete it in ten minutes or less. Further details 
 .on the instrument are included in Appendix C. 
 
 FIGURE 3.1 
 
 Results of Phone Screening and Interviewing, 1977 
 
« 
 
 3.2 Home Interview Survey 
 
 The home interview survey instrument is included as Appendix C in 
 this report. It consists of four sections. The first asks a series of 
 background questions on the household and the impacts of the tornado on 
 its travel behavior. This section is administered to the head of the 
 household, if possible; the remaining three sections were administered 
 to two persons in the household, if possible. The second portion 
 contains questions on personal characteristics, previous transit usage, 
 and perceptions of auto and transit tradeoffs. The third form is a one- 
 day trip diary used to develop aggregate trip generation, destination, 
 and mode choice statistics to form the data base to attempt to estimate 
 a disaggregate (logit) travel demand model and to provide a data base 
 for forecasting with the direct value assessment procedure described in 
 this report. 
 
 Finally, the direct assessment experiment was administered to two 
 persons in the household as the last part of the survey. Eighteen 
 alternatives were listed on a single page, with an 11-point numerical 
 scale beside each (see Appendix C). Each alternative is defined by 
 four variables: type of transit service, cost, walking distance, and 
 travel time. The respondent is asked to rate each of the 18 combinations 
 on the scale according to how frequently he or she might use each. 
 
 The surveyor explains the instructions in detail for this section of the 
 survey, and may assist in interpreting the first combination. After 
 that, the respondent completes the remainder of the form without 
 
 3-3 
 
assistance from the interviewer, who is administering the survey to the 
 second member of the household. In 30 percent of the households, two 
 persons were surveyed, while only one was surveyed in the remainder. 
 
 The survey was under 30 minutes in length and was administered by 
 interviewers hired by the City of Xenia. The interviewers had no 
 previous experience; they were trained and supervised by a person from 
 a local marketing firm. Thus, survey administration was a fairly simple 
 process. Of 670 persons surveyed, 451 completed the direct utility 
 assessment section of the survey in a manner allowing a unique utility 
 equation to be estimated. An additional 70 respondents gave the same 
 
 i 
 
 numerical response (generally "never used transit") to all combinations; 
 obviously no tradeoff information could be gained from these responses. 
 
 The remainder of the sample (about 20 percent) did not complete all 18 
 combinations which is necessary to develop the direct utility model. 
 
 No major problems were encountered by the respondents to the survey. 
 While there had been concerns expressed before the survey was administered 
 about people’s ability to rate 18 combinations, only a small portion of 
 the sample had diffficulties with the ta.sk. The majority found it easy 
 and were able to complete it in ten minutes or less. Further details 
 ;on the instrument are included in Appendix C. ~, r ana iy Z i n g multi- 
 attribute judgment data of the type obtained in the survey. It is de¬ 
 scribed in Appendix A and will be briefly described in this chapter. 
 
 A modification of functional measurement procedures was employed in 
 
 hich the usual reliance on factorial or fractional factorial designs 
 
 (Chapter 4, Section 4.3) was retained, but multiple linear regression was 
 
 3-4 
 
 
4. METHODOLOGY 
 
 % 
 
 4.1 Overview of Research Objectives 
 
 This component of the Xenia evaluation was designed to focus on data 
 obtained from the responses to the direct utility experiment in the Xenia 
 Home Interview Survey. The experiment contained a set of 18 alternative 
 transit systems that were described by different combinations of type of 
 service and headways, costs, walking distances, and travel time differ¬ 
 ences of auto over transit. For example, one transit description might 
 be: fixed-route bus service every 15 minutes; 50-cent fare; bus stop lo¬ 
 
 cated three blocks from the user's home; and travel time ten minutes 
 longer by transit than by auto. Other descriptions contained different 
 combinations of levels for these four attributes. The attribute combin¬ 
 ations were selected so that they had the following properties: (1) Each 
 vector or set of attribute values is minimally correlated with every 
 other vector. Had it been possible, the attributes would have been made 
 totally uncorrelated. (2) The levels of the attributes span as much as 
 possible of the variation in actual transit system experience during the 
 Xenia demonstration. 
 
 Respondents were instructed that each of these 18 different transit 
 systems either already had been or could be implemented in Xenia. They 
 were then asked how frequently they might use each if each was available. 
 Some respondents were asked for their reactions regarding a work trip to 
 the central business district (CBD); others were asked about a shopping 
 trip to the CBD. Respondents were provided a frequency scale to make 
 
their responses; the scale had five major categories: "never,” "rarely," 
 
 system (in front, 3 blocks, 6 blocks). Paratraasit was 
 
 "half the time," "more than half the time," and "all the time." However, 
 
 between attributes in the experiment. t s the level 
 
 as shown on the survey form, an eleven-point scale was used to record the 
 
 response to walk. 
 
 actual responses. The eleven-point scale was chosen as the maximum level 
 
 T^ is the difference in travel time of auto minus transit 
 
 of distinctions that could reasonably be made. An odd number is generally 
 
 of time difference squared, used to detect non-linearity 
 
 used to allow exactly neutral responses (a six on an eleven-point scale) 
 
 which would not be possible with an even number. The choice of the words 
 "how frequently" to describe the scale may have been unfortunate, as the 
 
 intent of the study is to examine mode choice. This wording is ambiguous, 
 
 The regression equation expressed above as Equation 4.0 is described 
 
 particularly for shopping trips, and trip generation issues may be con- 
 
 in detail in Chapter 5, A unique equation of the form of 4.0 was estimated 
 
 founded with the desired mode choice issues. A better wording would be 
 
 for every respondent who completed the survey which is technically a 
 
 "how likely are you to use transit." The objective in collecting these 
 
 controlled experiment with sufficient observations to estimate all the 
 
 responses was to permit the derivation of a unique mathematical expression 
 
 coefficients 3 of - Equation 4.0 if all responses are complete. The co- 
 
 for each individual to explain his/her responses. 
 
 efficients of each individual 1 s equation describe how each individual is 
 
 These individual equations express the tradeoffs each individual is 
 
 willing to trade off each of the four attributes. Moreover, any non- 
 
 willing to make among the four attributes. Based upon previous work with 
 
 linearity in an individual's response to any of the attributes, holding 
 
 functional measurement, certain methods and analytical procedures dis- 
 
 the remaining ones constant (termed the "marginal response") is 
 
 cussed later in this chapter were selected to construct the quantitative 
 
 detectable. . • " 
 
 expressions to describe and explain each individual’s tradeoffs. Func- 
 
 Xt is important to note at this point that the individual’s pre- 
 
 tional measurement is a theoretical procedure for analyzing multi— 
 
 dieted judgment values are assumed to correspond to the rank order of 
 
 attribute judgment data of the type obtained in the survey. It is de- 
 
 their choices. Because of this correspondence, choices can be predicted 
 
 scribed in Appendix A and will be briefly described in this chapter. 
 
 by assuming.that whichever of two alternatives is assigned a higher 
 
 A modification of functional measurement procedures was employed in 
 
 which the usual reliance on factorial or fractional factorial designs 
 
 These squared terms are technically termed 'othogonal polynomials." 
 
 (Chapter 4, Section 4.3) was retained, but multiple linear regression was 
 
 formations that render x and x*"- uncorrelated to permit efficient 
 estimates to be derived. 
 
 4-2 
 
used as an approximation to the true tradeoff function. Previous work 
 discussed in Dawes and Corrigan (1974) has demonstrated that even if the 
 true function is not a linear, additive equation, a linear regression 
 specification will still capture the rank order of the preferences very 
 well. Thus, if individuals select that alternative which yields the 
 highest expected likelihood of use, it is only necessary to predict the 
 rank order of likelihood of use to predict the number of individuals who 
 will select a particular alternative. 
 
 For every individual who completed the experiment in the Xenia 
 Home Interview Survey (and who didn't give the same numerical response 
 to all alternatives, such as never take the alternative), a linear 
 regression equation was estimated that had the following form: 
 
 Li - B 0 + B 1 (TS i )+ B 2 (bh 1 )+ B 3 (th 1 )-»- B 4 (F i )+ 8 5 (F i 2 ) (4.0) 
 
 + e 6 (v i )+ 6 7 (w i 2 )+ 6 9 (T 1 2 )+ e t . 
 
 where Li is the expected likelihood of using any one of the 
 
 i (-18) transit alternatives; 
 
 T$i is the type of transit system (bus or paratransit); 
 
 BH^ is the headway for fixed-route bus (15 or 30 minutes) 
 
 TH^ is the headway for paratransit (taxi) system (call- 
 
 ahead 2 hours or on-demand) 
 
 F^ is the level of fare charged (free, 50 cents, $1) 
 
 F^ is fare level squared, used to detect non-linearity 
 in the response to fare. 
 
 4-3 
 
is the level of walk distance to the nearest fixed-route bus 
 system (in front, 3 blocks, 6 blocks). Paratransit was 
 always equal to "in front 1 '; this is the only correlation 
 between attributes in the experiment. W^2 i s the level 
 of walk squared, used to detect non-linearity in the 
 response to walH.-j, like lihood of use value for i; 
 
 is the difference in travel time of auto minus transit 
 (same, 10 minutes more, 20 minutes more); Ti^ is the level 
 of time difference squared, used to detect non-linearity 
 in the resDonse to time.l 
 
 This likelihood'-of-use value is best thought of as an approxima- 
 
 3 are regression coefficients 
 
 0 y 
 
 £. is the error term 
 
 Drr the 
 
 ' 
 
 The regression equation expressed above as Equation 4.0 is described 
 in detail in Chapter 5. A unique equation of the form of 4.0 was estimated 
 for every respondent who completed the survey which is technically a 
 controlled experiment with sufficient observations to estimate all the 
 coefficients $ of Equation 4.0 if all responses are complete. The co¬ 
 efficients of each individual’s equation describe how each individual is 
 willing to trade off each of the four attributes. Moreover, any non¬ 
 linearity in an individual’s response to any of the attributes, holding 
 the remaining ones constant (termed the "marginal response") is 
 detectable. 
 
 It is important to note at this point that the individual’s pre- 
 
 \ 
 
 i * i . ( 
 
 dieted judgment values are assumed to correspond to the rank order of 
 the’ir choices. Because of this correspondence, choices can be predicted 
 by assuming that whichever of two alternatives is assigned a higher 
 
 These squared terms are technically termed ’’othogonal polynomials." 
 Orthogonal polynomials are discussed in Appendix B. They are trans¬ 
 formations that render x and x2 uncorrelated to permit efficient 
 estimates to be derived. 
 
 4-4 
 
numerical score by the individual, then that one would be chosen over the 
 other when both are present. This provides a method for developing a 
 forecasting system based on the analysis of the experimental data. If 
 every individual has an equation and the values of each of the alterna¬ 
 tives on each of the important attributes are known, one can generate ex¬ 
 pected likelihood-of—use values for each alternative for each indivi¬ 
 dual by using their equations. Then by assuming that choice is rank- 
 order related to the predicted likelihood-of-use values, the individual 
 must choose that alternative that yields the highest predicted value. 
 
 One can then generate an aggregate forecast by simply counting the num¬ 
 ber of assignments to each alternative, and expanding the sample to the 
 total population. 
 
 The forecasting system, therefore, consists of a set of individual 
 linear equations. Each equation has coefficients (Equation 4.0) for the 
 important attributes; and each alternative has a different value for each 
 of the attributes. To forecast aggregate transit shares (or shares for 
 other modes), one substitutes the appropriate attribute levels for a given 
 alternative in each individual’s estimated form of Equation 4.0 and 
 generates a predicted likelihood-of-use value. These likelihood-of-use 
 values are compared for each of the available alternatives and it is as¬ 
 sumed that each individual chooses that alternative with the highest 
 values. Thus, the choice rule is: 
 
 { 1.0 if is max (L^ 
 
 (4.1) 
 
 0 otherwise 
 
 4-5 
 
where 
 
 ? (i | A) 
 
 is the probability of the individual choosing alternative 
 i from the set of A available; 
 
 L, 
 
 1 
 
 is the expected likelihood of use value for i; 
 
 Max(L^) 
 
 is the maximum of the L. in A. 
 
 of all systems. Th%s, the desired me 
 
 is tl 
 
 ..... - r c 3 > i. 
 
 This iikelihood-of-use value is best thought of as an approxima- 
 
 V 
 
 tion to that numerical value that the individual respondent would have 
 
 given on the response scale in the survey had that particular alterna¬ 
 
 tive been one of the 18 described. If there are M alternatives, one 
 
 Section 1.2 this approach is ad hoc and is based on previous exo< 
 
 generates M expected numerical likelihood of use responses. The M 
 
 mav ovei 
 
 ate their 
 
 itent to. 
 
 values are searched to find the highest value which is used to assign 
 
 use transit. This approach implicitly adjusts for this effect. 
 
 the individual to that alternative. After all individuals have been 
 
 assigned to an alternative, all individuals assigned to a particular 
 
 alternative are summed and then divided by the total number of indivi- 
 
 duals in the sample to obtain an estimate of the aggregate mode share or 
 
 >d with 
 
 :at 
 
 .oeit demand model regarding their pre¬ 
 
 proportion of total ridership that should be assigned to that mode. 
 
 abi-li 
 
 . 
 
 bee 
 
 50 mere 
 
 was little, of no variation in 
 
 Once these sets of individual coefficients have been estimated, it 
 
 the four attributes discussed above in the real Xenia environment* it, 
 
 seems logical to ask whether the coefficients of individuals are related 
 
 impossible to estimate a logit model that could he compared, to 
 
 to measures that can be observed on the individual respondents and their 
 
 ier forms. Hence the only results reported in this document 
 
 households: social, demographic, environmental, experiential, and situ- 
 
 refer to the models based on the direct likelihood of use estimates• 
 
 ational measures. It is hypothesized that differences in coefficients 
 
 Xhfi forthconiins sections py tin 0 Vcnrious snslvticsl in 0 £Tio(ls ©iq"'* 
 
 are systematically associated with differences in these interpersonal 
 
 iective of these sections is to 
 
 
 1 
 
 Sometimes an alternative will be one of the 18 used in the survey. 
 
 4-6 
 
measures. There is some data in the home interviews to examine this 
 question, and the results are reported in Chapter 5. 
 
 Finally, it seems logical to compare the individual level fore¬ 
 casting system discussed above with a more aggregate approach based on 
 a representative or average individual. Two different aggregate ap¬ 
 proaches to forecasting were developed: 
 
 1. Equation 4.0 was estimated from the average likelihood of use 
 responses calculated for each of the 18 alternatives. It can be shown 
 that the coefficients estimated just on the average responses over all 
 individuals are the same as if they were estimated from all the indivi¬ 
 dual data points, and are the same as the average of the individual co¬ 
 efficients if certain conditions are met, which are approximately met in 
 the Xenia data.'*' Thus, the first aggregation method involves an esti¬ 
 mate of the average coefficients of Equation 4.0, which estimates the 
 expected (or average) likelihood of use value across all individuals for 
 each of the alternatives. 
 
 2. Equation 4.0 was estimated from the proportion of self- 
 estimated regular users for each of the 18 alternatives. The proportion 
 of regular users is determined by calculating the number of responses of 
 seven (7) or greater on the response scale used in the survey for each 
 of the 18 alternatives. This frequency is converted to a relative 
 
 These conditions are satisfied by balanced combinations of alterna¬ 
 tives. This means that all levels of all attributes appear an equal 
 number of times and all attributes are uncorrelated. As will be ex¬ 
 plained later in Chapter 4, these conditions are almost satisfied in 
 the Xenia data. 
 
 4-7 
 
frequency of responses seven or greater by dividing the absolute number 
 of sevens or greater for all alternatives. The logic for this adjust¬ 
 ment is that if all 18 alternatives were available, the individual ob¬ 
 viously couldn’t be a regular user (which is what seven or greater on 
 the scale means) of all systems. Thus, the desired measure is the pro¬ 
 portion of regular user responses for alternative i relative to all 
 regular user responses for the i alternatives. This interpretation is 
 very similar to the concepts expressed in a disaggregate behavioral 
 choice model (see Domencich and McFadden, 1975). As noted earlier in 
 Section 1.1, this approach is ad hoc, and is based on previous experi¬ 
 ence which indicates that respondents may overstate their intent to 
 use transit. This approach implicitly adjusts for this effect. 
 
 Note that in both aggregation procedures the same functional form 
 of Equation 4.0 is estimated. Only the dependent variable changes in 
 
 S’ 
 
 each case. In addition, it was expected that these function forms could 
 be compared with a disaggregate logit demand model regarding their pre¬ 
 dictive ability. However, because there was little or no variation in 
 the four attributes discussed above in the real Xenia environment, it 
 
 , i . . i -i i . i ^ r 
 
 proved impossible to estimate a logit model that could be compared to 
 
 ♦ i T .11 j i . j . i , i j » r r __ . • 
 
 these other forms. Hence the only results reported in this document 
 refer to the models based on the direct likelihood of use estimates. 
 
 The forthcoming sections review the various analytical methods em¬ 
 ployed in the research project. The objective of these sections is to 
 
 are held constant at some level. Each attribute has a marginal value 
 
 4-8 
 
provide a simple introduction to the concepts and methods required to 
 develop tradeoff functions and to analyze data related to such functions. 
 To do this, it is necessary to discuss the following subjects: 
 
 1. Direct value assessment is the method by which individuals re¬ 
 spond to alternatives and by which their tradeoffs can be directly de¬ 
 rived. Derivation of different functional forms is discussed, as is 
 their interpretation. 
 
 2. Factorial sampling plans are the method by which the set of 
 alternatives is created for individuals to evaluate. It provides the 
 vehicle to implement the statistical theory of direct value assessment. 
 
 3. The Xenia sampling plan is the specific factorial sampling plan 
 employed to implement the direct value assessment method in Xenia. The 
 properties of this plan and their interpretation are discussed. 
 
 4. Forecasting procedures from individual tradeoff functions are 
 also described in a brief section. 
 
 4-9 
 
4.2 Direct Value Assessment: An Overview 
 
 Direct value assessment seeks to develop a quantitative expression 
 or model of a single individual's or some group's relative values such 
 as their likelihood to use a set of multi-attribute alternatives. Di¬ 
 rect value assessment, therefore, develops a model of the process by 
 which attributes are combined by an individual or group to assign a 
 value to a bundle or collection of attributes. Because any alternative 
 such as a transportation mode can be expressed as a combination (or 
 bundle) of levels of different relevant attributes, the responses of 
 individuals to alternatives can be expressed as values for the attri¬ 
 butes (and their levels) of the alternatives. Interest centers on the 
 shape (functional form) of each separate attribute function and the 
 
 i ^sA /\b 
 
 manner by which the individual combines the separate preferences (or 
 values) for the attributes into an overall value for the alternative(s) 
 (the model or specification). 
 
 Technically, it is necessary to describe both the marginal and 
 
 joint value functions. A marginal function refers to the shape or be- 
 
 i i 1 i 1 _ | | 
 
 havior of an individual's responses to a single attribute, while all 
 
 9*ou 5>./5 $1.00 $1.50 $2.00 
 
 other attributes are held constant. For example, if cost, type of ser¬ 
 vice and headway, walk distance, and travel time difference are primary 
 
 FIGURE 41 
 
 attributes which determine choice (or rejection) of travel modes, then 
 
 A11eman 1 ve £oss.ihilities for * 
 
 the marginal response (or value) for cost would be the shape of the re- 
 
 2l .,^ ar giQal Preferences for Levels of Fare 
 
 sponse curve for various levels of cost, while the remaining attributes 
 
 \ 
 
 are held constant at some level. Each attribute has a marginal value 
 
 4-10 
 
 4-12 
 
function. An example of various possible marginal value functions is 
 given in Figure 4.1 for the case of cost. 
 
 Figure 4.1 depicts marginal preference responses for cost (fare) 
 P(F_^), as a function of cost, F^. Curve A illustrates a marginal pre¬ 
 ference response function that decreases at a decreasing rate; i.e., 
 each successive increase in fare leads to less decrease in preference: 
 
 $2 is almost equivalent to $1. Curve B suggests that preferences de¬ 
 crease at an increasing rate; i.e., each unit change in fare leads to 
 increasing declines in preference. Finally, curve C represents a con¬ 
 stant decline in preference for any unit change in fare. The shapes of 
 these curves are unique to individuals; however, one can approximate any 
 individual's curve by a polynomial expansion^ such as: 
 
 where: 
 
 P <V * s oi + Vi 
 
 pup 
 
 S 0i’ S li’ and S 2i 
 
 A.i 
 
 6 A 4 
 21 i 
 
 (4.2) 
 
 is the marginal preference function of A^; 
 are individual value coefficients; 
 
 is attribute i; 
 
 « 
 
 is attribute i squared. 
 
 In practice, polynomial expansions of degree two (squared, as in the 
 example above) are generally sufficient. Although this modelling strat 
 egy has the power to determine unique shapes of marginal response 
 curves for individuals, one can also hypothesize that individuals with 
 
 ^ A technical note in the appendix details polynomial approximations, 
 especially with reference to multiple regression. 
 
 4-11 
 
t-V'* * 
 
 point that almos 
 
 all potential forms or specifica- 
 of interest caji be transformed to a'linear additive form; hence., 
 the assumption of a jointly additiv e value specification is not unduly 
 restrictive. For example, if one proposes an alternative form, say 
 multiplication, which often emerges from an analysis of individual to 
 alternatives, this can be vrri-ttsn as. 
 
 
 
 where 
 
 V 
 
 P 
 
 P(F.) 
 
 Now. 
 
 Equation 
 
 V. - $ 0 + p ( A ^» 
 
 i. 
 
 (4.5) 
 
 ed in 
 
 •, \is the nargins^^rei 
 
 \ the product overSiJ^i marginal preferences; and 
 
 B 
 
 c 1CVCXS ° f Sn ; 
 
 JL- 
 
 I t 1_1_l 
 
 (4.6) 
 
 0 $.25 $.50 $.75 $1.00 $1.50 $2.00 
 
 where all terms are as defined in Equations 4.0 and 4.5. Equation 4. 
 
 states that, if expanded by cross-multiplication, the multiplicative form 
 is actually a jointly additive involving single terms (mar- 
 
 Altemative Possibilities for the Behavior 
 
 of Marginal Preferences for Levels of Fare 
 
 • terms, i.e., expressions in 
 
 1 Note that in a jointly multiplicative function, any constants on the 
 P(A.) would "wash out" by common multiplication to some constant such 
 i o 11 
 
» 
 
 similarly shaped curves may have similar interpersonal situations and 
 characteristics. The shapes of these curves are determined by the co¬ 
 efficients 8 li and 8 2 ^ in the example above. Thus, if these coefficients 
 can be estimated for each individual, then it is possible to test for 
 differences in these coefficients among individuals as a function of 
 differences in interpersonal characteristics. This is discussed in 
 Section 4.4. 
 
 A joint value (or ^response) function represents the manner in which 
 the separate marginal values are combined to produce an overall value 
 for the bundle (i.e.,. as noted above, the value or response for an alter¬ 
 native). For example, it is assumed in most travel demand models that 
 the joint function is: 
 
 V j “ V + 6 l' P(A i ) + B 2 P(A 2 ) + ... + 6 n , P(A n ) (4.3) 
 
 where 
 
 V. is the value assigned to alternative j (bundle or com- 
 
 ^ bination j); 
 
 P(A^) represents the marginal preference responses for attri¬ 
 bute i; 
 
 Bq* , . . . are constants. 
 
 In particular, most travel demand models assume that: 
 
 P(A 1> - 8 0i + 6 u A i 
 
 (4.4) 
 
 That is, the marginal value function is assumed to be linear; this is 
 equivalent to assumption C in Figure 4.1. The joint value function is, 
 therefore, also linear and additive in the B’s. 
 
 4-13 
 
Note at this point that almost all potential forms or specifica¬ 
 tions of interest can be transformed to a linear additive form; hence, 
 the assumption of a .jointly additive value specification is not unduly 
 
 restrictive. For example, if one proposes an alternative form, say 
 
 PC . H. V 1 'V y XS 
 
 multiplication, which often emerges from an analysis of individual to 
 alternatives, this can be written as: 
 
 VV 
 
 where 
 
 V. 
 
 j 
 
 v, - £ 0 + 8-, n p(a ), 
 
 J 1 
 
 is as defined in A.3; 
 
 H(l) 
 
 H(2) 
 
 HC3) 
 
 (A.5) 
 
 $2 
 
 P(A.) is the marginal preference for attribute i; 
 2- 
 
 n 
 
 3 0 " and B 1 ” 
 
 is the product over all i marginal preferences; and 
 are. constants. 
 
 Now, if one substitutes the relationship between the levels of an 
 
 '• I ■’ V' \ : 
 
 attribute and the marginal values for these levels (Equation A.O) into 
 P(F i , H*) X \ 
 
 Equation A.5, restricting it to degree two, this yields: 
 
 X \ 
 
 vj - 6 0 " + 6!" n (6 0 \ e, a. + b 2 .a. 2 ) 
 
 i i i i i 
 
 (A.6) 
 
 where all terms are as defined in Equations A.O and A.5. Equation A.6 
 states that, if expanded by cross-multiplication, the multiplicative form 
 
 *H(3) 
 
 is actually a jointly additive expression involving single terms (mar¬ 
 ginals) , i.e., expressions in A^, alone ; and cross-product 
 
 terms, i.e., expressions in A., • A^ * A^, etc. The single terms are 
 
 FIGURE A.2 
 
 H(l) 
 
 P(Fi, 
 
 Note that in a jointly multiplicative function, any constants on the 
 P(A.) would "wash out" by common multiplication to some constant such 
 as 1 8 1 ". 
 
 1!> minutes 
 
 the combined or joint pi4rfk4*ence for combinations F* and Rj . 
 
 A-16 
 
commonly referred to as "main effects" because they represent the 
 "main" contribution of each attribute, independent of other attributes 
 to the overall preference or value; the latter terms are commonly re¬ 
 ferred to as "interactions" because they represent any effects that two 
 or more attributes have in combination above and beyond the sum of their 
 main effects. If one considers how preference responses change as one 
 .jointly changes fare and headway, the strictly linear additive form of 
 Equation 4.3 states that no matter what the level of fare , a unit change 
 in headway will make a unit change 82 * in the response value, if is 
 fare and is headway. By contrast, in Equation 4.6, the contribution 
 of headway to differs for different levels of fare . A simple illus¬ 
 tration of this is given in Figure 4.2. 
 
 Both panels of Figure 4.2 show that cheaper fares are always pre¬ 
 ferred to more expensive ones and shorter headways to longer; the mar¬ 
 ginal functions for fare averaged over headways in fact are similar in 
 both panels. To obtain the marginal function for fare, one simply aver¬ 
 ages the values of P(F^,H^) observed at a given value of F^. In this 
 example, there are three values to average to obtain the marginal func¬ 
 tion for F^ because there are three levels of ( 1 , 2 , and 3 ) repre¬ 
 sented by the three curves. The joint functions in panels a and b of 
 Figure 4.2, however, are very different, even though the marginal func¬ 
 tions are very similar. 
 
 Figure 4.2a shows that there is always a constant difference be¬ 
 tween the curves at each level of . This suggests that regardless of 
 
 4-15 
 
remaining 
 
 attributes might be. These results have consistently sug- 
 
 for evaluation and 
 which one can esti- 
 leir potential joint 
 H(l)issessment 
 H(2lection to ob- 
 H(3)£ considered 
 
 $2 
 
 'Bated o . 
 
 0 $1 
 (free) 
 
 F. 
 
 ize its chances for success. Obviously, the success of 
 this strategy depends on the consistency of the behavioral Intentions 
 
 found in t 
 
 \ 
 
 have been t 
 
 P(F ± , H.) 
 
 couragi J 
 
 actual behavior once 
 •.idles that ; 
 technique have been an- 
 
 timate Equation 4.5 for 
 rarely possible to do 
 
 H(l) 
 
 H(2) 
 
 h(3) 
 
 0 $1 $2 
 (free) 
 
 p wishes, to assess a person’s 
 
 i 
 
 ilue.function or combinations (or bundles) or fare, headway, walking 
 
 FIGURE 4 2 
 
 >Lance, and travel time differences. One would begin by assigning i 
 
 Additive and Multiplicative Joint Function 
 
 4-18 
 
 H(l) = 15 minutes; H(2) = 30 minutes; H(3) = 60 minutes in headways. 
 P(Fp,Hj) is the combined or joint preference for combinations F-j_ and Hj . 
 
the value of F^, the move from H(l) to H(2) always produces a constant 
 
 change (at a given point on F^) in P(F i> Hj). In fact, this is the 
 
 very notion of statistical independence—the effects, as measured by 
 
 the response of P(F.,H.) of one or more attributes in combination with 
 
 * 1 
 
 one or more other attributes are entirely independent of the levels of 
 one another. This is what one explicitly hypothesizes when one esti¬ 
 mates functions like Equation 4.3 in a travel demand model. 
 
 By way of contrast, Figure 4.2b shows that there is a non-constant 
 difference between the curves at each level of . Indeed, one might 
 
 note that as drawn, the difference between each H, curve is inversely 
 * J 
 
 proportional to the value of F^. That is, as F^ increases, the space 
 between the curves decreases. There is an intuitive way of explaining 
 this preference structure: If F^ is very expensive (very low prefer¬ 
 ence), values of make little, if any, difference; however, as values 
 of F i improve, differences in H^. begin to matter. 
 
 So to illustrate. Figure 4.2a suggests that even beyond $1, good 
 headways can make up for high fares. Figure 4.2b says, beyond $1, al¬ 
 most no improvement in headway can compensate for high fare. Previous 
 studies of individuals (see Louviere, Levin and Norman references) have 
 
 repeatedly found that Figure 4.2b is a better description than Figure 
 
 4.2a for a wide array of attributes in addition to fare and headway. 
 Similarly, it has been found that when three or more attributes are con¬ 
 sidered simultaneously (jointly), if one attribute is at a very unde¬ 
 sirable level, it makes little difference what the values of the 
 
 4-17 
 
remaining attributes might be. These results have consistently sug¬ 
 gested model forms such as Equation A.5. 
 
 These kinds of results can be extremely useful, for evaluation and 
 planning studies because they provide a means by which one can esti¬ 
 mate threshold levels for attributes and determine their potential joint 
 effects prior to initiating service. The direct utility assessment 
 technique can be applied as part of the ’’before" data collection to ob- 
 tain the expected response to the ranges of variables being considered 
 for the demonstration. Based on these results, a preferred service type 
 vehicle, headway, and fare can be established for the demonstration 
 which can maximize its chances for success. Obviously, the success of 
 
 this strategy depends on the consistency of the behavioral intentions 
 found in the direct value technique with people's actual behavior once 
 the demonstration is initiated. The limited validation studies that 
 have been done with the direct value assessment technique have been en- 
 
 couraging in this respect. 
 
 \ev, utility theory in economics and 
 
 In general, one would like to be able to estimate Equation A.5 for 
 single individuals. In practice, however, it is rarely possible to do 
 so because this requires an individual to make a very large number of 
 
 responses or value judgments in most cases. Although this problem is 
 
 cnaxce cnGOiry xn GcunOmGuirxcs ioir 3 . posrsxrDXG crroxcG^ ur icxsuxottstiip 
 
 considered in detail in Section A.3, it is useful to review the con- 
 cepts briefly at this stage. Suppose one wishes to assess a person's 
 value function for combinations (or bundles) or fare, headway, walking 
 distance, and travel time differences. One would begin by assigning i 
 
 Wilson, ! 
 
 A-18 
 
levels to fare, j levels to headway, k levels to walking distances, and 
 i levels to travel time. There would, therefore, be (i • j • k • i) 
 total bundles needed to completely specify all joint and marginal func¬ 
 tions, plus a second (i • j • k • l) or replication set of judgments 
 
 to yield sufficient variation for an error analysis. For example, if 
 
 4 
 
 i = j = k = Jl = 4, there are 4 possible bundles, or 256 total judg- 
 
 * 
 
 ments, even without a replication set which would add another 256. 
 
 Such a large number of judgments is clearly impractical in applied 
 work. As a result, a number of ways to reduce the total number of judg¬ 
 ments required have been developed, which are explored in Section 4.3 
 (Factorial Sampling Plans). It is sufficient to note that all of these 
 methods for reducing the number of bundles required for judgment result 
 in a loss of some information regarding joint functions. Thus, any 
 reduction in the total number of combinations means that one must make 
 assumptions about the joint functions that cannot be observed. This 
 is, however, not a serious restriction, just as the earlier assumptions 
 of linear and additive values were not restrictive. Appendix A demon¬ 
 strates that it is always possible to determine the true marginal value 
 functions from a reduced number of alternatives. If one knows the mar¬ 
 ginal value functions, one can then substitute in Equations 4.3 or 4.5 
 and test the joint result against the observed judgment data. That 
 specification which fits best across all individuals is accepted as the 
 ’’best" representation.^" 
 
 1 Although it is possible that different individuals have different 
 joint specifications, one usually assumes that all individuals share 
 the group form. In fact, previous work has demonstrated this to be 
 a reasonable assumption. 
 
 4-19 
 
An alternate way of approaching the problem exists as well (see 
 
 ut 
 
 Dawes and Corrigan, 1974). Specifically, it has been demonstrated 
 
 that even if a multiplicative form is true, a simpler form such as the 
 
 . 
 
 linear equation (Equation 4.3) will still correlate very highly with 
 
 ,, . . i K-r*=>«cnfvrfc.flMon data sets consists of bundles of attributes: 
 
 the observed data. In particular, it has been demonstrated that a 
 
 linear additive specification will reproduce the correct rank order of 
 
 r , rvF nafnrp of the real world the vectors of attributes are 
 
 the observed data, even if the true specification is multiplicative. 
 
 This can be seen by noting that graphs 4.2a and b imply identical rank 
 
 piit-n rn<?t and transit fare are almost certainly related to distance, 
 
 orders. Intuitively, this can be understood by noting that as long as 
 
 the marginal value functions are the same (as drawn in 4.2a and b), a 
 
 -j v A aiihset of the oossible alternatives are available to individuals, 
 
 linear function will preserve the approximate rank order. 
 
 also limiting the inferences that can be made. Thus, the sampling plans 
 
 The above discussion has stressed prediction of rank order” because 
 
 : ' .\- •' ! d-ita collection 
 
 this is all that is necessary to predict order of choice. If one assumes 
 
 rn improve alt Lve and attribute descrip 
 
 that the individual will most frequently select that alternative with the 
 
 highest predicted value, it is clear that this can be predicted from know- 
 
 •It is theoretically possible, although not practical in most in¬ 
 ledge of rank order alone. Further, utility theory in economics and 
 
 inces. hese sampling plans guarantee ndependen of all 
 
 psychology merely requires the assumption that utility (or value) and 
 
 attributes and to permit estimation of a large number of joint effects 
 
 probability of choice are order-related; it makes no statements about the 
 
 £ hese attributes...Such.a possibility would be realized by conducting 
 
 functional relationship, if any. One may, however, turn to discrete 
 
 one.or more of a number of possible controlled experiments. For example, 
 
 choice theory in econometrics for a possible choice of relationship 
 
 various bus sy bems could be developed as combinations of fare, headway, 
 
 such as the logit model. It has been empirically demonstrated that 
 
 ... .. . . . , , , , . 
 
 individual value equations yield values that appear to be related to 
 
 , q .each, attribute Ae .#., iare—25 cents. 50 cents; headways— 
 
 real world choices in much the same way as logit and other economic 
 
 ^yery 15 minutes, every .30 minutes; walking distance—-3 blocks or less, 
 
 models suggest (see Louviere, Wilson, and Piccolo, 1978; Louviere and 
 
 cases 
 
 Levin, 1978; and Louviere and Wilson, 1978). 
 
 4-20 
 
This is an important conclusion because the remainder of the re¬ 
 search relies upon it: if one can estimate a linear value equation for 
 each of a sample of individuals as an approximation to predicting how 
 the values of their responses will vary over bundles of attributes, one 
 can forecast choices over any alternatives that can be characterized by 
 these attributes. Thus, this theory is the basis of the remainder of 
 the research tasks. These tasks are to develop a reduced set of attri¬ 
 bute bundles that describe alternative transportation modes, to sample 
 individuals' likelihoods of using responses for each bundle on a numer¬ 
 ical judgment scale, and then to predict their most frequent choice of 
 
 • »< • # 
 
 • • • • 
 
 real alternatives by using the attribute levels of fare, headways, walk- 
 ing distance and travel time difference for available modes in each 
 person's equation. The choice prediction consists of assigning each 
 individual to the mode alternative that yields the highest value. Aggre 
 gate choice proportions are then obtained by counting the number of 
 assignments to each mode alternative (see Section 4.5). 
 
 The next section explores the reduction methodology for simpli¬ 
 fying the number of bundles necessary and the development of the 
 forecasting system. 
 
 4-21 
 
4.3 Factorial Sampling Plans 
 
 Modelling individuals' responses to multi-attribute alternatives 
 involves selecting bundles (combinations) of attributes for individuals 
 
 to evaluate. In a similar manner, the observations of choice employed in 
 
 factorial combinations m Bus Service Example 
 
 traditional transportation data sets consists of bundles of attributes; 
 i.e., combinations of observations of attributes at different levels. 
 
 Because of the nature of the real world, the vectors of attributes are 
 
 ■?ianqe 
 
 almost always correlated. For example, auto and transit travel time, 
 
 auto cost, and transit fare are almost certainly related to distance, 
 
 ; 0 - : o' k s 
 
 and will be strongly correlated for most trips. Moreover, in many cases 
 
 only a subset of the possible alternatives are available to individuals, 
 
 : 0 minute > 3b'oc! 
 
 also limiting the inferences that can be made. Thus, the sampling plans 
 
 m|nutes <? 3 b^o^ks 
 
 discussed in this section are extensions of traditional data collection 
 
 ! SOc j ] ■) minutes ! > 3 blocks 
 
 efforts which improve alternative and attribute descriptions over those 
 
 available in actual data. 
 
 50? 
 
 '30 minutes 
 
 <3 blocks 
 
 t b]ocks 
 
 It is theoretically possible, although not practical in most in- 
 stances, for these sampling plans to guarantee independence of all 
 attributes and to permit estimation of a large number of joint effects 
 of these attributes. Such a possibility would be realized by conducting 
 
 one or more of a number of possible controlled experiments. For example, 
 
 • ' * • ' . . : ‘ 
 
 various bus systems could be developed as combinations of fare, headway, 
 
 and walking distance to bus stop. If one assumes that two levels are al- 
 
 n 
 
 lowed for each attribute (e.g., fare—25 cents, 50 cents; headways— 
 every 15 minutes, every 30 minutes; walking distance—3 blocks or less, 
 
more than 3 blocks), there are eight possible bus systems given by 
 these combinations as shown in Table 4.1. Note that each combina¬ 
 tion provides information regarding all of the attributes. This 
 sampling plan is very effective because it provides information about 
 all separate (marginal) effects and all joint (interaction) effects. 
 
 As stated previously, a factorial experiment consisting of all com¬ 
 binations of all variables is required in order to be able to estimate 
 all possible main effects and interactions. The approach to using this 
 kind of plan from which the present study is derived, is to present 
 individuals with sets of alternatives similar to those in Table 4.1 and 
 have them express some response of interest, such as preference or like¬ 
 lihood of use. This permits one to simulate the same experiment, as 
 if it has been run in the real world. The term for this kind of sam¬ 
 pling design, or experimental design, is a factorial design. It is 
 called a factorial design because all combinations of all levels of all 
 attributes are employed. Technically, the sampling design in Table 4.1 
 
 3 
 
 is a 2 factorial. If one assigns each attribute a third level, say $1 
 for fare; every 60 minutes for headway; 0-3 blocks, 3-6 blocks, more 
 
 3 
 
 than 6 blocks for walking distance, the design would become a 3 fac¬ 
 torial, or 27 alternatives. 
 
 It should be clear that as one increases the number of attributes 
 or the number of levels, or both, the total number of possible combinations 
 grows rapidly. Two examples of this problem might be five attributes at 
 three levels which is a 3^ or 243 combinations, or an asymmetrical design 
 
 ) 
 
 4-23 
 
he. marginal 
 
 response values only: (ii) yields more information than (i)> in that 
 
 TABLE 4.1 
 
 Factorial Combinations in Bus Service Example 
 
 or asymmetric fractional design’. 
 
 \ether one can employ a 
 
 
 a cat a 
 
 £■' ~ .v 1 
 
 02 T£1C 
 
 system 
 
 Fare 
 
 Headway 
 
 Walking Distance 
 
 
 
 
 
 
 
 i 
 
 25c 
 
 15 minutes 
 
 < 3 blocks 
 
 is me 
 
 j u only bi 
 
 nse some trz 
 
 .asportation rese 
 
 
 
 2 
 
 25c 
 
 15 minutes 
 
 > 3 blocks 
 
 using ( 
 
 
 Bsent less t 
 
 han all attribut 
 
 is at a time. In 
 
 
 3 
 
 25C 
 
 30 minutes 
 
 < 3 blocks 
 
 
 liar, one type 
 
 3f design ir 
 
 which attribute 
 
 ■ a\ nr35 t wo at ‘a 
 
 
 4 
 
 25c 
 
 30 minutes 
 
 > 3 blocks 
 
 
 aalad tradeoff 
 
 analyses) i 
 
 as been frequent! 
 
