College and Research Libraries


Research Notes 
A Conjoint Analysis of Reference 
Services in Academic Libraries 
Gregory A. Crawford 

Conjoint analysis has been used by market researchers for the development of 
many products and services. This article displays the potential of conjoint 
analysis for evaluating reference services in academic libraries. Six dimensions 
of reference service are included in the analyses: definitiveness of answer, in line 
wait times, service time, number of items given to patron, hours of service, and 
cost of service. Of greatest importance to users are cost of the service and the hours 
during which reference is available. Most users prefer that all reference services be 
free and that reference help be available at all times the library is open. 

ne of the primary goals of 
marketing is to provide prod-
ucts or services that will ap-
peal to the preferences of 

consumers. Preferences, however, may 
vary among individuals and the final 
product or service offered must involve 
trade-offs of specific levels of the attrib-
utes desired by consumers. As a result, 
marketing strives to develop products 
or services that match the preferences of 
the consumers in such a way as to make 
them as appealing as possible to the larg-
est number of consumers. 

Similarly, one of the primary goals of 
libraries is to provide services that meet 
the information needs and desires of 

their patrons. In a recent RQ editorial 
addressing the promotion of libraries, 
Susan S. DiMattia says, "Our promo-
tional efforts tell about how libraries 
make information affordable, accessible, 
and available. We keep telling the public 
how wonderful we are, but we don't ask 
the public what they need. We don't 
seem to listen too much, we're so busy 
talking." 1 

Because patrons will vary in their 
preferences for specific library services, 
a method to assist in developing services 
to meet the desires of patrons could 
prove to be a great benefit to library 
administrators when used in conjunc-
tion with statements of the mission and 

Gregory A. Crawford is Head of Public Services at the Pennsylvania State University-Harrisburg, 
Harrisburg, Pennsylvania. This research was done while the author was a student in the Ph.D. program in 
Communication, Information and Library Studies at Rutgers University. The author would like to express his 
appreciation to Paul Kantor (Rutgers University) for his valuable insights and assistance in this project and 
for his helpful comments on the final draft of this paper. Special thanks go to Valerie Manusov (Rutgers 
University), Robert Pruznick (Warren County Community College), and Lori Toedter (Moravian College) for 
pennitting their students to participate in this research. 

257 



258 College & Research Libraries 

goals of the individual library. Conjoint 
analysis is one possible method of as-
sessing such preferences. 

As a technique, conjoint analysis can 
trace it roots to an article by R. Duncan 
Luce and John W. Tukey in 1964.2 Luce, 
a pioneer in decision theory, and Tukey, 
a leader in statistical analysis, developed 
a mathematical way to transform user's 
rankings of arbitrary combinations of at-
tributes into a scale of measurement, 
usually called utility. Thus, the impor-
tance of different features can be reex-
pressed on additive interval scales 
whose units are equivalent and whose 
zero point is arbitrary. The result is a 
model of the user's preferences when all 
the attributes are considered together. 
Such a model has ready application in 
many fields of the behavioral sciences. 
Marketing researchers have developed 
the technique of conjoint analysis, based 
upon the work of Luce and Tukey, and 
have used it heavily for the planning and 
development of new products. 

In library and information science re-
search, there have been only a few 
attempts to use conjoint analysis in 
the study of library services. 

Within marketing research, conjoint 
analysis is based in large part on the 
work of Paul E. Green. Over the last 
two decades, Green and his coauthors 
have published numerous articles on 
the use of conjoint analysis in consumer 
research.3 In 1971 Green and Vithala R. 
Rao introduced conjoint measurement 
as one method for quantifying judg-
mental data that could be useful for mar-
keting researchers.4 In a 1972 empirical 
study Green, Frank J. Carmone, and 
Yoram Wind concluded that additive 
conjoint models provided good descrip-
tions of data from their study even when 
compared to descriptions produced by 
more complex models.5 

By the late 1970s, conjoint analysis had 
spawned a growing body of theoretical 
literature, as reviewed and later up-
dated by Green and Srinivasan.6 Com-

May1994 

mercia! uses of conjoint analysis have 
been reviewed by Philippe Cattin and 
Dick R. Wittink.7 Cattin and Wittink 
identified 698 uses of the methodology 
by marketing research firms during the 
ten years following the first commercial 
project involving conjoint analysis in 1971. 
In their follow-up article, Wittink and Cat-
tin documented at least 1,062 projects us-
ing conjoint analysis over the five-year 
period from 1981 to 1985, and estimated 
that there may in actuality have been ap-
proximately 2,000 such projects. 