 .y employed (e.g., se 
 
 
 5 
 
 50C 
 
 15 minutes 
 
 < 3 blocks 
 
 
 
 
 
 
 
 6 
 
 50c 
 
 15 minutes 
 
 > 3 blocks 
 
 
 escion (c) is i 
 
 lot discusse 
 
 d in this study. 
 
 an plans to do not-';, • 
 
 
 7 
 
 50c 
 
 30 minutes 
 
 < 3 blocks 
 
 permit, 
 
 nil of the fac 
 
 tors simulta 
 
 heously (or "at 1« 
 
 last a large subset) 
 
 
 8 
 
 50c 
 
 30 minutes 
 
 > 3 blocks 
 
 nave s: 
 
 gnificant disai 
 
 Ivantages. 
 
 Therefore, only ( 
 
 Questions (a) and (b) 
 
 
 V:-.'. 
 
 assessment study. It is impossible to provide a general rule for the 
 select Lon o: fractional designs and it is recommended that the interested 
 researcher examine some basic texts, such as Winer (1973) or Cochran and 
 (1957). An excellent cookbook to guide selection of plans for any 
 specific problem is provided by Hahn and Shapiro (1966). 
 
such as a 4 x 3 x 2^, or 1536 combinations. (Designs in which all attri¬ 
 butes are at the same number of levels are termed symmetric; all others 
 are asymmetric.) As a result of this property of factorial designs, one 
 will usually want to use considerably less than all possible combinations, 
 given any set of attributes and their levels of interest. This requires 
 that one understands the methods available for reducing the number of 
 combinations required; these are termed "fractional factorial designs." 
 
 4.3.1 Fractional Factorial Sampling Designs 
 
 If one is interested in estimating value functions as described in 
 Section 4.2, the following questions must be answered before any design 
 can be selected (see Green, 1974): 
 
 a. What type of information does one require for modelling 
 purposes: 
 
 i) main effects only 
 
 ii) main effects plus selected interaction effects 
 
 iii) all main and interaction effects 
 
 b. What is the nature of the levels of each attributes? 
 
 i) all attributes have equal numbers of levels 
 
 ii) different attributes have different levels 
 
 c. How many attributes does the researcher want to vary in any 
 
 single set of combinations? 
 
 i) all attributes 
 
 • ii) some subset of attributes 
 
 Question (a) essentially determines the complexity of the information one 
 can obtain from the individual: (i) yields the least information. 
 
 4-25 
 
essentially equivalent to providing information about the marginal 
 response values only; (ii) yields more information than (i), in that 
 it permits one to examine selected joint value terms; (iii) provides 
 
 complete information, but is rarely practical. Question (b) concerns 
 
 ~ ~ A is r i A B G 
 
 whether one can employ a symmetric or asymmetric fractional design. 
 
 Although it is easier to obtain symmetric designs in available sources, 
 
 ! 
 
 a catalog produced by Hahn and Shapiro (1966) covers a very wide range 
 of both types of designs and should ordinarily suffice. Question (c) 
 
 ! « -i r> o 1 
 
 is included only because some transportation researchers have advocated 
 using designs^ that present less than all attributes at a time. In 
 
 t 
 
 particular, one £ype of design in which attributes are varied two at a 
 time (called tradeoff analyses) has been frequently employed (e.g., see 
 
 Eberts and Koeppel, 1977). 
 
 | 
 
 • . ,. i 
 
 Question (c) is not discussed in this study, as plans to do not 
 permit, all of the factors simultaneously (or at least a large subset) 
 
 4 
 
 have significant disadvantages. Therefore, only Questions (a) and (b) 
 are considered truly relevant to the design of a direct response or value 
 assessment study. It is impossible to provide a general rule for the 
 selection of fractional designs and it is recommended that the interested 
 researcher examine some basic texts, such as Winer (1973) or Cochran and 
 Cox (1957). An excellent cookbook to guide selection of plans for any 
 specific problem is provided by Hahn and Shapiro (1966). 
 
 4-26 
 
As a rule of thumb, a common multiple that is larger than the sum of 
 each of the levels minus the number of attributes represents a potential 
 plan that is at least a main-effects plan. For example, if one has seven 
 attributes with the following levels 4x3x3x4x2x2x2 (» 1,152 
 
 combinations), then common multiple of potential interest are 24, 36, 
 
 \ • 
 
 and 48 (one certainly would not want more than 48!). Twelve is not pos¬ 
 sible because there are an insufficient number of observations to estimate 
 all main effects (4+3+34-4+2+2+2-7= 13). Thus, 24 combina¬ 
 tions is the minimum design, and will allow some interactions to be esti¬ 
 mated; 36 and 48 would allow more interactions to be analyzed. 
 
 It is frequently desirable to develop designs which provide that 
 all main effects and two-way interactions can be estimated independently 
 of one another and all other interaction effects. To illustrate this 
 idea, consider a 3 x 3 x 3 sampling plan. The levels are labelled 1, 2, 
 and 3. The full design is given in Table 4.2 for hypothetical attributes 
 A, B, and C. If one wishes to fractionate this design so that one can 
 infer the main effects of A, B, and C, one needs to know which terms or 
 effects are correlated with which others. For example, if one wants to 
 estimate the main effects independently of one another, it is obviously 
 desirable that all main effects be uncorrelated with each other. So, one 
 would want to choose a fractional sampling plan that guaranteed this inde¬ 
 pendence. Likewise, if one suspected that there would be significant 
 interaction effects, one would want to try to minimize correlations with 
 these effects as well. As a rule of thumb, less and less variation is, 
 accounted for by interactions after main effects have been accounted for, 
 
 4-27 
 
 I 
 

 
 0 
 
 
 TABLE A. 2 
 
 2 i j i A i 5 l 
 
 (4.7) 
 
 Full Factorial Coding Example 
 
 + 6,C 2 + £ W A. * B, + By, A, > B. 2 + B„A , 2 * B. 
 
 + B 
 
 2 . Tj 2 
 
 • lJ . 
 
 O A 
 
 8 1 i 
 
 cl + 3, aA. 
 
 9 1 
 2 
 
 i 
 
 A 
 
 B 
 
 C 
 
 
 
 _ 
 
 • 
 
 A 
 
 B 
 
 C 
 
 — 
 
 — 
 
 A B 
 
 C 
 
 
 
 
 
 
 
 
 
 
 
 
 
 1 
 
 1 
 
 1 
 
 
 
 
 • 2 
 
 1 
 
 1 
 
 i 
 
 3 1 
 
 1 
 
 1 
 
 1 
 
 2 
 
 
 
 . . 
 
 C. ^ 4 
 1 
 
 1 
 
 P 17 
 
 2 
 
 * C. 1 
 i 
 
 f „i3 - . 1- 
 
 . 
 
 2 
 
 i 
 
 1 
 
 1 
 
 3 
 
 
 
 
 2 
 
 1 
 
 3 
 
 
 3 1 
 
 3 
 
 
 
 
 
 
 ♦ 
 
 B ’C 
 
 
 
 A ‘ • B 
 
 
 A • B ’ 2 
 
 1 
 
 O 
 
 L. 
 
 1 
 
 
 
 i 
 
 2 
 
 2 
 
 1 
 
 i i 
 
 3 2 
 
 H 
 
 H- 
 
 M 
 
 H- 
 
 1 
 
 2 
 
 2 
 
 
 
 
 2 
 
 2 
 
 2 
 
 
 3 2 
 
 2 
 
 % 
 
 
 
 
 
 i • 
 
 B. ”• 
 
 
 + 
 
 ^23 A i 
 
 • B n . * C, 
 
 
 1 
 
 2 
 
 3 
 
 
 
 
 2 
 
 2 
 
 3 
 
 
 3 2 
 
 3 
 
 1 
 
 3 
 
 1 
 
 + 
 
 
 2 
 
 i 
 
 2 
 
 c i 
 
 1 
 
 
 • B, 2 3' C. 3 
 
 1 
 
 1 
 
 3 
 
 2 
 
 
 
 2 
 
 2 
 
 3 2 
 
 2 
 
 
 3 3 
 
 2 
 
 
 
 
 
 
 1 
 
 • b_. 
 
 
 
 
 
 
 1 
 
 3 
 
 . 3 
 
 
 
 
 2 
 
 3 
 
 3 
 
 
 3 3 
 
 3 
 
 
 
 V' . 
 
 
 . . 
 
 , v 
 
 where: 
 
 
 $o ~ 8 0 ^ are regression coefficients (a total of 27: one con¬ 
 stant, 6 main effects, 12 two-way interactions, and 8 
 higher-order terns); 
 
 is the linear effect of Factor A; A^ is its quadra¬ 
 tic effect; 
 
 B 
 
 2 . 
 
 
 
 i 
 
 is the linear effect of factor B; B. is its quadra¬ 
 tic effect; 1 
 
 is the linear effect of factor C; C/ is its quadra¬ 
 tic effect; 
 
 All remaihging terms are the cross-products or interaction ef¬ 
 fects of the above three factors. 
 
 
 A-28 
 
 
 
even if the interactions are significant. It is usually the case, in 
 fact, that two-way (e.g., A x B) interactions account for less variance 
 than main effects, but more than three-way interactions (e.g., A x B x C). 
 Hence, one usually wants to collapse across as many interaction effects 
 that are three-way or larger as possible. In fact, one tries to minimize 
 correlations with two-way effects because these could be large and affect 
 interpretation of results. 
 
 To understand how one selects such a fraction, it is instructive to 
 return to a multiple linear regression format. For the design in Table 
 4.2, the following regression equation may be specified. 
 
 4-29 
 
JB 
 
 2 ' : '' V 2 
 
 snai 
 
 v i ■ s 0 + 6 1 A i + B 2 a.- + B 3 b 1 + B 4 B i + 6 5 c i 
 
 > create all" the terms of interest' by replacing the codes 
 
 + 6.C. 2 + 8_A. * B. + B C A • B. 2 + 6 q A. 2 • B, 
 
 01 7i i oi i y i i 
 
 (4.7) 
 
 > e new 
 
 + ^10 A i 
 
 wherever the value 1" appears,oreplaci 
 
 B i + e il A i ’> %P^ ^12 A i * C i 
 
 n -jr t » 
 
 iratic esrect, wnereverotne value I appears, create 
 
 + S 13 A i ‘ C i + 8 14 A i ’ C i + 6 15 B i * C i 
 
 + Vi 
 
 reate'. 
 
 Ml If 
 
 *? ? 2 2 
 
 c, + e,_Bi« * c. + b,„b/ • c/ 
 
 i i7 1 i 
 
 18 i 
 
 
 ;t Ot 
 
 + b 19 a 1 
 
 + ^2 A i 
 
 + B 24 A i 
 
 2 2 
 E i ’ C i + 6 20 A i ' E i • C i + 8 2lV B i 
 
 ictor equals zero. The correlation between 
 
 2 2 
 B i • c i + 
 
 ft a ^ . ■r . r 
 P 23 i i C i 
 
 . of .oi --<- wf i.hon a row (observation) ; 
 
 B, • CO + B„A 2 • B 2 • C 
 
 ider 1 is -1, 25 l i l 
 
 -prodt 
 
 + • Bi 2 . Ci 2 + £i , 
 
 = +1. All cross- 
 
 rodiict s are £ormed in 
 
 same wav. 
 
 can 
 
 :ac 
 
 ore 
 
 
 Tabid 4;-'S shows that there are six independent main effects which 
 where: 
 
 ted; similarly, by expansion, one could find 20 interacjjj|)|,- 
 
 are regression coefficients (a total of 27: oneffcop- 
 stant, 6 main effects, 12 two-way interactions, and 8 
 higher-order terms); 
 
 tsti 
 
 or 
 
 comb 
 
 A. 
 
 l 
 
 B. 
 
 i 
 
 C, 
 
 l 
 
 itions, is used. The 2 ~k observations 
 
 is the linear effect of Factor A; A^ is its quadra¬ 
 tic affB.ctcJoeff icients, although no deg] 
 
 is of f 
 
 2 . 
 
 is the linear effect of factor B; B_. is its quadra¬ 
 tic effect; 
 
 )n< 
 
 :rovri thf cotal 27; y exam?le of a 
 
 is the linear effect of factor C; is its quadra¬ 
 tic effect; - u 
 
 All remainging terms are the cross-products or interaction ef¬ 
 fects of the above three factors. ^ 
 
 hey are not independent of other terms not shown . For example, 
 
 4-30 
 
In order to create a fraction which has the properties discussed in 
 
 the preceeding paragraphs, one would usually want to select a sampling 
 
 2 2 2 
 
 plan that leaves A., A. , B J , B. , C. and C. uncorrelated with one 
 
 l i i i i l 
 
 another and uncorrelated with as many of the two-factor cross-products 
 
 (interactions) as possible. In fact, one usually tries to collapse such 
 
 that terms like 8 through 8^ are correlated with one another and some 
 
 of the 8^ through 8^ terms, but not with the 8^ through 8^ terms. This 
 
 permits one to get estimates of the main effects ( 8. through 8 ) uncon- 
 
 founded by correlations with other potentially highly significant terras. 
 
 However, in order to obtain efficient estimates of these terms, one 
 
 2 2 2 
 
 needs to be able to specify A and A , B and B , and C and C in such a 
 way that they are uncorrelated. In this case a factorial or fractional 
 factorial sampling plan is a necessary but not sufficient condition to 
 
 ' V ‘ 0 > y ^ M • 
 
 4 
 
 guarantee this independence. Unless one can transform A, B, and C into 
 
 2 
 
 separate uncorrelated linear and quadratic terms, A and A will be 
 
 2 2 
 
 correlated, as will be B and B and C and C . The creation of independence 
 is accomplished by means of a transformation procedure termed the "method 
 of orthogonal polynomial transformations." This method is discussed in 
 Appendix B; briefly, it consists of a mathematical function to trans¬ 
 form the levels of any factor into terms that represent linear and 
 
 \ 
 
 squared effects and which are independent. This method is used in 
 the research tasks to insure independence of main linear and quadratic 
 effects in analyzing the responses to the Xenia sampling plan. 
 
 4-31 
 
For example, in Table A. 2, one can use the method of orthogonal 
 
 i 
 
 polynomials to create all the terms of interest by replacing the codes 
 in Table A.2 with the following: il Plan, Polynomially Cod ed 
 
 a. For a linear effect, wherever the value "1" appears, replace 
 it with "-1"; replace "2" with "0"; replace "3” with "1 M 
 
 b. For a quadratic effect, wherever the value "1" appears, create 
 a value of a new vector equal to "1"; for the level "2," 
 create a value of ”-2"; for the level "3," create a "1" 
 
 These new codes are given in Table A.3. In Table A.3, L stands for a 
 
 linear effect; Q for a quadratic effect. Note that in each column the 
 
 sum of the elements of the vector equals zero. The correlation between 
 
 each pair of vectors is zero. The cross-product, A(L) *. B(L) is formed 
 
 simply by multiplying each pair of elements within a row (observation); 
 
 thus, the first observation under A(L) is -1, and under B(L) is -1; 
 
 their cross-product is A(L) x B(L), or (-1) x (-1) «= +1. All cross- 
 
 products are formed in the same way. 
 
 1 ...... 1 __ 1 . ._* __ 
 
 Table A.3 shows that there are six independent main effects which 
 
 can be estimated; similarly, by expansion, one could find 20 interaction 
 
 terms, all of which are independent. This is possible because the full 
 
 3 
 
 factorial design, 3 or 27 combinations, is used. The 27 observations 
 are sufficient to estimate 27 coefficients, although no degrees of free¬ 
 dom remain for estimating error. Now consider the effect of taking a 
 
 « 
 
 one-third fraction of 9 combinations from the total 27; an example of a 
 one-third fraction is given in Table A.A. 
 
 All of the terms shown in Table A.A are independent of each other; 
 however, they are not independent of other terms not shown . For example, 
 
 A~32 
 
TABLE A.3 
 
 Orthogonal Coding of the Design of Table 4.2 
 
 
 
 Main 
 
 Effects 
 
 
 
 Some Example Interactions 
 
 A(L) 
 
 A(Q) 
 
 B(L) 
 
 B(Q) 
 
 C(L) 
 
 C(Q) 
 
 A(L)B(L) 
 
 B(Q)C(L) 
 
 A(Q)B(Q)C(L) 
 
 -1 
 
 +1 
 
 -1 
 
 +1 
 
 -1 
 
 +1 
 
 +1 
 
 -1 
 
 -1 
 
 -1 
 
 +1 
 
 -1 
 
 +1 
 
 0 
 
 ' -2 
 
 +1 
 
 0 
 
 0 
 
 -1 
 
 +1 
 
 -1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 -1 
 
 +1 
 
 o • 
 
 -2 
 
 -1 
 
 +1 
 
 • 
 
 0 
 
 +2 
 
 +2 
 
 -1 
 
 +1 
 
 0 
 
 -2 
 
 0 
 
 -2 
 
 0 
 
 0 
 
 0 
 
 -1 
 
 +1 
 
 0 
 
 -2 
 
 +1 
 
 +1 
 
 0 
 
 -2 
 
 -2 
 
 -1 
 
 +1 
 
 +1 
 
 +1 
 
 -1 
 
 +1 
 
 -1 
 
 -1 
 
 -1 
 
 -1 
 
 +1 
 
 +1 
 
 +1 
 
 0 
 
 -2 
 
 -1 
 
 0 
 
 0 
 
 -1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 -1 
 
 +1 
 
 +1 
 
 0 
 
 -2 
 
 -1 
 
 +1 
 
 -1 
 
 +1 
 
 0 
 
 -1 
 
 +2 
 
 0 
 
 -2 
 
 -1 
 
 +1 
 
 0 
 
 -2 
 
 0 
 
 0 
 
 0 
 
 0 
 
 -2 
 
 -1 
 
 +1 
 
 +1 
 
 +1 
 
 0 
 
 +1 
 
 -2 
 
 0 
 
 -2 
 
 0 
 
 -2 
 
 -1 
 
 +1 
 
 0 
 
 +2 
 
 -4 
 
 0 
 
 -2 
 
 0 
 
 -2 
 
 0 
 
 -2 
 
 0 
 
 0 
 
 0 
 
 0 
 
 -2 
 
 0 
 
 -2 
 
 +1 
 
 +1 
 
 0 
 
 -2 
 
 +4 
 
 0 
 
 -2 
 
 +1 
 
 +1 
 
 -1 
 
 +1 
 
 0 
 
 -1 
 
 +2 
 
 0 
 
 -2 
 
 +1 
 
 +1 
 
 0 
 
 -2 
 
 0 
 
 0 
 
 0 
 
 0 
 
 -2 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 0 
 
 +1 
 
 -2 
 
 +1 
 
 +1 
 
 -1 
 
 +1 
 
 -1 
 
 +1 
 
 -1 
 
 -1 
 
 -1 
 
 +1 
 
 +1 
 
 -1 
 
 +1 
 
 0 
 
 -2 
 
 -1 
 
 0 
 
 0 
 
 +1 
 
 +1 
 
 -1 
 
 +1 
 
 +1 
 
 +1 
 
 -1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 0 
 
 -2 
 
 -1 
 
 +1 
 
 0 
 
 +2 
 
 +2 
 
 +1 
 
 +1 ' 
 
 0 
 
 -2 
 
 0 
 
 -2 
 
 0 
 
 0 
 
 0 
 
 +1 
 
 +1 
 
 0 
 
 -2 
 
 +1 
 
 +1 
 
 0 
 
 -2 
 
 -2 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 -1 
 
 +1 
 
 +1 
 
 -1 
 
 -1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 0 
 
 -2 
 
 +1 
 
 0 
 
 0 
 
 +1 
 
 +1 
 
 l 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 4-33 
 

 the range of past, current and near-future term possibilities In Xenia. 
 
 TABLE 4.4 
 
 
 1/3 Fractional Plan, Polynomially Coded 
 
 
 nea 
 
 ■ 
 
 
 
 
 
 
 
 
 
 A(L) 
 
 Main Effects 
 
 A(Q) BU) B(Q) C(L> CCQ) 
 
 AB 
 
 Some 
 
 AB 2 
 
 Two-Way Interactions 
 
 2 2 2 
 
 AB AB AC BC 
 
 ABC 
 
 -1 
 
 +1 
 
 -1 +1 -1 
 
 +1 
 
 +1 
 
 -1 
 
 -1 
 
 +1 
 
 +1 
 
 +1 
 
 -1 
 
 -1 
 
 +1 
 
 Tsxi • 
 
 0 -2 +1 
 
 (1) call— ahead two 
 
 +1 
 
 hour 
 
 0 
 
 +2 
 
 0 
 
 -2 
 
 -1 
 
 o • 
 
 0 
 
 -1 
 
 ■fl 
 
 +1 +1 0 
 
 _2 
 
 -1 
 
 -1 
 
 • +1 
 
 -hi 
 
 0 
 
 0 
 
 0 
 
 0 
 
 -2 
 
 -1 +1 +1 
 
 +1 
 
 0 
 
 0 
 
 +2 
 
 -2 
 
 0 
 
 -1 
 
 0 
 
 0 
 
 -2 
 
 0-2 0 
 
 -2 
 
 0 
 
 0 
 
 0 
 
 +4 
 
 0 
 
 0 
 
 0 
 
 0 
 
 -2 
 
 +1 +1 -1 
 
 +1 
 
 0 
 
 0 
 
 -2 
 
 -2 
 
 0 
 
 -1 
 
 0 
 
 +1 
 
 +1 
 
 -1 +1 0 
 
 -2 
 
 -1 
 
 +1 
 
 -1' 
 
 +1 
 
 0 
 
 0 
 
 0 
 
 +1 
 
 +1 0 -2 -1 +1 
 
 frgo 
 
 
 0 
 
 -2 
 
 -1 
 
 0 
 
 0 
 
 +1 
 
 +1 
 
 +1 +1 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 +1 
 
 fc, 3 blocks 
 
 c. 6 blocks 
 
 4. Travel Time Difference Between Transit and Auto: 
 
 a. same by transit and auto 
 
 b. 10 minutes longer by transit than by auto 
 
 c. 20 minutes longer bv transit than by auto 
 
 Technically, this is a 4 x 3 >: 3 x 3 factorial design with headway nested 
 within type of service. Hence, the effect for type of service (bus, taxi) 
 and headway nested within it contains four levels. There are 108 total 
 possible combinations in a complete design with these factors. 
 
 at destination w£s34ssumed to be small. 
 
 4-36 
 
2 2 
 
 AC = AB C and AB = ABC . This can be found by the multiplication of the 
 corresponding columns in Table 4.3. In general, the sum of the inner 
 products of two vectors must be zero for them to be independent of one 
 another. Many of the possible terms not shown in Table 4.3 fail to meet 
 this criterion. Thus, one could estimate an effect or coefficient for 
 AC or AB, but one would not be able to ascribe the coefficient to either 
 effect. Note that in traditional demand data there will rarely be ob- 
 servations over all oossible combinations, nor will there be balance in 
 
 the number of observations of each level of each factor. Thus, tradi¬ 
 tional travel demand data constitutes a non-orthogonal (non-independent) 
 sampling plan with some effects confounded in a manner which cannot be 
 determined prior to data collection. After data collection, an analysis 
 of the independent variables in the data set could be performed to exa¬ 
 mine which main effects and interactions were estimable, as if the data 
 were an experiment. In practice most of the interactions would be con¬ 
 founded, and perhaps even some main effects. On the other hand, fac¬ 
 torial sampling plans permit one to know £ priori what the confoundment 
 structure is and what effects can be reliably estimated. 
 
 4.3.2 The Xenia Sampling Plan 
 
 The particular sampling plan employed to derive combinations (alter¬ 
 natives) in the Xenia study represents one of many possible "main effects 
 only" plans and also represents a compromise for the sake of realism. 
 There were four factors of interest in the Xenia design: (1) headway 
 and mode of operation; (2) fare; (3) walking distance; and (4) travel 
 time differences. These attributes were given values to generally cover 
 
 4-35 
 
the range of past, current and near-future term possibilities in Xenia. 
 
 Thus, the following levels were selected: 
 
 I li rH CM 
 
 CM O O' 
 
 1. Headway and mode 
 a. Bus: 
 
 m 
 
 o 
 
 « 
 
 o 
 
 o 
 
 ♦ 
 
 o 
 
 o 
 
 © 
 
 b. Taxi: 
 
 ( 1 ) 
 ( 2 ) 
 
 
 m 
 
 N 
 
 
 o 
 
 
 fO 
 
 
 
 o 
 
 0N 
 
 1/^1 
 
 CM 
 
 
 o 
 
 
 co 
 
 OS 
 
 
 o 
 
 15 
 
 minutes 
 
 CM O 
 
 o 
 
 It. 
 
 o 
 
 * 
 
 30 
 
 minutes 
 
 . o © 
 
 l 
 
 o 
 
 ■ rH 
 
 |P <1 
 
 
 
 o 
 
 
 cc 
 
 . Os 
 
 
 © 
 
 
 
 • 
 
 
 o 
 
 
 2 . 
 
 Fare: 
 
 Oi 
 
 a. 
 
 call-ahead two hours 
 on-demand, immediate call 
 
 J ' <tr\ © 
 
 O O 
 
 m O 
 
 O'. O 
 
 So rH 0.0 O © 
 
 O 
 
 o 
 
 free 
 
 o 
 
 '• 
 
 o 
 
 b. 
 
 50 cents 
 
 T 
 
 <u 
 
 c. 
 
 $1 
 
 so 
 
 o 
 
 m 
 
 o 
 
 o 
 
 o 
 
 © 
 
 o 
 
 o 
 
 o 
 
 o 
 
 o 
 
 o 
 
 ip 
 
 ♦ 
 
 o 
 
 r-. 
 
 CM 
 
 as 
 
 CM 
 
 © 
 
 3. Walking Distance to Transit from Home : 
 
 a. 
 
 b. 
 
 c. 
 
 . u o 
 
 3 blocks 
 
 1 00 
 
 6 blocks 
 
 © 
 o 
 c 
 o 
 o 
 
 o 
 
 o 
 
 o 
 
 o 
 
 o 
 • ' 
 
 o 
 
 (0 blocks) 
 
 .... O ; , © -H[ ' 
 
 ; O., 
 
 0.0 
 
 o 
 
 . o.:. 
 
 0.0 
 
 0*0 
 
 tH 
 
 © 
 
 OS 
 
 
 rH 
 
 m 
 
 rH 
 
 CM 
 
 o 
 
 m 
 
 
 Os 
 
 CM 
 
 OS 
 
 SC 
 
 o 
 
 rH 
 
 
 CM 
 
 in 
 
 CM 
 
 O 
 
 o 
 
 in 
 
 
 CM 
 
 cn 
 
 CM 
 
 O 
 
 0 
 
 o o 
 
 • • 
 
 o 
 
 • 
 
 o 
 
 CM 
 
 o 
 
 CM 
 
 • 
 
 4. 
 
 Travel Time Difference Between Transit and Auto: 
 
 mT f-— hJ-—;---:-----—.1 
 
 a. 
 
 b. 
 
 c. 
 
 same by transit and auto 
 
 £ u r" o o m 
 
 m 
 o 
 
 co 
 O'. 
 
 cm 
 
 00 
 
 m 
 
 Ov 
 
 os 
 
 SO 
 
 Mf- 
 
 10 minutes longer by transit than by auto 
 
 flO-iH rH O O O O O O 
 
 t 
 
 20 minutes longer by transit than by auto 
 
 g o th th m w . 
 
 o 00 OQ so <N 
 
 ■ „ _ _ _ . 
 
 o 
 o 
 . o 
 to 
 
 § 
 
 M 
 
 CU 
 
 O 
 
 c 
 
 <y 
 
 u 
 
 a) 
 
 co 
 
 0) 
 
 
 
 4-f 
 
 c 
 
 
 
 
 •H 
 
 >, 
 
 Qi 
 
 'O 
 
 . > 
 
 
 U 
 
 
 
 in £ 
 
 c 
 
 O 
 
 ■ 0; 
 
 
 CO 
 
 
 £ cc 
 
 *C QJ 
 
 (0 
 
 
 vO 
 
 o 
 
 o 
 
 -H <D > 
 
 within type of service. Hence, the effect for type of service (bus, taxi) 
 
 J ... . ’ '' ,l ' ' .'••V- ‘ “ r —II I I J. *J ,jO .4J SM ^ 
 
 and headway nested within it contains four levels. There are 108 total 
 
 I was 
 
 : 00 H s h 
 
 possible combinations in a complete design with these factors. 
 
 Walking distance at destination was assumed to be small. 
 
 4-36 
 
Note, however, that whenever the transit service is taxi, walking 
 distance must be zero. Thus, walking distance can be eliminated from all 
 taxi combinations, so there are technically 2 x 3 x 3 x 3 bus combinations 
 and 2x3x3 taxi combinations, or 72 instead of 108. Nonetheless, this 
 is a large number of combinations and it was felt that a caximum of 20 for 
 this trial of the method was appropriate. 
 
 A compromise sampling plan was selected that employed 18 of the 
 72 combinations (see Table 4.6). in such a way that if desired, separate 
 bus and taxi response functions could be estimated for each respon¬ 
 dent. Thus, within a type of service level (bus, taxi) each attri¬ 
 bute vector is independent of every other attribute vector, but between 
 types of service there is a consistent correlation with walking dis¬ 
 tance because of the requirement that all taxi combinations have zero 
 walking distance. (In most applications of this technique, there would 
 be no correlations among any variables.) The correlation matrix for 
 these attributes for all 18 combinations is given in Table 4.5, which 
 contains nine effects or terms. These terms arise because it is possi¬ 
 ble to partition the variation (as in an analysis of variance) in a 
 dependent response measure into "main" and "interaction" effects. For 
 example, it can be shown that the sum of the squared deviations (sum 
 of squares of SS) about the mean of the dependent response (termed the 
 "total sum of squares" or sum of all the squared deviations) can be 
 partitioned in the following manner if all 108 combinations are used: 
 
 Total Sum of Squares * SS (main effects) + SS(interaction (4.8) 
 
 effects) + SS(error) 
 
 4-37 
 
on, 
 
 
 travel time.. 
 
 way/mode , two ii 
 
 
 const: 
 
 da 
 
 c 
 
 OC 
 
 •H 
 
 to 
 
 as 
 
 a 
 
 CO 
 
 •H 
 
 c 
 
 <D 
 
 X 
 
 c 
 
 •H 
 
 in d 
 
 co 
 
 o 
 
 <D 
 
 lH 
 
 W 
 
 W 
 
 (-3 IM 
 
 PQ O 
 
 < 
 
 H X 
 
 •H 
 n> 
 
 I 
 
 .'S' £} 
 
 o 
 
 CO 
 
 rH 
 CD 
 
 out the ma£fy eff 
 
 o 
 
 u 
 
 as follows: 
 
 
 rH 
 
 PN 
 
 in 
 
 
 
 IN. 
 
 O 
 
 
 CM 
 
 rH 
 
 CM 
 
 
 
 fN 
 
 o 
 
 CM 
 
 pn 
 
 NT 
 
 ON 
 
 ON 
 
 m 
 
 co 
 
 
 
 CM 
 
 ON 
 
 ere arg three 
 
 3 
 
 MT 
 
 in 
 
 o 
 
 
 o 
 
 CM O 
 
 o o o 
 
 
 • 
 
 • 
 
 • 
 
 
 • 
 
 • • 
 
 • • • 
 
 l COf 
 
 O 
 
 I 
 
 o 
 
 ° 
 
 al 
 
 ° 
 
 o o 
 
 o rH O n . 
 
 
 
 CM 
 
 rH 
 
 
 
 ares lis 
 
 o 
 
 
 
 00 
 
 ON 
 
 
 qu 
 
 © 
 
 CM 
 
 
 m 
 
 CM 
 
 
 
 
 © 
 
 
 NT 
 
 CM 
 
 
 
 
 © 
 
 
 o 
 
 NT 
 
 CM 
 
 
 o 
 
 © o 
 
 © © © 
 
 
 • 
 
 ui U( 
 
 • 
 
 
 • 
 
 • • 
 
 . 
 
 
 o 
 
 © 
 
 o 
 
 
 o 
 
 o o 
 
 rH © © 
 
 
 
 m 
 
 
 
 
 © 
 
 
 
 
 © 
 
 
 
 
 o 
 
 
 
 
 co 
 
 
 
 
 o 
 
 
 H 
 
 o 
 
 rH 
 
 o 
 
 se 
 
 o 
 
 o © 
 
 © o © 
 
 
 • 
 
 • 
 
 • 
 
 
 • 
 
 • • 
 
 • • • 
 
 
 o 
 
 o 
 
 o 
 
 
 o 
 
 O rH 
 
 © © © 
 
 f s< 
 
 can 
 
 CP 
 
 H 
 
 The reason for 
 levels of a fac 
 can be exactly 
 
 exactly fit- by 
 (data points) minus one 
 
 CN 
 
 H 
 
 to be 
 
 VO 
 
 rH 
 
 
 NT 
 
 ON 
 
 
 rH 
 
 CM 
 
 
 rH 
 
 CM 
 
 o 
 
 iH 
 
 CM 
 
 • 
 
 • 
 
 • 
 
 © 
 
 © 
 
 © 
 
 iec 
 
 r~> 
 
 o 
 
 o 
 
 © 
 
 © 
 
 o 
 
 o 
 
 o 
 
 o 
 
 o 
 
 o 
 
 o 
 o 
 o 
 o 
 o 
 
 3 
 
 the 
 
 ti¬ 
 
 the 
 
 the number 
 and mode of 
 Lctuct- ;■ ■ aijd three for. 
 
 terms in head- 
 
 m 
 
 ID 
 
 *<r 
 
 in 
 
 v£> 
 
 O' 
 
 m 
 
 rH 
 
 in 
 
 o 
 
 on 
 
 in 
 
 — 
 
 o 
 
 o 
 
 o 
 
 — 
 
 — 
 
 m o 
 
 o o o o o 
 
 • • • • • 
 
 O O 1H O O 
 
 Laf or 
 
 r-~ 
 
 in. 
 
 CN 
 
 os 
 
 CNJ O 
 
 
 o 
 
 o r t 
 
 Lnear regression 
 fie analysis 
 
 o 
 
 © 
 
 o 
 
 o o o © 
 
 ■ .: lareranee regarding 
 
 o o o 
 
 rH 
 
 CO nj © 
 
 B 
 
 .ti’v 
 
 tor, 
 
 f n ♦- 
 
 JL X U 
 
 _ 
 
 00 vO 
 
 pn o 
 
 ® °. 
 
 o o 
 
 I 
 
 o 
 o 
 
 °. 
 
 rH 
 
 © © 
 
 © 
 
 © 
 
 © 
 
 » • 
 
 • 
 
 • 
 
 • 
 
 © © 
 
 © 
 
 © 
 
 © 
 
 ON 
 
 rH 
 
 m 
 
 rH 
 
 in 
 
 ON 
 
 CM 
 
 ON 
 
 rH 
 
 CM 
 
 m 
 
 CN 
 
 m 
 
 CM 
 
 CO 
 
 CM 
 
 © © 
 
 • • 
 
 CM 
 
 • 
 
 © 
 
 • 
 
 CM 
 ► ^ 
 
 o © 
 
 © 
 
 © 
 
 © 
 
 
 
 
 
 
 
 
 
 
 
 
 rH 
 
 o 
 
 © 
 
 
 ON 
 
 m 
 
 CM 
 
 rN 
 
 vD 
 
 
 CO 
 
 © 
 
 o 
 
 
 m 
 
 © 
 
 00 
 
 rH 
 
 <r 
 
 BH 
 
 oo 
 
 © 
 
 © 
 
 
 rH 
 
 co 
 
 m 
 
 ON 
 
 rH 
 
 
 o 
 
 o 
 
 
 m 
 
 ON 
 
 ■o- 
 
 ON 
 
 rH 
 
 © 
 
 o 
 
 © 
 
 o 
 
 © 
 
 rH 
 
 NT 
 
 m 
 
 rH 
 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 
 O 
 
 1 
 
 rH 
 
 rH 
 
 © 
 
 © 
 
 i 
 
 o 
 
 O 
 
 o 
 
 © 
 
 
 © 
 
 
 
 
 
 
 
 
 
 
 © 
 
 rH 
 
 rH 
 
 
 in 
 
 
 
 rH 
 
 
 CO 
 
 © 
 
 00 
 
 00 
 
 
 vD 
 
 
 
 CM 
 
 
 © 
 
 00 
 
 00 
 
 
 
 
 
 PN 
 
 
 rr 
 
 © 
 
 IN. 
 
 JN. 
 
 
 m 
 
 
 
 <r 
 
 
 
 o 
 
 © 
 
 © 
 
 © 
 
 vD 
 
 © 
 
 © 
 
 <r 
 
 © 
 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 c 
 
 • 
 
 • 
 
 
 rH 
 
 © 
 
 o 
 
 © 
 
 © 
 
 © 
 
 © 
 
 © 
 
 mde 
 
 S3 
 
 CQ 
 
 stood as follows: (a) if 
 
 = n r e: t 
 
 H 3 
 
 CN CN CN 
 
 H fe 3 H 
 
 one straight line or ore term (co 
 r el3, there are three data .: ints and 
 
 ---:___■_ 
 
 is 4 . Sa and b 
 ly estimable if 
 sence, the F-tests 
 
 test the ratio 
 
 . . . 
 
 .1 sum of squares 
 
 acause each of the 
 
 o 
 
 earate and 
 
 on 2 can write 
 
 o 
 
 CO 
 
 E 
 
 0 
 
 u 
 
 tross-product <u 
 
 c 
 
 CD 
 
 H 
 
 (D 
 
 > 
 
 U 
 
 CD 
 
 O Z 
 X 
 CO 
 
 CD CO T3 
 CO ^ CO 
 'O CD 
 VH CO X 
 
 o id: 
 
 x 
 
 <D -H 
 a co * 
 
 •H 
 CD T3 
 O 
 
 C 01 
 CO g 
 ■U *H 
 CO 4J 
 •H 
 
 'O rH 
 
 CD 
 
 0) X > 
 
 H rH CO 
 
 >■> 3 CO CO CO U 
 •U X 4J 14-1 p U 
 
 I I 
 W S3 
 
 I 
 
 h M H h 
 
 I I 
 
 I 
 
 Two data"points 
 
 efficient); (b) if 
 these can be 
 number of levels 
 
 determines the number of terms. 
 
These sums of squares can be expressed as additive terms as follows for 
 the Xenia design: 
 
 SS(main effects) = SS(headway and mode of operation) (A.8a) 
 
 + SS(walking distance to stop) 
 
 + SS(fare) 
 
 + SS(travel time difference transit/auto) 
 
 SS(interaction 
 effects) 
 
 * SS(headway and fare) (4.8b) 
 
 + SS(headway and walking distance to stop) 
 
 + SS(headway and travel time) 
 
 + SS(fare and walk) 
 
 + SS(fare and time) 
 
 + SS(walk and time ) 
 
 + SS(headway and fare and walk) 
 
 + SS(headway and fare and time) 
 
 + SS(headway and walk and time) 
 
 + SS(fare and walk and time) 
 
 + SS(headway and fare and walk and time) 
 
 Because there are only 18 of the 108 total combinations, it is not possi¬ 
 ble to estimate all of these effects. Specifically, one must sacrifice 
 all of the interaction effects because the particular 18 combinations 
 selected have the property that all interaction effects are so highly 
 correlated that it is virtually impossible to distinguish one from an¬ 
 other. However, the main effects can be expressed as nine terms in a 
 multiple regression equation. The nine terms arise because one can es¬ 
 timate a number of terms within each main effect . In fact, the number 
 of terms one can estimate ie equal to the number of levels within a main 
 effect minus one. Thus, if there are Z levels within an attribute, the 
 main effect of that attribute requires (£ - 1) terms in a multiple 
 
 4-39 
 
F*W) 
 
 (4*10) 
 
 regression equation to be completely specified. Recalling the number 
 of levels for the Xenia design, there are four for headway and mode of 
 operation, three for cost, three for walking distance, and three for 
 travel time. Applying the (£ - 1) rule, there are three terms in head¬ 
 way/mode, two in cost, two in walk and two in time, or nine. 
 
 Thus, each of the sums of squares listed in equations 4.8a and b 
 
 constitutes an effect, and each is uniquely and independently estimable if 
 
 xj-Lictu a linear pore ion or trie interaction 
 
 ef* 
 
 data are obtained over the entire factorial design. In essence, the F-tests 
 
 F*V. T is the linear x quadratic portion of the 
 
 of analysis of variance (or those for multiple regression) test the ratio 
 
 a.s the ''u' • r; y ■ port 5 on cf fVrp 
 
 of the sum of squares contributed by each term to the total sum of squares 
 
 against zero or some measure of pure error. A multiple linear regression 
 
 or 
 
 
 i.nteraction effeet,' 
 
 model can provide additional information beyond that of the analysis of 
 
 variance test of the effects because it provides an inference regarding 
 shape or form of the functions. The reason for this is because each of the 
 
 sums of squares in Equation 4.3 can be decomposed into separate and 
 
 -Lims, sxmxxar^to a mam errect, any interaction may be decomposed 
 
 independent (additive) regression terms. For example, one can write 
 
 xn^o separate aaaitlve polynomials such as those in Equation 4.10. Note 
 
 out the main effects (those sums of squares terms with no cross-products) 
 
 tnar tnere are exactly as many terms as the product (£ - 1)•C ^2 " 1)* or 
 
 as follows: 
 
 tns pfoduct of the levels minus one. To truly estimate and diagnose a 
 
 unique model form for each individual requires the individual to respond 
 ; a ^ x combinations,in a factorial design. However, this is rarely 
 
 Cenia design,*many less than all possible 
 
 • •* ' "'[ \ 
 
 -- so that respondents could complete all responses 
 
 The reason for this can be understood as follows: (a) if there are two 
 levels of a factor, there are in effect two data points. Two data points 
 can be exactly fit by one straight line or one term (coefficient); (b) if 
 there are three levels, there are three data points and these can be 
 exactly fit by a linear plus a squared term. Hence, the number of levels 
 (data points) minus one determines the number of terms. 
 
 4-40 
 
(4.9) 
 
 SS (total) = 8 Q + B^TS) + 8 2 *(BH) + 8 3 -(TH) + 8 4 *(F) + 
 
 8 5 *(F 2 ) + 6 6 (W) + B 7 (w2) + 6 8 (T) + 6 9 (t2) + ? 
 
 where 8 - 8 = coefficients and the variables are as de¬ 
 
 fined in Table 4.5. 
 
 A separate regression coefficient can be estimated for each of the 
 (£- 1) terms in each attribute. Headways and mode of operation is de¬ 
 composed into 8^ for type of service (bus or taxi); 6^ for "if bus and 
 every 15 minutes (or every 30 minutes)"; 8^ for "if taxi and on demand (or 
 call-ahead)". Similarly, 8^ is the coefficient for the linear effect of 
 cost, 8- is for its quadratic, 8, is for the linear effect of walking 
 
 D O 
 
 distance, 8_ is for its quadratic; 8 is for the linear effect of travel 
 / o 
 
 time, and gg is for its quadratic. So, for example, if the marginal rela¬ 
 tionship (holding other factors or attributes constant) between the depen¬ 
 dent response (likelihood to use) and cost is linear, 8^ should be 
 statistically significant, while 8^ should not. 
 
 In a like manner, one can decompose any cross-product or interaction 
 effect into separate, independent components for estimation. Thus, if 
 attribute A has a levels and attribute B has b levels, there are always 
 (a - 1)(b - 1) separate terms which can be estimated. Similarly, if 
 there is an attribute C with c levels, there are (a - l)(b - 1)(c - 1) 
 possible separate three-way terms which can be estimated. Thus, for cost 
 (F) and walking distance (W) we would have: 
 
 4-41 
 
(A.10) 
 
 that 
 
 ipendentl'' 
 
 where 
 
 3’s 
 
 SS (F and W interaction) - Bj.; (F-W) 
 
 + B 2 -(F-W 2 ) 
 
 n effects are as uncorrelated as passible, with any. other 
 
 + 3 3 -(F 2 -W) 
 
 i^?, t such that main effects can be estimated totally iti- 
 
 + £ 4 -(f *w z ) 
 
 f i ti ' ■ si tx st.icn one sacrifices 
 
 + error 
 
 11, interaction (cross-product) effects by confounding 
 
 another. Many interaction terms are perfectly correla¬ 
 te separate co 
 
 thers and. tl 
 
 .one 
 
 are separate coefficients 
 
 cherefore, their estimation provides little 
 
 F-W 
 
 rmation. 
 
 F.W“ 
 
 Kenxa 
 
 is the linear x linear portion of the interaction 
 effect 
 
 is the linear x quadratic portion of the 
 interaction effect 
 
 * V H, 
 
 F-W is the quadratic x linear portion of the 
 
 interaction effect 
 
 Thus, the confounding pattern is more difficult to estimate ti.an in the 
 
 _2 tt 2 
 F * W 
 
 example of 
 
 is the quadratic x quadratic portion of the 
 interaction effect. 
 
 straight-forward and can be. determined a priori » This is. important 
 
 There would be eight terms in a three-way interaction among variables 
 
 o ecitionai analyses of travel choice data, tl 
 
 having three levels, 
 
 '■ exact analytical''sti 
 
 as shown by terms £ 1Q through 3 
 
 in Equation 4.7. 
 2d 
 
 ail 
 
 Thus, similar to a main effect, any interaction may be decomposed 
 
 into separate additive polynomials such as those in Equation A.10. Note 
 that there are exactly as many terms as the product (£ - 1 )•- 1), or 
 
 the product of the levels minus one. To truly estimate and diagnose a 
 
 it ceh D0 reaaxj.y odsbxv 6u uns.t icvcis vx 
 
 unique model form for each individual requires the individual to respond 
 
 to all the combinations in a factorial design. However, this is rarely 
 possible. In the case of the Xenia design, many less than all possible 
 
 r b y j r 
 
 ft 
 
 combinations were desired so that respondents could complete all responses 
 
 without problems. However, it was also necessary to select a sufficient 
 
 S IdulXariy DlaScU. Uui.y LraVci. UXulc uXbpiciy & out l xciciiL veuxotiwu 
 
 number of combinations so that information about the main effects could 
 be obtained. 
 
 4-42 
 
The Xenia sampling plan, which consists of 18 combinations, repre¬ 
 sents a one-sixth fraction of the total design. The design is not a 
 regular fraction. A regular fraction is one in which the number of 
 levels are balanced in how frequently they appear. For example, the 
 level of walking distance in front appears in 12 out of the 18 combina¬ 
 tions; 3 blocks and 6 blocks only appear three times each. Thus, walk 
 levels are unbalanced and the coefficients for walking distance are not 
 independent of type of service of headways within type of service. How¬ 
 ever, if one analyzes each type of service separately, both fractions 
 are regular and balanced. 
 
 Because there are only 18 out of all 108 combinations, it is clear¬ 
 ly not possible to estimate all of the effects contained in Equation 
 4.8, or the regression coefficients represented in Equations 4.9 and 
 4.1Q as noted above. This point is important because it also bears on 
 the limitations of traditional transportation demand data. Designs such 
 
 I 
 
 as the 18-combination Xenia plan are commonly referred to as "confounded" 
 or "aliased" designs. These expressions are particularly descriptive of 
 the situation. Many of the terms in the regression equation would be 
 perfectly correlated (r - 1.0) and, therefore, the interpretations of 
 one term is confounded by its perfect correlation with another term. 
 
 In thj.s situation, only one term of the correlation group can be esti¬ 
 mated, and it cannot be known which of the several perfectly correlated 
 effects it stands for. This confounding occurs because one collapses 
 the full factorial design over most of the interactions in order to 
 
 4-43 
 
develop fractions. Typically, one desires to develop fractions such 
 that the main effects are as uncorrelated as possible with any other 
 effects; that is, such that main effects can be estimated totally in¬ 
 dependently of all other effects. In this situation one sacrifices 
 
 k ’ 1 ~ " L B , 
 
 many, if not all, interaction (cross-product) effects by confounding 
 them with one another. Many interaction terms are perfectly correla¬ 
 ted with others and, therefore, their estimation provides little 
 inf ormation. », > .• 
 
 The Xenia design has nine main effect terms as noted in Equation 
 4.9; however, it is not completely orthogonal nor is it exactly balanced. 
 Thus, the confounding pattern is more difficult to estimate than in the 
 example of Table 4.4. Yet the pattern of estimates is still relatively 
 straight-forward and can be determined _a priori . This is important 
 because in contrast to traditional analyses of travel choice data, the 
 exact analytical structure is known in advance. Furthermore, because 
 
 — * T i 
 
 all respondents face the same alternatives and the same analytical struc¬ 
 ture, they all can be compared. 
 
 The combinations for each type of service are listed in Table 4.6. 
 
 It can be readily observed that all 50-cent levels of fare occur within 
 
 the 15-minute level of headway. Effectively, this means that no esti- 
 
 • 
 
 mate of the middle effects of either fare or headway can be estimated 
 in an interaction. Similarly, all observations of 3 blocks occur with¬ 
 in the 15-minute level of headway. Hence, the middle level of walk is 
 similarly biased. Only travel time displays sufficient variation to 
 
 permit the middle level to be examined. In the case of the taxi 
 
 4-44 
 
combinations, sufficient variation occurs within both fare and travel 
 time across headway to allow an examination of the interaction struc¬ 
 ture. In both cases, however, it should be noted that each interaction 
 cell has only a single observation; it is not a very reliable test in 
 this respect, either. These problems were not envisioned in the design 
 of the experiment, but occurred because such a large emphasis was 
 placed on minimizing the number of alternatives. A design of 27, which 
 would have avoided these problems, would have been preferred and has 
 proved feasible in other real-world applications. 
 
 4-45 
 
TABLE 4.6 
 
 Combinations within Types of Service for the Xenia Design 
 
 
 BUS PLAN 
 
 TAXI PLAN 
 
 
 Headway 
 
 Fare 
 
 Walk 
 
 Time 
 
 Headway 
 
 Fare 
 
 Time 
 
 1 
 
 15 
 
 min. 
 
 free 
 
 3 blk 
 
 +10 min. 
 
 On Demand 
 
 free 
 
 +10 min. 
 
 2 
 
 15 
 
 min. 
 
 50c 
 
 6 blk 
 
 same 
 
 On Demand 
 
 50c 
 
 same 
 
 3 
 
 15 
 
 4( 
 
 min. 
 
 50c 
 
 in front 
 
 +10 min. 
 
 On Demand 
 
 50c 
 
 +20 min. 
 
 4 
 
 15 
 
 min. 
 
 50c 
 
 3 blk 
 
 +20 min. 
 
 On Demand 
 
 $1.00 
 
 +10 min. 
 
 5 
 
 15 
 
 min. 
 
 $1.00 
 
 3 blk 
 
 same 
 
 Call Ahead 
 
 free 
 
 » . 
 
 < * 
 
 i ' same 
 
 6 
 
 30 
 
 min. 
 
 free 
 
 in front 
 
 same 
 
 Call Ahead 
 
 free 
 
 30 cents 
 
 +20 min. 
 
 7 
 
 30 
 
 min. 
 
 free 
 
 6 blk 
 
 +20 min. 
 
 Call Ahead 
 
 50c 
 
 +10 min. 
 
 8 
 
 30 
 
 1( 
 
 min. 
 
 $1.00 
 
 6 blk 
 
 +10 min. 
 
 Call Ahead 
 
 $1.00 
 
 +20 min. 
 
 9 
 
 30 
 
 min. 
 
 $1.00 
 
 inf rent 
 
 +20 min. 
 
 Call Ahead 
 
 $1.00 
 
 same 
 
 L 
 
 
 
 
 
 
 
 
 
 
 ELS OF FACTOR FREQUENCY (MINUTES) 
 
 > FIGURE 4,3a 
 
 a 
 
 Average Sample Response to Combinations of Frequency of 
 Service and Fare, Holding Walking Constant 
 
 4-46 
 
 4-48 
 
4.4 Analytical Techniques for Model Specification 
 
 Bearing the caveats about confounding in mind, one can attempt a 
 graphic representation of the data*to illustrate both the notion of main 
 effects and interactions. As explained in the formal development of 
 direct value assessment, knowledge of these various effects leads di¬ 
 rectly to model specification because it pinpoints the appropriate func¬ 
 tional form. As an example, if the four attributes combine linearly, 
 
 « 
 
 then no interactions can be significant. This is equivalent to the 
 expectation that plotting the response surface for headway at various 
 levels of fare (or vice versa) would yield a series of parallel curves. 
 If the model is strictly additive, each attribute contributes a con¬ 
 stant amount to the response independent of the remaining attributes. 
 This implies that all points on any two curves are separated by the 
 same constant amount; hence., the. curves must be parallel. Any departure 
 from parallelism implies non-additivity, or a significant interaction. 
 Thus, an interaction test is tantamount to a rejection criterion for 
 additivity: if interaction exists, additivity is rejected; if it does 
 not, additivity is retained. 
 
 I 
 
 These concepts can be illustrated with reference to the data from 
 an earlier study in Iowa City, Iowa, and illustrate how it can be used 
 to guide specification (Figure 4.3). The data in Figure 4.3 are from 
 a 3 x 3 x 3 factorial design in which a number of respondents judged 
 each of the 27 combinations several times. The attributes were fare 
 (15, 30, and 45 cents), walking distance (1, 3, and 9 blocks), and 
 
 4-47 
 
SUBJECTIVE PROBABILITY 
 
 \ 
 
 FIGURE 4.3a 
 
 Average Sample Response to Combinations of Frequency of 
 
 Service and Fare, Holding Walking Constant 
 
 4-48 
 
SUBJECTIVITY PROBABILITY 
 
 » 
 
 15 cents 
 30 cents 
 
 45 cents 
 
 LEVELS OF WALK DISTANCE (BLOCKS) 
 
 FIGURE 4.3b 
 
 % 
 
 Average Sample Response to Combinations of Fare 
 And Walking Distance, Holding Service Constant 
 
 4-49 
 
which 
 
 is always associated with 15-minute headways. This interaction 
 is given, in Table 4.7, and graphed in Figure 4.A, The results clearly 
 indicate curves similar to those illustrated in Figure 4.3, In parti- 
 cular, one can note that if. the cost is $1, travel time difference is 
 virtually irrelevant. The effect of travel time depends upon the level 
 of fare. They are not independent of one another in their joint effects 
 
 50 
 
 multiplicative ba 
 
 ^ 40 
 
 such as th^S in r 
 
 t—i 
 
 PQ 
 
 specifi « 
 
 U 
 
 (- 
 Je 
 
 ^ can write 
 
 where 
 
 o 
 
 & . 
 
 30 
 
 a. 
 
 >-* 
 
 H 
 
 + k. 
 
 M 
 
 > 
 
 20 
 
 M 
 
 H 
 
 a 
 
 k. Q UG 
 
 w 
 
 
 ■-5 
 
 PQ 
 
 P 
 
 10 
 
 tn 
 
 
 (TS) + k 2 U(TH,BH) 
 
 ) + k / U(T)U(F) 
 
 C4.li) 
 
 15 minutes 
 30 minutes 
 45 minutes 
 
 ility of.the alternative 
 
 J ——I- 
 
 
 «• •» 
 
 .5 3 
 
 are coefficients 
 
 LEVELS OF WALK DISTANCE (BLOCKS) 
 
 The overall response or value observed in cell i is an additive func¬ 
 tion of the marginal value for type of mode, headways within each mode. 
 
 walking distance within the bus mode, and times and costs common to 
 
 specification FIGURE from the values of the 
 
 Average Sample Response to Combinations of Walking Distance 
 
 And Frequency of Service, Holding Fare Constant 
 
 This is.only an assumption, not an inference, but it is supported 
 
 by many previous studies. 
 
headway (15, 30, and 60 minutes). As can be noted from Figure 4.3, 
 each set of two-way interactions has the same form: when values are 
 good, the curves diverge, as they become increasingly bad, the curves 
 converge. This clearly implies that the joint effects of these attri¬ 
 butes are non^additive. Indeed, the statistical tests of the interac¬ 
 tions were all significant, confirming the visual information. In 
 particular, this pattern of interaction curves implies a multiplicative 
 value equation. As demonstrated in the appendix, multiplication implies 
 that the amount of separation of the curves is directly proportional 
 to the marginal value of the response at each level of the two attri¬ 
 butes. This implies the pattern which is observed in Figure 4.3: in¬ 
 creasing value of response to levels of one attribute should cause di¬ 
 vergence in the curves of the second attribute. There are statistical 
 tests for this expectation, which are discussed in the appendix. It 
 should be clear, however, that the data curves in Figure 4.3a through 
 4.3c conform to this expectation. This means that users do not trade 
 off service and cost at all levels of these variables. When one varia¬ 
 ble is at a poor level, improvements in other variables do not offset 
 it, and the overall attractiveness of the alternative remains low. This 
 interpretation may be quite relevant to the Xenia experience, as dis¬ 
 cussed in Chapter 5. 
 
 In the case of the Xenia data, these interactions cannot be relia¬ 
 bly estimated because the factorial design is incomplete. It is possi¬ 
 ble, however, to estimate the joint effect of time and cost, holding all 
 else constant, if one discounts the middle level of cost (50-cents) 
 
 4-51 
 
which is always associated with 15-minute headways. This interaction 
 is given in Table 4.7, and graphed in Figure 4.i4. The results clearly 
 indicate curves similar to those illustrated in Figure 4.3. In parti¬ 
 cular, one can note that if the cost is $1, travel time difference is 
 virtually irrelevant. The effect of travel time depends upon the level 
 
 i ( 
 
 of fare. They are not independent of one another in their joint effects. 
 
 One can assume that the ether joint effects of interest are also 
 multiplicative based o\i this observation and other previous evidence 
 
 DC 1 
 
 such as that in Figure 4.3. x In this case, one can write the following 
 
 + 10 MIN + 20 MIN 
 
 specification: 
 
 TRAVEL TIME DIFFERENCE 
 
 U = k Q + k-jUCTS) + k 2 U(TH,BH) 'RIP RESPONSE (4.11) 
 
 + k 3 U(W) + k 4 U(T)U(F) 
 
 where L 
 
 U is the utility of the alternative 
 
 . o : 6 =.v. 
 
 U(x) is the marginal utility of variable x 
 
 free 
 
 V 
 
 V 
 
 The overall response or value observed in cell i is an additive func¬ 
 tion of the marginal value for type of mode, headways within each mode, 
 
 walking distance within the bus mode, and times and costs common to 
 
 + 10 MIN + 20 MIN 
 
 both modes. This specification is estimated from the values of the 
 
 TRAVEL TIME DIFFERENCE 
 
 attribues in Table 4.6 using the following logic: 
 
 FIGURE 4.4 
 
 __ Interaction Graphs of Time x Cost 
 
 This is only an assumption, not an inference, but it is supported 
 by' many previous studies. 
 
 4-52 
 
TABLE 4.7 
 
 Time x Cost Interaction Means 
 
 Work Trips 
 
 Transit 
 
 Travel Time Difference (Transit - Auto) 
 
 Cost 
 
 Same 
 
 + 10 min. 
 
 + 20 min. 
 
 free 
 
 5.17 
 
 4.97 
 
 3.23 
 
 $1.00 
 
 3.28 
 
 3.18 
 
 2.98 
 
 Shopping Trips 
 
 Transit 
 
 Travel Time Difference (Transit - Auto) 
 
 
 
 
 Cost 
 
 Same 
 
 + 10 min. 
 
 + 20 min. 
 
 free 
 
 6.31 
 
 5.36 
 
 3.56 
 
 $1.00 
 
 3.68 
 
 3.52 
 
 3.48 
 
 Entries in table are mean rankings by respondents on an 11 point scale. 
 
 4-53 
 
Equation 4.13 was estimated for the 18 cell means for the cor’ J 
 
 coefficients estimated from the means yield the same val- 
 _ •• ds ‘ us-l. Only coefficient 8- 
 
 free 
 
 regressed" 
 
 w 
 
 a 
 
 O 
 w 
 
 H 
 
 nif i 
 
 rH 
 
 o 
 
 4- 
 
 as: 
 
 this blast. 
 
 j H ill I H M«— M l I II I MU 
 
 $1 
 
 i t> q was then estimated, for all individuals. Thus, it 
 
 2 ... 
 
 
 + 10 MIN 
 
 ions were additive, 
 
 TRAVEL TIME DIFFERENCE 
 
 + 20 MIN 
 
 4.4a SHOPPING TRIP RESPONSE 
 
 TRAVEL TIME DIFFERENCE 
 4.4b WORK TRIP RESPONSE 
 FIGURE 4.4 
 
 Interaction Graphs of Time x Cost 
 
 5 4-54 
 
(4.12a) 
 
 U(TS) * a ]_ + b^(-l if taxi; +1 if bus) 
 
 U(BH) - a 9 + b (-1 if 15; +1 if 30 min; 0 if taxi) (4.12b) 
 
 U(TH) = a,, + b~(-l if on-demand; +1 if call-ahead; (4.12c) 
 
 0 if bus) 
 
 (4.12d) 
 
 U(W) - a. + b,(-1 if 0, 0 if 3, 1 if 6 ) + b (1 if 0, -2 if 3, 
 
 5 1 if 6 ) 
 
 • (4.12e) 
 
 U(T) - a + b.(-l if 0, 0 if 10, 1 if 20) + b ? (l if 0, -2 if 
 ° 10 , 1 if 20 ) 
 
 (4.12f) 
 
 U(C) - a, + b Q (-l if 0, 0 if 50, 1 if 100) + b Q (l if 0, 
 
 6 8 ' 9 -2 if 50, 
 
 1 if 100 ) 
 
 The coding for W, T, and C is the linear (b^, b^ and bg) and quadratic 
 (bg, b^ and b^) coding from the method of orthogonal polynomials, dis¬ 
 cussed earlier and in Appendix B. Note that substitution of Equations 
 4.12a to 4.12f in Equation 4.11 gives an expanded form of 4.11: 
 
 U - 8 0 + 8 1 *(TS) + 6 2 *(BH) + 8 3 *(TH) + 8 4 -(W) + (4.13) 
 
 6 5 *(W 2 ) + 8 6 *(F) + 8 7 *(F 2 ) + 8 8 -(T) + 8 9 -(T 2 ) + 
 
 6 10 - (T • F) + 6 U * (T • F 2 ) + S 12 * (T 2 • F) + 
 
 8 13 *(T 2 • F 2 ) + e 
 
 It should be noted, however, that the quadratic term for cost (8y) is 
 confounded with headway and mode in the case of bus ( 62 )* 30 one can 
 at best get an approximate test over the 18 combinations. 
 
 4-55 
 
Equation A.13 was estimated for the 18 cell means for the combina¬ 
 tions in Table A.6. Because the design is approximately orthogonal, 
 the regression coefficients estimated from the means yield the same val¬ 
 ues as if all of the individual data were used. Only coefficient B^q 
 from the set 8 -^q through B^ approached significance and it was not sig- 
 nificant at the .05 level; however, this test is biased, as noted above. 
 As an approximation and a simplification, an equation with only coeffi¬ 
 cients Bq through B^ was then estimated for all individuals. Thus, it 
 was assumed that all individuals could have non-linear marginals, but 
 their joint functions were additive. 
 
 A-56 
 
 58 
 
4.5 Forecasting From Individual Utility Models 
 
 I 
 
 One can estimate either the full or some reduced form of Equation 
 4.13 for each individual or at the aggregate level. To forecast choices 
 one must assume that there is a rank-order relationship between the val¬ 
 ues that are predicted by the individual equations and the likelihoods 
 of choice, i.e.: 
 
 A 
 
 P(A ) = Max[M(V )] 
 
 (4.14) 
 
 where: 
 
 p( Aj) 
 
 M 
 
 is the probability of alternative j being selected 
 by an individual i (perhaps individual i's long¬ 
 term frequency of choice); 
 
 is the response or likelihood of use value that an 
 individual’s i Equation 4.13 will predict for alter¬ 
 native j; and 
 
 is some monotonic (or rank-order) mapping. 
 
 Equation 4.14 implies that one must search through the for 
 each individual i and assign that individual to that alternative j 
 
 with the highest Vj. 
 
 Thus, there are two approaches to forecasting: 
 
 1 
 
 1. Using a random sample of individuals, simulate their choice 
 of mode for observed trips by substituting the levels or values of 
 each attribute in each individual's equation for each trip for each trip 
 purpose. Each individual is assigned to that mode alternative with the 
 highest predicted value. Community-wide choice proportions are obtained 
 
 4-57 
 
by calculating the proportions assigned to each mode in the sample and 
 
 expanding to the population. Any sub-group can be similarly estimated. 
 
 a. 
 
 oO ' 00 ao | W . 
 
 2. Using the mean or average equations estimated over all indivi- 
 
 2 
 
 duals, repeat the simulation as in method 1. A single equation is used 
 
 to represent each individual. 
 
 CM 
 
 tt|-cr¬ 
 
 ew 
 
 rather 
 
 than a separate equation 
 
 -- cr«M ryev 8 * cm 1 
 
 to each individual. Both 
 
 »n a tt 
 
 3 ■ * i ^ ^ 
 
 approaches are discussed in Chapter 
 
 unique 
 
 5. 
 
 w 
 
 o| 
 
 
 C 
 
 t—1 
 
 <c 
 
 o 
 
 CO 
 
 u 
 
 •H 
 
 +-> 
 
 CO 
 
 •H 
 
 ■U 
 
 23 
 
 C/3 
 
 <U 
 
 > 
 
 •H 
 
 CM 
 
 ♦H 
 
 u 
 
 u 
 
 CO 
 
 0) 
 
 p 
 
 0i 
 
 03 
 
 a. 
 
 u 
 
 u 
 
 CD 
 
 O 
 
 X 
 
 <u 
 
 &c 
 
 ct 
 
 V- 
 
 21 
 
 > 
 
 < 
 
 SM*' 
 
 
 -o ? 
 
 V QJ 
 
 to o 
 
 ‘■g; 
 
 §>■ 
 
 o' 
 
 i-i 
 
 w 
 
 © 
 
 ♦ 
 
 * 
 
 * 
 
 3' 
 
 « 
 
 IOC 
 
 © 
 
 © 
 
 {SI 
 
 -O' 
 
 (SJ 
 
 CM 
 
 
 CM 
 
 CT> 
 
 <C OIK 
 IO Cl Oi 
 © CM 
 
 0 , 0 - 
 
 I 
 
 to 
 
 
 - ci 
 
 in © 
 
 © O 
 
 o © 
 f I 
 
 
 
 /»*S 
 
 
 
 Sw- 
 
 
 
 tt X 
 
 z 
 
 
 
 
 
 O BQ 
 
 H 
 
 lu 
 
 u 
 
 
 
 •H P* 
 
 w 
 
 
 •H 
 
 
 
 > 
 
 
 
 U 
 
 p. 
 
 
 P 
 
 Ss 
 
 
 tt 
 
 C 
 
 
 tt tt 
 
 to 
 
 
 Vrf 
 
 
 w 
 
 to 5 
 
 
 0) 
 
 X 
 
 tI 
 
 a 
 
 •c 
 
 TJ 
 
 c 
 
 tt 
 
 V 
 
 •P 
 
 a/ 
 
 V 
 
 U-s tt 
 
 0 tt 
 
 • 10 
 
 "W 
 
 
 ■ 
 
 P 
 
 
 X . 
 
 l 
 
 
 tp 
 
 a; 
 
 tt 
 
 
 <u 
 
 CJ 
 
 tt 
 
 P 
 
 a » 
 
 X 
 
 |y 
 
 
 0 
 
 
 
 « 
 
 tt 
 
 •tt 
 
 o 
 
 
 
 © 
 
 
 
 CM 
 
 CM 
 
 00 
 
 
 
 O' o 
 
 CM ® 
 
 cc 
 
 f> — 
 
 b 
 
 oo m 
 
 CM 
 
 
 6 ~ 
 
 
 •m m 
 
 .© © 
 
 CO. 
 
 cc - 
 
 or. 
 
 o ^ 
 
 © m 
 
 r'v 
 
 <N OS 
 
 •tt ■ 
 
 \G O 
 
 as Q 
 
 
 cs» cm 
 
 
 *—i sD 
 
 •f» P 
 
 ■’ • 
 
 ■>: . '* 
 
 • ■ ■ 
 
 .♦ • * • 
 
 * • # 
 
 
 © © 
 
 © ' 
 
 
 © © 
 
 c 
 
 oo 
 
 *»■ 
 
 ie 
 
 
 1 CEUCJ-CV-C 
 
 tt (d w cto C’-o v 
 
 ■ 
 
 X C s^ ' c '««' C V" 
 
 ■{ Cl C - —< — 
 
 K-* O *•=*< 
 
 P3 
 
 tt 
 
 rs. 
 
 ■* 
 
 O * 
 • * 
 © 
 i 
 
 > U 
 
 l- to 
 
 « I- 
 V TJ 
 C « 
 •w D 
 ~J © 
 
 (-0 r— 
 
 m to 
 3t 3 
 
 » — 
 © © 
 
 © © 
 
 © CM 
 
 m © 
 
 . © O 
 
 .© a 
 
 it 
 
 © * 
 © * 
 * * 
 
 © 
 
 u*> 
 
 
 CN 
 
 <N 
 
 CN . 
 
 JJ 
 
 
 u*. 
 
 
 ■H 
 
 T“< 
 
 
 
 
 
 
 in s 
 
 •X ■?*. 
 
 
 * 
 
 •3' 
 
 H is 
 
 H U- 
 
 ^ , 
 
 H 
 
 
 
 
 4 J 
 
 4 J 
 
 p 
 
 4 J 
 
 
 
 
 
 u 
 
 P 
 
 p 
 
 P 
 
 
 
 
 
 C 
 
 o 
 
 c 
 
 o 
 
 
 CN 
 
 
 VP 
 
 Ip 
 
 1*4 
 
 WM 
 
 
 
 H 
 
 4 J 
 
 
 
 
 
 
 
 W' 
 
 
 iO 
 
 vO 
 
 
 P< 
 
 4 J 
 
 .--v 
 
 
 P 
 
 o 
 
 C\ 
 
 »n 
 
 
 
 H 
 
 
 0 
 
 m 
 
 CO 
 
 rn 
 
 o 
 
 P 
 
 
 •H 
 
 IW 
 
 • 
 
 • 
 
 • 
 
 • 
 
 o 
 
 
 U 
 
 
 
 <SI 
 
 m 
 
 in 
 
 
 P 
 
 tt 
 
 vO 
 
 
 
 
 
 
 tt 
 
 P 
 
 CO 
 
 1 
 
 
 i 
 
 \ 
 
 •f 
 
 a> 
 
 •c 
 
 • 
 
 
 
 
 in 
 
 
 c 
 
 tt 
 
 
 H 
 
 vO 
 
 sD 
 
 
 •H 
 
 •3 
 
 
 'C 
 
 O 
 
 os 
 
 m 
 
 § 
 
 
 ■ 5 * 
 
 1 
 
 CO 
 
 <n 
 
 00 
 
 
 f 
 
 
 
 • 
 
 .» 
 
 » 
 
 * 
 
 
 4 ) 
 
 
 o 
 
 
 CM 
 
 CM 
 
 m 
 
 in 
 
 S 
 
 £ 
 
 
 B 
 
 li 
 
 II 
 
 ii 
 
 0 
 
 •H 
 
 
 
 <N 
 
 rs 
 
 •<r 
 
 m 
 
 \D 
 
 
 
 * 
 
 
 
 
 
 
 ,e 
 
 o 
 
 c 
 
 0) 
 
 -I 
 
 w 
 
 •In 
 
 0 
 
 ii 
 
 C 
 
 tt 
 
 w 
 
 0» 
 
 iw 
 
 c 
 
 Id 
 
 01 
 
 1M 
 
 <0 
 
 > 
 
 V 
 
 Q <0 
 
 . 41 
 
 T5 tt 
 
 U tt 
 
 to W 
 
 •K * 
 
 * * 
 
 * 
 
 fed 
 
 o 
 
 0I\ 
 
 V 
 
 K* 
 
 op. 
 
 01 
 
 "O 
 
 .e 
 
 In 
 
 O 
 
 
 
 
 tt. 
 
 w 
 
 *2 
 
 44 
 
 
 
 .tt ■ 
 
 0) 
 
 <L> 
 
 
 
 
 
 V 
 
 p 
 
 P 
 
 tt 
 
 rH 
 
 
 
 CM 
 
 »• 
 
 3 
 
 
 
 1! 
 
 
 ^ 0) 
 
 r. 
 
 C 
 
 tt 
 
 r' 
 
 tt) 
 
 
 o 
 
 •4 
 
 
 
 a> 
 
 0 
 
 
 
 O 
 
 £ 
 
 
 
 
 
 *0 <u 
 
 
 
 •O 
 
 
 
 
 C U 
 
 
 •• 
 
 
 a» 
 
 m 
 
 
 
 
 tt 
 
 *.H 
 
 u 
 
 c 
 
 0) 
 
 
 o ■ 
 
 OJ' 
 
 P 
 
 3 
 
 11 
 
 p 
 
 O tt 
 
 P 
 
 p 
 
 tt 
 
 tt 
 
 
 0 
 
 T5 P 
 
 
 3 
 
 > 
 
 U 
 
 K 
 
 c 
 
 £ 
 
 C ■ 
 
 e. 
 
 
 *P 
 
 
 
 C V 
 
 *H.. 
 
 *r< 
 
 
 vp 
 
 H 
 
 £ 
 
 c o 
 
 £ 
 
 3 
 
 0! 
 
 ■H 
 
 
 
 
 
 
 
 C 
 
 
 
 
 
 
 ' *P 
 
 ttO 
 
 
 
 
 
 
 rP 
 
 p 
 
 in 
 
 -s- 
 
 « 
 
 x 
 
 01 
 
 u 
 
 Id 
 
 ii 
 
 D. 
 
 in 
 
 © 
 
 0) 
 
 s 
 
 to 
 
 > 
 
 01 
 
 J= 
 
 © 
 
 © 
 
 cn 
 
 CM 
 
 I 
 
 4-58 
 
5. RESEARCH RESULTS 
 
 5.1 Individual Level Response Specifications 
 
 The initial analytical task was to estimate separate regression 
 equations to explain the tradeoffs among level-of-service and cost 
 attributes for each individual who completed the direct utility assess¬ 
 ment experiment of the Home Interview Survey. Out of the 670 persons 
 surveyed, 451 supplied sufficient datq to permit separate equations to 
 be derived. The results of this analysis are summarized in Table 5.1, 
 which lists the average values of the calibrated coefficients, their 
 standard deviations, and a tabulation of the number of observed t 
 statistics in various categories ranging from not significant to highly 
 significant. The average coefficient values are given for both raw 
 attribute units (cents, minutes, blocks) and orthogonal polynomial 
 units.^ Standard deviations refer only to orthogonal polynomial units 
 because the equations were estimated in those units. One can readily 
 
 convert to raw attribute units through algebraic manipulations relating 
 
 % 
 
 the orthogonal coding (see Equation 4.12) to the raw attribute values. 
 
 Results indicate that, relative to the size of the average 
 coefficient, individuals were most variable regarding the quadratic 
 terms and bus headway. The variance in quadratic terms is to be 
 expected because quadratic shapes could vary widely among individuals; 
 
 ^Orthogonal polynomial units refer to transforms of the attributes 
 which permit linear and squared terms to be estimated. They are 
 discussed in Chapter 4 and the Appendix. 
 

 
 In order to test this hypothesis, 
 cross-classification analysis was perf 
 
 
 multiple contingency or 
 
 individual t value 
 
 ye, A 
 
 on th 
 
 C 
 
 oo 
 
 c 
 
 -o 
 
 o o 
 
 V 
 
 V 
 
 CO 
 
 u 
 
 TJ 
 
 <0 
 
 u 
 
 u 
 
 CO 
 
 TJ 
 
 co 
 
 II • 3/^ 3 rv 5^ 
 
 CO -i CM CTfS CTM 
 
 3 * | »—« "H 
 
 X — II "3 • *C • *U • 
 
 i -o c ««• c «o- c <■ 
 
 * II C « <0 CO 
 
 —4 • <o 
 
 i cewc^cwc 
 iTt#«o , cjo , *er 
 ■H £ T3 11 u re a li u 
 
 X C C C W 
 
 CO lO C *H *n< «W 
 
 frN — O nN X 
 
 - 
 
 II 
 
 -O 
 
 re 
 
 2 tions o: 
 
 tM CNI 
 
 IS (0 
 
 01 01 
 
 3 
 
 C 
 
 e £ 
 
 I e 
 
 u 
 
 ■n CN 
 
 II ► CO 
 
 CO O U 
 
 2 12 
 -o re n 
 
 CO • 
 
 » w re re 
 o re e - oi 
 
 ii u re re oj 
 
 ■H 3 >3 u 3 
 
 X c c c 
 re ih c re *h 
 
 H e c o 6 
 
 fferent individua 
 
 re 
 
 re 
 
 3 
 
 C 
 
 rsi tN cn §5 
 
 a 3 
 
 The secc id re 
 that differer :es 
 
 the indivic 5 
 
 u 
 
 H 
 
 H 
 
 in Coefficients 
 
 4N U 4N 
 
 In U >4 U ' 
 
 o o o o 
 
 vO vO iTi -n *j 
 
 WOC'l/1-3- 
 
 o re co re o u 
 *M *- • ■ • O 
 
 n n re re o 
 
 v£> 
 
 00 I I 1 I + 
 
 • « 
 in h \o \o re ^ 
 io o o> re ij 
 i eo <re oo cn o 
 
 O ^ n m re re 
 B It II II II II 
 
 n n n>j rec 
 
 « 
 
 c 
 
 re 
 
 T1 
 
 re 
 
 o 
 
 o 
 
 re 
 
 x 
 
 to be 
 
 were 
 
 S 
 
 u 
 
 re 
 
 •f 
 
 4J 
 
 o 
 
 c 
 
 "c 
 
 a) 
 
 « 
 
 re 
 
 »~i 
 
 A 
 
 CO 
 
 5 ?! 
 
 •o 
 re 
 
 £ 
 
 10 
 
 c 
 
 o 
 
 X 
 
 <W 
 
 In 
 
 O 
 
 re 
 
 x 
 
 ui 
 
 O' 
 
 re 
 
 3 
 
 h^ fyypotgesis 
 
 re 
 
 w 
 
 > 
 
 V 
 
 a 
 
 In 
 
 re 
 
 N 
 
 response 
 
 00 
 
 * 
 
 * 
 
 c 
 
 re 
 
 w 
 
 m 
 
 w 
 
 « 
 
 ♦ 
 
 ♦ 
 
 co 
 
 -N -H 
 u-i 
 
 <f vO 
 © 
 It ere 
 
 2 cs 
 
 etc 
 
 In 
 
 ■der to evaluate this 
 
 of transit usage, auto ownership, 
 hypothesis, a multiple analysis of'variance was performed. A. multiple 
 analysis of variance permits one to test the hypothesis that one or 
 
 5-4 
 
 5-2 
 
the variation in bus headway coefficients is less interpretable, but 
 indicates a wide range of reaction to that type of service/headway factor. 
 
 Also included in Table 5.1 are the percents of coefficients with 
 various t values observed over the 451 respondents.* As indicated, 
 
 2.306 is the cutoff for significance at the 95 percent level; hence, 
 it is interesting to note that many respondents had non-significant 
 fare (48.1 percent) or time (56.4 percent) coefficients, while most 
 had significant type of service/headway (63.8 percent) and walk dis¬ 
 tance (63.2 percent) coefficients. This finding supports commonly held 
 notions about the aggregate "importance' 1 of out-of-vehicle time effects 
 (walk and headway/wait times) in travel demand data. 
 
 A more interesting and important analysis of these coefficients 
 involves asking whether they are independent of one another. That is, 
 if one observes a respondent with a significant or high t value for 
 cost, is it also likely that this respondent has a high t value for 
 time? This information can have important planning implications if it 
 can be shown that the t values are dependent (related) because it would 
 imply that individuals that are strongly influenced by one attribute 
 are strongly influenced by one or more others as well. This would indi¬ 
 cate that the evaluation of the system from the individual’s perspective 
 would depend on both or three or all four of the attributes. The 
 hypothesis is that all are strongly interrelated so that a high c on 
 one attribute should imply a high t on the remaining attributes. 
 
 *In a controlled experimental plan such as the Xenia design, t values 
 reflect not only precision of estimate but also relative effect or 
 significance. 
 