As an analytical technique, conjoint 
analysis strives to decompose or sepa-
rate a set of predetermined "stimuli," 
which are attributes of some item (e.g., a 
commercial product or service) so that 
the utility of each level of stimulus or 
attribute can be inferred from the overall 
evaluations given by subjects.8 The pro-
cedures used in particular conjoint 
analyses· vary, but they all maintain a 
factorial design in which all levels of 
each stimulus or attribute can be com-
pared to all levels of all other stimuli or 
attributes. 

Richard M. Johnson, cited by many 
later articles, used a variation of conjoint 
measurement to study the value systems 
of individual consumers.9 He called his 
method trade-off analysis. The basis of his 
argument rested on the assumption that 
the choice behavior of individual con-
sumers is governed by trade-offs that 
"may be revealed by choices among 
product concepts having characteristics 
which are varied in systematic ways." 10 
The result of such analysis is a model of 
preferences for product or service char-
acteristics stated in the form of utilities 
for each attribute. It is then possible to 
determine the optimal configuration of 
product characteristics by selecting the 
highest rated attributes. 

In library and information science re-
search, there have been only a few at-
tempts to use conjoint analysis in the 
study of library services. Michael 
Halperin and Maureen Strazdon exam-
ined preferences for reference services 
using conjoint analysis.11 They used 
eight factors: completeness and accu-
racy of answer (4levels), database serv-



A Conjoint Analysis of Reference Services 259 

ice (4levels), interlibrary loans (2levels), 
time needed to answer question (2 lev-
els), attitude of librarians (3 levels), 
hours of reference service (2 levels), 
knowledge of librarians (2 levels), and 
wait for service (2 levels). The authors 
concluded: "Conjoint analysis is a tech-
nique that allows us to quantify some of 
the seemingly intractable qualitative as-
pects of library service. In doing so it 
represents a new and potentially fruitful 
method of relating library services to 
user requirements."12 

In another article, Halperin discussed 
the potential of conjoint analysis to help 
inform library administrators of user 
preferences for information services.n 
Although they did not actually perform 
a study using conjoint analysis, Kenneth 
D. Ramsing and John R. Wish illustrated 
the use of the technique to determine the 
service preferences of library users, with 
their main example being online search-
ing.14 Thus, Halperin, Strazdon, and 
Ramsing and Wish have shown that con-
joint analysis has potential applicability 
to library and information science. This 
potential has yet to be tapped. 

In brief, conjoint analysis could be 
used to determine the attributes of ali-
brary or information service that are 
most important to library patrons. Utili-
ties or "part-worth" weights can be de-
rived for all levels of each attribute from 
reports by respondents completing the 
conjoint instrument. Attributes could then 
be segmented by different user groups in 
order to meet the desires of those groups 
as expressed by their preferences, for ex-
ample, by gender, age, institutional af-
filiation, or amount of library use. 

Since the derived part-worths or util-
ity scores are additive, it should be pos-
sible to determine the level of patron 
satisfaction with current library and in-
formation services by simply adding the 
utility scores of those levels of attributes 
that the services currently possess. This 
number could then be compared to the 
total of utility scores of alternative 
schemes of services. A library with a 
larger total could be said to meet better 
the preferences of users while a library 
with a lower total could be said to be 

deficient in some area or areas. The area 
or areas of deficiency could then be ame-
liorated or eliminated by striving to pro-
vide the service attributes preferred by 
patrons. Conjoint analysis, therefore, 
could provide a method for predicting 
patron satisfaction if current services are 
changed in specific ways. 

The major caveat about conjoint 
analysis is that it focuses on elicited con-
sumer preferences among hypothetical 
choices. Such preferences may or may 
not translate into behavior. For example, 
preference for less expensive database 
searching may not necessarily mean 
that, if the library lowered fees for 
searching, more patrons would use the 
service. In addition, altering services to 
meet patron desires may necessitate re-
structuring of funding, reassignment of 
staff, changing library hours, etc. These 
may negate the projected increases in 
utility to patrons. Finally, since patrons 
differ on sociodemographic and psy-
chographic variables, it may be difficult 
to change library or information services 
to meet the preferences of all patrons. 
Nonetheless, conjoint analytic tech-
riiques may be very beneficial in the 
marketing of library services to specific 
segments of the population or in estab-
lishing new services aimed at these spe-
cific user segments. It can help in 
decision making because it can provide 
data on patron preferences that were not 
obvious before. 

METHODOLOGY 

This research applied conjoint analy-
sis to the user preferences for reference 
service in the academic library, similar to 
the research of Halperin and Strazdon. 
Six attributes of reference services were 
selected for this study: (1) Definitiveness 
refers to the likelihood that an answer to 
a question can be found in rna terials 
given to a patron by a librarian; (2) 
Hours of service describes the times dur-
ing which a librarian is available to assist 
patrons; (3) Cost of service means the 
actual monetary charges that will have 
to be paid for interlibrary loan (ILL) and 
the searching of online databases (db 
searching); (4) In line wait time refers to the 



260 College & Research Libraries 

length of time patrons must stand in line 
or wait for a librarian to assist them; (5) 
Number of items indicates how many 
physical items (books, magazines, maps, 
etc.) a librarian helps a patron find or 
gives to a patron in order to answer his 
or her question(s); and (6) Service time 
refers to the length of time it takes a 
librarian to answer a question or assist a 
patron. The levels of each attribute are 
given in table 1. 

This study includes four attributes simi-
lar to those of Halperin and Strazdon, but 
utilizes a different number of levels for 
each: "Hours of service," "In line wait 
time" ("Wait for service" in Halperin 
and Strazdon), ''Definitiveness" (called 
"Completeness of answer"), and "Service 
time" (''Time needed to answer"). The 
"Cost of service" attribute incorporates 
two of theirs: database searching and in-
terlibrary loans. The rationale for this com-
bined attribute is that database searching 

TABLE 1 
REFERENCE SERVICES: 

ATIRIBUTES AND THEIR LEVELS 
1. Definitiveness 

(1) definite answer 
(2) possible answer 

2. Hours of Service 
(1) any time library open 
(2) specified times only 
(3) by appointment only 

3. Cost of Service 
(1) all services free 
(2) less than $5 for interlibrary loan 

and database searching 
(3) over $5 for interlibrary loan and 

database searching 
4. In Line Wait llme 

(1) less than 5 minutes 
(2) 5 to 15 minutes 
(3) more than 15 minutes 

5. Number of Items 
(1) 1 item 
(2) 2 to 5 items 
(3) 6 or more items 

6. Service Time 
(1) less than 5 minutes 
(2) 5 to 15 minutes 
(3) more than 15 minutes 

May1994 

and interlibrary loan are the two serv-
ices provided by the reference or public 
services department with which charges 
are often associated. Two attributes in 
the Halperin and Strazdon study, atti-
tude of librarians and knowledge of li-
brarians, are not included in the present 
study to avoid problems in patron 
judgments of librarian knowledge and 
attitude. One new attribute is included: 
number of items. The attributes in-
cluded in the present study can all be, at 
least to some extent, quantified or meas-
ured within the library; librarian knowl-
edge and attitude cannot be so easily 
quantified. 

To test for differences among students 
in academic institutions that vary in size 
and type, undergraduate students from 
three colleges and universities were sur-
veyed: Rutgers University, Moravian 
College, and Warren County (New Jer-
sey) Community College. In addition, 
Rutgers M.L.S. students taking a re-
search methods class completed the con-
joint instrument to provide comparative 
data for graduate students. All respon-
dents were asked the following demo-
graphic questions: age, gender, major 
field of study, and the number of times 
they visit the library each week. A total 
of 100 usable instruments were col-
lected. (See table 2 for a breakdown by 
institution.) Four instruments were in-
complete and, therefore, unusable. 
Three of these were from Warren County 
Community College and one was from 
Moravian College. All of the unusable 
instruments contained incomplete data 
for the trade-off matrices involving the 
cost of service attribute. 

Data analyses included t-tests for each 
level of every attribute to test for differ-

TABLE2 
BREAKDOWN OF 

SUBJECTS BY INSTITUTION 

Moravian College 32 

Rutgers (undergraduates) 40 

Rutgers (M.L.S.) 15 

Warren County Community College 13 

Total 100 



A Conjoint Analysis of Reference Services 261 

ences between the genders and one-way 
analysis of variance tests for all levels of 
each attribute to test for differences be-
tween respondents stratified by age, in-
stitution, and frequency of library use. 
The .05level of significance was used for 
all inferential statistical tests. Due to the 
total number oft-tests and ANOVAs in-
cluded in the analysis, chance alone may 
account for approximately three signifi-
cant results. 