 5-3 
 
In order to test this hypothesis, a multiple contingency or 
 cross-classification analysis was performed on the individual t values 
 using the classes in Table 5.1. Results indicated that the dependence 
 hypothesis should be rejected in part: cost or fare was found to be 
 independent of the remaining attributes; but the other three were all 
 dependent. Thus, a high t value for walking also implies a high t 
 value for travel time and for headway. All relationships were positive, 
 indicating that all three jointly increase (or decrease). The finding 
 
 that evaluations of cost are independent of evaluations of service is 
 interesting and potentially important if found in future work; it 
 implies that the other three attributes constitute the "level-of- 
 
 service" component and that the influence of cost is independent of 
 their influences. Hence, cost affects different individuals in 
 different ways, which are explored in the following analyses of the 
 individual coefficients. 
 
 i-he vertical axis is the value of the weight coefficient, and the 
 
 5.1.1 Analysis of Individual Differences in Coefficients 
 
 factor. The 
 
 The second research task was concerned with testing the hypothesis 
 
 interpretation of these results is as follows. 
 
 that differences in the regression coefficients of individual response 
 
 Figure 5.1.a 
 
 equations are associated with differences that can be observed about 
 
 (a.l) This figure indicates that the average self-estimated prob- 
 
 the individuals: differences in age, sex, income, previous history 
 
 ability (the intercept or grand mean) of using any of the 18 transit 
 
 of transit usage, auto ownership, etc. In order to evaluate this 
 
 alternatives decreases with income, but not in a simple linear manner. 
 
 hypothesis, a multiple analysis.of variance was performed. A multiple 
 
 obability drops off rapidly after the first income category and 
 
 analysis of variance permits one to test the hypothesis that one or 
 
 decreases slowly thereafter. 
 
 
 5-4 
 
 5-6 
 
more independent variables account for differences in two or more 
 (multiple) dependent variables. In addition, this analytical technique 
 allows one to examine the effects of the independent variables on each 
 of the dependent variables separately as well as together. 
 
 In the present context interest centered on examining differences 
 
 in the individual regression coefficients as a function of differences 
 
 in interpersonal measures such as auto ownership, income, sex, age, 
 
 « 
 
 previous history of transit usage and the like. In addition, interest 
 focused on differences in individual weights as a function of these 
 interpersonal differences. Weight refers to how much influence an 
 attribute has on an individual's responses when evaluating the alter¬ 
 natives; the higher the weight, the more influence an individual accords 
 an attribute in making judgments. Technically, one measures weight 
 as the proportion of "explained variation" from the overall regression 
 equation that can be attributed to variation in a particular attribute. 
 Weights are therefore standardized so that a weight of 1.0 means that 
 an individual considers that attribute to the exclusion of all others, 
 while a weight of 0.0 means that a particular attribute is ignored. 
 Intermediate weight values refer to intermediate influence and the sum 
 of the attribute weights is always unity. 
 
 Thus, the two analyses are complementary: the analysis of the 
 regression coefficients supplies information about variation in shape 
 of individual functions as a function of variation in individual 
 difference measures; the analysis of weights supplies information re¬ 
 garding variations in influence that are associated with variations in 
 interpersonal variables. 
 
 5-5 
 
t 
 
 <u 
 
 0) ® 
 
 W C 
 
 C 0) 
 
 a) O 
 
 The full list of interpersonal variables and a summary of results is 
 
 given in Appendix D. (In Table D.l, any significant interpersonal dif- 
 
 I] / w 
 
 ference effects are noted for both weights and regression coefficients. 
 
 sO § I ; « 
 
 2 33 4-j i 
 
 Also, one can examine overall or global effects of the interpersonal 
 
 ✓ 
 
 factors on the weights or coefficients taken as a whole by examining 
 
 the "overall test" rows in Table D.l.) 
 
 (Cl 
 
 _ 
 
 £3 
 
 Because there are many significant effects, attention will be con¬ 
 
 fined to discussion of some of the most significant effects to illus- 
 
 ?«S 
 
 IT. 
 
 . 25 , 
 
 trate the power of the procedure to detect important differences in 
 
 O *v I J 0 I I 
 
 traveller responses as related to differences among traveller charac- 
 
 teristics. These results are given in Figure 5.1, which shows the 
 
 H , ns o| 
 
 0) K> 
 
 Cm 
 
 O 
 
 results uncovered for selected effects. In Figure 5.1.a (parts 1 through 
 
 4J C3 
 •H . W 
 
 22), the vertical axis is the value of the regression coefficient for a 
 
 41 
 
 Cm 
 
 o 
 
 particular attribute, while the horizontal axis is the value of the 
 
 ■■ ■ j y -1 «? . * © - -» o 
 
 s 
 
 interpersonal difference factor. In Figure 5. l.b (parts 1 through 6), 
 
 j | / p ^ 3*| i ' j W 
 
 2 : v' - r ! < g y 
 
 the vertical axis is the value of the weight coefficient, and the 
 
 w 
 
 K i\ M <1 
 
 horizontal axis is the level of the individual difference factor. The 
 
 O OQ CM SO 
 
 - 
 
 X 
 
 interpretation of these results is as follows. 
 
 Figure 5.1.a 
 
 
 o 
 
 in 
 
 •I 
 
 to o 
 
 S' 
 
 (a.l) This figure indicates that the average self-estimated prob- 
 
 CO 
 
 CO <0 
 * (lit 
 
 41 CO 
 © * 
 
 ability (the intercept or grand mean) of using any of the 18 transit 
 
 2 
 
 / 5 , v 0 
 
 alternatives decreases with income, but not in a simple linear manner. 
 
 03 
 
 
 The probability drops off rapidly after the first income category and 
 
 decreases slowly thereafter 
 
 COEFFICIENT VALUE 
 
 5-6 
 
 5-8 
 
Grand Mean Grand Mean Grand Mean 
 
 I-1-1-1- 
 
 o sO CM 00 
 
 CM r* >-l 
 
 o 
 
 I 
 
 S 
 
 O 
 
 vO 
 
 o 
 
 o 
 
 CM 
 
 UJ 
 
 o 
 
 < 
 
 COEFFICIENT VALUE 
 
 COEFFICIENT VALUE 
 
 
 w 
 
 o 
 
 l-l 
 
 u, 
 
 5-7 
 
 Results of Analysis of Interpersonal Effects 
 
to 
 
 *—< TO 
 * V 
 TO 52 
 
 in 
 
 & 
 
 O "O 
 -4 TO 
 • TO 
 TO S 
 
 in to 
 
 m 
 
 >> 
 
 TO 
 
 CTi 3 
 
 • TO3 
 TO TO 
 
 • TO 
 
 in a 
 
 TO 
 
 co co 
 
 TO iw 
 • O 
 m 
 
 TO 
 
 o. 
 
 >• 
 
 TO 
 3 
 •a 
 •o TO 
 t-t TO 
 • * 
 TO 
 
 • vi 
 
 in x 
 TO 
 H 
 
 - o 
 z 
 
 co 
 
 w 
 
 >i 
 
 w 
 
 CO 
 
 a H 
 
 H 
 CO CO 
 3 Z 
 
 > H 
 U 
 
 K te. 
 P- O 
 
 CB 
 
 i * 
 
 ■4 
 
 CO 
 
 ■TO 
 
 CO CO 
 
 a a 
 o 
 
 J 
 
 o 
 
 o 
 
 •TO 
 
 o 
 
 «N 
 
 m 3 
 p a 
 
 • TO 
 TO TO 
 
 • a 
 
 m 
 
 •H 
 
 x 
 
 TO 
 
 H 
 
 W 
 
 W 
 
 < 
 
 s \ 
 
 \ \ 
 
 \) 
 
 • u 
 JO TO 
 - O 
 in C-J 
 
 O 
 
 z 
 
 o 
 
 m 
 
 / 
 
 TO 
 
 vi vi 
 TO C C 
 
 •to s*a- 
 
 w 
 
 < 
 
 u 
 
 V’S £ V 
 
 X -S? H / 
 
 ro 
 
 cni - 
 
 'P in ^ 
 
 • * « 
 
 WEIGHT VALt 
 
 co 
 
 P 
 
 «M TO 
 
 ^ 35 
 
 • i3 
 TO TO 
 
 ‘ A! 
 in a 
 
 co 
 
 9 
 
 03 
 
 -TO 
 
 I 
 
 00 
 
 I 
 
 04 
 
 v4 
 
 I 
 
 VO 
 
 P 
 
 I 
 
 0) 
 
 CO 
 
 c 
 
 0) 
 
 a 
 
 0) 
 
 co 
 
 c 
 
 a) 
 
 o 
 
 •H 
 
 ►J 
 
 o 
 
 u 
 
 a 
 
 o 
 
 o 
 
 w 
 
 5 
 
 o 
 
 z 
 
 & 
 
 >• 
 
 
 H 
 
 
 TO 
 
 
 
 
 3 
 
 
 . 
 
 
 *o 
 
 
 3 -Vi 
 
 / 
 
 •TO TO 
 
 
 . 
 
 / 
 
 rH a) 
 
 
 
 
 • a 
 
 m 
 
 > a 
 
 
 TO 
 
 • ^ 
 
 
 / 
 
 
 pO"»H 
 
 
 
 in x 
 
 • TO 
 
 Ok 
 
 
 TO 
 
 m a 
 
 
 
 H 
 
 
 
 ^te* 
 
 
 
 _ 
 
 
 g 
 
 co 
 
 •TO 
 
 r~s0' 
 
 O 
 
 § 
 
 • TOV 
 CN 
 
 a 
 
 b- p 
 
 ° § 
 w u 
 nj co 
 p a 
 co o 
 
 •to cn . ri p, o 
 
 •TO 
 • .X 
 P 
 
 I/*) 
 
 w 
 
 p 
 
 w 
 
 *1 m 
 
 m 
 
 fM 
 
 H •■■* 
 
 oo 
 
 sr o 
 
 vo m •to m 
 
 b. J 
 O O 
 
 w w 
 
 NS CO 
 
 m a 
 co o 
 
 * 
 
 TD 
 
 <D 
 
 c 
 
 c 
 
 o 
 
 o 
 
 un 
 
 W 
 
 es 
 
 & 
 
 o 
 
 CN P © 
 
 •J g 
 
 z 
 
 CO 
 •• W 
 
 « >S, 
 
 TO 
 
 •a c 
 
 TO TO 
 TO Vi 
 
 a H 
 
 VO 
 
 / 
 
 o 
 
 •TO 
 
 O 
 
 CN 
 
 tad 
 
 O 
 
 < 
 
 00 
 
 •TO 
 
 I 
 
 CO 
 
 P 
 
 o 
 
 <L> 
 
 u-i 
 
 <P 
 
 W 
 
 rt 
 
 C 
 
 o 
 
 CO 
 
 p 
 
 <u 
 
 p 
 
 <u 
 
 4 -) 
 
 e 
 
 CO 
 
 •H 
 
 CO 
 
 >4 
 
 CO 
 p 
 r—I 
 2 
 CO 
 Cl) 
 PS 
 
 n3 
 
 a) 
 
 p 
 
 e 
 
 o 
 
 o 
 
 ih 
 
 pi 
 
 r—> 
 
 ♦—< 
 
 o 
 
 >p 
 
 Pi 
 
 co 
 
 p 
 
 o 
 
 o> 
 
 p 
 
 VM 
 
 w 
 
 CO 
 
 .c 
 
 o 
 
 to 
 
 p 
 
 0) 
 
 a, 
 
 P 
 
 GJ 
 
 P 
 
 d 
 
 w 
 
 vp 
 
 0 
 
 CO 
 
 'fit 
 
 CO 
 
 TO*5 
 
 rPl 
 
 cc 
 
 d 
 
 <3 
 
 *p 
 
 o 
 
 CO 
 
 p 
 
 CO 
 
 <D 
 
 PS 
 
 COEFFICIENT VALUE 
 
 COEFFICIENT VALUE 
 
 5-10 
 
/ 
 
 1 
 
 a 
 
 V V 
 
 Dd 
 
 S3 
 
 u 
 
 V 
 
 3J 
 
 w 
 
 ■ 
 
 
 J in 
 
 . CN 
 
 "T3 
 
 dJ 
 
 3 
 
 C 
 
 C 
 
 O 
 
 o 
 
 m 
 
 W 
 
 o 
 
 H-t 
 
 Ck 
 
 03 
 
 C 
 
 o 
 
 co 
 
 u 
 
 0) 
 
 0| 
 
 u 
 
 <u 
 
 o 
 
 CO 
 
 •H 
 
 CO 
 
 >J 
 
 03 
 
 c 
 
 < 
 
 3 
 
 co 
 
 <u 
 
 OcS 
 
 tM vO 
 
 I I I 
 
 COEFFICIENT VALUE 
 
 COEFFICIENT VALUE 
 
 5-9 
 
LiUX JL Qj. 
 
 
 to Figure 5.1.7, this figure shows that the value 
 is a U-shaped function of the number of adults in 
 
 the household. Negative values favor bus and positive values favor taxi. 
 Thus, bus is favored regardless of household size over the range 
 illustrated; however, very small and very large households prefer bus 
 
 less strongly. 
 
 H 
 
 M 
 
 to 
 
 U ot vo int 
 
 E> E> * g 
 
 © ^ 8 
 
 03 M 
 
 fl) >• u"i H 
 
 (5 
 
 PS 
 
 Pm 
 
 us* headway mean that 15 -flipu.te. 
 
 ser 
 
 . i i 
 
 01 
 
 i 
 
 0) 
 
 tfa 
 
 ilationship 
 
 
 «J W 
 
 jfee is preferred.to 
 
 fa 
 
 ja w 
 
 • o 
 © 
 
 vittua L1y no &f f ect • (zero 
 
 -I—u 
 
 J-1-L 
 
 \C> tri O ts fH O 
 
 . • * « 
 
 ' c 
 
 *S co 
 
 © 3 3 
 
 \\ 2 
 
 \ ' H W C 
 
 O W o 
 
 cn Z >* 
 
 O 
 m w 
 
 A < o « 
 
 « 
 
 ui 3 
 
 O 
 
 fM 
 
 
 onwership. . so in n n *-*, © 
 
 • • • • • * 
 
 ^s.10 - a.1" • . . ■ 
 
 : ag previou 
 
 . in w 
 
 ■ 
 
 I I 
 
 -J L. 
 
 CD 
 
 Cl 
 
 U 
 
 a> 
 
 •u 
 
 C 
 
 • u 
 A m 
 
 • o 
 ^ © 
 
 / 
 
 i>.\© m «» m 01 .h o 
 
 O 
 
 - .a- tn J 
 
 ° ° 
 
 ^ w w . 
 
 . ~ ss 
 
 V) O 
 
 X 
 
 W 
 
 3 
 
 o 
 
 CO 
 
 •H 
 
 CO 
 
 M ^ 
 
 pti 
 
 O 
 
 vD 
 
 transit 
 
 O W 
 vj* O 
 
 y 
 
 ja -t 
 * 
 
 ms 
 
 O 
 
 CN 
 
 
 vo m n n h o 
 
 -jL.,a— l.u— i —l 
 
 I l l 
 
 ^ ^ - •» 3 
 
 n °g 
 
 1 «N N W 
 
 rL Crt O 
 S 
 
 LW 
 
 CO 
 
 ■u 
 
 CO 
 
 a) 
 
 dS 
 
 .° WEIGHT VALUE 
 
 WEIGHT VALUE 
 
 age and previous usage (a.12) indicates that^most of the efiect con¬ 
 fined to previous transit users—the slope is approximately zero -for 
 those who have never used Xenia public transportation. 
 
 (a.13) This figure indicates that there is a significant differ¬ 
 ence in slope for taxi headway by sex. Negative taxi headway values are 
 
 m-r 5-10 as opposed to call-ahead 
 
(a.2) This figure indicates that the average self-estimated 
 probability of use of the 18 alternatives decreases linearly with age. 
 
 (a. 3) This figure shows the joint effects of income and age 
 on the average self-estimated probability of use which differs signifi¬ 
 cantly by both age and income. It appears that income effects are 
 lessened by increasing age. 
 
 (a.4) This figure shows that previous experience with prior 
 
 systems in Xenia significantly raises the average self-estimated 
 
 » 
 
 probability of use. 
 
 (a.5) This figure shows that the average self-estimated probability 
 of use at first decreases with ownership of up to two autos, then 
 rapidly increases. The values for three and four autos, however, are 
 based on very few observations or correlated with one of the other 
 attributes in a manner not determinable from these data. 
 
 (a.6) This figure shows that there are significant differences 
 in the average self-estimated probability of use of the 18 alternatives 
 according to whether the individual has had previous experience with 
 transit. No previous experience with Xenia public transportation 
 results in a lower curve except for "no autos". As in Figure 5.a.5, 
 the three and four auto observations are limited. 
 
 (a.7) This figure shows that the average self-estimated probability 
 of use of the 18 systems is a U-shaped function of the number of adults 
 in the household. At first, probability decreases until one reaches 
 three adults and then rapidly climbs again. There are few four- and 
 five-adult households in the sample, however. 
 
 5-11 
 
first lnci than decrease after two autos. If the respondent is 
 
 (a.8) Similar to Figure 5.1.7, this figure shows that the value 
 
 1 • *» . t ' *| I /■ 
 
 for type of service is a U-shaped function of the number of adults in 
 
 rapidly increases, indicating that cost is ignored. 
 
 the household. Negative values favor bus and positive values favor taxi. 
 
 fa. iv) rs thcit b6com6 l 0 ss 86nsitiv6 
 
 Thus, bus is favored regardless of household size over the range 
 
 illustrated; however, very small and very large households prefer bus 
 
 less strongly. 
 
 autos. Again, observations of four or more 
 
 3utos drs Drobaaly limitfid! hsncG thss^ ooiLiitis should be intGrorstfid 
 
 (a.9) This figure illustrates an interesting joint relationship 
 
 between auto ownership, previous usage of transit and bus headway. Nega- 
 
 tive slopes for dus headway mean that 15-mmute service is preferred to 
 
 _ in the time slopes by income, auto* ownership, possession of a driver’s 
 
 30-minute service. Thus, previous experience with Xenia transit service 
 
 licer se and previou transit usage. &y inspection, these effects 
 
 is associated with increasingly negative slopes as auto ownership m- 
 
 creases; however, lack of experience with transit is associated with 
 
 more 
 
 .n the 
 
 r one 
 
 
 virtually no effect (zero slope) for bus headway, regardless of auto 
 
 „sion 
 
 -S' 
 
 onwership. 
 
 :o a real "effect." In essence, 
 
 graphs show 
 
 / 1 _ mi. .-11 l , . . 
 
 (a.1C - a,12) These figures illustrate the separate and joint 
 
 interest is the finding that the slope for time is largely inde¬ 
 effects of age and previous transit usage on the slope for bus headway. 
 
 pendent cf income (a.20). 
 
 As age increases, individuals become more sensitive to headway on the 
 average; similarly, previous use of transit is associated with slightly 
 
 Tt 
 
 
 . 1 
 
 
 p'D t" 
 
 less negative slopes for headway on the average. The joint effect of 
 
 amount oi effect of each .attribute. , . , . 
 
 age and previous usage (a.12) indicates that most of the effect is con- 
 
 -. j • ;e shows yuuiiisei, miu. uxuexr iiiqivxciu£ijl& macc 
 
 fined to previous transit users—the slope is approximately zero *for 
 
 
 and old< 
 
 lv: 
 
 .most weight on cost on .the average. However figure b 
 
 those who have never used Xenia public transportation. 
 
 Dec 
 
 2 indicates that 
 maLfi.q place, jopre 
 
 (a.13) This figure indicates that there is a significant differ- 
 
 we 
 
 :ha 
 
 iut ole 
 
 ence in slope for taxi headway by sex. Negative taxi headway values are 
 
 cost than males. 
 
 associated with preferences for on^demand as opposed to call-ahead 
 
 5-12 
 
service. Thus, males appear to be more sensitive to this difference 
 than females and prefer on-demand service; indeed, females appear in¬ 
 different on the average because their value is zero. 
 
 (a.14) This figure suggests that the taxi headway slope differs 
 significantly by trip purpose. On-demand service is preferred for 
 shopping trips on the average; work trips are largely indifferent, 
 with a slight tendency toward call-ahead service. 
 
 (a.15) This figure illustrates taxi headway slope differences as 
 a function of income. Lower and upper incomes appear to be most sensi¬ 
 tive to the difference in call-ahead and on-demand. 
 
 (a.16) This figure shows that the slope for taxi headway differs 
 significantly by both income and possession of a driver's licence. For 
 licensed drivers, as income increases, there is little effect on slope; 
 and, in fact, slope is approximately zero. In the case of unlicensed 
 individuals, as income increases, slopes for taxi headway decrease 
 indicating increasing dislike for call-ahead service. 
 
 (a.17) This figure demonstrates contrasting age and sex effects 
 on the slope for cost. As females age, their slopes for cost increase 
 on the average indicating increasing sensitivity to cost. On the other 
 hand, as males get older, cost slopes increase indicating less 
 sensitivity to cost on the average. 
 
 (a.18) This figure displays the joint effect of auto ownership and 
 possession of a driver's license on the slope for cost. As auto owner¬ 
 ship increases, if the respondent is a licensed driver, cost slopes at 
 
 5-13 
 
first increase than decrease after two autos. If the respondent is 
 unlicensed, the cost slope decreases from zero to one auto and then 
 rapidly increases, indicating that cost is ignored. 
 
 (a.19) This figure suggests that females become less sensitive 
 to walking distance up to ownership of three autos, while males become 
 more sensitive up to three autos. Again, observations of four or more 
 autos are probably limited; hence, these points should be interpreted 
 
 with cautionres ults of this analysis, therefore, 
 
 (a.20 - a.22) These figures illustrate significant differences 
 in the time slopes by income, auto^ ownership, possession of a driver’s 
 
 i 
 
 license and previous transit usage. By inspection, these effects 
 appear minor and the "significance” of these results is probably due 
 
 more to low variance in the slopes of these groups (or high precision 
 
 5.1.2 Individual Forecasting System Results 
 
 of estimate) than to a real ’'effect.” In essence, the graphs show 
 
 ia final research task - which Involves.';. the- analysis of individual 
 
 that these groups have virtually zero slopes for time. Perhaps of 
 
 response functions is the construction and testing of an individual 
 
 most interest is the finding that the slope for time is largely inde¬ 
 forecasting system. This is an important and necessary step to vali- 
 
 pendent of income (a.20). 
 
 date the method. If the system appears to perform well with histor- 
 
 Figure 5.b. 
 
 ical data, it can be used to develop elasticity estimates and to 
 
 The remaining figures (b.l - b.9) refer to the "weight” or 
 
 evaluate demonstration projects both before and after ttieir imple- 
 
 amount of effect of each attribute. 
 
 mentation. Thus, if the system validates well, it can be used tb 
 
 (b.l) This figure shows that younger and older individuals place 
 
 compute measures for service level and cost, which can extend the 
 
 most weight on cost on the average. However, figure b.2 indicates that 
 
 analysis in the primary evaluation report and assist in reaching 
 
 this U-shaped relationship differs by sex. Younger males place more 
 
 conclusions regarding service and cost tradeoffs, 
 
 weight on cost than females, but older females place more weight on 
 
 cost than males Transit Service Demonstration Project*: Transit 
 
 a "'a h'T rat rar.sit ‘Services for a Snail Urban Area,. 
 
(b.3) This figure indicates that weight differs significantly 
 by both previous use of transit and possession of a driver's license. 
 Respondents without a driver's license and with a previous history of 
 transit usage place more weight on cost as do licensed drivers with no 
 previous transit usage. Previous transit usage and a license, however, 
 are associated with lower weight, as is no license and no experience 
 with transit. 
 
 (b.4) This figure shows that as household size increases from one 
 
 to two or more, the weight for walking distance drops and then remains 
 
 • * 
 
 constant. Thus, one-person households place more weight on walking 
 distance. If, however, one examines the joint effect of household size 
 
 * 0 * » * 
 
 and transit usage (b.5), it is clear that previous transit users 
 weight walking less as household size increases. On the other hand, 
 respondents with no transit experience, weight walking more as house¬ 
 hold size increases. 
 
 (b.6) This figure indicates that females place considerably more 
 weight on travel time differences than males in responding to the 18 
 alternatives. This is a function of trip purpose with a greater propor¬ 
 tion of females' trips being for non-work purposes than for males. 
 
 These selected results indicate the power of the direct value 
 assessment techniques to detect important differences in traveller 
 groups' responses to mode choice attributes. Because separate coeffi¬ 
 cients can be estimated for separate individuals, the analyses described 
 above can be performed. Although all of these analyses could have been 
 performed by means of a single expanded regression analysis in estimating 
 
 5-15 
 
the utility function with socioeconomic variables, the number of 
 
 possible terms to be investigated makes this an unattractive alter- 
 
 ni-npr than the 18 transit alternatives. Because virtually ail Aena 
 
 native. For example, each separate regression analysis involved 117 
 
 Automob 1 gs it ils reasonable to assume tnat Luis < 
 
 t.ppv-LUcllUS llo ^ "*■* 5 
 
 terms; thus a full regression analysis would require 1170 terms (10 
 
 iusted fieure represents the auto term, xhus, ii the top of the r« 
 
 coefficients times 117 terms), currently practically impossible to 
 
 sponse scale were 11, and the grand mean of the transit alternatives 
 
 solve on many (if not all) computers. The strategy adopted in this 
 
 •• f . . fas C-\ - V>JU: would L'O <?> ■ 1 1 ~ '■ l) . 
 
 research task, therefore, is recommended where the number of possible 
 
 In order to forecast, it is necessary, to assume that the individual 
 
 terms of interest is large. The results of this analysis, therefore, 
 
 is most likely to select thst alternative which receives the highest 
 
 clearly indicate the power of the technique to examine interpersonal 
 
 predicted value computed from the individual s set or coelxioients* 
 
 differences in responses to proposed service and pricing changes. 
 
 Then, by counting how many individ' re prer elect each 
 
 Likewise, groups likely to be significantly impacted or which are of 
 
 ternativd, one cai irate aggregate mode shares. The results or the 
 
 particular policy interest can be singled out for analysis. 
 
 ( - t'-Ip i £ g.iv en 
 
 5.1.2 Individual Forecasting System Results 
 
 re- 
 
 The final research task which involves the analysis of individual 
 
 response functions is the construction and testing of an individual 
 
 n 
 
 ovi q* 
 
 forecasting system. This is an important and necessary step to vali- 
 
 .977 
 
 date the method. If the system appears to perform well with histor- 
 
 r'/) j -f i-Ji a Jr nr- rmnrmfini? nnct of 3AltO 
 
 ical data, it can be used to develop elasticity estimates and to 
 
 evaluate demonstration projects both before and after their imple- 
 mentation. Thus, if the system validates well, it can be used tt> 
 compute measures for service level and cost, which can extend the 
 
 1 
 
 analysis in the primary evaluation report and assist in reaching 
 conclusions regarding service and cost tradeoffs. 
 
 variable refl 
 
 between auto and transit. However, 
 
 l ori . v .- verr.c 1 e time dxtier enc< 
 
 The Xenia, Ohio Model Transit Service Demonstration Project: Transit 
 
 and Paratransit Services for a Small Urban Area. 
 
 
 
 
 
 5-16 
 
The individual level forecasting system consists of two components: 
 (1) a set of individual coefficients for 451 of 670 respondents who 
 completed the experiment in the home interview survey; and (2) a set of 
 engineering or physical measures for type of service and headway level, 
 cost of service, walking distance to service and travel time difference 
 of transit minus auto. These measures are computed for each intra-Xenia 
 home-based work, and home-based shop trip observed in the travel pattern 
 section of the home interview survey. Hence, for each intra-Xenia trip, 
 each of the levels of the four attributes were measured using available 
 data. These measures were then substituted into each individual's set 
 of coefficients and an expected "likelihood of use" value can be gener¬ 
 ated for each alternative mode. For example, for the 1977 paratransit 
 system, one can generate attribute measures for auto, taxi on demand 
 and taxi call-ahead service by measuring network times, transit wait 
 times, etc. These measures were substituted into an individual's set 
 of coefficients and an expected value derived for each of these three 
 alternatives. 
 
 Since "auto" is the base alternative in the Xenia study against 
 which all the transit options are compared, it is necessary to establish 
 
 <4 
 
 a value for auto to use as the basis for comparison. This is accom¬ 
 plished by subtracting the intercept (or grand mean) from the top of the 
 response scale (the value "11") plus one. 1 This is done because the 
 intercept has the interpretation that it is the "average likelihood of 
 using" any of the 18 alternative transit systems. Thus one minus this 
 
 1 The value "1" is added because the base of the response scale is 1, 
 not 0. 
 
 5-17 
 
average probability yields the average probability of using some mode 
 other than the 18 transit alternatives. Because virtually all Xenia 
 respondents have automobiles, it is reasonable to assume that this ad¬ 
 justed figure represents the auto term. Thus, if the top of the re- 
 
 . . ... ; 1 .> r v- • .-i-r •. • vable 
 
 sponse scale were 11, and the grand mean of the transit alternatives 
 
 were 4, the auto value would be 8 (11 - A+ 1). 
 
 In order to forecast, it is necessary to assume that the individual 
 
 ■ 
 
 is most likely to select that alternative which receives the highest 
 
 predicted value computed from the individual’s set of coefficients. 
 
 )ti— if6leitjLonslijLp bBtwscjn the £orBCcistinfi iiesuits for tti6 work 
 
 Then, by counting how many individuals are predicted to select each al¬ 
 and t choices., A similar relationship. 
 
 ternative, one can generate aggregate mode shares. The results of the 
 
 -jt shopping trips is not available because of unavailability, of an 
 
 application of this procedure to the Xenia home interview data are given 
 
 :curate estimate of shopping trips as a proportion of total non-work 
 
 in Table 5.2. This table displays the results of applying the fore¬ 
 casting procedure to develop mode shares for the first fixed-route bus 
 
 '.ported in ioiis cf this tech 
 
 system (197A), the second fixed-route bus system (1975), the third fixed- 
 
 .on of. this result is that a simple rank—order transformation exists lq,„ 
 route bus system (1975), the 1976 shared-ride taxi system, and the 1977 
 
 paratransit system. All these options are described in Chapter 2. 
 
 lplies that the forecasting system is predicting results consistent 
 
 Table 5.2 includes four options for computing the cost of auto 
 
 -th observed behavior in Xdnia. Thus the results of the analysis of 
 
 and the travel time for transit. The intent of the experiment had 
 
 if individual coerricients can be tentatively accepted as represents— 
 
 been for respondents to consider only transit cost explicitly, as it 
 
 civs] (ii-Mrinct the Xsnis dsmonstest) on 
 
 was believed that short auto trips in a small city have a low perceived 
 
 cost. Also, it was intended for the headway variable to include all 
 
 wait time considerations, with the travel time variable reflecting 
 only in-vehicle time differences between auto and transit. However, 
 
 because it was possible that respondents viewed these attributes in a 
 
 5-18 
 
K 
 
 CN 
 
 un 
 
 w 
 
 ►J 
 
 9 
 
 H 
 
 e 
 
 <u 
 
 u 
 
 CD 
 
 > 
 
 C/3 
 
 GC 
 
 c 
 
 •H 
 
 4-1 
 
 co 
 
 03 
 
 U 
 
 (U 
 
 V- 
 
 o 
 
 U- 
 
 03 
 
 3 
 
 03 
 
 T> 
 
 C 
 
 
 
 y 
 
 
 
 
 
 
 
 
 
 
 
 Sj 
 
 
 
 C 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 —« 
 
 
 
 
 
 
 
 
 
 
 
 
 
 • 1) 
 
 H 
 
 
 
 
 
 
 
 
 
 
 
 
 
 ^ & 
 CO -H 
 
 ±J 
 
 
 
 
 
 
 
 
 
 
 
 
 
 O H 
 
 •H 
 
 
 
 
 
 
 
 
 
 
 
 
 
 U 
 
 d 
 
 
 
 
 
 
 
 
 
 
 
 
 
 rH 
 
 3 
 
 
 
 
 
 
 
 
 
 
 
 
 
 O 4) 
 U > 
 
 UH 
 
 
 
 
 
 
 
 
 
 
 
 
 
 3 <8 
 
 < V4 
 
 o 
 
 nO 
 
 ■H 
 
 ao 
 
 
 NO 
 
 r* 
 
 m 
 
 r~. 
 
 
 in 
 
 
 
 H 
 
 y 
 
 o 
 
 rH 
 
 o 
 
 
 o 
 
 O 
 
 o 
 
 o 
 
 o 
 
 O 
 
 dfl 
 
 
 u 
 
 > 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 
 • 
 
 • 
 
 • 
 
 ee. 
 
 MB 
 
 
 t4 U 
 
 hH 
 
 
 
 
 
 
 
 
 
 
 
 
 CJ «H 
 
 33 
 
 
 
 
 
 
 
 
 
 
 
 
 fH CO 
 
 3 
 
 
 
 
 
 
 
 
 
 
 
 U ■ «! 
 
 
 C 
 
 rH 
 
 
 
 
 
 
 
 
 
 
 
 z 
 
 
 CL 03 
 
 o 
 
 
 
 
 
 
 
 
 
 
 
 4-t 
 
 
 x u 
 
 3 
 
 
 
 
 
 
 
 
 
 
 
 
 
 W H 
 
 HH 
 
 
 
 
 
 
 
 
 
 
 
 s 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 ■D » 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 *-* 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 >> 4 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 73 M 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 s 
 
 
 
 
 
 
 
 
 
 
 
 u H 
 
 
 
 hh 
 
 
 
 
 
 
 
 
 
 
 
 •h 
 
 
 • 
 
 H 
 
 
 
 
 
 
 
 
 
 
 
 □ H 
 
 
 lj £ 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 CO tH 
 
 U 
 
 
 
 
 
 
 
 
 
 
 
 u 
 
 
 O H 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 U 
 
 03 
 
 
 
 
 
 
 
 
 
 
 
 "^1 dl 
 
 
 rH 
 
 3 
 
 
 
 
 
 
 
 
 
 
 
 1) *w 
 
 
 0 4) 
 
 
 
 
 
 
 
 
 
 
 
 
 X * 
 
 
 -> > 
 
 UH 
 
 
 
 
 
 
 
 
 
 
 
 -—* •■ » 
 
 
 3 03 
 
 0 
 
 m 
 
 CM 
 
 rH 
 
 
 o 
 
 Os 
 
 ao 
 
 ao 
 
 Os 
 
 r* 
 
 '-4r 
 
 
 < u 
 
 
 o 
 
 rH 
 
 rH 
 
 rH 
 
 rH 
 
 o 
 
 O 
 
 O 
 
 o 
 
 O 
 
 
 
 H 
 
 y 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 a ■ 
 
 
 u 
 
 > 
 
 
 
 
 
 
 
 
 
 
 
 
 
 ■H 4-» 
 
 ^H 
 
 
 
 
 
 
 
 
 
 
 
 
 
 y -l 
 
 33 
 
 
 
 
 
 
 
 
 
 
 
 33 
 
 
 CO 
 
 3 
 
 
 
 
 
 
 
 
 
 
 
 ■ 
 
 
 -I C 
 
 
 
 
 
 
 
 
 
 
 
 
 1) 
 
 
 a. 03 
 
 o 
 
 
 
 
 
 
 
 
 
 
 
 L>-1 VI 
 
 
 x u 
 
 X 
 
 
 
 
 
 
 
 
 
 
 
 1) 1 4 
 
 V 
 
 W H 
 
 m 
 
 
 
 
 
 
 
 
 
 
 
 
 a 
 
 
 
 
 
 
 
 
 
 
 
 
 
 ^ Jr 
 
 -3 
 
 
 
 
 
 h—^rr 
 
 
 
 
 
 
 
 
 K 
 
 m 
 
 
 
 
 
 
 
 
 
 
 
 
 
 w K 
 
 V 
 
 1 
 
 
 
 
 
 
 
 
 
 
 
 
 • : 
 
 T3 
 
 • C 
 
 
 
 
 
 
 
 
 
 
 
 
 u ■ 
 
 2 
 
 4-* HH 
 
 CO 
 
 O V 
 
 Si 
 
 
 
 
 
 
 
 
 
 
 
 i 
 
 4J 
 
 U s 
 
 h4 
 
 
 
 
 
 
 
 
 
 
 
 CB ‘ i 
 
 **H 
 
 •0 
 
 H 
 
 
 
 
 
 
 
 
 
 
 
 X 
 
 CO 
 
 O H 
 
 
 
 
 
 
 
 
 
 
 
 
 1) 
 
 c 
 
 4H 
 
 u 
 
 
 
 
 
 
 
 
 
 
 
 
 03 
 
 3 rU 
 
 hH 
 
 
 
 
 
 
 
 
 
 
 
 
 u 
 
 < 41 
 
 03 
 
 
 
 
 
 
 
 
 
 
 
 n ■ 
 
 M H 
 
 > 
 
 3 
 
 
 
 
 
 
 
 
 
 
 
 '■4H 
 
 
 u 03 
 
 — « U 
 
 UH 
 
 vO 
 
 a 
 
 ^H 
 
 
 Os 
 
 o 
 
 06 
 
 08 
 
 vO 
 
 vO 
 
 . 1 
 
 
 y H 
 
 o 
 
 o 
 
 rH 
 
 *H 
 
 »H 
 
 o 
 
 
 o 
 
 o 
 
 9 f- 
 
 
 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 D » 
 
 
 -H 4H 
 
 <u 
 
 
 
 
 
 
 
 
 
 
 
 lj 
 
 
 CL ^ 
 
 > 
 
 
 
 
 
 
 
 
 
 
 
 39 
 
 
 X 33 
 
 w c 
 
 hH 
 
 
 
 
 
 
 
 
 
 
 
 ^ 1 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 03 
 
 3 
 
 
 
 
 
 
 
 
 
 
 
 
 
 0 Li 
 
 rH 
 
 
 
 
 
 
 
 
 
 
 
 y 
 
 
 Z H 
 
 u 
 
 
 
 
 
 
 
 
 
 
 • 
 
 u M 
 
 3 * 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Q ■ 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 1 
 
 
 
 
 
 
 
 
 
 
 
 
 y 
 
 
 • X 
 
 
 
 
 
 
 
 
 
 
 
 
 X 
 
 
 i-i 'Jj 
 
 
 
 
 
 
 
 
 
 
 
 
 ■r* 
 
 
 OB 
 
 O 0) 
 
 § 
 
 
 
 
 
 
 
 
 
 
 
 ^ i-' 
 
 
 u a 
 
 **H 
 
 
 
 
 
 
 
 
 
 
 
 AJ 
 
 
 
 H 
 
 
 
 
 
 
 
 
 
 
 
 JS 
 
 
 O H 
 u 
 
 u 
 
 
 
 
 
 
 
 
 
 
 
 u B 
 
 
 3 -H 
 
 hH 
 
 
 
 
 
 
 
 
 
 
 
 o 
 
 
 •< 4) 
 
 
 
 
 
 
 
 
 
 
 
 
 4H 
 
 
 > 
 
 3 
 
 vO 
 
 o 
 
 ao 
 
 ^H 
 
 Os 
 
 o 
 
 CO 
 
 08 
 
 o 
 
 ao 
 
 
 
 *-♦ aj 
 
 
 o 
 
 rH 
 
 O 
 
 rH 
 
 o 
 
 
 o 
 
 
 O 
 
 a 
 
 
 ^ > 
 
 UH 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 • 
 
 u 
 
 
 y H 
 
 o 
 
 
 
 
 
 
 
 
 
 
 
 y 
 
 
 hH 
 
 
 
 
 
 
 
 
 
 
 
 
 UH 
 
 
 rH 4J 
 
 V 
 
 
 
 
 
 
 
 
 
 
 
 V 
 
 
 Q. -M 
 
 > 
 
 
 
 
 
 
 
 
 
 
 
 u 
 
 
 X « 
 
 «rH 
 
 
 
 
 
 
 
 
 
 
 
 
 
 ■m a 
 
 30 
 
 
 
 
 
 
 
 
 
 
 
 X 
 
 
 03 
 
 3 
 
 
 
 
 
 
 
 
 
 
 
 a 
 
 
 O Li 
 
 rH 
 
 
 
 
 
 
 
 
 
 
 
 3) 
 
 
 Z H 
 
 y 
 
 
 
 
 
 
 
 
 
 
 
 i 
 
 «-» 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 c 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 y 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 L» 
 
 
 
 
 
 
 
 
 .X 
 
 
 
 
 
 
 a 
 
 
 
 
 
 
 
 
 u 
 
 a 
 
 -X 
 
 CL 
 
 -X 
 
 CL 
 
 Cl 
 
 
 
 
 
 
 
 
 o 
 
 0 
 
 u 
 
 0 
 
 u 
 
 a 
 
 
 
 
 
 
 
 
 
 3 
 
 J3 
 
 o 
 
 jC 
 
 O 
 
 A 
 
 C 
 
 
 
 
 
 
 
 
 
 C/J 
 
 3 
 
 CO 
 
 3 
 
 CO 
 
 hH 
 
 
 
 
 H 
 
 a 
 
 -X 
 
 a. 
 