Respondents included 77 women and 
23 men. Ages ranged from 18 to 46 (two 
subjects did not give their ages), with 66 
percent of the total respondents being 18 
to 21 years of age. Ages were grouped 
into the following spans: 18-20, 21-30, 
31-40, and 41 and above. For the institu-
tional AN OVA test, the groups are War-
ren County (New Jersey) Community 
College, Moravian College, Rutgers 
University undergraduates, and Rut-
gers University M.L.S. students. 

The respondents used the library an 
average of 2.3 times per week (std. dev. 
1.64), with a range of zero to ten times 
per week. Over half of the respondents 
(55) said that they used the library once 
or twice a week. Eight said that they 
never used the library or used it less than 
once per week. Thirty-six used the li-
brary three or more times a week. One 
respondent did not answer this ques-
tion. For an additional ANOVA test of 
preferences, the respondents were clas-
sified into three categories of library use: 
(1) less than one use per week (n = 8), (2) 
one to two uses per week (n =55), and 
(3) three or more uses per week (n = 36). 

Academic majors of the undergradu-
ate respondents represented twenty-
nine different majors or combinations of 
majors. No discernible pattern of majors 
was observed. All the Rutgers under-
graduate respondents were enrolled in a 
communications class and were major-
ing in either communication or a closely 
related field. The respondents from 
Mora vi an College and Warren County 
Community College were taking an in-
troductory psychology class. Moravian 
respondents represented fourteen dif-
ferent majors while the community col-
lege students reported six majors. 

The resulting sample of respondents 
constitutes a nonprobability, conven-
ience sample, drawn from classes whose 
instructors agreed to participate. The 
lack of random selection of subjects may 
decrease the external validity of the re-
sults of the study. The total sample, how-
ever, does consist of a wide variety of 
students as evidenced by the large num-
ber of majors, the differences in ages, the 
differences in institution types, and the 
differences in library usage. 

In the "full profile" approach, subjects 
are asked to rank descriptions of all 
possible combinations of levels of 
attributes of services or products. 

Because of the length of time it takes 
to complete, the "full/.rofile" conjoint 
approach was not use . In the full pro-
file approach, subjects are asked to rank 
descriptions of all possible combina-
tions of levels of attributes of services or 
products. One method of reducing the 
number of descriptions to be ranked is 
the use of orthogonal arrays of attributes 
in which combinations of factors are se-
lected in such a way as to balance the 
contribution of each factor. 15 Because of 
the ease of instrument construction and 
administration, the two-factor method 
was chosen for this study. 

The reference service preference data 
were collected by Johnson's two-factor-
at-a-time method. The six attributes be-
ing examined yield fifteen different 
trade-off matrices, each of which was 
printed on a half sheet of paper (8.5 
inches by 5.5 inches). See appendix A for 
a sample trade-off matrix from the con-
joint instrument. 

Respondents were given a data-collec-
tion instrument consisting of a cover 
sheet with directions, the fifteen trade-
off matrices assembled randomly for 
each instrument, and a final sheet asking 
the demographic questions. Respondents 
were directed to rank their preferences for 
the combinations of the levels of two at-
tributes represented in each trade-off ma-
trix on either a one-to-nine or a one-to-six 



Attribute 

Level: 