 « 
 
 
 
 
 
 
 
 
 
 
 u 
 
 0 
 
 u 
 
 0 
 
 r-s 
 
 \ ■ 
 
 x-s 
 
 
 
 ^-s 
 
 a 
 
 
 
 
 O 
 
 A 
 
 0 
 
 - 
 
 <*> 
 
 m 
 
 CM 
 
 CM 
 
 
 rH 
 
 u ■ 
 
 
 
 
 3 
 
 in 
 
 3 
 
 CO 
 
 >-✓ 
 
 1 
 
 
 
 w 
 
 
 J I 
 
 
 
 
 
 r* 
 
 NO 
 
 NO 
 
 
 m 
 
 m 
 
 sn 
 
 
 -y 
 
 1 
 
 
 
 
 
 r«* 
 
 r* 
 
 
 n* 
 
 rM 
 
 rn 
 
 
 
 
 3 1 
 
 
 
 
 <7» 
 
 os 
 
 os 
 
 Os 
 
 Os 
 
 Os 
 
 OS 
 
 Os 
 
 Os 
 
 Os 
 
 z 
 
 
 
 
 *■4 
 
 
 rH 
 
 
 rH 
 
 
 
 rH 
 
 rH 
 
 rH 
 
 LI 
 
 5-19 
 
different way than anticipated, alternative assumptions were also 
 
 rcis€ . hen be done.to 
 
 tested; namely, that the cost of auto trips was perceived at near 
 
 assess the adjusted model s validity. If successful, the results of 
 
 10 cents per mile, and that the transit travel time variable included 
 
 this kina of analysis can oe used _a priori to guide demonstration pro- 
 
 both wait and in-vehicle time. However, there does not seem to be any 
 
 jects and detect differences in potential responses and impacts for 
 
 significant systematic effect from these different measures observable 
 
 different groups of individuals as well as to evaluate the same factors 
 
 in Table 5.2. 
 
 »r demonstrations. 
 
 Because there seems to be no systematic trend in the data, the 
 
 results of the base option are graphed against the observed mode choice 
 
 data and displayed in Figure 5.2. This figure reveals a potentially 
 
 non-linear relationship between the forecasting results for the work 
 
 trip and the observed Xenia work mode choices. A similar relationship 
 
 for shopping trips is not available because of unavailability of an 
 
 accurate estimate of shopping trips as a proportion of total non-work 
 
 travel. The relationship is consistent with a number of similar results 
 
 reported in previous applications of this technique. The interpreta- 
 
 * 
 
 tion of this result is that a simple rank-order transformation exists to 
 transform predicted into observed values. Moreover, this relationship 
 implies that the forecasting system is predicting results consistent 
 with observed behavior in Xenia. Thus the results of the analysis of 
 the individual coefficients can be tentatively accepted as representa¬ 
 tive of traveller responses during the Xenia demonstration. A stronger 
 test of this assertion would be to use the experimentally derived 
 utilities as the independent variables in, say, a disaggregate logit 
 formulation to derive scale factors or other corrections. If these 
 
 5-20 
 
 5-22 
 
OBSERVED CHOICE % (WORK) 
 
 .15 
 
 
 .10 
 
 .05 
 
 0 
 
 • •. 
 
 
 J_I 
 
 .05 .10 
 
 PREDICTED CHOICE 
 
 
 FIGURE 5.2 
 
 I 
 
 . 15 
 (WORK) 
 
 Observed Versus Predicted Work Mode Split 
 
n f f, 7va&n] AT fll f PTf lflflV g J rflTTm fl TIPn to PM 
 
 parameters were reasonable, a validation exercise could then be done to 
 assess the adjusted model’s validity. If successful, the results of 
 
 *- ■ fepp fi^pHnn C* T JliniPt 1 nn A 1 1V wore fit 
 
 this kind of analysis can be used _a priori to guide demonstration pro¬ 
 jects and detect differences in potential responses and impacts for 
 
 different groups of individuals as well as to evaluate the same factors 
 during and after afe rf&rf&kt&ffis attributes (headway and type-of-service, 
 
 cost, walking distance and travel time differences) are listed in 
 Table 5.2. The. final estimated specifications of these equations are 
 
 given below: 
 
 
 vY * 3.5089 + 0.3(TS;) + 0.2(BH.) - .65(TH.) - 1.33(F.) - 1.22(F, 2 ) 
 1 x 1 1 1 i 
 
 - 0. 3200 - 0.010/) - 0.05(T.) - 0.0013(1^), 
 
 
 (5.la) 
 
 .* ■ 
 
 - 
 
 V* * 3. 832 + 0.16;as 4 > + 0.44(8^) - 0.5(TH^) 0*4 
 
 - 0.43(W i ) - 0.0180./) - 0.0b5(Tj - 0.0026(1^), 
 
 (5.1b) 
 
 P\ = 0.175 + 0.23(TS i ) + .024(BH i ) - ‘.84(1^) - .142(1^) - .105(F i 2 ) 
 
 - .032(W i ) + .006(W i 2 ) - .004(T/ - .00034(1^), 
 
 - .0370/ - .000620. ) - .006(T.)‘ - .0001(T,. 2 ), 
 
 where 
 
 is the number of alternatives (i = 1 - 18); 
 
 W S “ 
 
 and are the expected average response values for work and 
 shop, respectively; 
 
 
 
 
 (5.2a) 
 
 V s . = 0.176 + 014(TS ) + ,063(BH ) - .102(TH.) - .148(F.) 0 .165(F i 2 ), 
 
 (5.2b) 
 
 5-22 
 
 5-24 
 
5.2 Aggregate Level Response Specifications 
 
 The major effort in this portion of the research was to develop 
 alternative aggregate tradeoff expressions for the data derived from the 
 home interview survey. In order to accomplish this task objective, the 
 data were analyzed in two ways. 
 
 (1) The average "likelihood of use" value observed over all 
 respondents was calculated for e^ch of the 18 alternative transit 
 systems. These 18 averages were employed as the "expected response" 
 of the sample to each alternative and a regression equation was fit to 
 these 18 data points in exactly the same manner as done for the indi¬ 
 vidual respondents (see Section 5.1 and Equation 4.11).* Separate 
 equations were estimated for work and shop trip purposes. 
 
 (2) The number of responses observed to be greater than or equal 
 to category seven (7) on the response scale was calculated for each of 
 the 18 alternatives. This procedure was used because category seven 
 represents a self-estimated "likelihood of using" an alternative of over 
 50 percent. Hence, the total of these responses for an alternative 
 divided by the total possible responses for that alternative can be 
 interpreted as an estimated of the relative proportion of "regular 
 users" of an alternative. When normalized by dividing the number of 
 
 responses of seven or greater over all alternatives, this yields the 
 
 % 
 
 *The reason that one can use the overall averages for each of the 18 
 alternatives is because it can be shown that if a balanced sampling 
 plan was used to collect the observations, the coefficient estimates 
 will be identical to these estimated from the entire set of individual 
 data points. Because the Xenia sampling plan is almost balanced, this 
 property is expected to be approximately true. A considerable savings 
 in computer processing is therefore realized using this approach. 
 
 5-23 
 
relative proportion of "regular users" of alternative i compared to all 
 
 
 m 
 
 + 
 
 other "regular users" of all other alternatives. Separate regression 
 
 
 * 
 
 P" © 
 « (r> 
 P 
 
 equations for work and shop (see Section 5.1 and Equation 4.11) were fit 
 
 t I O H 
 
 to these 18 adjusted cell proportions. 
 
 // 
 
 / 
 
 -O' H _ 
 
 <N © 
 
 + J M 
 
 W 
 
 The data used to generate the regression equations and the ortho- 
 
 /; g 7/ 1 §5 
 
 ! o rj o 
 
 gonal coding values of the four attributes (headway and type-of-service, 
 
 baj ST I I- 
 
 y j h w . 
 
 cost, walking distance and travel time differences) are listed in 
 j I 1 « f\ ■ ! 
 
 Table 5.2. The final estimated specifications of these equations are 
 
 given below: 
 
 © 
 
 i 
 
 M 
 
 
 vO fJw 
 
 8 8 
 
 V) 
 
 10 H 
 PC M 
 ro y m 
 
 a 
 
 o 
 
 S'- 
 
 
 ft 
 
 ms c 
 
 © H 
 H »5 
 
 yj H 
 Sc M 
 
 © CO 
 
 t*i s § 
 
 O 
 
 , = 3.5089 + 0.3(TS.) + 0.2(BH.) - ,65(TH.) - 1.33(F.) - 1.22(F. ) 
 
 1 11 -i • ill 
 
 - 0.32(5^) - O.OKWjj 2 ) - 0.05(Tp - 0.0013(1^), . 
 
 w ■ 
 
 <U| 
 
 G 
 
 o 
 
 P4 
 
 to 
 
 4) 
 
 OS 
 
 rH 
 
 ' 
 
 (5 .la) 
 
 x 
 
 tiS 
 
 a{. 
 
 0! 
 
 AJ 
 
 C5 
 
 feQ 
 
 V = 3.832 + 0.16(TS.) + 0.44(BE.) - 0.5(TH.) - 1.52(F.) - 1.5(F. ) 
 
 1 1 1 1 1 ... 1 
 
 . .± 11 . - 2 . - . - 2 ( 5 . 1 b) 
 
 < 
 
 - 0. 43(W ) - 0.018(W ) - 0.065(T.) - 0.0026(T, ), 
 
 i i l i 
 
 t 
 
 i 
 
 l 
 
 // 
 
 • h 
 
 P* = 0.175 + 0.23(TS i ) +^024(BH i ) - ,.84(TH ) - .142(F) - .105(F i 2 ) 
 
 O 
 
 ITi 
 
 - .032(W,) + .006(W. 2 ) - .004(T,) - .00034(T 2 ), 
 
 11 1 g 1 // 1 
 
 to 
 
 >-< 
 
 2 
 
 (5.2a) 
 
 CO 
 
 I 
 
 I / 
 
 P. = 0.176 + ,014(TS.) + .063(BH,) - .102(TH 
 
 So 
 
 2 , 
 
 I.. 
 
 w 
 
 CO 
 
 o- 
 
 o u<< 
 
 148(F ) 0 .165(F ), 
 
 where 
 
 - .037(W,) - .00062(W. ) - .006(T.)' - .0001(T, ), 
 i l l i 
 
 if J i ij 
 
 is the number of alternatives (i = 1 - 18); 
 
 (5.2b) 
 
 V 
 
 m 
 
 CM 
 
 T w 
 
 1/ 
 
 and are the expected average response values for work and 
 shop, respectively; 
 
 Cr m 
 
 m cm 
 
 EXPECTED RESPONSE 
 
 © 
 
 cs 
 
 O 
 
 ni 
 
 it 
 
 h- 
 
 H-i 
 
 EXPECTED PROPORTION OP REGULARS 
 
w s 
 
 P i and are the unadjusted predicted proportion of 
 regular users (7 or greater) for work and 
 shop, respectively; 
 
 TS. is a dummy variable representing type of service 
 
 (-1 if taxi; +1 if bus); 
 
 BEL is a dummy variable representing headway nested 
 
 within bus service (-1 if every 15 minutes, +1 
 if every 30 minutes; 0 if taxi service); 
 
 TH. is a dummy variable representing headway nested 
 
 within taxi service (-1 if on demand, +1 if call 
 slhead, 0 if bus); 
 
 2 
 
 and F^ are fare or cost of service and the square of 
 cost, respectively; 
 
 2 
 
 and are walking distance to transit stop and the 
 square of walking distance, respectively; and 
 
 2 
 
 and are travel time difference and its square, 
 respectively. 
 
 2 2 2 
 
 F^, ^i * ^i * ^i* ^i are co< * ec * us ^ n 8 the orthogonal polynomials. 
 
 Equations 5.1a - 5.2b indicate that on the average, the Xenia 
 sample preferred buses to taxis (positive coefficient for type of 
 service); preferred 30-minute headway to 15-minute headway (positive 
 
 i 
 
 coefficient for headway), although this latter observation may be due 
 
 l 
 
 to confounding of 15-minute headway (see Section 4.3 and Table 4.6); 
 and prefer lower costs, shorter walks and lower travel time differences. 
 These relationships are displayed in Figure 5.3 for cost, walk and . 
 travel time; type of service and headway are omitted in the graphs 
 because they are dummy variables. The graphs in Figure 5.3 are 
 derived directly from the equations listed above. These graphs represent 
 the marginal response values which are the expected values of the 
 respective aggregate response measures for a single attribute, holding 
 
 5-25 
 
scale--in effec-t it rep: 
 
 / 
 
 Mil Cb 
 U / / O 
 Of / £ 
 
 V ^ 
 
 !/ 
 
 tions predict/ho one wi 
 
 v z 
 
 / 
 
 / 
 
 isiactorvf, f Individual 
 
 i 
 i 
 
 © 
 
 CO 
 
 •f 
 
 :s the average ”likelihood of using 
 
 o. / 
 
 O / > 
 
 u thd •§■ /X 
 
 I; Y % 
 
 HO " S 
 
 o 
 
 <N 
 
 4* h 3 fi 
 
 / 
 
 / 
 
 // level-o %8 tfic characte; // «Sv 
 
 / / gj / / 0 « X 
 
 X H ✓ / 
 
 J / / 
 
 £ s 17 
 
 55 f 
 
 I £ H // 
 
 / 
 
 mode u—Li_._ 
 
 L 
 
 m 
 
 8 
 
 o 
 
 n 
 
 + 
 
 X O 
 « H 
 
 ° H 
 (N O 
 + H H 
 
 > 
 
 ' Q 
 W 
 
 o _ 
 
 0 X 0 
 + to 
 
 H to 
 
 X 
 
 8 
 
 
 . The second validity and usefulness assessment procedure involves 
 
 
 users { 
 
 adi 
 
 *> a, proport »/ «u» - ve o 
 
 CO 
 
 W M 
 
 - enow y 
 
 are pre »g;rom syste^*^'''''^averages calculat ed 
 
 O H 
 H CO 
 
 52 M 
 
 U CO 
 
 o z 
 o < 
 « c2 
 
 ^ 1 w '—•f'- 1 * w 
 
 <=* o o s. vaaues usecbin this© 0 ar 6 
 
 CM • CM i-t 
 
 PO 
 
 m 
 
 co 
 a > 
 co 
 
 fi 
 
 o 
 
 at 
 
 co 
 
 <u 
 
 os 
 
 a> 
 
 T3 
 
 w 
 D 
 
 omw -| z> * t re- -l-H w 
 
 , 0 / / X „ 
 
 ll £ / / CD 
 
 resmLts in Table 5.4 are encouraging in thajt/within the separate 
 
 / 
 
 ' / / 
 
 between bodes. (Confour 
 
 -ems—bus land taxi—dhe results are rank-ordered/Aorrectly 
 
 O 
 
 m 
 
 I |j // | 
 
 CO 
 o 
 
 O Pb 
 
 f> 
 
 /> 
 
 I ?st€ esti 0.3 sepata'te -time /and cost 
 
 taxiUho* noted Pin Chapter 4 , fbfere is considerabl u 
 
 i j • i 
 
 } I “ S - eve ^ s withi, •// os, despit no conf 
 
 dment means that fjbst and time levels .£ 
 
 Ij 
 
 i _^_ « , / , 
 
 0 ) 
 ■u 
 o 
 (ft 
 
 <U 
 
 u 
 
 Oi 00 
 
 oc 
 
 o 
 
 tTi 
 
 work 
 
 u 
 
 o 
 
 M-l 
 
 O 
 
 >4 
 
 M 
 
 CO 
 
 3 
 
 C/3 
 
 co 
 
 *> 
 
 © Pb 
 
 JC ©QUS.. , . 
 
 H 
 CO 
 
 O 
 
 o 
 
 $ 
 
 re corre- 
 
 r^ason, 
 
 -—~ — .. I . — — . w 
 
 O iri © ©O O 4 " 1 
 
 • « © CX ^-t 
 
 ^ ctittns for time and cOst_ 2 a node iApxp not- ♦oor TH-ie 
 
 EXPECTED RESPONSE EXPECTED PROPORTION OF REGULARS 
 
 is a simple matter to correct in future applications by insuring that 
 
 ■J (tV 
 
 5-26 
 
other attribute levels constant. If the tradeoff process were strictly 
 additive (and Figure 4.4 suggests it is not), then one could interpret 
 these graphs as the marginal change in the aggregate response value 
 as one changes the levels of a particular attribute, while holding all 
 other attributes at their average valued 
 
 The validity and usefulness of these equations nay be assessed in 
 two ways: 
 
 1. One can assume that all individuals share these common speci¬ 
 fications and then use them to forecast choices by systematically apply¬ 
 ing them to the trip data for all intra-Xenia trips in exactly the same 
 manner as for the individual forecasting system. For example, to fore¬ 
 cast mode shares for 1977, one substitutes values for taxi service and 
 headway, cost of taxi, travel time difference and walk distance for 
 each observed trip into the appropriate equation to generate expected 
 values for call-ahead, on-demand and auto mode use. As with the 
 individual forecasting system, the individual is assigned to that mode 
 with the highest predicted valud. 
 
 Auto mode is represented by taking the top category value, adding 
 one, and subtracting the grand mean (12 minus the overall average or 
 intercept term). Auto equals the difference between the average 
 "likelihood of using" transit and the adjusted top of the response 
 
 ^"Average values were zero for all dummy variates, 50 cents for cost; 
 three blocks for walking distance and 10 minutes longer by transit 
 for travel time difference. 
 
 5-27 
 
scale—in effect it represents the average "likelihood of using" some 
 mode other than transit. When this adjustment is computed, all equa¬ 
 
 tions predict no one will use transit for anv of the five systems under 
 study. This shows, as expected, that the use of an average response 
 
 model based only on level-of-service characteristics analysis is unsat- 
 
 - j.h are aiso consistent with the 
 
 isfactorv. Individual level equations, market segments, or at least 
 
 the inclusion of socioeconomic variables must be used to model travel 
 
 ” xrn crnairges is captured py ^oth the 
 
 mode choice decisions. 
 
 niwuciiiug ei;x.orc. Anotner way of assessing 
 
 2. The second validity and usefulness assessment procedure involves 
 
 using the equations to predict the adjusted proportions of "regular" 
 
 J WiLn ODScivea 6-L<iS£lCit IBS ffOID th6 
 
 users (responses of 7 or greater) in a sketch planning format. The 
 
 adjusted shares are predicted from systemwide averages calculated for 
 each of the four attributes. The values used in this exercise are 
 listed in Table 5.A. Also displayed in Table 5.4 are data regarding 
 observed mode shares and shares predicted from equations 5.2a and 5.2b. 
 
 * ojoL-cai *»L-ila.tlge • JLilBSS V&J-UCS C3TI 
 
 The results in Table 5.4 are encouraging in that within the separate 
 
 i rom me od served patronage 
 
 systems—bus and taxi—the results are rank-ordered correctly for work 
 
 trips. Yet, the results are inconclusive between systems; this implies 
 
 re -noretriTi latne oy5,^ne elasticities 
 
 that bus and taxi values might, be different for .times and costs. This 
 
 ■•wixn-TnE xanK.-T)TTier Jattern or observed 
 
 could be tested by estimating separate time and cost functions for bus 
 
 ently under contract from the DOT Program of University Research 
 
 and taxi; however, as noted in Chapter 4, there is considerable con- 
 
 “ to answer tnls ana a number 
 
 -V. questions stemming from this research project and related work 
 
 foundment of cost and time levels within modes, despite no confoundment 
 
 ° ~(y -v ) A +A 
 
 between modes. (Confoundment means that cost and time levels are corre- 
 
 wnere 
 
 (AV +V 
 
 lated with levels of other attributes within a mode.) For this reason, 
 
 yx-witxexr xe-spunsB vaiue ana A is me attribute 
 
 value. 
 
 separate functions for time and cost by mode were not estimated. This 
 
 5~ 30 
 
 is a simple matter to correct in future applications by insuring that 
 
Aggregate Sketch Planning Results 
 
 
 T3 
 
 
 
 
 
 c 
 
 
 
 
 
 CO 
 
 
 
 
 
 m 
 
 
 CO 
 
 
 
 r- 
 
 
 CO 
 
 
 
 O' 
 
 
 
 
 
 ^H 
 
 
 "3 
 
 
 
 
 
 a) 
 
 
 
 u 
 
 
 4-1 
 
 
 
 o 
 
 
 CO 
 
 
 
 
 
 rH 
 
 
 
 
 
 3 
 
 
 
 
 
 CJ 
 
 
 
 • 
 
 
 rH 
 
 
 
 e 
 
 
 cO 
 
 
 
 aj 
 
 • 
 
 CJ 
 
 
 
 4-1 
 
 4-1 
 
 
 
 
 CO 
 
 0) 
 
 3 
 
 
 
 
 CO 
 
 C 
 
 
 
 CO 
 
 
 CO 
 
 
 
 
 CO 
 
 
 
 
 s 
 
 4-1 
 
 
 
 
 o 
 
 cO 
 
 i-- 
 
 
 
 u 
 
 T3 
 
 O' 
 
 
 
 14-1 
 
 
 
 
 • 
 
 
 rH 
 
 1 
 
 
 <u 
 
 TJ 
 
 CO 
 
 'O 
 
 
 u 
 
 s 
 
 c 
 
 n- 
 
 
 iH 
 
 CO 
 
 •H 
 
 O' 
 
 
 > 
 
 
 <4-1 
 
 r-H 
 
 
 u 
 
 0 
 
 
 
 
 V 
 
 4-1 
 
 a) 
 
 >H 
 
 
 cn 
 
 
 j= 
 
 o 
 
 
 
 
 4-1 
 
 <4-1 
 
 
 
 rH 
 
 
 
 
 3 
 
 cO 
 
 3 
 
 4J 
 
 • 
 
 CO 
 
 5 
 
 •H 
 
 c 
 
 CO 
 
 e 
 
 
 
 CO 
 
 3 
 
 0) 
 
 4-1 
 
 CO 
 
 4H 
 
 4-1 
 
 Q 
 
 CO 
 
 <U 
 
 CO 
 
 3 
 
 i 
 
 3 
 
 00 
 
 c 
 
 3 
 
 o 
 
 6 
 
 «0 
 
 o 
 
 •H 
 
 o 
 
 
 }-i 
 
 CJ 
 
 E 
 
 
 <U 
 
 cu 
 
 
 
 n 
 
 c 
 
 > 
 
 * 
 
 3 
 
 0 
 
 0 
 
 CO 
 
 o 
 
 •H 
 
 u-i 
 
 
 
 4J 
 
 
 
 CO 
 
 T> 
 
 3 
 
 
 0k 
 
 
 0) 
 
 CO 
 
 CO 
 
 Q 
 
 o 
 
 i-l 
 
 
 B 
 
 O 
 
 o 
 
 CO 
 
 CO 
 
 CU 
 
 
 »H 
 
 rH 
 
 3 
 
 4J 
 
 • A 
 
 X 
 
 3 
 
 3 
 
 CO 
 
 <U 
 
 
 CJ 
 
 — 
 
 >, 
 
 cj 
 
 rH 
 
 rH 
 
 S 
 
 CO 
 
 •H 
 
 CO 
 
 CO 
 
 
 
 > 
 
 4J 
 
 o 
 
 4J 
 
 m 
 
 M 
 
 o 
 
 
 •H 
 
 
 <u 
 
 4-1 
 
 0) 
 
 CO 
 
 O' 
 
 on 
 
 
 
 3 
 
 
 
 a) 
 
 4-1 
 
 3 
 
 l 
 
 -a 
 
 J3 
 
 
 U 
 
 sr 
 
 CO 
 
 4J 
 
 0) 
 
 4-1 
 
 r>» 
 
 0) 
 
 
 u 
 
 
 O' 
 
 XI 
 
 o 
 
 CO 
 
 o 
 
 
 < 
 
 4-1 
 
 
 4-1 
 
 
 i 
 
 
 a) 
 
 
 u 
 
 rH 
 
 CO 
 
 CO 
 
 CD 
 
 0 
 
 rH 
 
 u 
 
 <U 
 
 V-i 
 
 CM 
 
 CO 
 
 at 
 
 X 
 
 0) 
 
 
 CJ 
 
 4-1 
 
 u 
 
 <4H 
 
 CO 
 
 
 a) 
 
 
 
 3 
 
 u 
 
 in 
 
 0k 
 
 >H 
 
 00 
 
 0 
 
 
 co 
 
 
 3 
 
 4-1 
 
 c 
 
 B 
 
 c 
 
 »H 
 
 
 g 
 
 CD 
 
 § 
 
 a) 
 
 CO 
 
 3 
 
 4-1 
 
 3 
 
 > 
 
 "O 
 
 i - 4 
 
 CO 
 
 rH 
 
 
 c 
 
 o 
 
 >s 
 
 O 
 
 
 CO 
 
 CJ 
 
 CO 
 
 CJ 
 
 e 
 
 4-1 
 
 
 
 
 0) 
 
 CO 
 
 CO 
 
 sr 
 
 CO 
 
 4-1 
 
 
 •H 
 
 r^ 
 
 •H 
 
 CO 
 
 < 
 
 .3 
 
 O' 
 
 .3 
 
 >4 
 
 CJ 
 
 H 
 
 r^ 
 
 H 
 
 CO 
 
 
 * 
 
 
 •K 
 
 
 
 ■K 
 
 
 « 
 
 
 
 
 
 ■K 
 
 
 Predicted 
 Shop Share 
 
 7.55X 
 
 M 
 
 S 
 
 on 
 
 7.16Z 
 
 M 
 
 vO 
 
 On 
 
 H 
 
 to 
 
 o 
 
 ON 
 
 V 
 
 
 
 
 
 
 -3 5 
 
 
 
 
 
 
 25 
 
 S a 
 
 2 
 
 5 
 
 3 
 
 2 
 
 2 
 
 3 5 
 
 
 
 
 
 
 CO 
 
 
 
 
 
 
 V 
 
 
 
 
 
 
 V « 
 
 
 
 
 
 
 U CO 
 
 >r 
 
 S 
 
 3 
 
 8 
 
 to 
 
 IH 
 
 •g-2 
 
 • 
 
 • 
 
 
 • 
 
 
 
 
 
 
 
 
 « 
 
 
 
 
 
 
 u 
 
 M 
 
 H 
 
 H 
 
 M 
 
 
 S 
 
 O 
 
 O 
 
 N 
 
 IH 
 
 ON 
 
 «s 
 
 - 
 
 N 
 
 *> 
 
 «A 
 
 - 
 
 3 
 
 
 
 
 
 
 m 
 
 
 
 
 
 
 a 
 
 
 
 
 * 
 
 
 H a 
 
 
 
 IN 
 
 
 IN 
 
 •H (V 
 
 o 
 
 © 
 
 to 
 
 O 
 
 sT\ 
 
 > « 
 
 IH 
 
 IH 
 
 m 
 
 
 IH 
 
 £2 
 
 
 
 
 
 
 a 
 
 
 
 
 
 
 a- . 
 
 
 
 
 
 
 — « 
 
 
 
 
 to 
 
 iH 
 
 ■3 2 
 
 
 
 • 
 
 • 
 
 
 CO 
 
 
 
 
 
 
 to 
 
 
 
 
 
 
 
 1 MO — 
 
 
 
 
 
 0 a 
 
 S I 2 
 
 V o 
 
 o 
 
 o 
 
 5f 
 
 
 
 X a 
 
 jH 
 
 (A 
 
 O 
 
 
 1 X - 
 
 
 
 
 NM 
 
 
 .pi. r. </» 
 
 
 
 
 
 
 CT 
 
 
 
 
 
 K 
 
 S 
 
 *1* 8 
 
 
 8 , 
 
 8 . 
 
 8 . 
 
 1 
 
 1 d 8 
 
 to 
 
 d 8 
 
 K 2 
 
 3 3 
 
 3 2 
 
 t ; 
 
 ’i i 
 
 V 
 
 J 3 
 
 2 2 
 
 
 
 
 • 
 
 • 
 
 « 
 
 Is 
 
 U 
 
 3 9 
 
 • 
 
 9 
 
 to 
 
 kJ 
 
 l 
 
 w 
 
 3 
 
 4J 
 
 9 
 
 O 
 
 « 
 
 0 >s 
 « CO 
 
 - H 
 
 to • 
 
 « 1 ^ 
 
 41 1 % 
 
 „ 1 2 
 
 <r 1 * 
 
 >• 
 
 to u u 
 
 S -2 3 
 
 £ jj 3 
 
 to M • 
 
 to M • 
 
 
 — > to 
 
 2 1? a 
 
 2-2 3 
 
 •* to a 
 
 *• to to 
 
 5-29 
 
future sampling plans permit one to derive separate estimates for attri¬ 
 butes within modes. Thus, the observation that bus and taxi tradeoffs 
 may be different must await the correction suggested above to determine 
 
 its validity.1 observe* in behavior during the demonstration 
 
 Note that the results in Table 5.4 are also consistent with the 
 
 positive and negative changes in patronage observed during the demon¬ 
 stration. Thus, the basic pattern of changes is captured by both the 
 aggregate and individual modelling effort. Another way of assessing 
 
 validity is to examine the elasticity values which the aggregate 
 
 sge 
 
 forecasts yield and compare these with observed elasticities from the 
 
 
 errent to use. Thus, the 
 
 demonstration. 
 
 These elasticities may be computed directly from the graphical 
 
 data in Figure 5.3. For example, arc elasticities may be computed 
 
 2 
 
 for each attribute. Using the data from Figure 5.3, the arc elasticity 
 for cost can be computed for each system fare change. These values can 
 be compared with the elasticities calculated from the observed patronage 
 variation collected during the demonstration. This comparison is 
 summarized in Table 5.5. As can be noted in Table 5.5, the elasticities 
 
 mUSX cTWclXu ITUCUlv WULN i 171 VaJL xud u XlTir • y UCLdUDw 0 
 
 for bus and taxi systems agree with the rank-order pattern of observed 
 
 able amount of the choice behavior in Xenia may Be due to these exoge 
 
 Work currently under contract from the DOT Program of University Research 
 
 is being conducted at the University of Iowa to answer this and a number 
 
 of other questions stemming from this research project and related work. 
 
 Thus, some resolution of these issues should soon be available. 
 
 2 -(V r V 2 ) A x +A 2 
 
 Arc Elasticity is computed using the formula N - —r-—r* * where 
 
 \ ^2 / v v ^ 
 
 V is the observed or predicted response value and A is the attribute 
 value. 
 
 :t the r: 
 
 in 
 
 5-30 
 
 number of individuals 
 
 5-32 
 
5-31 
 
system results. As with Table 5.4, however, it appears that bus and 
 taxi values may be different. Again, it can be seen that the aggregate 
 forecast data yield results which correspond in rank order to those cal¬ 
 culated from observed changes in behavior during the demonstration 
 period. 
 
 It is important to note that it is not possible to account for 
 differences in choice behavior due to seasonal variation or to replace¬ 
 ment of damaged or destroyed vehicles in the observed Xenia data. Also, 
 a very large proportion of early fixed-route ridership was school-age 
 children, who were not included in the home interview survey, and for 
 whom a 25-cent fare might be a significant deterrent to use. Thus, the 
 direct assessment models may be forecasting actual proportional changes 
 more closely than the comparisons indicate. A reasonable interpretation 
 of the direct assessment predictions therefore is that they represent 
 the expected average behavior of a transit system with given levels of 
 the four attributes, holding natural disaster effects and seasonal ef¬ 
 fects constant. This supposition cannot be tested on the Xenia data and 
 must await future work for validation. Nonetheless, because a consider¬ 
 able amount of the choice behavior in Xenia may be due to these exoge¬ 
 nous influences, it deserves mention. Additionally, differences in mone¬ 
 tary value of the costs of the various systems were not taken account of 
 in forecasting back to each system. Thus, future work should adjust for 
 inflation when calculating cost figures for retrospective prediction. 
 
 A final adjustment should be made for data which were omitted but 
 which affect the final estimates of shares. A number of individuals 
 
 5-32 
 
responded with category "1" or "2" on the response scale to all of the 
 18 alternatives. No equation could be estimated for these respondents 
 because they have constant responses regardless of attribute levels; 
 hence, they were removed from the forecasting system. Yet it is clear 
 that these respondents are saying that they will probably never use 
 transit. Their omission from the data set is tantamount to biasing the 
 shares in favor of transit. Predicted transit shares should all be 
 adjusted downward by the proportion of such responses observed in the 
 survey. This result would imply a lower initial response to the free 
 bus, which would have probably been true if so many autos had not been 
 destroyed. Likewise, the lower predicted values for 1977 would be 
 much more consistent with observation in Xenia. 
 
 It appears that this preliminary test of the direct value assess¬ 
 ment method has yielded positive results, although many unresolved 
 questions remain. Some of these questions and potential extensions 
 are explored in Chapter 6. 
 
 5-33 
 
be calculated by counting: the number of responses to each cell that 
 were greater than seven and dividing by the total numbei offrespondents! 
 
 
 of Tel 6.1 appear to agree: relatively speaking, walking distance 
 
 rips and their effects are Similar in magnitude However, in the case 
 of shopping trips, the impact of walking distance becomes preponderant, 
 being tvn.ee 01 travel rime, almost four times that of headway, and 
 or- third more influential than cost, /This finding.parallels the re--- > 
 
 mlts of Norman (19* 0, who reported similar effects in Tuscaloosa, Ala- 
 
 1 
 
 
 >ac i, among coliege students. Table 6.1 is aiso interesting In that it 
 
 tual data: the ela 
 
 
 the regular user should be comparable to existing data (since it is 
 based largely on the numbers of regular users before and aftef a 
 , are of the same magnitude as values reported in the 
 
 evaluation. 
 
 Caution must be exercised in the it lerpretation of these figures 
 because they are indeed relative: that is, these comparative values 
 
 nr only to the given domains and ranges spanned by the attr ibutes : 
 
 • ent values’. One can say, however, that for those systems whose operation-. 
 
 • al levels fall within these ranges, one would expect (if Xenia is repre¬ 
 sentative) that these relative impacts could be expected. A positive 
 
 feature of the methodology is that it readily permits derivation and cal- 
 culation of these figures. Likewise, if one suspects that a particular 
 
 
6. IMPLICATIONS FOR DEMONSTRATION EVALUATION 
 
 6.1 Relative Importance of Cost and Service Tradeoffs in Xenia 
 
 The results of the direct value assessment approach permit a number 
 of conclusions about relative tradeoffs. In particular, one can examine 
 two measures which reflect these relative effects: 
 
 1. The proportion of total explained variation accounted for by 
 each of the attributes can be derived. These measures can be compared 
 because of the controlled, balanced nature of the sampling plan and be¬ 
 cause all respondents completed the same survey. The logic behind the 
 separability of these sources of variation is explained in Chapter 4 
 (see, e.g.. Equations 4.6, 4.8, and 4.9). 
 
 2. The elasticities of the responses with respect to the attri¬ 
 
 butes can also be estimated. Two measures are available for this com¬ 
 parison: (a) the arc elasticity of the marginal response means from 
 
 the experimental data with respect to the corresponding attribute levels, 
 and (b) the arc elasticity of the marginal proportion of respondents who 
 self-estimated themselves to be regular users. A "regular user" is 
 defined as an individual who judged that they would use any one of the 
 
 18 alternatives more than half the time. Thus, regular users can appear 
 in several different cells—technically, any response number greater 
 than seven (7) on the scale was used as an indication that the indivi¬ 
 dual would be a regular user for that system. Because the total number 
 of respondents is known, the expected proportion of regular users can 
 
 6-1 
 
be calculated by counting the number of responses to each cell that 
 were greater than seven and dividing by the total number of respondents. 
 
 The results of these manipulations are given in Table 6.1. The 
 data of Table 6.1 appear to agree; relatively speaking, walking distance 
 and cost have the greatest impacts on responses in the case of work- 
 trips and their effects are similar in magnitude. However, in the case 
 of shopping trips, the impact of walking distance becomes preponderant, 
 being twice that of travel time, almost four times that of headway, and 
 one-third more influential than cost. This finding parallels the re¬ 
 sults of Norman (1977), who reported similar effects in Tuscaloosa, Ala¬ 
 bama, among college students. Table 6.1 is also interesting in that it 
 demonstrates an interesting link to actual data: the elasticities of 
 the regular user should be comparable to existing data (since it is 
 based largely on the numbers of regular users before and after a change), 
 and indeed, are of the same magnitude as values reported in the 
 evaluation. 
 
 Caution must be exercised in the interpretation of these figures 
 because they are indeed relative: that is, these comparative values 
 pertain only to the given domains and ranges spanned by the attributes: 
 hence, any differences in domains and ranges yields potentially differ¬ 
 ent values. One can say, however, that for those systems whose operation¬ 
 al levels fall within these ranges, one would expect (if Xenia is repre¬ 
 sentative) that these relative impacts could be expected. A positive 
 feature of the methodology is that it readily permits derivation and cal¬ 
 culation of these figures. Likewise, if one suspects that a particular 
 
 6-2 
 
TABLE 6.1 
 
 Relative Importance Comparisons 
 
 Trip 
 
 Purpose 
 
 Attribute 
 
 (Factor) 
 
 Method of 
 
 Computing Influence 
 
 Explained Variance 
 
 * r 
 
 n of 
 Response 
 
 n of 
 
 Reg. User 
 
 Work 
 
 Headway 
 
 .228 
 
 N/A** 
 
 N/A 
 
 
 Cost 
 
 .310 
 
 -.25 
 
 -.41 
 
 
 Walk 
 
 .312 
 
 -.28 
 
 -.49 
 
 
 Time 
 
 .151 
 
 -.14 
 
 -.25 ' 
 
 Shop 
 
 Headway 
 
 .106 
 
 N/A 
 
 N/A 
 
 
 Cost 
 
 .294 
 
 -.22 
 
 -.44 
 
 
 Walk 
 
 .408 
 
 -.34 
 
 -.66 
 
 
 Time 
 
 .192 
 
 -.17 
 
 -.35 
 
 * n is the symbol for elasticity. 
 
 **N/A means "not applicable" because its calculation is arbitrary. 
 
 0 
 
 6-3 
 
area is a unique case, or at least different from Xenia, it is possible 
 to derive new estimates quickly and cheaply for the area in question. 
 This is discussed later in this chapter. 
 
 Thus, a major conclusion is that walking distance to the transit 
 stop plays a particularly important role in transit choice; this is 
 especially true in the case of shopping trips. By implication, there¬ 
 fore, route structure and coverage emerge as critical attributes to be 
 considered in the design and evaluation of demonstrations. This result 
 parallels the observed transit choice behavior in Xenia. That is, as 
 route coverage and fare were changed to less desirable levels for the 
 second and third fixed-route bus systems, patronage steadily declined. 
 The combined effect of both the fare and route structure change (as 
 reflected in average walking distances—see Table 5.4) was to signifi¬ 
 cantly reduce patronage from fixed-route one to two; similarly, the 
 slight increase in average walking distance caused by the route struc¬ 
 ture change for fixed-route three resulted in further patronage reduc¬ 
 tions. These reductions parallel the results predicted by both of the 
 direct response model systems (disaggregate and aggregate). 
 
 « 
 
 patronage for that individual will be zero. Because these values Re¬ 
 present constraints or boundary conditions, it should permit one to 
 solve two important problemsV (1) prediction of the feasible choice 
 • set—-only those alternatives in which all levels of all influential 
 attributes are within the interior of the feasible boundary; and (2) 
 
 6-4 
 
6.2 General Applicability of the Technique to Other Demonstration 
 
 Projects 
 
 The technique of direct response assessment explored in this re¬ 
 search project appears to have application to a variety of demonstration 
 questions which follow: 
 
 i 
 
 1. It can be used to assess the likely consequences of the project 
 prior to committing all project resources. Because one can use this approach 
 to anticipate the likely outcomes of the proposed demonstration beforehand , 
 an important application would be to conduct a "before" study to examine 
 not only the consequences of a particular alternative under considera¬ 
 tion, but to examine a range of alternatives. In this way one could 
 spot potential disappointments prior to total commitment of resources, 
 and likewise one could identify potentially positive alternatives, as 
 well. Similarly, questions regarding optimization of resources could 
 
 be addressed. It is recommended, however, that a series of "before/ 
 after" studies be conducted in which this technique is applied indepen¬ 
 dently of the demonstration decision(s) to see if its accuracy of predic¬ 
 tion is good and whether the results are consistent with system perfor¬ 
 mance. In this way, confidence in the technique can be gradually 
 achieved, although some applications and their benefits can be achieved 
 now. 
 
 2. Following from 1, it is possible to consider optimizing the 
 configurations of cost and service to maximize community participation, 
 private operator profits, etc., for given levels of resource allocation. 
 This is because one has the ability to simulate both costs of operating 
 various alternatives and some response measure of interest—patronage. 
 
 6-5 
 
for example. Thus, it is possible to conduct controlled simulation ex¬ 
 periments easily and cheaply which will permit one to model both the 
 response and cost surfaces and begin to address optimization issues. 
 