Moravian College 

Mean .51 

2 

.04 .59 

St. dev. .36 .21 .37 

Minimum 0 0 0 

Maximum 1.65 1.18 1.22 

Rutgers University-Undergraduate 

Mean .79 .03 .67 

St. dev. 1.38 .1 0 .35 

Minimum 0 0 0 

Maximum 7.99 .55 1.27 

Rutgers University-M.LS. Students 

2 

2 

.35 

.29 

0 

1.33 

0 

.34 

.25 

.78 

TABLE3 
SUMMARY UTILITIES BY INSTITUTION 

3 

.07 .91 

.22 .36 

0 0 

1.04 1.56 

.05 . . 93 

.16 .39 

0 0 
.93 1.49 

3 

2 

.52 

.24 

0 

1.32 

.47 

.24 

0 

1.11 

4 

3 2 

.07 .54 .31 

.27 .34 .16 

0 0 0 

1.34 1.17 .70 

.06 .64 .36 

.21 .29 .17 

0 0 0 

1.24 1.08 .62 

3 

.03 

.09 

0 

.40 

.05 

.18 

0 

.82 

.19 

.23 

0 

.70 

.14 

.23 

0 

.73 

s 
2 3 

.26 .31 

.18 .37 

0 0 

.65 1.12 

.27 .28 

.21 .27 

0 0 
.73 .91 

.46 

.27 

0 

.88 

.48 

.25 

0 

1.05 

6 

2 

.28 

.18 

0 

0 

.61 

.24 

.14 

.59 

3 

.08 

.17 

0 

.61 

.05 

.14 

0 

.54 

Mean .53 0 .77 .40 0 .99 

.25 

.46 

.51 0 .77 .41 .01 .15 .40 .30 .49 .29 .01 

St. dev. .21 0 .25 

Minimum 

Maximum 

.04 0 

.85 0 

.28 

1.09 

Warren County (NJ) Community College 

.20 0 

.05 0 

.69 0 

Mean .31 .03 .74 .30 .03 

St. dev. .31 .08 .42 .21 .11 

Minimum 0 0 0 .01 0 

.19 0 

.22 0 

1.26 .98 0 

.77 .42 .04 

.46 .27 .15 

0 0 0 

.24 .25 .04 .24 .24 .33 .28 .22 ,03 

.17 0 0 0 .03 0 0 0 0 

1.07 1.07 .15 .69 .85 .91 .92 .69 .08 

.51 .37 .02 .25 .22 .18 .46 .30 .06 

.40 .22 .05 .28 .17 .29 .34 .23 .17 

0 0 0 0 0 0 0 .03 0 
Maximum .82 .28 1.35 .66 .38 1.34 .71 . . 55 1.34 .66 .14 .75 .61 .88 .91 .86 .57 

Total Group 

Mean 

St. dev. 

Minimum 

Maximum 

.60 

.92 

0 

7.99 

.03 

.14 

0 

1.18 

.67 .35 

.35 .25 

0 0 
1.35 1.33 

05 

.16 

0 

1.04 

.91 .49 .05 

.37 .24 .21 

0 0 0 

1.56 1.32 1.34 

.61 .35 .04 .17 .28 .28 .47 .27 .06 

.32 .19 .13 .24 .21 .31 .27 .18 .15 

0 0 0 0 0 0 0 0 0 
1.34 .80 .82 .75 .85 1.12 1.05 .86 .61 



A Conjoint Analysis of Reference Services 263 

TABLE4 
RANGES AND RELATIVE IMPORTANCE OF EACH ATTRIBUTE 

Utility 
Attribute Scores 

Cost of service 

All services free .91 

Less than $5 .49 

$5 and over .05 

Hours of service 

Any time library is open .67 

Specified times only .35 

By appointment only .05 

In line wait time 

Less than 5 minutes .61 

5 to 15 minutes .35 

More than 15 minutes .04 

Definitiveness 

Definite answer .60 

Possible answer .03 

Service time 

Less than 5 minutes .47 

5 to 15 minutes .27 

More than 15 minutes .06 

Number of items 

1 item .17 

2-5 items .28 

6 or more items .28 

Total Range 

scale, matching the number of pairs to be 
compared. Completion of the instru-
ment took approximately twenty minutes. 

Utility calculations were performed 
for each respondent using the trade-off 
program, part of the PC-MDS package of 
programs written by Scott Smith of 
Brigham Young University.16 Other 
analyses were completed using SPS5-
PC+ Studentware software. 

RESULTS 

Table 3 presents summary utility data 
for each of the three institutions in-
volved in this research, with Rutgers 
University represented twice, once for 
undergraduates and once for M.L.S. stu-
dents. The mean, standard deviation, 
minimum, and maximum of the stu-
dents' utility scores are given for all lev-

%of Total 
Range Range Rank 

.86 27 1 

.62 20 2 

.57 18 3.5 

.57 18 3.5 

.41 13 5 

.11 4 6 

3.14 100 

els of each attribute. In addition, the 
means, standard deviations, minimum, 
and maximum utility scores are given 
for the entire sample. 