 3. If policy sensitivity of various groups to proposed alterna¬ 
 tives is an important issue, it is possible to isolate groups of interest 
 
 directly and develop models for them easily and cheaply and assess the 
 
 6.3.1 Alternative Survey Approaches 
 
 likely impacts of alternatives on these groups, as well as the general, 
 
 It has been frequently suggested informally that a direct value 
 
 randomly sampled public. Because the method should prove relatively 
 
 ent study re- controlled home interview or lab- 
 
 inexpensive with some further refinements discussed later (Section 6.3), 
 
 oratory format. This is not true and the misconception is probably the 
 
 one could update forecasts on a regular, short-term basis, if necessary. 
 
 fault of the researchers who have advocated its use: zealousness to ex— 
 
 4. Because the data are obtained entirely at the individual re¬ 
 toll, the virtues of controlled experimentation has undoubtedly conveyed 
 
 spondent level from instruments whose measurements are not in question 
 
 the impression that "controls'* extend to all facets of the study. While 
 
 (all respondents judge fixed levels of the attributes), it is possible 
 
 this is certainly desirable, it is rarely, if ever, practical. The 
 
 to begin making comparisons across demonstration projects. For example, 
 
 notion cf controls as it is used in this report : has to do mainly with 
 
 are differences in responses due to differences between urban areas or 
 
 the design of the experimental plan and how the response data can be 
 
 transit systems? 
 
 interpreted rather than extending controls .to interviewing, measurement 
 
 5. Because it is possible to estimate "threshold" effects for 
 
 and the like, _ _ - 
 
 individuals, it is possible to predict those levels above or below which 
 
 Hence, some practitioners have mistakenly fostered the impression 
 
 patronage for that individual will be zero. Because these values re- 
 
 that alternative survey formats were inappropriate. As mentioned in 
 
 present constraints or boundary conditions, it should permit one to 
 
 Section 6.2, however, the majority of applied work with the technique 
 
 solve two important problems: (1) prediction of the feasible choice 
 
 has been through the use of self-administered forms: drop-off or mail- 
 
 set—only those alternatives in which all levels of all influential 
 
 out surveys. Sampling is based upon the requirements of particular 
 
 attributes are within the interior of the feasible boundary; and (2) 
 
 studies, but will frequently require a random and/or choice-based 
 
 procedure. 
 
 \ 
 
 
 6-6 
 
 6-8 
 
determination of whether these boundary conditions should be the subject 
 of policy manipulation because their impact is shown to be different 
 for different groups such as the elderly, the wealthy, solo drivers, etc. 
 
 6. It should also be noted that very strict limits were placed 
 'on the technique in this application. In most cases, respondents can be 
 asked to judge up to 32 combinations, which typically allow the analyst 
 to incorporate seven or eight variables in the experiment. This is as 
 large or a larger number of variables than is typically included in 
 other models. Current research in Wisconsin is applying this technique 
 to a joint-trip frequency and mode choice model, so this approach can 
 handle multiple dimensions of travel behavior. The technique can handle 
 new mode and new scenario (e.g., gas unavailability) issues better than 
 many other approaches, as it is not dependent on an existing data set. 
 Thus, the technique offers the promise of great flexibility in many 
 applications. 
 
 6-7 
 
 I 
 
6.3 Issues to be Addressed in Future Applications 
 
 Although the results of this application are encouraging, a number 
 
 of issues which bear on interpretation of results and ease of use should 
 
 !> s ion • od analysis of va lance :.nat were empioyea i.n tms sau, . 
 
 be considered in future work. A number of these salient issues are con- 
 
 These include conjoint analysis procedures developed xor ramc-oraer 
 
 siaered below. 
 
 response data 
 
 nd the rank-order analogs to analysis of variance, and 
 
 , 0 . A - in c analc to regression and analysis of variance 
 
 6.3.1 Alternative Survey Approaches 
 
 fn-s. mwl rittomlal- resDonse data In all cases, these methods yield inter* 
 
 It has been frequently suggested informally that a direct value 
 
 ntel ions, j zous to those-discussed under factorial designs and the 
 
 assessment study requires a carefully controlled home interview or lab¬ 
 
 oratory format. This is pot true and the misconception is probably the 
 
 , obtained and its treatment. Nonetheless, the bas 
 
 fault of the researchers who have advocated its use: zealousness to ex- 
 
 varinc: lie of estimating,and testing effects is common^amqng all these 
 
 toll the virtues of controlled experimentation has undoubtedly conveyed 
 
 l P L CI 
 
 ment £see appendix). "The difference i 
 
 -{ n 
 
 the impression that "controls" extend to all facets of the study. While 
 
 •oble 
 
 this is certainly desirable, it is rarely, if ever, practical. The 
 
 r al 
 
 , . _ 
 
 emp. 
 
 5, 7, 9 
 
 • ^ • 
 
 , . 
 
 
 notion of controls as it is used in this report has to do mainly with 
 
 ^ ily • „a-different pe rspective. One can eat each 
 
 the design of the experimental plan and how the response data can be 
 
 pflt-ooftm ac ffprpnra value nr a croup of indifference values, 
 
 interpreted rather than extending controls to interviewing, measurement 
 
 and the like function can be derived that will maxi- 
 
 jjrr prp _ npfi Vietween categories based on the sjet of predictor 
 
 Hence, some practitioners have mistakenly fostered the impression 
 
 cj Thi c _ . mstruet compos e.indifference 
 
 that alternative survey formats were inappropriate. As mentioned m 
 
 .. ... es can .. flenr^^nd^ treat. the da.ta frpBL# new perspective. 
 
 Section 6.2, however, the majority of applied work with the technique 
 
 onstri. 
 
 has been through the use of self-administered forms: drop-off or mail- 
 
 n rpr 
 
 out surveys. Sampling is based upon the requirements of particular 
 
 studies, but will frequently require a random and/or choice-based 
 procedure. 
 
 6-8 
 
If the instructions are carefully prepared and pretested, there 
 should be little problem with self-administration. As a matter of pro¬ 
 cedure, names and contact telephone numbers should be included so that 
 anyone with problems can easily have questions answered. To date, how¬ 
 ever, there have been virtually no questions, and the follow-up sampling 
 that has been done has consistently indicated that respondents have few 
 difficulties. The singular exception is this Xenia study. The response 
 scale probably was confusing to the respondents not only because of the 
 placement of the verbal labels, but because it is difficult for respon¬ 
 dents to self-estimate probabilities or likelihoods. 
 
 This is a minor technical problem that can be rectified in several 
 ways: (a) change the response scale to one which is more easily under¬ 
 
 stood, such as degree of preference, or to a discrete choice format 
 where the respondent must select that mode that they would "most likely" 
 use; (b) change the scale to reflect a comparison between one fixed mode 
 such as auto alone and other choices; this could also be done in a 
 discrete choice format. 
 
 In any case, in the future, a variety of cheaper ways of obtaining 
 
 t 
 
 responses other than home-interview techniques should be explored, as 
 should a variety of procedures to facilitate respondents * understanding 
 of and responding to the alternatives. A moderate amount of preliminary 
 or pilot work that was not possible in the Xenia project could be very 
 valuable to future work and should be undertaken prior to any future 
 applications so that potential problems can be assessed before actual 
 data collection. 
 
 6-9 
 
non—exi 
 
 responi 
 
 in tne real world and to introduce new alternatives„ the 
 
 If one considers alternative means of obtaining response data, a 
 
 :o 
 
 nai 
 
 & not vec been assessed. These strengths complement 
 
 variety of analytical devices are available in addition to multiple re- 
 
 the weaknesses of traditional travel demand modelling approaches# 
 
 gression and analysis of variance that were employed in this study. 
 
 i r>e weaknesses of the approach are transparent. One must assume a 
 
 These include conjoint analysis procedures developed for rank-order 
 
 rank-order correspondence between predicted values and real behavior. 
 
 response data and the rank-order analogs to analysis of variance, and 
 
 Although this is apparently not an unreasonable assumption, it warrants 
 
 discrete multivariate analogs to regression and analysis of variance 
 
 j - r ; e. ce■ us, so no con" ?.si art mapping 
 
 for multinomial response data. In all cases, these methods yield inter 
 
 f • " ~ vv ; : 3-ur^s -v.- '.1 ;es : n"o re..?] choices or frequency of 
 
 pretations analogous to those discussed under factorial designs and the 
 
 choices is needed Also, appropriate transformations of physical or 
 
 theory of functional measurement (see appendix). The difference is in 
 
 engineering measures are necessary to yield the appropriate estimates of 
 
 the type of data obtained and its treatment. Nonetheless, the basic 
 
 ?st, time etc., in the models. The Xenia results suggest that this • 
 
 principle of estimating and testing effects is common among all these 
 
 may not be as much of a problem as first imagined, but additional atten- 
 
 approaches to the problem. 
 
 tion should be directed towards it. However, it is precisely in this 
 
 A final and potentially interesting approach is to employ 5, 7, 9 
 
 area that traditional travel modelling procedures are strong and 
 
 or 11 category scales commonly used in so-called attitude studies and 
 
 complementary. : <• . - •'' ' *’ ? 
 
 view their analysis from a different perspective. One can treat each 
 
 iet in Xenia, no disaggregate model could be successfully estimated. 
 
 category as an indifference value, or a group of indifference values, 
 
 Thus, the potential significance of this new approach is great if the 
 
 and ask whether a discriminant function can be derived that will maxi¬ 
 problems posed in the above paragraph can be solved. it represents a 
 
 mize the differences between categories based on the set of predictor 
 
 practical, straightforward way to obtain demand estimates even in situ- 
 
 attributes. This would permit one to construct composite indifference 
 
 ations where real-world variation is a problem. 
 
 surfaces over all respondents and treat the data from a new perspective 
 
 « 
 
 Obviously, if one has the indifference surface, one can reconstruct the 
 response or value surface of dgiHflfSSfets and £inles were employed in 
 
 tine choice simulations, it would be useful to know which of these mea- 
 
 - s or 
 
 tant que 
 
 m b( 
 
 •ther measures are most appropriate. This is an impor- 
 
 cause' one cannot tell whether the forecasting errors are 
 
 6-10 
 
A positive feature of these methods is that they are very complemen- 
 
 * 
 
 tary. If one is careful, one can obtain data that can be analyzed from 
 a variety of perspectives and establish "convergent validity," the 
 ability of all techniques to reach similar conclusions. 
 
 6.3.2 Interface and Comparison with Other Disaggregate Modelling 
 
 Strategies 
 
 A major strength of the present approach is that it can interface 
 well with existing methods. In particular, the approach can contribute 
 significant new information regarding appropriate functional forms for 
 the arguments of disaggregate models (Lerman and Louviere, 1978). Even 
 in situations where less-than-complete information is available regard¬ 
 ing all effects from a complete factorial, one can still make inferences 
 about marginal functions and usually about one or more joint effects. 
 
 Such information, while incomplete, is still better than the usual 
 assumptions regarding linearity and additivity. Likewise, a number of 
 alternative forms can be tested on the response data easily and cheaply 
 to guide estimation of a revealed behavior-based disaggregate model. 
 
 Furthermore, because one has the ability to test for differences in 
 coefficients as a function of differences among individuals on various 
 market segmentation criteria, one can assist in the specification of ex¬ 
 panded models for demand analysis by specifying a priori those terms 
 which are likely to be significant in the revealed behavior-based model. 
 Additional strength is contributed by the ability of the approach(es) 
 to consider attributes whose range of variation is either limited or 
 
 6-11 
 
non-existent in the real world and to introduce new alternatives, the 
 response to which has not yet been assessed. These strengths complement 
 the weaknesses of traditional travel demand modelling approaches. 
 
 The weaknesses of the approach are transparent. One must assume a 
 rank-order correspondence between predicted values and real behavior. 
 Although this is apparently not an unreasonable assumption, it warrants 
 additional testing before being accepted. Thus, some consistent mapping 
 of self-conceived measures or values into real choices or frequency of 
 choices is needed. Also, appropriate transformations of physical or 
 
 * 1 . >1 1 * i *1 * • 1 • , _ . 4 
 
 engineering measures are necessary to yield the appropriate estimates of 
 
 fnnrfinnfi &£■£ ozvf- e A rvm’4,' rrTn A..i im , jnf fnnn t- -? r>rvc ( i t ^ a rj* 
 
 cost, time, etc., in the models. The Xenia results suggest that this 
 may not be as much of a problem as first imagined, but additional atten¬ 
 tion should be directed towards it. However, it is precisely in this 
 
 area that traditional travel modelling procedures are strong and 
 
 complementary l'- 1 ” ii - nce large multi-attribute designs has' demonstrated 
 
 Yet in Xenia, no disaggregate model could be successfully estimated. 
 Thus, the potential significance of this new approach is great if the 
 problems posed in the above paragraph can be solved. It represents a 
 practical, straightforward.way to obtain demand estimates even in situ¬ 
 ations where real-world variation is a problem. 
 
 The study in question involved two residential preference assessment 
 
 studies conducted in Laramie, Wyoming. Both involved random samples 
 
 6.3.3 Finding Appropriate Measures for Attributes 
 
 Data are very reliable. 
 
 Although alternative measures of costs and times were employed in 
 the choice simulations, it would be useful to know which o£ these mea¬ 
 sures or which other measures are most appropriate. This is an impor- 
 tant question because one cannot tell whether the forecasting errors are 
 
due to inadequacy of the models or the measurements or both. It would 
 clearly be beneficial to separate these two error problems out in future 
 work. 
 
 6.3.4 Expanding the Number of Alternatives Evaluated 
 
 The number of attribute combinations is an important issue for 
 future work. Virtually all previous experiences with direct assessment 
 procedures have-used considerably more than 18 combinations (alterna- 
 tives or scenarios). As discussed in Chapters 4 and 5, the small number 
 of combinations, as well as some peculiarities of the Xenia transit en¬ 
 vironment, led to confounding or correlation of attributes which pre¬ 
 vented the estimation of mode-specific effects for times, costs and 
 walking distances, A slight modification in design strategy could over¬ 
 come these limitations and vastly expand modelling capability. 
 
 In particular, several different sampling plans could be used, 
 each assigned on an equal probability basis, to develop rather complete 
 information about all (or most) marginal and joint effects of the attri¬ 
 butes. For example, one survey form could capture as much bus informa¬ 
 tion as possible, a second could capture paratransit and a third could 
 capture both. By pre-specifying exactly what information is desired, it 
 is almost always possible to develop alternative designs to obtain the 
 information. It is encouraging that a single sampling plan obtained as 
 much information as was reported in Chapter 5, but slight changes could 
 yield considerably more information. 
 
 6-13 
 
6.3.5 Expanding the Number of Attribute Dimensions 
 
 It is also possible to greatly expand the number of attributes or 
 
 variables which are investigated. Research completed this past year in¬ 
 
 dicates that at least 13 attributes can be examined and reliable data 
 
 . 6.3.7. problems of.Missing Pata 
 
 obtained from respondents without requiring a very large number of com- 
 
 In fhe future it should be stressed to all respondents thaw **.*... 
 
 binations. In fact, the study at the University of Iowa is employing 
 
 - *[ 1 
 
 questions should 
 
 
 poss 
 
 ■no c 
 
 „ - 
 
 a 10-attribute sampling plan in which respondents will make 38 to 40 
 
 :a in 
 
 the Xenia survey increased the difficulty of analysis and, interpretation 
 
 judgments. A preliminary version of this instrument is included in the 
 
 d ,. r€ rru . 1 ,. cr enable amount of data, which was onlv 
 
 appendix. there will be five different instruments, each randomly 
 
 partially complete. A few simple, reminders in the survey or from, the 
 
 assigned to sample individuals; within each instrument, all marginal 
 
 - iter viewers would largely obviate^ this problem.,. * 
 
 functions (main effects) and some joint functions (interactions) are 
 
 estimable. Across all five instruments, all effects are estimable. 
 
 Thus, it can be seen that increasing the number of attributes need not 
 necessarily greatly increase the number of combinations. Moreover, 
 previous experience with large multi-attribute designs has demonstrated 
 
 that individuals readily reduce complexity by selecting only those at¬ 
 tributes salient to them and ignoring the remainder. This has the side 
 benefit of providing data regarding which attributes are considered in 
 choice and by whom. 
 
 The study in question involved two residential preference assessment 
 studies conducted in Laramie, Wyoming. Both involved random samples 
 of individuals from Laramie; one used 11 attributes, the other used 13. 
 Data are very reliable. 
 
 1 
 
6.3.6 Alternatives to Home Interviews 
 
 Other types of survey or interview formats in addition to home or 
 personal interviews should be examined. Most previous applications of 
 the technique have employed a self-administration format rather than an 
 interview. This can be accomplished in a number of ways: 
 
 1. Contact prospective respondents by telephone and explain the 
 study to them and ask them to participate and to make a commitment to 
 complete a survey which will be sent by mail. Immediately send the 
 survey with detailed and easy-to-understand instructions regarding self¬ 
 administration. Include phone numbers which respondents can call if they 
 need something explained. Previous work with this procedure has usually 
 resulted in response rates of 70 percent or higher with few refusals. 
 
 2. Contact individuals personally at home and elicit a commitment 
 
 to participate if the survey is left with the respondent to be completed. 
 
 . * 
 
 As in procedure 1, response rates are usually high and refusals low. 
 Response rates in excess of 85 percent have been achieved when the 
 contact person makes an appointment to return and collect the completed 
 survey rather than letting the respondent mail it back. 
 
 3. Mail-out and mail-back formats have also been used in the past, 
 but the usual low response rate of 25-35 percent has served to discourage 
 its use. When used, however, there has been little indication that 
 respondents had any more or less difficulty than with the other pro¬ 
 cedures. However, no complex designs which have been used in recent 
 
 6-15 
 
 
studies have been administered in this manner. Therefore, one can say 
 that moderately complex designs yielded satisfactory results, but there 
 is no available data on complex designs. 
 
 Tr t-v-ip direct response assessment survey approach can be accompj. ■ shed 
 
 6.3.7 Problems of Missing Data 
 
 able partner to more expensive travel 
 
 In the future it should be stressed to all respondents that all 
 
 modelling efforts that employ detailed home interview data on travel 
 
 questions should be completed to the extent possible. Missing data in 
 
 choices. 
 
 the Xenia survey increased the difficulty of analysis and interpretation 
 
 4, The research results suggest that the direct response, assess- 
 
 and resulted in a loss of a considerable amount of data which was only 
 
 roent approach has the potential to be developed as a-disaggregate fore- 
 
 partially complete. A few simple reminders in the survey or from the 
 
 casting system, different in some respects from current disaggregate 
 
 interviewers would largely obviate this problem. 
 
 modelling methods. It appears 
 
 A. w w L.UXO W X. w JL- w. 6 
 
 v ' ~ ;: '■ b s t.hrj Oth or ap~ 
 
 proach is weak, such as the ability to build variation into the alterna¬ 
 tives which is not present in the real world and to include variables 
 whose effects-new can be only speculated upon either, because there is 
 oirfentiy only ; constant variation in attribute levels or the attributes 
 
 are not yet influential in the choice process. 
 
 5. The, research results also suggest that one or the strong ad\ r an~ 
 
 « . • «* 
 
 tages of the direct response assessment approach is its ability to 
 uncover non-linear marginal and joint effects of choice variables. Thus 
 it offers the potential to generate a priori information regarding non- 
 linearities to planners and choice modellers. The results also demon¬ 
 strate that the method of orthogonal polynomials has great potential ^or 
 detecting non-linearities which should be explored in current methods, 
 as well as future applications, of this approach. 
 
 6-16 
 
6.4 Summary of Conclusions 
 
 The following conclusions may be reached regarding the results of 
 this pilot research: 
 
 1. The results permit the conclusion that the direct response 
 assessment approach can play a useful role in the design and evaluation 
 of demonstration projects. Because surveys can be taken before as well 
 as after a demonstration, and even during the demonstration, it is pos¬ 
 sible to guide design, monitor performance and evaluate results. A 
 number of useful measures of likely system performance can be obtained, 
 such as predictions of patronage, elasticities with respect to service, 
 and cost tradeoffs and differences in response and impacts for different 
 traveller groups. Although tentative, results are highly suggestive of 
 this potential. 
 
 2. It is clear that one can directly estimate response functions 
 for individuals and that these functions do bear some relationship to 
 their real choice behavior. Indeed, previous research has already dem¬ 
 onstrated that fact, and this study confirms it in a new context. Some 
 
 • i 
 
 I 
 
 mechanism needs to be found, however, for deciding how to eliminate some 
 individuals from the forecasting system whose equations are clearly 
 wrong, biased or error-laden. One suggestion is to eliminate all equa¬ 
 tions that do not have overall F values significant at least at the .1 
 or .2 level. 
 
 3. The research results demonstrate that utility coefficients vary 
 systematically with various market segment measures, including a number 
 of segmenting measures not previously considered. More importantly, the 
 
 6-17 
 
results—at least tentatively—permit the conclusion that an a_ priori 
 analysis of this type could suggest a variety of expanded demand models 
 that might be examined within several different modelling approaches. 
 
 If the direct response assessment survey approach can be accomplished 
 cheaply, it might prove to be a valuable partner to more expensive travel 
 modelling efforts that employ detailed home interview data on travel 
 choices. 
 
 4. The research results suggest that the direct response assess¬ 
 ment approach has the potential to be developed as a disaggregate fore¬ 
 casting system, different in some respects from current disaggregate 
 modelling methods. It appears to have strengths where the other ap¬ 
 proach is weak, such as the ability to build variation into the alterna¬ 
 tives which is not present in the real world and to include variables 
 whose effects .now can be only speculated upon either because there is 
 currently only constant variation in attribute levels or the attributes 
 are not yet influential in the choice process. 
 
 5. The research results also suggest that one of the strong advan¬ 
 tages of the direct response assessment approach is its ability to 
 uncover non-linear marginal and joint effects of choice variables. Thus, 
 it offers the potential to generate a. priori information regarding non- 
 linearities to planners and choice modellers. The results also demon¬ 
 strate that the method of orthogonal polynomials has great potential for 
 detecting non-linearities which should be explored in current methods, 
 
 as well as future applications, of this approach. 
 
 6-18 
 
6. If future research confirms the validity of the approach, it 
 should be possible to apply these methods very quickly and cheaply to 
 selected "market panels" of individuals in local areas. These panels 
 would be so chosen as to be representative of the population,and re¬ 
 sponse equations could be updated as often as desired with this panel 
 to permit one to continuously make short-range predictions, for two to 
 five years ahead, for example. A side benefit of this type of approach 
 would be the collection of a considerable amount of longitudinal data 
 on travel choices, utility coefficients, and interpersonal measures 
 which would permit one to begin to realistically assess dynamics. 
 
 6-19 
 
■u 
 
 i 
 
 k 
 
 f k 
 
 Si 
 
 h 
 
 is the vector (U r3 . U ) 
 
 *■ 5 ac ' ;ua ' choice or behavior in a nonexperimental situation 
 is the number of available alternatives 
 is the number of variables 
 
 are all functions. . ' f” 4 . 
 
 a many cases, the s may include factors for which thje correspond- 
 mg x ki s are difficult to measure or not well understood. For example, 
 
 . 
 
 21 5 are not wall known. Such factors are treated in this 
 
 as distinct, qualitative variables and are part of the x, .’s 
 
 ki 
 
 As developed above, this theory allows for responses, perceptions. 
 
 and behavior ever any set of discrete alternatives, indexed as i - 1,..., I 
 ^ mr. example, one might be interested in mode choice behavior, in which 
 
 there are different factors influencing the desirability of driving alone, 
 spooling, taki ng,transit, etc. In many situations, however, the behavior 
 ■ >Lr : £ is continuous and involves only one alternative. In these 
 ir ances » the theory often can be reduced to the case 1=1, and the i 
 subscript can be deleted. However, because in this case study we are 
 oncerned with people s choices among discrete alternatives, we will retain 
 
 the ful] notation except as noted. 
 
 ♦ 
 
 Each of these assumptions is restated more formally below, and the 
 case of additive and multiplicative utilities is explored in detail. 
 
APPENDIX A 
 
 Algebraic Theory for Direct Value Assessment 
 
 Overview of Functional Measurement 
 
 As used in this context, the term functional measurement describes an 
 approach to modelling individual behavior which is characterized by two 
 aspects. First, functional measurement is based on an explicit theory of 
 how people reach decisions. Second, it uses laboratory-like experimental 
 measurement methods to estimate models rather than observations on people’s 
 revealed preferences. 
 
 Functional measurement is based on theoretical and empirical research 
 in mathematical psychology and related fields, where there is extensive 
 
 i 
 
 support for the following assumptions: 
 
 \i ■ f ki (x ki> 
 
 (1) 
 
 U i ■ X 21 . x ki ) 
 
 (2) 
 
 B - h(U) 
 
 (3) 
 
 where: 
 
 are physically measurable attributes of the alternative under 
 study; 
 
 x^ ci are the values of as perceived by individuals; 
 
 U. is some level of response (such as numerical judgments, rankings, 
 or choices) which are observed in an experimental context for 
 alternative i. (For the purpose of this paper, we shall refer 
 to this response as utility.) 
 
 A-l 
 
u 
 
 B 
 
 is the vector (U , ...» U^) 
 
 Ki‘ 
 
 (5) 
 
 is an actual choice or behavior in a nonexperimental situation 
 
 The vector of responses (U) is connected to the observed travel behavior 
 
 i is the number of available alternatives 
 
 by means of some algebraic function. Hence, if we agree to call the 
 
 k is the number of variables 
 
 observed behavior B, then we can write: 
 
 f. 
 
 -ki 
 
 Si 
 
 h 
 
 Then 
 
 are all functions. 
 
 »stitu 
 
 In many cases, the X^.'s may include factors for which the correspond- 
 
 E * h(U) (7) 
 
 ing x^ 's are difficult to measure or not well understood. For example, 
 
 h(g(x)) 
 
 automobile safety may affect a person's choice of auto type, but its 
 
 - h;g(: (5,0;) 
 
 physical referrents are not well known. Such factors are treated in this 
 theory below as distinct, qualitative variables and are part of the x,.'s. 
 
 As developed above, this theory allows for responses, perceptions, 
 
 practical application, it is necessary ao matce explicit assumptions aDott 
 
 and behavior over any set of discrete alternatives, indexed as i = 1,..., I 
 
 l,. g, anc n aim aeauce tneir .ccnpequjenues. me re»tiits ieau to a general 
 
 For example, one might be interested in mode choice behavior, in which 
 
 •paraalp. naiysis oi Denavior wnicn nas grov.-ing emi 
 
 there are different factors influencing the desirability of driving alone, 
 
 XTi j X J / 0 anc -u 7 / • / 
 
 carpooling, taking transit, etc. In many situations, however, the behavior 
 
 ine critical component oi tnis tneory ror tne purposes oi aeveiopmg 
 
 of interest is continuous and involves only one alternative. In these 
 
 instances, the theory often can be reduced to the case 1=1, and the i 
 
 a 5 . 9 • • • j X- . / • miaiyais or variance provides a suxaignr— 
 
 x xi 2i ki 
 
 subscript can be deleted. However, because in this case study we are 
 
 concerned with people's choices among discrete alternatives, we will retain 
 
 aiLcuwuvc i.uutuinicu xuLiub . in uirs btuuy, we wixi cunsxcrer potiruie 
 
 the full notation except as noted. 
 
 r e are two key conditions involved in 
 
 Each of these assumptions is restated more formally below, and the 
 case of additive and multiplicative utilities is explored in detail. 
 
 A-4 
 
 A-2 
 
Assumption 1 
 
 For any observed travel behavior there exists a set of independent 
 factors which are functionally connected to its occurrence or the magni¬ 
 tude of its occurrence. Each factor may be either quantitative or 
 qualitative in nature. We shall denote the set of J quantitative factors 
 by S i = (S^> S 2 i » ...» Sj^) and the set of L qualitative factors by 
 Q i " (Q li\ Q 2i’ Qlf); J + L - K. The entire vector is simply 
 
 and Q^. 
 
 Assumption 2 
 
 Associated with each quantitative and qualitative factor is a corre¬ 
 sponding value or quantity of its magnitude which may be obtained by one 
 of several psychological measurement procedures. We shall let the utility 
 of this quantity provided by one or a group of subjects by (u.^, u^,..., 
 u^). Because there may be K different values or corresponding utilities 
 for each of the K factors, we may represent the utilities as u^. For¬ 
 mally, we postulate that 
 
 f 
 
 ki 
 
 (x ) 
 ki 
 
 (4) 
 
 Assumption 3 
 
 In an experimental context we observe a response to a combination of 
 
 (S , S , ...» S , Q,., ... Q ) on a psychological measurement scale. 
 
 2 i J i Li 
 
 We assume that this response measure is connected to the utility of the 
 experimental factors according to some algebraic combination rule. If we 
 agree to let represent the response to the i C ^ alternative. 
 
 A-3 
 
(5) 
 
 U. " 8 (x-m. x , ...» x ) 
 
 i - 1 - 1 2i Ki 
 
 cant, it signaxs ciiau 
 
 The vector of responses (U) is connected to the observed travel behavior 
 by means of some algebraic function. Hence, if we agree to call the 
 observed behavior B, then we can writers when plotted against e: 
 
 U values on 
 n 
 
 le abscissa. To see why, assume the linear form to be 
 
 B = h (U) (6) 
 
 correct,, and consider the effect of subtracting level 1 from level 2 of 
 
 Then by substitution, 
 
 the first factor. This yields* 
 
 
 in 
 
 B = h(U) 
 
 ( U 2 + U n^ " " "n 
 
 = h(g(x)) 
 
 1 + U 2 ) + (e 
 
 2n 
 
 e ln ) 
 
 (7) 
 
 = h (g (f (S ,Q))), 
 
 1 TT 4 . f cr - c 
 
 vt 2n 
 
 ’In 
 
 ) 
 
 This is too general a formulation for modelling purposes; in a 
 
 where l and l are the utility values assigned to levels 1 and of 
 
 practical application, it is necessary to make explicit assumptions about 
 
 factor one, respectively. Thus, the difference between the points when 
 
 f, g, and h and deduce their consequences. The results lead to a general 
 
 takes on any value' is always a constant--(except for disturbances) , 
 
 paradigm for the analysis of travel behavior which has growing empirical 
 
 hence, the graph should yield a series of parallel line». 
 
 support. (See Levin, 1976 and 1977.) 
 
 Note that this is true regardless of the forms we assume for the 
 
 The critical component of this theory for the purposes of developing 
 
 marginal relationships (i.e., U m * f. (X J and = f 2^ Z n ^’ 
 
 appropriate functional forms for travel demand models is the specification 
 
 ffect or utility (the so-called 
 
 :he average ei 
 
 demonstrated that a measure of 
 
 of U. = g (x^, x , ..., x, .). Analysis of variance provides a straight- 
 
 marginal utilities) of each of the two variables is given by their marginal 
 
 forward means of implementing the theory and diagnosing and/or testing 
 
 means. We now demonstrate that this is true for any multi¬ 
 alternative functional forms. In this study, we will consider both the 
 utility model, confirming thereby that it holds for any more restricted 
 
 linear and multiplicative cases. There are two key conditions involved in 
 
 form such as simple addition or multiplication. 
 
 the application of analysis of variance that must be satisfied: 
 
 A-4 
 
( 1 ) 
 
 The pattern of the statistical significance (or nonsignificance) 
 of the utility responses to various combinations of the inde¬ 
 pendent variables must be of a specific nature so as to permit 
 inference (diagnosis) or testing of model form; and 
 
 (2) corresponding graphical evidence must support the inference or 
 test. 
 
 Consider the hypothesis that individuals in the experiment outlined 
 in the survey will trade off time and cost of travel independently of one 
 another. That is, they combine the effects of these two variables 
 linearly. This hypothesis may be tested directly by an analysis of 
 variance. If for clarity we suppress the subscript "i" and write: 
 
 U 
 
 mn 
 
 + U 
 
 n 
 
 ^mn 
 
 ( 8 ) 
 
 where: 
 
 are the utility values assigned to the m c ^ level of the first 
 factor (say, time) in a factorial experimental plan, 
 
 2 th 
 
 are the utility values assigned to the n level of the 
 
 second factor (say, cost), 
 
 is the overall utility assigned by individuals to combinations 
 of levels of factors 1 and 2, and 
 
 is a random error term with zero mean, 
 mn 
 
 The test for independence of the two effects (time and cost) corre¬ 
 sponds to the test of the significance of the interaction "effect" of 
 1 2 
 
 U m • U£. In an analysis of variance, this is a global test for any and 
 
 all interaction effects between travel time and cost. If the interaction 
 
 1 2 
 
 is not significant (i.e., the hypothesis that and combine linearly 
 cannot be rejected), then the linear form may be accepted; if the interaction 
 
 A-5 
 
is significant, it signals that some form other than a simple linear 
 combination is appropriate. This test is accompanied by a graphical plot 
 of the interaction. If the hypothesis of linearity is correct, the data 
 
 should plot as a series of parallel lines when plotted against either 
 
 2 Umri U m. U .' £mn 
 
 or U values on the abscissa. To see why, assume the linear form to be 
 
 r tV^ y—, TT J 
 
 correct, and consider the effect of subtracting level 1 from level 2 of 
 the first factor. This yields: 
 
 U. - U. 
 2n in 
 
 ■ (U 2 + U n> * (U 1 + °n> + < e 2 n ' E ln> 
 
 (9) 
 
 are as defined in Equation 13, except to^ k which is a 
 
 - ui - U. + (e - e. ) 
 
 21 rrfttrary zero point 6n the utility 
 
 1 1 
 
 where and are the utility values assigned to levels 1 and 2 of 
 factor one, respectively. Thus, the difference between the points when U 
 
 n 
 
 takes on any value is always a constant (except for disturbances); 
 
 ive 
 
 w- 
 
 hence, the graph should yield a series of parallel lines. 
 
 Note that this is true regardless of the forms we assume for the 
 
 ’ ( -os 
 
 1 Hi 
 
 (15) 
 
 marginal relationships (i.e., U = f.(X ) and U = f_(Z )). It can be 
 
 m 1 m n ^ n 
 
 demonstrated that a measure of the average effect or utility (the so-called 
 
 marginal utilities) of each of the two variables is given by their marginal 
 
 ' 0nry source of var iation in U and U is that due 
 
 . n 
 
 :o the 
 
 means. We now demonstrate that this is true for any multi-linear 
 
 travel time and error. Thus, 
 
 utility model, confirming thereby that it holds for any more restricted 
 
 form such as simple addition or multiplication. 
 
 U 
 
 mn 
 
 uhe lwo 1 actors combine additively, or 
 
If the data were obtained from a factorial design in which factor one 
 is the row factor (subscripted m) and factor two is the column factor 
 (subscripted n), we may write the most general multi-linear form as follows 
 
 ( 10 ) 
 
 where all terms are as defined previously and the k’s are scaling constants 
 .Additional factors simply add additional one-, two-, three-, and higher-way 
 terms. Now, if we average the factorial data over the second subscript n 
 (i.e., the column factor), we would have: 
 
 U 
 
 m. 
 
 k.ui 
 
 1 m 
 
 + k 2 U n 
 
 + 
 
 J m 
 
 TJ 2 + £ 
 
 n m. 
 
 (ID 
 
 where IT is the average over the column factor. Thus, Equation 11 
 reduces to: 
 
 V * K o + K i u » + e -. (12) 
 
 where the K's are collected terms. Equation 12 demonstrates that the 
 marginal row means (in general, the marginal means for any subscript), are 
 equal to the marginal utilities up to a linear transformation. Hence, they 
 are as "good" as any other estimate measured on an interval scale. 
 
 Equation 12 is important because it demonstrates that an estimate of the 
 marginal utility for any factor may be obtained by manipulating that factor 
 as part of a factorial or fractional factorial design so long as any 
 multi-linear utility function can be assumed to have generated the data. 
 
Returning to the reduced strictly additive form, it may also be 
 demonstrated that these marginal means relate to the overall utility value 
 of cell m,n as follows: 
 
 U + U 
 m. 
 
 (13) 
 
 where U is the grand average utility (mean). Similarly, for a strictly 
 • • 
 
 multiplicative form, it may be demonstrated that the following is true: 
 
 u mn = k + - k)(U_ ffi - k)/«I _ - k)] -f ^ (14) 
 
 where all terms are as defined in Equation 13, except for k which is a 
 scaling constant which represents the arbitrary zero point on the utility 
 scale. 
 
 . Now, in the assumption that Equation 12 is true, we may write the 
 following expressions by assigning levels of cost to the rows and levels 
 of travel time to the columns: 
 
 U = f (cost ) and (15) 
 
 m. i m 
 
 o - f 2 (time ), (16) 
 
 because the only source of variation in U and U is that due to the 
 
 m • • xi 
 
 levels of cost and travel time and error. Thus, 
 
 U = f-, (cost _) + f 0 (time ) - U + e 
 mn 1 ® ^ n .. mn 
 
 (17) 
 
 if the two factors combine additively, or 
 
{[freest m ) • f 2 (time n )]/u__} + t m 
 
 (18) 
 
 u. 
 
 mn 
 
 if the factors combine multiplicatively and we assume that k in Equation 14 
 equals zero as a first hypothesis. 
 
 Following our previous logic. Equation 18 is testable statistically 
 and graphically. In particular, Equation 18 requires that all interaction 
 effects be statistically significantly different from zero and that the 
 graph of the interaction must consist of a series of diverging curves. An 
 exact statistical test may be obtained by using the marginal means as the 
 independent values, estimating k (usually done by iterative methods) and 
 performing the following linear regression: 
 
 ln(U - k) - In(U - k) + ln(U - k) - ln(U - k) (19) 
 
 rai m. .n 
 
 If Equation 18 is true, the coefficients of the cost and travel time 
 terms should not be significantly different from 1.0. 
 
 Thus, we have demonstrated than an algebraic and statistical theory 
 to diagnose and test any multi-linear utility form exists. In order to 
 derive a model in the units of the original variables (e.g., miles, 
 
 % 
 
 minutes, dollars), it is necessary to first diagnose the overall form of 
 Equation 10, and then make assumptions about the functions in Equations 17 
 and 18 (or a more general form given by Equation 10, if appropriate). 
 
 A-9 
 
to one another and therefore linearizable and testable. The problem 
 arises, there ore, in non-balanced data collection schemes such as the 
 design in the Xen h is nearly orthego 
 
 "revealed behavior data such as that obtained in the detail travel or 
 trip data in the survey. That is, it is usually the case that travel 
 tiroes and costs are correlated in real data and there is not a balanced 
 sampling of high costs and high times, high costs and low times, low 
 costs and high times, and low costs and low times, with equal numbers of 
 observations in each of these conditions. It would be desirable to be 
 able to estimate a non-linear marginal effect for both time and cost, 
 independently of the linear effect of time and cost, as well as any 
 interactions between the two factors, as they may affect the dependent 
 variable. While it cannot remove, the collinearity between the linear 
 
 3St, the method o 
 
 • •. . 
 remove all correlation between time and time squared and between cost and 
 
 cost squared and reduce the correlations between the various cross- 
 T 'ducts and those other effects, as well. In particular, the method of 
 orthogonal polynomials has the property’ that it provides an exact test 
 ^- e highest-order effect . Thus, if a Squared term is appropriate in 
 the jnodel, this method will find it with a high degree of confidence. 
 
 Lower terms—for example, linear terms—are estimated with less pre¬ 
 cision, because, as we shall demonstrate, the higher terms are really com¬ 
 posites ,of the lower terms; hence, the presence of a significant squared 
 term implies that there is a real quadratic effect, yet if the linear term 
 
 B-2 
 
APPENDIX B 
 
 Purpose and Method of Construction of Orthogonal 
 Polynomial Contrasts 
 
 The purpose of orthogonal polynomial contrasts is to permit one 
 
 \ 
 
 to make inferences regarding polynomial terms in an expanded linear 
 equation, as well as to make inferences regarding the significance of 
 
 cross-product (interactions) terms. That is, it is not generally 
 
 2 2 2 2 2 2 
 possible to test x^, x^, , x^.x^, x^ .X 2 » x j * x 2 » • x 2 » 
 
 each as separate terms in a model which includes all of- these effects 
 
 because of the collinearity problems which result. Although it is 
 
 frequently alledged that collinearity is a "sample size" problem, this 
 
 is at best misleading. Rather, collinearity should properly be viewed 
 
 as a problem of confounding of effects, such that one cannot really say 
 
 whether one is estimating x^ or some other term that might be highly 
 
 correlated with x^. Hence, for various reasons, not the least of which 
 
 is interpretability, collinearity should be reduced to as minimal a 
 
 level as possible. The method of orthogonal polynomials accomplishes 
 
 i 
 
 this in a straightforward manner, but at the sacrifice of "direct" 
 interpretation in the original units of measurement. However, as we shall 
 note, it is always possible to return to the original units in a straight¬ 
 forward manner and to interpret some of the effects directly, as well. 
 