The ranges and relative importance 
for each attribute level are given in table 
4. This table reveals that the total range 
of scores is 3.14. The importance of each 
attribute is given by the percentage that 
it contributes to this total range. 

The cost of service is the single most 
important variable in the student per-
ception of academic reference services, 
accounting for 27 percent of the total 
range of scores. Most patrons prefer that 
all services be free, including interli-
brary loan and database searching. For 
comparison, in the Halperin and 
Strazdon study, cost of database service 
ranked second in importance and cost of 



264 College & Research Libraries May1994 

TABLES 
SUMMARY OF SIGNIFICANT DIFFERENCES IN UTILITY SCORES 

Group 

1. Gender 

2. Gender 

Attribute 

Definitiveness 

Number of items 

Level 

T-Tests 

2 (possible answer) 

3 (6 or more items) 

p 

2.36 .02 

2.57 .01 

d/f 

98 

98 

Analysis of Variance Tests 

3. Age Service time 

4. Age In line wait time 

5. Library Use Definitiveness 

6. Institution Variable = library use 

interlibrary loan ranked fifth. Together, 
these two factors represent 27.5 percent 
of the total range of scores yielded by 
their study. The second most important 
factor in the current research is the hours 
during which reference service is avail-
able (20 percent of total range), with ref-
erence service available at any time the 
library is open the preferred mode. This 
factor ranked fourth in Halperin and 
Strazdon, representing 10 percent of 
their total range. Tied for third place in 
the current study are definitiveness of 
the answer and in line wait time (18 
percent each). Students overwhelmingly 
prefer a definite answer to a possible 
answer and also prefer to wait less than 
five minutes for service. In comparison, 
Halperin and Strazdon report that 
"Completeness of answer" ranked first 
in their study (34.5 percent of total 
range) and "Wait for service" ranked 
eighth (3.6 percent). In fifth place, serv-
ice time accounts for 13 percent of the 
range, with most patrons preferring to 
have a librarian help them for less than 
five minutes. Halperin and Strazdon' s 
factor, "time needed for answer," ranked 
seventh. Number of items given to the 
patron ranks sixth and provides little 
influence on the perception of reference 
service, accounting for only 4 percent of 
the range. No similar factor was in-
cluded in Halperin and Strazdon. 

Assuming that the utility scores are 
additive, the highest possible mean 
score for all attributes taken at one time 

1 (less than 5 minutes) 2.71 .05 3/94 
1 (less than 5 minutes) 3.75 .01 3/94 
2 (possible answer) 3.51 .03 2/96 

3.89 .01 3/95 

is 3.54. The mean high utility score for 
each individual institution is as follows: 
3.32 for Moravian College, 3.79 for Rut-
gers undergraduates, 3.95 for Rutgers 
M.L.S. students, and 3.29 for Warren 
County Community College. Since these 
scores are known, it would be possible 
to examine the reference services pro-
vided by the library in each of the stud-
ied institutions in order to compare their 
current level of service to the level of 
service preferred by patrons as indicated 
in the derived utility scores. 

In addition, the utility scores can pro-
vide an indication of how satisfaction 
may change if an attribute of the service 
is changed. For example, if a library rou-
tinely charges a fee less than $5 for inter-
library loan or database searching, it can 
potentially increase patron satisfaction 
by .42 if these services could be provided 
free. On the other hand, if it became 
necessary for a library to charge for such 
services, overall satisfaction levels 
might be maintained if other attributes 
are changed so that their utility scores 
compensate for the potential reduction 
in satisfaction associated with costs. 

T-tests were run on all levels of each 
attribute to determine if any derived 
utility scores were significantly different 
between the genders. Analysis of vari-
ance (ANOVA) tests were performed on 
the utility scores to assess differences of 
individual institutions, age groups of re-
spondents, and library use groupings. 
T-tests and ANOVAs were also run on the 



A Conjoint Analysis of Reference Services . 265 

library use variable to determine differ-
ences by gender and by group of respon-
dents. A summary of the statistically 
significant results is given in table 5. 

Men indicated a small but signifi-
cantly higher preference for possible an-
swers (level 2 of the attribute 
"Definitiveness") (t = 236, p = .02, df = 98). 
The mean utility scores for both men (.09) 
and women (.01), however, were negli-
gible. Both genders overwhelmingly pre-
ferred a definite answer (level 1 of the 
attribute "Definitiveness"), with men giving 
it a utility score of 0.62 and women 0.59. 