 In the situation in which one were to have a balanced factorial 
 sampling plan for obtaining observations, the method of orthogonal poly¬ 
 nomials guarantees that all possible testable effects will be orthogonal 
 
 B-l 
 

 
 The construction of the functions f (>:- ) is accomplished by application 
 
 to one another and therefore linearizable and testable. The problem 
 
 the liru ciX and quadratic cerins y . 
 
 arises, therefore, in non-balanced data collection schemes such as the 
 
 r ■£ f , \ — f - — ' . • \ 
 
 design in the Xenia survey (which is nearly orthogonal) and the usual 
 
 "revealed behavior" data such as that obtained in the detail travel or 
 
 \ v 2 _ /_, /y\ _ fy_ x) (LA/EB) 
 
 trip data in the survey. That is, it is usually the case that travel 
 
 times and costs are correlated in real data and there is not a balanced 
 sampling of high costs and high times, high costs and low times, low 
 costs and high times, and low costs and low times, with equal numbers of 
 observations in each of these conditions. It would be desirable to be 
 able to estimate a non-linear marginal effect for both time and cost, 
 independently of the linear effect of time and cost, as well as any 
 
 
 interactions between the two factors, as they may affect the dependent 
 
 
 variable. While it cannot remove the collinearity between the linear 
 
 c „ j 1 . . uariafA fhflt' hAS thfi ya of + tllfc • JOBcUIj When all 
 
 components of time and cost, the method of orthogonal polynomials can 
 
 • • /h -". v, . ' . -■ • , 3 ' ‘. 
 
 remove all correlation between time and time squared and between cost and 
 
 , , , , the zero order effect or the 
 
 cost squared and reduce the correlations between the various cross- 
 
 ' Vv-^r - r-r- 1 ‘I TT JV^V. 1 tiplVitlfi 
 
 products and those other effects, as well. In particular, the method of 
 
 . . _,•« jujpet-inru. •foraaine the, mean of that term and 
 
 orthogonal polynomials has the property that it provides an exact test 
 
 for the highest-order effect . Thus, if a squared term is appropriate in 
 
 ates 
 
 aiding 
 
 or b 1 
 
 the model, this method will find it with a high degree of confidence. 
 Lower terms—for example, linear terms—are estimated with less pre- 
 
 the linear 
 
 cision, because, as we shall demonstrate, the higher terms are really com- 
 
 ta used 
 
 estimate 
 
 posites of the lower terms; hence, the presence of a significant squared 
 
 :hese forms and 
 
 ns t 
 
 the ease of teturning 
 
 term implies that there is a real quadratic effect, yet if the linear term 
 
 B-4 
 
 B-2 
 
is not significant in this case it does not mean that it is not signifi¬ 
 cant. This is because the quadratic effect contains linear terms which 
 
 would appear in the expanded equation. On the other hand, if both linear 
 and quadratic terms were non-significant, this would imply non¬ 
 significance for the entire effect. Thus, interpretation is somewhat 
 cumbersome in that the entire term must be expanded to know its form, but 
 this seems like little sacrifice in order to be able to estimate these 
 higher order effects. 
 
 The particular approach that we adopted in this research project v/as 
 patterned after an article by D.S. Robson ( Biometrics , June, 1959, 
 187-191: "A Simple Method for Constructing Orthogonal Polynomials When 
 
 the Independent Variable is Unequally Spaced"). As Robson demonstrates, 
 a least-squares regression equation of the form: 
 
 may be expressed in the following form: 
 
 n > r 
 
 where f^(x.) is a polynomial of degree i in x and where f q, f^...f r 
 
 are normal orthogonal functions; i.e., where 
 
 n 
 
 0 if i j i f • 
 
 l 
 
 1 if i - i' 
 
 j-1 
 
 B-3 
 
The construction of the functions f (x^) is accomplished by application 
 
 he high!st order terms, but the lover order 
 
 of the following formulae (for the linear and quadratic terms): 
 
 Linear Term f T (x ) = (x - x) 
 
 L j j 
 
 dome Interview Survey are given in Table 4.3. 
 
 Quadratic Term f^Cx^) = x^ 2 - l/n(ZC) - (x^ - x)(ZA/ZB) 
 
 where: 
 
 ZA = Zx„ 2 (x. - x) 
 i 1 
 
 i - 2 
 
 ZB = Z(x i - x) 
 i 
 
 ZC = Zx ± 2 
 i 
 
 Note that these formulae may also be applied to dummy variates, yielding 
 a transformed dummy variate that has the values of + the mean. When all 
 variates are transformed to orthogonal form in this manner by centering 
 about their means, the intercept estimates the zero order effect or the 
 mean of Y. Higher polynomials may be derived by simply multiplying the 
 linear term times the term in question, forming the mean of that term and 
 subtracting it from each observation. Thus, the cubic is formed by 
 multiplication of the linear and quadratic terms or by cubing the linear 
 term and subtracting its mean from every observation. 
 
 This procedure may be readily applied to travel data used to estimate 
 disaggregate choice models by transforming all variates to these forms and 
 expanding the model as appropriate. To demonstrate the ease of returning 
 
 B-4 
 
to the original units, we show the two forms of the work purpose model 
 estimated on the cell averages from the survey. Recall the original 
 orthogonal form of the model is: 
 
 U* - 3.5089 + 0.3(15^ + 0.2(8^) - 0.65(TH ) - 1.33(F ± ) 
 
 - 1.22(F 2 ) - 0.32(W^) - 0.01(W. 2 ) -0.05(T.)- 0.0013(T. 2 ) 
 
 By expanding the model form with the appropriate meai\s, we derive the 
 following expression: 
 
 U” * 5.459 + 0.3(TS i > + 0.2(6^) - 0.65(TH i > - 0.111(F i ) 
 
 -1.22(F i 2 ) - 0.253(W 1 ) - 0.01(W i 2) - 0.0470^) - O.OOIBC^ 2 ) 
 
 Note that the dummy variates TS, BH, and TH are unaffected because they 
 may be coded as convenient, and the quadratic terms have the same coeffi¬ 
 cients as above. As mentioned earlier, the quadratic terms are exact, but 
 the linear terms must be adjusted to take account of the formulae at the 
 top of the immediately preceeding page. The dependence of the linear 
 term is essentially the square of the linear term, subtracting the mean 
 of the quadratically transformed observation from each observation. Hence, 
 we have: 
 
 (x - x) 2 - ([x 1 - x]/n), 
 
 and there is clearly an unbiased x^ term, but x^ also contains -2x(x^) 
 which is involved in this quadratic term. This algebra clearly shows that 
 
 B-5 
 
one gets a ,: real M test on the highest order terms, but the lower order 
 terms must be reconstructed. 
 
 The orthogonal codes and the cell means for the design in the Xenia 
 Home Interview Survey are given in Table 4.3. 
 
 B-6 
 
APPENDIX C 
 
 Xenia Home Interview Survey 
 
 C-l 
 
XENIA HOME INTERVIEW SURVEY 
 SCREENING INSTRUCTIONS 
 
 There will V<- two phases to the survey: 
 
 X - Telephone Screening 
 
 XI - Borne interview 
 
 1. Cen-ral Instructions - Telephone Screening 
 
 1. The telephone screening will be conducted between 5:30 p.m. and 9:30 p.m. on veeknlghts 
 only, before the home Interviews begin. 
 
 2. Bath Interviewer will bsve a list of number* to call, this list should be transcribed 
 onto the Telephone Screening Term, If the number reached cannot be asked the 
 
 scre ening question the interviewer should record the reason by marking eh: . paces 
 provided; or ;>wtaple, If the number reached is a business phone, or the line is 
 busy, the Interviewer should place ar "X” ox checkmark under the eclat® headed 
 
 phone or ’busy. tse line ie busy or there is no answer, this number 
 
 should be placed on * "call back" list and called again later. 
 
 3. The Interviewer ahould speak to ar adult, and ask if the household’s residence Is 
 vltb.,n th<- Xenia city lircita. If the household does not reside within the eity limits, 
 tb i-Titervi .»ver ahoui d Indicate this in the appropriate space and terminate the 
 interview. 
 
 *' If the household does reside within the Xenia city limits, the interviewer should ask 
 the following three key questions. 
 
 - Were you a resident of Xenia when the tornado struck in April. 1974? 
 
 Has a member of youx household used the X-Line or Flexicat public transportation 
 system in the past year? 
 
 - Bow maay automobiles does your household own or operate? 
 
 
 e interviewer should record the answers to the three key questions on the Telephone 
 Sciesnipg Form by marking the appropriate box for each of the three questions asked. 
 
 After toe ar.swers to the three questions have been recorded, th* interviewer ehould refer 
 to the Category matrix at the bottom of the form, which Indicates-the appropriate 
 categorization of selected combinations of answers to the three questions, Bach Set of 
 will/it into one of the six categories indicated in the right hand column 
 oi this matrix. If to* respondent's answers place him or her ie a category that has 
 not reached ire target sample size, attempt to arrange an appointment, preferably 
 doling the evening. About one-half hour will be required, and whenever possible, there 
 s,.ouxd be at least .two respondents per household. The Fiexicab user, if any, in the 
 household, must be present. Strongly request the presence of the head of household, 
 although it is not mandatory. 
 
 >nt, after being asked, refuses to be interviewed at home, mark ’'refused" in 
 * space for address. 
 
 If respondent falls into category which already has a sufficient number of interviews 
 scheduled, ask questions on Form X and terminate Interview. 
 
 5. Quotas for categories IV, V, VI listed below do not need to be met; any shortfall in 
 categories IV, V, or VI must be compensated for by additional scheduled interviews in 
 I, 11, or III, as indicated below. 
 
 Quotas for Scheduled Interviews in Each Category Shown on Form Tare ; 
 
 1. 94 (or nore) IV. 24 (or less): Categories I and IV aust together total US . 
 
 II- 141 (or more V. 35 (or less): Categories II snd V must together total 176. 
 
 _ 
 
 C-2, 
 
FIGURE 3.1: Results of Phone Screening and Interviewing. 1977 
 
 C-l*- 
 
XENIA HOME 
 
 INTERVIEW SURVEY 
 
 SCREENING 
 
 INSTRUCTIONS 
 
 : survey: 
 
 
 There will be two phases to the survey 
 
 I - Telephone Screening 
 
 II - Home interview 
 
 I. General Instructions - Telephone Screening 
 
 1. The telephone screening will be conducted between 5:30 p.m. and 9:30 p.m. on weeknights 
 only, before the home interviews begin. 
 
 ; fi • . • I 
 
 2. Each interviewer will have a liat of numbers to cell, this list should be transcribed 
 onto the Telephone Screening Form. If the number reached cannot be asked the 
 screening question, the interviewer should record the reason by marking the spaces 
 provided; for example, if the number reached is a business phone, or the line is 
 busy, the interviewer should place an "I" or checkmark under the column headed 
 “business phone" or "busy." If the line is busy or there is no answer, this number 
 should be placed on a "call back" liat and called again later. 
 
 C 'W • 
 
 t 
 
 cr 
 
 •a* 'j : 
 
 The Interviewer should speak to en adult, and ask If the household's residence is 
 within the Xenia city limits. Xf the household does not reside within the city limits, 
 the interviewer should indicate this in the appropriate space and terminate the 
 interview. 
 
 5. 
 
 Zf the household does reside within the Xenia city limits, the interviewer should ask 
 the following three key questions. 
 
 - Were you a resident of Xenia when the tornado struck in April, 1974? 
 
 - Has a member of your household used the X-Line or Flexicab public transportation 
 system in the past year? 
 
 - How many automobiles does your household own or operate? 
 
 The interviewer should record the answers to the three key questions on the Telephone 
 Screening Form by marking the appropriate box for each of the three questions asked. 
 
 After the answers to the three questions have been recorded, the interviewer should refer 
 to the "Category" matrix at the bottom of the form, which indicates the appropriate 
 categorization of selected combinations of answers to the three questions. Each set of 
 three answers will fit into one of the six categories indicated in the right hand column 
 of this matrix. If toe respondent's BDSwers place him or her in a category that has 
 not reached its target sample size, attempt to arrange an appointment, preferably 
 during the evening. About one-half hour will be required, and whenever, possible, there 
 should be at least two respondents per household. The Flexicab user, if any, in the 
 household, must be present. Strongly request the presence of the head of household, 
 although it is not mandatory. 
 
 If respondent, after being asked, refuses to be Interviewed at home, mark "refused" In 
 space for address. 
 
 ill-? rn —i. "r“T t 
 
 lf respondent falls into category which already has a sufficient number of interviews 
 scheduled, ask questions on Form X and terminate Interview. 
 
 Quotas for categories IV, V, VI listed below do not need to be met; any shortfall in 
 categories IV, V, or VI must be compensated for by additional scheduled interviews In 
 I, II, or III, as indicated below. 
 
 
 
 I. 94 (or more) IV. 24 (or less): ( 
 
 II. 141 (or more V. 35 (or less): ( 
 
 III. 188 (or more VI. 47 (or less): ( 
 
 TT*0 - ok \ 
 
 
 
 
 
 
 aOp/j»s 3 ok j 
 
 
 c-: 
 
 L 
 
 euoud SS9UT&TII 
 
 
 
 
 »ucq<j 
 
6. Phone numbers will be selected by rsndom digit dialing using a standard random number 
 table, except that users of the Flexlcab service will be found from a phone list drawn 
 from system records to the extent possible. 
 
 7. All screening Is to be carried out from a phone bank at _ 
 
 8. All information from this survey Is to be held in the srrlctest confidence. 
 
 C-3 
 
There will 
 
 04 
 
Form X 
 
 Termination Questions for Telephone 
 
 Screening 
 
 (To be used when a respondent's category 
 quota has been filled.) 
 
 XI. 
 
 SKIP XI IF RESPONDENT IS IN EITHER CATEGORY I OR IV. 
 
 Are you aware of the Flexicab/X-Line service currently 
 being operated by the Xenia Cab Company? 
 
 □ YES □ NO 
 
 TERMINATE CALL IF RESPONDENT IS IN CATEGORIES 
 
 V, OR VI 
 
 X2. 
 
 Were you aware of the Free Bus service after the tornado? 
 
 IF NO, TERMINATE CALL. 
 
 IF NO, TERMINATE CALL. 
 
 me 
 
 | | "It stopped being free." 
 
 | | "A car became available." 
 
 | | "No further need." 
 
 | [ "It was unreliable." 
 
 "It was too slow." 
 
 "It was inconvenient." 
 
 | | Other _ 
 
 "Thank you very much for your cooperation." 
 
 C-5 
 
XENIA HOME INTERVIEW SURVEY 
 INTERVIEWING INSTRUCTIONS 
 
 There will be two phases to the survey: 
 
 I, - Telephone Screening 
 
 II. - Home Interviewing 
 
 I . General Instructions - Home Interview 
 
 1. Interviewers should allow a half hour between interviews for travel time, delays, etc. 
 Thus, interviews should be scheduled at one hour intervals for any single interview. 
 
 Clock time (to nearest minute) should be filled in at points specified on the survey 
 forms during pretest, first day of interviewing, and on a spot basis afterwards. 
 
 2. The interviewer is to obtain trip information on the trips made the day of the inter¬ 
 view (i.e., if the interview takes place on Wednesday, obtain Wednesday only trips). 
 
 If the respondents are aware of trips that will be mxde during the evening, they 
 should also be included. 
 
 3. Form A should be administered to the more senior of the two household members; then 
 continue with forms B, C and D for that person. After starting them on form D, leave 
 them to complete it on their own, and begin with form B with the second person, con¬ 
 tinuing through forms C and D to complete the interview. 
 
 11 . Specific Questions - Home Interview 
 
 Clarification is provided for some questions that may require interpretation: 
 
 (General) If respondent cannot answer a certain question, write "DK" (don't know) as 
 response in the first blank provided for a response. 
 
 Number of Ve hicles Owned or Operated by Family 
 
 Ask for the total number of vehicles that are available to the household. This can 
 include all types of vehicles that are owned by the people including cars, trucks, 
 vans, etc., but not motorcycles. Also included in this category would be company 
 vehicles that are driven hose.- Inoperable,-antique, or /ju^k cars should not be counted. 
 
 A6. We are interested in knowing how many cars were destroyed or severely 
 damaged to the extent that they could no longer be driven. Minor 
 damages should not be counted. 
 
 AIO. Although a selection of responses are listed for question Alt), the interviewer 
 should not read these answers aloud ; the respondent should answer in his/her 
 own words , and the interviewer must determine if the respondent’s answer fits 
 any of the listed alternative responses. If the respondent's answer does not 
 correspond to one of the alternative responses listed, or if the interviewer is 
 unsure how to categorize the response, the Interviewer should record the 
 respondent's answer in the ''other” category. Just as the respondent says it, to 
 the extent possible. 
 
 B2. It is permissible on this question for the interviewer to prompt respondent 
 by asking how many trips were taken for a specific purpose, e.g. "how many 
 trips were for going to or from work?" as written on questionnaire. Both 
 respondent and interviewer may be unsure how to categorize a particular type 
 of trip. In this case, the interviewer should write the number of trips taken 
 of this type and a short description of the trips in the blank space alongside 
 the categorized responses, e.g.£3}- "pick up groceries." 
 
 C-6 
 
BA. In asking if che respondent has ever used the "Flexiceb/X - line nr Free bus", 
 we really want to know if the person ha* used ray of the City of r.n-:» public 
 transit services since the tornado. 
 
 BIO. Once egaia the interviewer is called upon to categorise an "open-ended" 
 
 response. The respondent should describe his/her "major activity" in his/her 
 own words, end eh* interviewer must determine if the response fits any of the 
 listed response categories. The interviewer any prompt the respondent if. 
 necessary. For example, if the respondent says "work" is hls/har major 
 activity, the interviewer may ask "Bow many hours par week do you work" to deter¬ 
 mine if employment la full- or part-time. 
 
 Bll. The interviewer should read through the question and the alternative responses 
 
 aloud completely at least ones, to famillarlsa the respondent with the choice of 
 answers. Then the Interviewer should re-read the question end each response aloud 
 slowly, giving tha respondeat e chance to designate his/her choice when he/she 
 feels reedy to do so. 
 
 B12 - B15. The incervlaver should reed the question end response alternatives aloud at 
 laaat ones. 
 
 B17. If two or mors persons in the household are taking the survey, the income 
 
 question should be asked of only one ; this person should be the heed of household, 
 if possible, or s responsible sdult, 1. *. the more "settler”" person. 
 
 C. Trip Sheet 
 
 The Trip Sheet is co be used to obtain ell che required information on the daily trips 
 of the two members of che household interviewed. A separate form-should.be used for 
 each household member Interviewed sad the Interview number inserted. Tech these 
 should also be nusibered conaecuclvely if more chan one cheat Is required. 
 
 After completing preliminary information about an individual respondent (form's),‘the 
 following information should be asked about each trip made by thee person coday. 3uc 
 first, what is s trip? 
 
 The most important eleasec of che survey is the trip. Therefore, it is vital that 
 each interviewer have a clear understanding of the exact definition of a trip. 
 
 A crip is defined as eh* one-way movement of a person by automobile or transit from 
 
 one primary activity to another chat is more chan three (3) blocks sway. 
 
 (E7AKPI.ES: - A person driving from borne co work directly le considered to be one trip. 
 
 - A person who drives to pickup £L other people before travelling to work 
 is considered co have mad* only on* trip. In other words, hie primary 
 activities were home and work; we ere not Interested in hie intermediate 
 etops. 
 
 - K person driving from work co home but scope to buy some groceries Is 
 
 considered co have mad* two trips: work co shop and shop to home. 
 
 - Trips mad* as part of a person's employment (l.e., real estate brokers, 
 delivery men, ate.) are not to be considered by the Interviewer, and 
 therefore nor marked on che trip sheet. 
 
 - A person may go to a neighbor's for coffee and then go shopping. This 
 would repreeenc two crips, one to coffee and the ocher to shop. This * 
 would be s result of che time leg between the movements. 
 
 - Tripe made on e school bus, bicycle, motorcycle or by walking (not a 
 regular transit bus), are not included. 
 
 C-7 
 
Trip Number 
 
 This Is simply to be used to identify each individual trip. The first trip will be 
 numbered 1 and so on. It is not important to maintain any particular sequence as to 
 which trip should come first. 
 
 Time of Day 
 
 Ask the respondent the time the trip started. (AH et PH ) 
 
 Trip Origin 
 
 The Interviewer is to ask for the origin of the trip (that is, where did the trip 
 start out from?). The location should be identified by specifying the nearest 
 intersecting streets, e.g. Main and 2nd, except if home is tne point of origin, in 
 which case the interviewer can simply write the word "home", since we already have 
 the respondent's address recorded. 
 
 Trip Destination 
 
 The same comments that have been made for the trip origins apply for the trip destina¬ 
 tion (where the trip ended). Remember that the more specific the address, the easier 
 it will be to code the information later. 
 
 Trip Purpose 
 
 For all trips, the respondent should be asked the activity at the origin and 
 destination. There are seven purpose codes used in thl6 survey. The interviewer 
 has the option of designating each of these seven trip purposes by the letter codes 
 indicated or in words, but if the interviewer chooses to write out the response in 
 words, he/she should use the exact code words indicated whenever possible. If the 
 interviewer has difficulty classifying the response as one of the listed codes, 
 be/she should record whatever the respondent says. 
 
 Note on trip purpose I), Chauffeuring : A trip activity should be recorded as 
 ’’chauffeuring" only when the driver is the respondent, and transporting a passenger 
 was the driver's only reason for making the trip. In this case, the Interviewer should 
 record code "D" or the word "chauffearing" and the passenger’s purpose , in the column 
 headed, "What was the activity at your destination"; hence, there will be two trip 
 activities appearing in the "activity at destination" column for this trip. The 
 "activity at crigin"for this trip would be the dri**a. a activity at origin. If the 
 driver starts a second trip from the place vt »r: he has chauffeured his passenger, 
 the "activity at origin" of this second trip would be code D, chauffeuring. 
 
 f) Mode of Travel 
 
 There are six numerically coded modes of travel that are acceptable for the survey. 
 (Remember, walking, bicycling, motorcycling and travelling by school bus are not 
 Included). The interviewer should use the identifying code number to record the mode 
 for each trip. 
 
Modi Definitions: 
 
 1. Auto - Tou drove alone: Tou drive e car in which there are no other passengers. 
 
 2. Auto - You drove with passenger: Tou drive a car in which there is at least 
 
 one additional passenger. 
 
 3. Auto - Tou rode as a passenger: Tou ride in a car which is driven by another 
 
 person. 
 
 A. Taxi - Side alone: "Kids Alone" taxi service. 
 
 3. Taxi - Advanced, shared ride: Rider calls two hours in advance; rider nay 
 share the cab with another person. 
 
 6. Taxi - lamedlace, shared ride: Taxi will pick up rider a few alnutes after 
 rider telephones; rider may share the cab with another person. 
 
 Sample responses on trip sheet: 
 
 a. A person uses the Flexicab lnnediate request, shared-ride service to travel from 
 hone to a store, at approximately 2:00 pa. 
 
 Trip would be recorded in the appropriate coluans on Fora C as follows. (Responses are 
 underscored). 
 
 At what ties did trip begin? 2:00 pa 
 
 Vhat are the nearest Intersecting streets to your origin? home (hone address recorded 
 
 elsewhere) 
 
 Vhat wee the activity at origin? A 
 
 What are the nearest intersecting a treats to your destination? Main and 8th 
 What was the activity at destination? £ 
 
 What aode of travel did you use? £ 
 
 b. After completing his shopping, this same person walks one block to pick up his car 
 at the garage, and at approxlaately 3:30 pa drives to pick up s friend and proceeds 
 to an athletic field, where he and his friend are aeeting others to play touch foot 
 -bell. 
 
 Trip would be recorded as follow. 
 
 At what time did trip begin? 3:30 
 
 ... nearest intersecting streets to origin? Main and 9th 
 . . . activity at origin? 0 
 
 . . . nearest intersecting streets to destination? Route 10 and Maple 
 . . . activity at destination? E 
 Uhat mode of travel did you use? 2. 
 
 e. After the game, at approximately 3:30 pa, the saae person 
 
 to the friend's music lesson, where the driver waits for his friend. 
 
 Trip recorded as follows. 
 
 At what time did trip begin? 3:30 pa 
 . . . nearest intersecting streets to origin? Roues 10 and Maple 
 . . . activity at origin? t 
 
 , . . nearest intersecting streets eo destination? Water and 7th 
 . . . activity at destination? D.G 
 What node of travel did you use? _2 
 
 C-9 
 
These '’ate are CONFIDENTIAL and will be used for travel studies only. 
 r ITY OF XENIA HOME INTERVIEW SURVEY - 1977 
 
 Form A 
 
 Household Data 
 
 (One for each household surveyed) 
 
 Al. 
 
 A2. 
 
 A3, 
 
 A4 . 
 
 A6 
 
 A8, 
 
 A9 
 
 Date 
 
 / 
 
 Time 
 
 Household I.D. 
 
 B5. Why did you stop using the Xenia public transit service' 
 
 Address 
 
 — 
 
 
 . ,. ■ — 
 
 ADMINISTER THIS FORM TO AN ADULT 
 
 __ 
 
 {Unreliable 1 
 
 How many persons are living here in this household? □ 
 
 How many of them are over 12 years old? □ 
 
 How many automobil es d oes your household own or 
 operate presently?! I 
 
 ■-- 
 
 Did you live in Xenia when the tornado struck in April 1974? 
 
 . s Head ‘ • 
 
 1 YES QjNO | IF NO, GO TO FORM~B 
 
 ^y ei V0r! Z 3 OJ y * 
 
 A5. ‘How many automobiles did your household own or operate at the 
 
 time of the tornado? 
 
 • . : Vv-7 
 
 □ 
 
 shared-household 
 
 
 
 Were any automobiles destroyed or made inoperable by the 
 tornado? 
 
 □ 
 
 I YES | 
 
 □ 
 
 |no 
 
 IF NO, GO TO FORM B 
 
 ,ent 
 
 i M \LE f EM/ LE 
 
 A7. Did you replace or repair any of them? 
 
 c 
 
 YES 
 
 I 
 
 NO IF NO, SKIP TO A10 
 
 How many automobiles did you replace or repair?^ 
 When did you replace or repair the damaged auto(s)? 
 
 auto #1 
 auto *2 
 
 J YES 
 
 Month 
 
 Month 
 
 Year 
 
 Year 
 
 r 
 
 GO TO FORM B 
 
 Of f ic 
 
 C-12 
 
 A10. why didn't you replace or repair your destroyed or damaged autos? 
 Transit available 
 Needed to rebuild home 
 Decreased income 
 Other 
 
These data are CONFIDENTIAL and will be used for travel studies only. 
 
 Form B 
 
 Pex-ssn Data 
 
 (One form for each person) 
 
 Time 
 
 Person No. 
 
 Household I.D. 
 
 Bl. ASK QUESTION ONLY IF PERSON WAS A RESIDENT OF XENIA DURING TORNADO 
 
 Were you aware of the free bus service after the tornado of 1974? 
 
 □ 
 
 o 
 
 YES 
 
 NO 
 
 f IF YES'J Did you use it? 
 
 ]yES I I NO 
 
 B2. Have you used the Flexicab/X-Line service in the last month? 
 
 1 
 
 H 
 
 □ 
 
 [if 
 
 yes} 
 
 YES 
 
 □ 
 
 NO 
 
 Of those, how many trips were for going to or from 
 work? 
 
 for going shopping? 
 for going to or from school? 
 
 U for any other reason? 
 
 
 IF YES, SKIP TO B6 ON NEXT PAGE 
 
 B3. Are you aware of the Flexicab/X-Line service currently being 
 operated by the Xenia Cab Company? 
 
 □ 
 
 YES 
 
 □ 
 
 NO 
 
 B4. Have you ever ridden Flexicab/X-Line Bus or the Free Bus? 
 
 □ 
 
 YES 
 
 □ 
 
 NO IF NO, SKIP TO B 6 ON NEXT PAGE 
 
 C-ll 
 

 iese data are CONFIDENTIAL and will be used in travel studies only. 
 
 These data are CONFIDENTIAL and will be used for travel studies only. 
 
 xilable ' 
 I NO [if 
 
 Of fi 
 
 ce Use 
 
 TO 
 
 iver; 
 
 iint 
 
 :rip, is your car 
 much more expena 
 
 B5. Why did you step using the Xenia public transit service? 
 
 Office Use 
 
 about the same cost 
 
 □ inconvenienfj , lightly cbupu 
 
 [□ Heard that it might stop 
 [ [unreliable 
 ~Jtoo Slow 
 
 Once agairfcffor your average shopping trip, is your car 
 
 I I Stopped being free 
 
 _ much faster 
 
 [TJ Other _ 
 
 — 
 
 j about the same time 
 
 BS. What is your relationship to the household? 
 
 5 i >- y s iov ' r 
 
 □ Head 
 
 r——1 _ 
 
 _I Spouse 
 
 Child over 12 years old 
 
 B17. 
 
 •I n your c>\* 
 use public 
 
 □ 
 
 Adult in shared household 
 
 wortrs, ■VhcTtr Tfrb tne reasons you ao or\ do not) 
 
 □ 
 
 Other 
 
 
 
 B7. OBSERVE 
 
 , DO NOT ASK| Sex of respondent 
 
 [ 
 
 1 
 
 MALE 1 IfEMALE 
 
 B8. How old 
 
 are you, approximately? 
 
 
 QUi 
 
 1STICW ONLY TO MOST SENIOR RESPONDENT.! 
 
 IF UNDER 16, GO TO B16 
 
 i.RD. 
 
 B9. Do you have a valid driver's license? 
 □I YES 
 
 □ no 
 
 • • ’ :' ;. v-i *vr. *.v; 
 
 GO TO FORM 
 
 C-14 
 
 C-12 
 
 
These data are CONFIDENTIAL and will be used in travel studies only 
 
 BIO. What is your major activity? 
 
 • Office Use 
 
 Work full-time (30 or more hours per week) - 
 
 Work part-time (8 to 29 hours per week) 
 
 Student (grade or high school) 
 
 Student (College) ' 
 
 Home Duties 
 
 Retired 
 
 Unemployed 
 
 Other _ 
 
 IF NOT A LICENSED DRIVER, SKIP TO B16. 
 
 II UNEMPLOYED, SKIP TO B13. _ 
 
 Bll. Let's talk about your trip to work. Do you generally have an 
 automobile available to drive to work? 
 
 >ES QnO IF NO SKIP TO 319 J 
 
 Now, I’m going to ask you to compare driving your car td 
 riding the Flexicab/X-Line service. For your information, 
 the average Flexicab fare is 75 cents. For your trip to 
 work, is driving your 
 
 (check one) 
 
 than taking Flexicab? 
 
 B12. For the same work trip, is driving your car 
 
 much taster 
 slightly faster 
 about the sane 
 slightly slower 
 much slower 
 
 H 
 
 (Check one) 
 
 than taking Flexicab? 
 
 car 
 
 much more expensive , 
 slightly more expensive, 
 about the same cost, 
 slightly cheaper, 
 much cheaper 
 
 □ 
 
 □ 
 
 □ 
 
 □ 
 
 □ 
 
 □ 
 
 □ 
 
 □ 
 
 C-13 
 
These data are CONFIDENTIAL and will be used in travel studies only. 
 
 B13. Now, let's talk about shopping trips. Do you generally 
 
 have an automobile available to drive for shopping trips? 
 
 IF NO SKIP*~TO B16~| 
 
 L_r s 
 
 □ 
 
 Office Use 
 
 NO 
 
 B14. For your average shoppino trip, is your car 
 
 (On 
 
 I READ TO THE RESPONDEN T:{ 
 
 Now, I'm going to i 
 
 than riding Flexicab? 
 
 include all trips, 
 to make. For each 
 
 « much more expensive 
 slightly more expensive 
 about the same cost 
 slightly cheaper 
 
 much cheaper 
 
 ou about all the trips you’ve made 
 
 (Average fare 75 cents) 
 
 even ones that may seem unusual for you 
 
 Of your trips, 
 
 .1 vsk you what time 
 
 B15. Once again, for your average shooping trip, is your car 
 
 itions,and the means of transportation 
 
 much faster 
 
 you 
 
 LX6S a w OOv 
 
 ed. Please 
 ,ting the 
 
 >PONDEN r 
 
 THE 
 
 "■ j . _ -ing streets in eact 
 
 slightly faster 
 
 about,the same time 
 
 IYITYi AND MODE OF TRAVEL CARD. 
 
 slightly slower 
 
 much slower 
 
 than riding Flexicab? 
 
 B16. in your own words, what are the major reasons you )do or do notj 
 use public transportation. 
 
 >egin 
 
 3Ur 
 
 ‘St trip today 
 
 --- 
 
 B17. ASK THIS QUESTION ONLY TO MOST SENIOR RESPONDENT. 
 
 HAND RESPONDENT THE INCOME CARD, 
 
 Please tell me the letter corresponding to the total income 
 of your household. 
 
 by 
 
 GO TO FORM C 
 
INCOME 
 
 A. Under $5,000 
 
 B. $5,000 - $9,999 
 
 C. $10,000 - $14,999 
 
 D. $15,000 - $19,999 
 
 E. $20,000 or over 
 
 C-15 
 
TIME: 
 
 TIME 
 
 Form C 
 
 Travel Diary 
 (One for each person) 
 
 (to nearest minute) 
 
 [READ TO THE .RESPONDENT^] 
 
 Now, I‘m going to ask you about all the trips you've made 
 today and any others you are planning to make later. Please 
 include all trips, even ones that may seem unusual for you 
 to make. For each one of your trips, I’ll ask you what time 
 you started, the place you started, the place you went, the 
 activities at both locations,and the means of transportation 
 you used. Please describe your trip origins and destinations by 
 indicating the nearest two intersecting streets in each case. 
 
 |HAND RESPONDENT THE ACTIVITY AND MODE OF TRAVEL CARD. 
 
 n TKE RESPONDENT IS EMPLOYED MORE THAN 30 HOURS/WEEK (FULL TIME) - SEE 
 
 CONTINUE .READING : j BOX AT TOP OF FORM D.' PRESENT FORM D FOR 
 
 > tfD PRESENT .-ART) TITLED m W 0RK TRIP — PART D ,f TO RESPONDED 
 
 uhon A i A vnn n vnnr f i ref frin a vj 
 
 After the respondent reads the card, the interyiever_assi^.ts.the r 
 spondent for the first three or four responses and then goes to Fora B 
 with second person. If Form D respondent needs more help, however, in¬ 
 terviewer should stay with them. If someone finds Form D impossible to 
 do, don t pressure them and just go on to the second person. 
 
 ?ora D, how to fill it out, and going through the first 
 
 Interviewer should be prepared to spend up to five minutes explaining 
 , ow to fill it out, and going through the first response. 
 
'O 
 
 r-i 
 
 o 
 
 je 
 
 « 
 
 CD 
 
 a 
 
 o 
 
 z 
 
 u 
 
 41 
 
 c 
 
 o 
 
 CD 
 
 U 
 
 V 
 
 a, 
 
 c 
 
 0 
 
 n 
 
 w 
 
 V 
 
 cu 
 
 Z 
 
 y 
 
 <0 
 
 41 
 
 u 
 
 o 
 
 41 
 
 c 
 
 o 
 
 What mode of 
 travel did 
 
 you use? 
 
 • 
 
 
 
 
 
 
 
 
 
 
 
 
 
 What was 
 the act¬ 
 ivity at 
 the dest¬ 
 ination? 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 What are the nearest 
 intersecting streets 
 to your destination? 
 
 
 • 
 
 
 
 
 
 
 
 
 
 • 
 
 - 
 
 
 What was 
 the act¬ 
 ivity at 
 the origin? 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 What are the nearest 
 intersecting streets 
 to your origin? 
 
 r 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 At What 
 time did 
 the trip 
 begin? 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Trip 
 
 No. 
 
 
 
 m 
 
 
 
 o 
 
 r- 
 
 00 
 
 O' 
 
 10 
 
 IT 
 
 12 
 
 13 
 
 w 
 
 
 
 <u 
 
 h 
 
 
 S> 4) 
 
 4) 
 
 C 
 
 O' 
 
 4) 53 
 
 V 
 
 c 
 
 •c •- 
 
 (A 
 
 41 
 
 -~1 u 
 
 X 
 
 cn 
 
 u 
 
 2 
 
 X 
 
 V 
 
 0. 
 
 <0 
 
 73 4 ) 
 
 o 
 
 0. 
 
 0 W 
 
 C Z 
 
 
 . (3 
 
 0 -u> 
 
 a 
 
 T Z 
 
 —* 
 
 
 z a 
 
 O 3 
 
 n 
 
 X 
 
 
 to o 
 
 • 
 
 
 05 
 
 u 
 
 a 
 
 o 
 
 u 
 
 « 
 
 CD 
 
 O 
 
 c 
 
 e 
 
 u 
 O' 3 
 C 4) 
 
 Q.u-1 
 (I JC 3.3 
 6 U. 0 <3 
 O O Z Z 
 = 2 05 Cj 
 
 CD 
 
 3 
 
 03 
 
 C 
 0 
 a 
 u 
 o 
 
 Cu O 
 
 u 
 
 « 
 
 £ 
 
 e- 
 
 u 
 
 < 
 
 < a o Q W 6u o 
 
 C-17 
 
Iie.fe?enc.e Datq 
 
 person) 
 
 HOURS/WEEK 
 
 PART D" 
 
 'WORK TRIP 
 
 30 HOURS/WEEK 
 
 RESPONDENT 
 
 pondent ne 
 If someone 
 
 red to spen, 
 going throu 
 
 
 
 
 
 
 
 *c> 
 
 4T\ 
 
 
 
 ■ 
 
 
 
 
 
 
 
 
 
 
 . 
 
 
 
 
 ; 
 
 
 
 
 
 - 
 
 
 
 
 
 >•» 
 
 
 •• 
 
 — 
 
 
 
 
 
 •*»* 
 
 
 o 
 
 C 
 
 . 
 
 a 
 
 o' 
 
 b 
 
 o 
 
 o 
 
 O 
 
 e 
 
 O 
 
 o 
 
 
 
 —* 
 
 ■“* 
 
 
 
 
 
 
 - 
 
 
 
 O' 
 
 
 CN 
 
 0*' 
 
 
 O' 
 
 
 O 
 
 O' 
 
 &■ 
 
 o> 
 
 O' 
 
 00 
 
 cC 
 
 90 
 
 .« 
 
 
 TIME: 
 
 * • 
 
 
 
 
 
 f*v 
 
 P” 
 
 
 " 
 
 r> 
 
 <« 
 
 :o nearest minute 
 
 
 4 «» 
 
 Sp 
 
 
 X 
 
 nO 
 
 >o 
 
 
 vO 
 
 «© 
 
 
 * 
 
 >o 
 
 
 lA 
 
 
 
 : 
 
 
 
 
 
 
 VV. 
 
 ’4T» 
 
 - 
 
 and i 
 
 fill t 
 
 >e use 
 
 i 
 
 * 
 
 for 
 
 travel s 
 
 studies onlj 
 
 I l 
 
 f. 
 
 <n 
 
 
 r"', 
 
 
 
 
 !■ - 
 
 IfH 
 
 
 
 
 r ’ 
 
 CITY OF XENIA HOME INTERVIEW SURVEY - 1977 
 
 Form D 
 
 " 
 
 To Che Incerviever: 
 
 
 (one form 
 
 « I 
 
 oC i 
 
 l g c 
 
 IF THE RESPONDENT IS EMPLOYED MORE 
 QUESTION B5 - CHECK "WORK" BOX AT 
 
 I • I « ' 
 
 WORK TRIP AND PRESENT CARD TITLED 
 
 ,©**•• Iwi w '.j o I . u 
 
 : > ''■■■• 
 
 Jh f 'U v 1 i i 
 
 IF RESPONDENT IS NOT EMPLOYED MORE 
 
 r—i ■ TT 
 
 BOX AT TOP OF FORM D, PRESENT FORM D 
 
 TITLED "SHOPPING TRIP - PART D" 1 
 
 $ 
 
 (FULL TIME) 
 
 Or 
 
 at 
 
 sC 
 
 <r 
 
 <n 
 
 - SEE 
 
 FORM D FOR 
 
 « ; ■ i *{“' 
 
 TO RESPONDENT. 
 
 , CHECK "SHOP" 
 AND PRESENT CARD 
 
 « 01 
 y e> 
 
 !_ i- 
 is-l 
 
 Afcer che respondent reads l 
 spondent for the first three or 1 
 with second person. If Form D r< 
 terviewer should stay with them, 
 do, don't pressure them and jus 
 c! I 6j « o E £ 
 
 Interviewer should be prep, 
 Form D, how to fill it out, and 
 
 e interviewer assists the re- 
 and then goes to Form B 
 eds more help, however, in¬ 
 finds Form D impossible to 
 e second person. 
 
 —i .cl jt\ rc ! 
 
 ©I C I lAl col i?) C m 
 
 up to five minutes explaining 
 
 h the first response. 
 
 c > 
 ctfc; 
 
WORK TRIP - HOW TO FILL OUT FORM D 
 
 We are interested in your opinion of various types of transit 
 services. We have listed on the accompanying form 18 cifferent 
 transit services. We would like you to imagine that you have a 
 job in downtown Xenia, and then tell us how often you would be 
 likely to use each of the alternative services to travel from 
 your home to your job downtown. Please rate each alternative 
 on a scale of “never" to “always* as shown on the accompanying 
 FORM D. For each of these services, we have indicated 
 (1) the type of transit service; (2) the cost of the transit 
 service; (3) how close the service passes to your home; (4) how 
 much longer it would take you to reach your destination downtown 
 using the service than it would if you drove your car. Each of 
 the terms used to describe the service are defined at the bottom 
 of this page. When considering each alternative, you should assume 
 that this service is the only public transit service available to 
 you. If you own a car, or have regular access to one, you should 
 assume that you have a choice of using the car or the transit 
 service described. Remember to respond to all 18 alternatives, 
 please. If you have any questions, please ask the interviewer. 
 
 eapimmgp or isms 
 
 1. “Every 15 (or 30) minutes;* 
 
 This refers to a regularly scheduled transit service, like a 
 bus route operating on a fixed route. Assume that you will know 
 the schedule and that the service will operate on time . 
 
 2. “CaU a t .. lg»yt jyq liSUEl In ftdvinsr « 
 
 You must call at least two hours before you want the transit 
 service to pick you up, but assume that it will be there when 
 you ask for it. 
 
 3. “On Demand*» 
 
 The transit service will arrive in 10 or fewer minutes after 
 you telephone for it, just like a taxi. 
 
 4. 
 
 This is the cost of each one-way trip on the transit service. 
 
 5. “ How close service comes to vour residence’: 
 
 The transit service will pick you up either 6 blocks from your 
 home, 3 blocks from your home, or else right in front of your home. 
 
 6. “ How much longer vour trip takes by the service *; 
 
 This represents how much longer the trip takes by the transit 
 service than if you drove a car. 
 
 C-19 
 
Household 
 
 a; 
 
 A 
 
 E 
 
 O 
 
 z 
 
 c 
 
 c 
 
 K 
 
 k 
 
 ( 1 ) 
 
 a. 
 
 er 
 
 e 
 
 •H 
 
 c. 
 
 a 
 
 o 
 
 ■C 
 
 Ui 
 
 u 
 
 z 
 
 o 
 
 x. 
 
 U 
 
 tJ 
 
 u 
 
 1 
 
 C-20 
 
APPENDIX D 
 
 Results of Multiple Analyses of Variance for 
 
 Individual Regression Coefficients and Weights 
 
TABLE D.l.a 
 
 Results of Multiple Analyses of Variance for 
 
 Individual Regression Coefficients and Weights 
 
 Individual 
 Coefficient Values 
 
 Interpersonal Measures 
 
 Trip 
 
 Purpose 
 
 Previous 
 Transit Use 
 
 Sex 
 
 Age 
 
 Possession 
 of License 
 
 Education 
 
 Overall Test 
 on Regression 
 Coefficients 
 
 
 • 
 
 kirk 
 
 
 
 
 i 
 
 Type of Service 
 
 * 
 
 
 
 
 
 
 Bus Headway 
 
 
 
 
 
 
 
 Taxi Headway 
 
 kk 
 
 
 k*k 
 
 
 
 
 Bare (Cost) * 
 
 
 
 
 
 
 
 Walk (Blocks) 
 
 % 
 
 
 
 * 
 
 
 
 
 Time Diff. 
 
 
 * 
 
 
 
 • 
 
 
 Fare Squared 
 
 
 
 
 
 
 
 Walk Squared 
 
 
 
 
 
 kkk 
 
 • 
 
 Time Squared 
 
 
 
 *★ 
 
 
 
 
 Intercept 
 (Grand Mean) 
 
 
 
 
 
 
 
 Overall Test 
 on Weight 
 Coefficients 
 
 
 
 kk* 
 
 
 
 • 
 
 Mode/Headway 
 
 
 % 
 
 kirk 
 
 
 
 • 
 
 Fare (Cost) 
 
 
 
 
 ** 
 
 
 
 Walk (Blocks) 
 
 • 
 
 i 
 
 
 
 
 
 Time Diff. 
 
 • 
 
 
 
 
 
 
 *90 percent significance level 
 **95 percent significance level 
 ***99 percent significance level 
 
 D-l 
 
TABLE ’D.l.b 
 
 Results of Multiple Analyses of Variance for 
 
 Individual Regression Coefficients and Weights 
 
 Individual 
 
 Coefficient 
 
 Values 
 
 
 louraaruinil Kuidmi 
 
 
 
 
 
 
 
 - 
 
 Income 
 
 
 Household 
 
 Size 
 
 
 Adult 
 
 Household 
 
 Size 
 
 
 Use and 
 
 Auto 
 
 
 
 Purpose & 
 Sex 
 
 
 
 
 ' 
 
 
 Omer ship 
 
 
 
 
 Overall lest 
 oc Regression 
 Coefficients 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Type of 
 Service 
 
 
 
 
 
 
 : 
 
 
 
 
 
 
 
 
 Bus Beadway 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Taxi Headway 
 
 
 
 
 
 
 * 
 
 
 
 
 
 - 
 
 
 Fare (Cost) 
 
 
 
 
 
 
 
 
 
 
 Walk (Blocks) 
 
 
 
 
 
 
 
 
 
 
 ’Time Diff. 
 
 
 ft | 
 
 
 
 
 
 #** 
 
 
 
 
 * 
 
 
 Fare Squared 
 
 
 
 
 
 
 
 * 
 
 
 i —. ..I .. ... ... 
 
 
 
 
 Walk Squared 
 
 
 
 
 
 * 
 
 
 
 
 
 
 
 Tine Squared 
 
 
 
 
 
 
 
 
 ... 
 
 
 Wet- - ■ ■ 
 
 
 Intercept 
 (Grand Mean) 
 
 
 
 
 
 ** 
 
 
 
 • 
 
 
 
 , 
 
 Overall Test 
 on Weight 
 Coefficients 
 
 
 
 
 
 
 
 
 *** 
 
 
 
 
 
 hode/Headvay 
 
 
 
 
 
 * 
 
 
 
 
 — 
 
 
 
 Fare (Cost) 
 
 
 
 
 
 * 
 
 
 
 
 
 Walk (Blocks) 
 
 
 
 
 
 
 
 
 
 
 Time Diff. 
 
 
 
 
 ---1 
 
 
 * 
 
 
 *** 
 
 
 
 
 
 
 *90 percent significance level 
 **95 percent significance level 
 ***99 percent significance level 
 
 D-2 
 
 D~4 
 
TABLE D.l.c 
 
 Results of Multiple Analyses of Variance for 
 
 Individual Regression Coefficients and Weights 
 
 Individual 
 
 Coefficient 
 
 Values 
 
 Interpersonal Measures 
 
 Purpose 
 
 and 
 
 Age’ 
 
 Purpose 
 
 and 
 
 License 
 
 Purpose 
 
 and 
 
 Education 
 
 Purpose 
 
 and 
 
 Income 
 
 Purpose and 
 Household 
 Slse 
 
 Purpose and 
 Adult 
 Household 
 
 Size 
 
 Overall Teat 
 on Regression 
 Coefficients 
 
 
 
 
 
 
 
 Type of 
 
 Service 
 
 
 
 
 
 
 • 
 
 Bus Headway 
 
 
 * 
 
 
 
 
 
 Taxi Headway 
 
 **• 
 
 
 
 
 
 
 Fare (Cost) 
 
 
 
 
 
 
 
 Walk (Blocks) 
 
 
 
 
 
 
 
 Tine Diff. 
 
 
 
 
 
 
 
 Fare Squared 
 
 
 
 
 
 *** 
 
 
 Walk Squared 
 
 
 
 
 
 
 
 Tine Squared 
 
 
 
 
 • 
 
 
 
 Intercept 
 (Grand Mean) 
 
 
 
 
 
 
 
 Overall Test 
 on Weight 
 Coefficients 
 
 
 
 
 
 
 
 Hods/Headway 
 
 * 
 
 
 
 
 
 
 Fare (Cost) 
 
 * 
 
 
 
 
 
 * 
 
 Walk (Blocks) 
 
 
 1 
 
 b 
 
 • 
 
 
 
 
 Tine Diff. 
 
 
 1 
 
 
 
 
 
 *90 percent 
 **95 percent 
 ***99 percent 
 
 significance 
 
 significance 
 
 significance 
 
 level 
 
 level 
 
 level 
 
 D-3 
 
TABLE D.l.d 
 
 Results of Multiple Analyses of Variance for 
 
 Individual Regression Coefficients and Weights 
 
 
 
 
 Interpersonal Measures 
 
 
 
 Individual 
 
 Coefficient 
 
 Values 
 
 
 
 interpersonal Measures 
 
 
 Purpose 
 and Auto 
 Ownersliip 
 
 
 Previous 
 Use end 
 Sex 
 
 
 
 
 Previous 
 Use and 
 License 
 
 
 Previous 
 Use and 
 Education 
 
 
 Previous 
 
 Use and 
 Income 
 
 
 
 
 
 
 
 
 
 
 
 
 Overall Test 
 on Regression 
 Coefficients 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Type of 
 Service 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Bus Headway 
 
 
 
 
 
 
 ** 
 
 
 
 
 
 
 Taxi Headway 
 
 
 *** 
 
 
 
 
 --- -- 
 
 
 
 
 
 
 
 
 
 Fare (Cost) 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Walk (Blocks) 
 
 
 
 
 
 
 
 
 Time Diff. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 * 
 
 
 
 
 
 
 
 
 
 
 
 
 Fare Squared 
 
 
 
 
 
 ** 
 
 
 
 
 
 
 ** 
 
 
 
 
 
 
 
 
 
 
 
 
 walk Squared 
 
 
 
 
 * 
 
 
 
 
 
 
 ' 
 
 - 
 
 Time Squared 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Intercept 
 (Grand Mean) 
 
 
 
 
 
 
 
 
 
 
 
 
 • 
 
 
 . 
 
 
 
 
 
 
 
 Overall Test 
 on Weight 
 Coefficients 
 
 
 
 
 
 
 
 
 * 
 
 
 
 
 
 
 
 Mode/Headway 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Fare (Cost) 
 
 ** H 
 
 
 
 
 
 
 ** 
 
 
 
 
 \ 
 
 
 
 Walk (Blocks) 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Time Diff. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 *90 percent significance level 
 **95 percent significance level 
 ***99 percent significance level 
 
 
 D-4 
 
 D-6 
 
TABLE D.l.e 
 
 Results of Multiple Analyses of Variance for 
 
 Individual Regression Coefficients and Weights 
 
 Individual 
 
 Coefficient 
 
 Values 
 
 Interpersonal Measures 
 
 Previous Use 
 & Household 
 Size 
 
 Previous Use 
 & Adult 
 Household 
 Size 
 
 Previous Use 
 & Auto 
 Ownership 
 
 Sex 
 
 & 
 
 *g« 
 
 Sex 
 
 & 
 
 License 
 
 Sex 
 
 & 
 
 Education 
 
 Overall Test 
 on Regression 
 Coefficients 
 
 • 
 
 ** 
 
 ** 
 
 * 
 
 
 
 Type of 
 
 Service 
 
 
 
 
 
 
 
 Bus Headway 
 
 
 
 #** 
 
 
 
 
 Taxi Headway 
 
 
 * 
 
 
 *+« 
 
 ** 
 
 
 Fare (Cost) 
 
 
 #* 
 
 * 
 
 *** 
 
 
 
 Walk (Blocks) 
 
 % 
 
 
 
 
 
 
 Time Dlff. 
 
 
 
 *** 
 
 
 
 
 Fare Squared 
 
 
 
 *** 
 
 
 
 * 
 
 Walk Squared 
 
 
 
 
 
 
 
 Time Squared 
 
 
 
 
 
 
 
 Intercept 
 (Grand Mean) 
 
 * 
 
 
 • 
 
 
 
 
 Overall Test 
 on Weight 
 Coefficients 
 
 ** 
 
 
 
 MM 
 
 
 
 Mode/He*dv«y 
 
 
 
 
 +* 
 
 
 
 Fare (Cost) 
 
 ** 
 
 * 
 
 
 #* 
 
 
 
 Walk (Blocks) 
 
 * 
 
 
 
 
 
 
 Time Dlff. 
 
 
 
 
 MM 
 
 
 
 *90 percent significance level 
 **95 percent significance level 
 ***99 percent significance level 
 
 D-5 
 
TABLE D.l.f 
 
 Results of Multiple Analyses of Variance for 
 Individual Regression Coefficients and Weights 
 
 
 . 
 
 Interpersonal Measures 
 
 
 Individual 
 
 Coefficient 
 
 Sex A 
 
 | Adult 
 
 Sex A 
 Household 
 
 License A 
 
 Sex & Adult 
 Household 
 
 1 
 
 License A 
 
 Sex A Auto 
 
 License & 
 
 Age A 
 
 License A 
 Household 
 
 Age & 
 
 
 Values 
 
 Income 
 
 Size 
 
 Size 
 
 Ownership 
 
 License 
 
 Education 
 
 
 Overall Test 
 on Regression 
 
 
 
 
 ** 
 
 
 
 
 Coefficients 
 
 
 
 
 
 
 
 
 Type of 
 
 
 
 
 
 
 
 
 Service 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Bus HfiACvsy 
 
 
 
 
 
 
 
 
 
 J - 
 
 
 
 
 
 
 
 
 
 
 
 
 
 a- v 
 
 
 
 
 
 
 
 
 Fare (Cost) 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Walk (Blocks) 
 
 
 
 
 
 * 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 a* 
 
 
 
 „ & ' 
 
 
 
 
 
 
 
 
 
 
 
 
 
 * 
 
 
 ** 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Intercept 
 
 
 
 
 
 * 
 
 
 
 
 
 
 
 
 
 
 Overall Test 
 on Weight 
 
 * 
 
 
 
 * 
 
 
 
 
 
 
 
 
 
 
 
 
 Mode/Headway 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 \ 
 
 rare (Lost) 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 walk (Blocks) 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 lime Dirt• 
 
 
 
 
 
 
 
 *90 percent significance level 
 
 *90 percent significance level 
 **95 percent significance level 
 ***99 percent significance level 
 
 D-8 
 
 D-6 
 
TABLE D.l.g 
 
 Results of Multiple Analyses of Variance for 
 
 Individual Regression Coefficients and Weights 
 
 Individual 
 
 Coefficient 
 
 Values 
 
 Interpersonal Measures 
 
 Age & 
 Income 
 
 Age & 
 Household 
 Size 
 
 Age i 
 Adult 
 Household 
 Size 
 
 Age & 
 
 Auto 
 
 Ownership 
 
 License & 
 Education 
 
 License & 
 Income 
 
 Overall Test 
 on Regression 
 Coefficients 
 
 
 
 
 
 
 
 Type of 
 
 Service 
 
 
 
 
 ftft* 
 
 
 
 Bus Headway 
 
 * 
 
 
 
 
 ftft 
 
 
 Taxi Headway 
 
 ftft 
 
 ftft 
 
 ftft 
 
 
 
 
 Fare (Cost) 
 
 
 
 ftft 
 
 
 
 
 Walk (Blocks) 
 
 * 
 
 
 
 
 
 
 Time Diff. 
 
 
 
 
 
 
 
 Fare Squared 
 
 
 
 
 ft 
 
 ftft 
 
 
 Walk Squarad 
 
 * 
 
 
 
 ft 
 
 ftft 
 
 
 Tine Squared 
 
 
 
 
 ftft 
 
 
 
 Intercept 
 (Crand Mean) 
 
 * 
 
 
 
 
 
 
 Overall Test 
 on Weight 
 Coefficients 
 
 
 
 
 ftft 
 
 
 
 Mode/Headway 
 
 ftft 
 
 
 
 
 ftft 
 
 
 Fare (Cost) 
 
 ftft 
 
 
 
 ftft 
 
 
 
 | Walk (Blocks) 
 
 
 
 
 
 
 
 | Tine Diff. 
 
 
 
 
 ft 
 
 
 
 *90 percent significance level 
 **9S percent significance level 
 ***99 percent significance level 
 
 D-7 
 
TABLE D.l.h 
 
 Results of Multiple Analyses of Variance for 
 Individual Regression Coefficients and Weights 
 
 
 
 Interpersonal Measures 
 
 - — | 
 
 Individual 
 
 Coefficient 
 
 Values 
 
 License & 
 Household 
 Size 
 
 License A 
 
 Adult 
 
 Household 
 
 Size 
 
 License A 
 Auto 
 
 Ownership 
 
 ise S 
 
 License A 
 Education 
 
 ise Hot 
 
 License A 
 Income 
 
 License A 
 Household 
 Size 
 
 rail 
 
 for 
 
 ession 
 Icitatg,.. 
 
 Overall Test 
 on Regression 
 Coefficients 
 
 reaslon 
 
 
 
 t 
 
 r** 
 
 
 
 
 
 
 
 
 
 Type of 
 
 Service 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Bus Headway 
 
 adway 
 
 
 
 
 
 
 
 
 kk 
 
 
 
 
 
 Taxi Headway 
 
 aadwey 
 
 
 * 
 
 
 
 
 itit 
 
 
 
 
 
 k 
 
 Fare (Cost) 
 
 
 
 
 
 
 
 
 
 
 
 
 zd 
 
 Walk (Blocks) 
 
 ** 
 
 
 
 
 * 
 
 
 kk 
 
 
 k 
 
 
 
 
 Time Diff. 
 
 ** 
 
 
 
 
 
 
 
 
 t 
 
 
 ** 
 
 * 
 
 . —r—— 
 
 Fare Squared 
 
 q uaTed 
 
 
 
 
 
 
 
 
 kk 
 
 
 kk 
 
 
 
 Walk Squared 
 
 ?viared 
 
 
 
 
 ** 
 
 
 
 
 kk 
 
 
 
 
 
 Time Squared 
 
 * 
 
 * 
 
 
 
 
 
 
 
 
 k 
 
 
 
 Intercept 
 (Grand Mean) 
 
 * 
 
 Ftear) 
 
 
 
 
 
 1 
 
 irk 
 
 
 
 
 * 
 
 • 
 
 
 Overall Test 
 on Weight 
 Coefficients 
 
 ght 
 
 cients 
 
 
 
 
 
 
 kk 
 
 
 
 
 
 ' 
 
 
 Mode/Headway 
 
 ■ 
 
 eadvay 
 
 
 k 
 
 
 ** 
 
 
 kk 
 
 
 
 
 k 
 
 
 
 Fare (Cost) 
 
 Cost) 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Walk (Blocks) 
 
 Blocks) 
 
 
 
 
 
 
 
 
 
 
 
 
 ** 
 
 Time Diff. 
 
 iff. 
 
 
 
 
 ** 
 
 
 
 
 
 
 
 
 _ 
 
 *90 percent significance level 
 **95 percent significance level 
 ***99 percent significance level 
 
 D-8 
 
TABLE D.l.i 
 
 Results of Multiple Analyses of Variance for 
 
 Individual Regression Coefficients and Weights 
 
 
 Interpersonal Measures 
 
 Individual 
 
 Coefficient 
 
 Values 
 
 & Adult 
 Househdld 
 Site 
 
 License & 
 Auto 
 
 Ownership 
 
 Education 
 & Income 
 
 Education & 
 Household 
 Size 
 
 Education & 
 
 & Auto 
 (Xmershlp 
 
 Income & 
 Household 
 
 Size 
 
 Overall test 
 on Regression 
 Coefficients 
 
 
 
 
 • 
 
 
 
 Type of 
 
 Service 
 
 
 
 
 
 
 
 Bus Beadvay 
 
 
 
 
 
 
 
 Taxi Headway 
 
 * 
 
 
 
 
 * 
 
 
 Pare (Cost) 
 
 
 ** 
 
 
 
 
 
 Walk (Blocks) 
 
 
 * 
 
 
 
 
 
 Time Diff. 
 
 
 ** 
 
 • 
 
 
 
 
 Fare Squared 
 
 
 
 
 • 
 
 
 . 
 
 Walk Squared 
 
 
 
 * 
 
 
 
 
 Time Squared 
 
 * 
 
 
 
 
 
 * 
 
 Intercept 
 (Grand Mean) 
 
 
 
 
 
 
 * 
 
 Overall Test 
 on Weight 
 Coefficients 
 
 
 
 
 
 
 
 Mode/Headway 
 
 * 
 
 
 
 
 
 
 Fare (Cost) 
 
 
 
 
 * 
 
 
 
 Walk (Blocks) 
 
 
 
 
 
 
 
 Tima Dlff. 
 
 
 #* 
 
 
 
 
 
 *90 percent significance level 
 **95 percent significance level 
 ***99 percent significance level 
 
 D-9 
 
H.H., 1976, "How Funct TABLE D. 1. j 
 
 ■val Scales of Mental Quantities," Journal of Applied 
 
 Results of Multiple Analyses of Variance for 
 
 Individual Regression Coefficients and Weights 
 
 he u il measu' 
 
 
 -Ai 
 
 hr; 
 
 "■p. (J 
 
 Individual 
 
 Coefficient 
 
 Values 
 
 rban Travc 
 
 Overall Test 
 on Regression 
 Coefficients 
 
 Type of 
 Service 
 
 Bus Headway 
 
 Taxi Headway 
 
 .mple, non-technlcaij ,_ 
 
 Interpersonal Measures 
 
 iction to his work. 
 
 exce- 
 
 »nt 
 
 Income 4 
 Adult 
 Household 
 Sire 
 
 rap 
 
 igrap'i 
 
 he 
 
 Results clearlv 
 
 " 
 
 Fare (Cost) 
 
 Walk (Blocks) 
 
 Time Diff. 
 
 Fare Squared 
 
 Walk Squared 
 
 Time Squared 
 
 Intercept 
 (Grand Mean) 
 
 Overall Test 
 on Weight 
 Coefficients 
 
 Mode/Headway 
 
 Fare (Cost) 
 
 
 Walk (Blocks) 
 
 Time Diff. 
 
 * 
 
 id Cox, C 
 
 _ 
 
 Bullei 
 
 — 
 
 — 
 
 Tt- 
 
 _ 
 
 income 
 & Auto 
 
 J*E£gaB 
 
 
 y pr£S6n 
 Salt 
 
 ite the i 
 
 • M, 
 
 — 
 
 — 
 
 1 
 
 m. 
 
 ** 
 
 -- 
 
 _ 
 
 ar models employ1 
 
 Household 
 Sire 
 & Adult 
 Sire 
 
 o the 
 Lake Cit 
 
 
 £ 
 
 PasSenge 
 
 l Travel Demand Mo 
 
 — 
 
 ** * 
 
 *** 
 
 Ht 
 
 
 - ; - 
 
 — 
 
 _ 
 
 Household 
 Sire 
 6 Auto 
 Ownership 
 
 Meetings 
 
 n grocer 
 
 ladequacy 
 
 rice ting. 
 
 ,* E:*** 
 
 — 
 
 — 
 
 _ 
 
 ng Simula :ion techniques 
 
 Adult 
 Household 
 Size & Auto 
 „,.Qwner.phtp . 
 
 in 
 
 Overall 
 Teat for 
 fcegreosion 
 .Coefficients 
 
 CiOE 
 
 £ t 01 
 
 of the 
 
 aeis 
 
 
 o 
 
 
 ---- N 
 
 
 — 
 
 — 
 
 — 
 
 _ 
 
 *90 percent significance level 
 **95 percent significance level 
 ***99 percent significance level 
 
 ts, P.M., and Koeppel, K.W.P., 1977, "The Trade-off Model: Empirical 
 and Structural Findings," NY/DOT Preliminary Research Report No. 123 
 
 n, p E., and Wind, Y., 1973, Multiattril. are Decisions in Marketing : 
 A Measurement Approach , Hinsdale, Illinois, Dryden Press. 
 
BIBLIOGRAPHY WITH SELECTED ANNOTATIONS 
 
 Albright, R.L.; S.R. Lerman and C.F. Manski, 1978, "Report on the 
 
 Development of an Estimation Program for the Multinomial Probit 
 Model," to be published in forthcoming Proceedings of NBER-NSF 
 Conference on Decision Rules and Uncertainty. 
 
 Anderson, N.H., 1970, "Functional Measurement and Psychophysical 
 Judgment," Psychological Review 77 : 153-170. 
 
 Anderson's seminal paper on functional measurement designed 
 to reach a wide audience in psychology, particularly those 
 interested in psychophysics and scaling. Good paper for 
 readers with some background in psychophysics. 
 
 _. , 1974, "Information Integration Theory: A Brief Survey," 
 
 in Contemporary Developments in Mathematical Psychology 2 . Edited 
 by D.H. Krantz, R.C. Atkinson, R.D. Luce and P. Suppes. San 
 Francisco: W. H. Freeman. 
 
 This is an excellent, well-written introduction to the con¬ 
 ceptual and statistical theory behind functional measurement 
 as it applies to multilinear models and their special cases — 
 additive and multiplicative equations. Considers applications 
 in a wide range of topics of interest to psychology. Those 
 with backgrounds in utility theory should find clear conceptual 
 links if they can "read past" the jargon differences. 
 
 _., 1974, "Algebraic Models in Perception," in Handbook of 
 
 Perception 2 . Edited by E.C. Carterette and M.P. Friedman. New 
 York: Academic Press. 
 
 Similar to above 1974 reference, but concentrates on perception 
 and psychophysics. 
 
 _., 1976, "Social Perception and Cognition," Center for Human 
 
 Information Processing Technical Report CHIP 62 , Center for Human 
 Information Processing, University of California at San Diego. 
 
 Similar to first 1974 reference, but concentrates on a very wide 
 range of issues in social psychology. Can by obtained by request 
 from Anderson. 
 
 OM I 
 
Anderson, H.H., 1976, "How Functional Measurement Can Yield Validated 
 Interval Scales of Mental Quantities," Journal of Applied 
 Psychology . 61, 677-692. 
 
 Destination Choice, Transportation Research Board meetings, 
 
 This paper sets out the underpinnings of functional measurement 
 in a simple, non-technical way. It constitutes an exce-lent 
 introduction to his work. 
 
 of ad hoc utility specifications in current travel demand 
 
 mo.dels. Demonstrates how utility specifications can be 
 
 Barnett, P., 1977, "The Application of Conjoint Measurement to Recurrent 
 Urban Travel," Paper presented to the Meetings in the Association 
 of American Geographers, Salt Lake City. 
 
 blend of functional measurement and more traditional econometric 
 
 Application of Conjoint Measurement to urban grocery store 
 choice. Results clearly demonstrate the inadequacy of the 
 
 tradeoff approach developed in marketing. 
 
 ;vin, I.P., 1977, "The Development of At tituainal Modelling Approaches i 
 
 Trcdeportation Research,' - paper presented to the .workshop on "Attitc 
 
 Ben-Akiva, M., 1974, "Structure of Passenger Travel Demand Models," 
 
 Transportation Research Record 526 , Transportation Research Board, 
 Washington, D.C. 
 
 Cochran, W.G., and Cox, G.M., 1957, Experimental Design , New York: John 
 Wiley. 
 
 Dawes, R.M., and B. Corrigan, 1974, "Linear Models in Decision Making." 
 Psychological Bulletin . 
 
 paper presented to the Special Sessions on Human Judgment and Spatial 
 
 An excellent review of the literature available on the applica¬ 
 tion of linear models to problems in human judgment and decision 
 making. It includes an important extension of the "robustness" 
 of linear models employing simulation techniques. 
 
 Review of the author’s work in the study of human decision 
 
 Domencich, T., and D. McFadden, 1975, Urban Travel Demand: A Behavioral 
 Analysis , North Holland, Amsterdam. 
 
 Eberts, P.M., and Koeppel, K.W.P., 1977, "The Trade-off Model: Empirical 
 and Structural Findings," NY/DOT Preliminary Research Report No. 123. 
 
 Green, P.E., and Wind, Y. , 1973, Multiattribute Decisions in Marketing : 
 
 A Measurement Approach , Hinsdale, Illinois, Dryden Press. 
 
 
Hahn, G.J. and S.S. Shapiro, 1966, "A Catalog and Computer Program for 
 the Design and Analysis of Orthogonal Symmetric and Asymmetric 
 Fractional Factorial Experiments," Technical Report 66-C-165 , 
 
 General Electric Information Sciences Laboratory, G*. E. Research 
 and Development Center, Schenectady, New York. 
 
 This is a powerful and comprehensive "cookbook" for individuals 
 who already understand the rudiments of experimental design. 
 
 It contains plans for selecting fractional factorial combinations 
 for virtually all sampling plans in which one is likely to be 
 interesetd. This document is a vital part of the library of 
 any person or organization interested in the design and analysis 
 of direct value assessment surveys. 
 
 Keeney, R.L., and H. Raiffa, 1976, Decisions with Multiple Objectives : 
 Preference and Value Tradeoffs, New York: John Wiley and Sons. 
 
 An introductory treatment of axiomatic utility theory, its 
 extensions and some important applications. Sophisticated 
 enough for all but the most sophisticated individuals. 
 
 Krantz, D.H., et al., 1971 Foundations of Measurement , New York: 
 Academic Press. 
 
 Highly technical and mathematical treatment of the bases for 
 measurement in science, with emphasis on psychology. Develops 
 the axiomatic basis for conjoint measurement. 
 
 _., and A. Tversky, 1971, "Conjoint Measurement Analysis of 
 
 Composition Rules in Psychology," Psychological Review 78 , 
 pp. 151-169. 
 
 A readable introduction to conjoint measurement and its 
 application. 
 
 Lancaster, K.J., 1966, "A New Approach to Consumer Theory," Journal of 
 Political Economy , Vo. 74, April 1966. 
 
 Lerman, Steven R., 1975, A Disaggregate Behavioral Model of Urban 
 
 Mobility Decisions , Center for Transportation Studies, Report No. 
 75-5, M.I.T., Cambridge, Massachusetts. 
 
Lerman, Steven R., and J.J. Louviere, 1978, "On the Use of Functional 
 Measurement to Identify the Functional Form of Travel Demand 
 Models," paper presented to the special session on Spatial 
 Destination Choice, Transportation Research Board meetings, 
 Washington, D.C., Forthcoming, Transportation Research Record . 
 
 A technical discussion of functional measurement and the problem 
 of ad hoc utility specifications in current travel demand 
 models. Demonstrates how utility specifications can be 
 diagnosed and tested independently of travel demand estimation 
 on revealed choice data. An empirical application to the 
 residential choice demonstrates the potential for a powerful 
 blend of functional measurement and more traditional econometric 
 approaches to travel analysis. 
 
 Levin, I.P., 1977, "The Development of Attitudinal Modelling Approaches in 
 
 Transportation Research," paper presented to the workshop on "Attitudes, 
 Attitudinal Measure,ent, and the Relationship between Behavior and 
 Attitudes," Third International Conference on Behavioral Travel 
 Modelling, Barossa Valley, Australia. Forthcoming in Proceedings 
 of the conference, 1978. 
 
 . s and K.L. Norman,. 1977, "Applications of Information 
 
 A comprehensive review of work in attitudinal modelling in 
 
 transportation with particular emphasis on functional measurement 
 and its advantages. 
 
 review o3u mucn ox. y \?q ttk. ^ri ois^nosis sdq ucbLiug vi uiuiit. .. 
 
 .ision" functions the functional measurement ■ 
 
 _., and M.J. Gray, 1976, Analysis of Human Judgment in Transportation, 
 
 paper presented to the Special Sessions on Human Judgment and Spatial 
 Behavior. Great Plains/Rocky Mountains AAG meetings, Manhattan, Kansas. 
 Published in Special Issue of the Great Plains/Rocky Mountains AAG: 
 
 Human Judgment and Spatial Behavior , 1977, edited by R. Reider and 
 J. Louviere. 
 
 Review of the author’s work in the study of human decision 
 making in various transportation contexts using the 
 functional measurement approach. 
 
 ersit 
 
 x>-'/ 
 
Louviere, J.J., 1978, "Modelling Individual Residential Preferences: A 
 
 Totally Disaggregate Approach,' 1 Technical Report No. 100 (forthcoming). 
 Institute of Urban and Regional Research, University of Iowa, Iowa 
 City, August. Accepted for presentation at 1979 TRP Meetings. 
 
 This is a conceptual treatment of the individual home buyer's 
 decision function which includes the report of several years' 
 work on modelling buyer decision functions. Paper demonstrates 
 that up to 13 attributes can be easily accommodated with a 
 modified experimental design and presents results demonstrating 
 consistent marginal utility functions from the respondents. Also 
 demonstrates that coefficients of individual functions are 
 systematically associated with interpersonal differences. 
 
 _, 1978, "Psychological Measurement of Travel Attributes," in 
 
 D.A. Hensber and M.Q.Dalvi (eds.), Determinants of Travel Choice , 
 London: Teakfield, Farnborough (Saxon House Studies) 
 
 This is a review and discussion of issues in the psychological 
 measurement of transportation attributes. Emphasizes functional 
 and conjoint measurement and proposes a paradigm for research 
 in this area. 
 
 _, 1974, "Predicting the Response to Real Stimulus Objects 
 
 from an Abstract Evaluation of their Attributes: The Case of Trout 
 Streams," Journal of Applied Psychology . 
 
 Brief discussion of an examination of the decision functions of 
 resource management personnel and the use of their decision 
 functions to predict their responses to actual situations. 
 
 _, et al., 1973, "Theory, Methodology and Findings in Mode 
 
 Choice Behavior," Working Paper 11 , Institute of Urban and Regional 
 Research, University of Iowa, Iowa City. 
 
 Report of the results of the initial attempts in 1972 to 
 estimate decision functions for mode choice. Contains a 
 discussion bearing on the results of some of the examples in 
 Chapter 4 of this report. 
 
 
Louviere, J.J., and I.P. Levin, 1977, ’’Functional Measurement Applied to 
 Consumer Spatial and Travel Decisions," paper presented to the 
 Special Session on Consumer Travel Behavior, Association for 
 Consumer Research, Cicago. Forthcoming i n Proceedings of the 
 conference, ACR. 
 
 I - V- • l, , .* ... r p__ I* £ 
 
 _., and R. Meyer, 1978, ”An Empirican Analysis of Potential 
 
 Consumer Responses to Aspects of President Carter's Proposed Energy 
 Policy," Technical Report (forthcoming). Institute of Urban and 
 Regional Research, University of Iowa, Iowa City. 
 
 i.datd 
 
 p 
 
 
 _., and E.M. 
 
 Travel Analysis," 
 pp. 1-9. 
 
 Wilson, 1978, Predicting Consumer Response in 
 Transportation Planning and Technology , 4, 1978, 
 
 - Ky 
 
 a Judgment and Spatial Behavior, 1977, edited 
 
 __., et al., 1974, "An Experiment to Derive Predictive Models 
 
 of Public Response to Policy Manipulations in Public Bus Transpor¬ 
 tation," Technical Report 35 , Institute of Urban and Regional 
 Research, University of Iowa, Iowa City. 
 
 __. , and K.L. Norman, 1977, "Applications of Information 
 
 Processing Theory to the Analysis of Urban Travel Demand," 
 Environment and Behavior, March, 1977. 
 
 Review of much of the work in diagnosis and testing of mode 
 choice decision functions 
 paradigm. 
 
 under the functional measurement 
 
 ing from rion-es 
 
 _., et al., 1976, "Travel Demand Segmentation: Some Theoretical 
 
 Considerations Related to Behavioral Modelling," in P.R. Stopher and 
 A.N. Meyberg (eds.). Behavioral Travel Demand Models , Lexington, Mass: 
 
 Lexington Books, D.C. Heath & Co. 
 
 Comprenensive review or recent developments m the developing 
 
 of models of decision making and choice. 
 
 __. , et al., 1977, "Theory and Empirical Evidence in Real-World 
 
 Studies of Human Judgment: Three Shopping Behavior Examples," 
 
 Research Paper No. 1 , Center for Behavioral Studies, Institute for 
 Policy Research, University of Wyoming. 
 
 This paper reports on the results of the first three studies of 
 the applicability of laboratory-type judgment models to predicting 
 real-world behavior. Demonstrates consistently high correlations 
 of models with behavior access studies. 
 
Louviere, J.J., E.M. Wilson, and P.M. Piccolo, 1977, "Applications of 
 
 Psychological Measurement and Modelling to Behavioral Travel Demand 
 Analysis," paper presented to the Third International Conference on 
 Behavioral Travel Modelling, Tanunda, Australia, April 1977. Forth¬ 
 coming in Proceedings , edited by P. Stopher and D. Hensher, 1978. 
 
 Review of current work in attitudinal modelling with primary 
 emphasis on development and extension of theory in fundamental 
 measurement applied »to travel and destination choice. Paper 
 also reports good correlations between laboratory derived 
 models and real choice behavior. 
 
 McFadden, Daniel, 1974, "Conditional Logit Analysis of Qualitative Choice 
 Behavior," in P. Zare’mbka (ed.), Frontiers in Econometrics, New York: 
 Academic Press. 
 
 Meyer, R.J., et al., 1978, "Issues in Modelling Travel Behavior in 
 
 Simulated Choice Environments: A Review," paper presented to the 
 Special Session on Spatial Destination Choice, National Meetings, 
 Association of American Geographers, New Orleans, April, 1978. 
 Forthcoming, Special Issue on Spatial Choice Problems, Economic 
 Geography , 1979. 
 
 Review and critical discussion of laboratory-type simulations of 
 choice behavior. Paper discusses the advances mode, limitations 
 of and unsolved problems related to modelling decisions in 
 simulated choice contexts. 
 
 _., I.P. Levin, and J.J. Louviere, 1978, "Functional Determinants 
 
 of Mode Choice," forthcoming. Transportation Research Record , Trans¬ 
 portation Research Board, Washington, D.C. 
 
 Report of a number of tests of laboratory derived decision models. 
 Results reveal high correlations between laboratory models and 
 the actual mode choices of the sample. 
 
 Norman, K.L., 1977, "Attributes in Bus Transportation: Importance Depends 
 on Trip Purpose," Journal of Applied Psychology , 62 (2), 1977. 
 
Norman, K.L., and J.J. Louviere, 1974, "Integration of Attributes in Public 
 Bus Transportation: Two Modelling Approaches," The Journal of 
 Applied Psychology 59 (1974): 753-758. 
 
 Both of the above two papers test for the additivity of decisions 
 models in mode choice. In both cases additivity must be rejected 
 in favor of multiplication. 
 
 Piccolo, J.M. and J.J. Louviere, 1976, "Information Integration Theory 
 Applied to Real-World Choice B^ravipr: Validation Experiments 
 Involving Shopping and Residential Choice," Paper presented to 
 the Special Session on Human Judgment AAG Meetings, Manhatten, KS, 
 October, 1976. Published in Special Issue of the Great Plains/Rocky 
 Mountains AAG: Human Judgment and Spatial Behavior, 1977, edited 
 by R. Reider and J. Louviere. 
 
 Paper which lays out approach to testing laboratory models on 
 real world choice behavior. Also demonstrates the approach 
 through two empirical comparisons, both of which exhibit 
 encouraging correspondence. 
 
 Shanteau, J., and R. Phelps, 1976, "Livestock Judges: How Much Information 
 Can the Experts Use?" Technical Report No. CKSU-HIPI), #76-15 , Depart¬ 
 ment of Psychology, Kansas State University, Psychology Series, 
 Manhatten, KS, December, 1976. 
 
 Important demonstration of the problem of inferring significance 
 of attributes in decision making from non-experimental data. 
 
 Slovic, P., B. Fischoff, and S. Lichtenstein, 1977, "Behayforal Decision 
 Theory," Annual Review of Psychology . 1 
 
 Comprehensive review of recent developments in the developing 
 of models of decision making and choice. 
 
 Winer, B.J., Statistical Principals in Experimental Design , New York: 
 McGraw-Hill Book Co., 1971 
 
 T>-/7 
 
 office 
 
 19096 
 
 •C.S. GOVZRNKEK? FHIiiT! 
 
 1960 0-622-593/1224 
 
JUN 0 4 RETD 
 
 The person charging- this material is re¬ 
 sponsible for its return to the library from 
 which it was withdrawn on or before the 
 Latest Date stamped below. 
 
 Theft, mutilation, and underlining of books are reasons 
 for disciplinary action and may result in dismissal from 
 the University. 
 
 To renew call Telephone Center, 333-8400 
 
 UNIVERSITY OF ILLINOIS LIBRARY AT URBANA-CHAMPAIGN 
 
 L161 — 0-1096 
 
UNIVERSITY OF ILLINOIS-URBANA 
 
 388 4042AN13 C001 
 
 AN ANALYSIS OF USER COST AND SERVICE TRA 
 
 3 0112 007229294