The only other significant difference 
revealed by the gender-based t-tests 
showed that women report a higher 
mean utility score on level 3, "six or 
more items," of the attribute Number of 
items (t = 2.57, p = 0.01, df = 98). The mean 
utility score on this attribute level for 
women was 0.32 and for men 0.14. In 
general, . the utility scores show that 
women preferred six or more items 
(0.59) to either one item (0.16) or two to 
five items (0.29). Men, in contrast, had 
higher utility scores on both one item 
(0.23) and two to five items (0.25) than 
on six or more items (0.14). 

Frequent users of the library may 
be more savvy in obtaining the 
information they need and may be 
more aware of the variety of sources 
in which to find their answers. 

The analysis of variance tests showed 
no significant differences between the 
respondents grouped by institution on 
any of the levels of the attributes. There 
is, however, a significant difference in 
library use between the groups (F = 3.89, 
p = .01, df = 3/95). Post hoc Scheffe tests 
reveal that this result is due only to the 
difference between the use of the library by 
the Rutgers undergraduates (mean = 1.8) 
and the Rutgers M.L.S. students (mean = 
3.43). No other differences between the 
groups are statistically significant. 

Analysis of variance tests by age cate-
gories reveal significant differences on 
two levels of attributes: Ievell of "Serv-

ice time" and level 1 of "In line wait 
time." Level 1 of the attribute "Service 
time" indicates that the respondents pre-
ferred reference service at any time the 
library is open. ANOVA tests reveal a 
significant difference between the four 
age categories of this level (F = 2.71, p = 
.05, df = 3/94). Post hoc Scheffe tests, 
~owever, fail to reveal statistically sig-
nificant differences among any two 
separate groups. 

ANOVA tests result in a significant 
difference between the four age group-
ings on level 1 of the "In line wait time" 
attribute, i.e., a wait of less than five 
minutes (F = 3.75, p = .01, df = 3 /94). Post 
hoc Scheffe tests show a significant dif-
ference in the utility scores of those aged 
31 to 40 (mean= .83) and those aged 20 
and under (mean= .52). 

ANOVA tests were also run on the 
utility scores of each level of all attrib-
utes with respondents grouped by 
amount of library use. Significant results 
were obtained only on level2 ("possible 
answer") of the attribute "Definitive-
ness" (F = 3.57, p = .03, df = 2/94). Post 
hoc Scheffe tests indicated that the util-
ity scores of those respondents using the 
library zero times per week (mean = .15) 
differed significantly from those using 
the library three or more times per week 
(mean = .01). 

DISCUSSION 

The results of these analyses indicate 
that most college students agree on their 
preferences for academic reference serv-
ices. As one might expect, they prefer a 
definite answer, reference service at any 
time the library is open, all services free, 
a less than five minute wait for service, 
two or more items, and less than five 
minutes of actually working with ali-
brarian. Of these preferences, the most 
important factors, in ranked order, are 
the cost of the service, the hours of serv-
ice, the length of wait for service, the 
definitiveness of the answer given, and 
the amount of service given. The num-
ber of items has little overall impact on 
preference. 

Men and women express similar pref-
erences on most levels of attributes as 



266 College & Research Libraries 

indicated by their utility scores. Only 
two statistically significant differences 
emerged in the analyses. First, while 
women and men both overwhelmingly 
desire a definite answer, men are more 
accepting of a possible answer than 
women. Second, women prefer to be 
given more items than men, giving a 
utility score of 0.32 for six or more items 
compared to men's score of 0.14. Men 
prefer to be given one to five items. 

Contrary to expectations, the institu-
tional affiliation of respondents did not 
affect the utility scores. No significant 
differences in any levels of any attributes 
appeared in the analyses. A significant 
difference did emerge in the amount of 
library use with Rutgers M.L.S. students 
using the library significantly more than 
Rutgers undergraduates. 

Analyses by age categories also failed 
to show many significant differences. 
The only significant differences discov-
ered were between those aged 31 to 40 
and those 20 and under. Older individu-
als indicated a significantly higher pref-
erence, as shown by their utility scores, 
for quick service, i.e., they wished to 
wait in line for no more than five min-
utes. This may be a reflection of their 
situation in life with more responsibili-
ties. Most of the older students are em-
ployed full-time and are enrolled in 
school part-time. Thus, when they are at 
the library, they wish to have services 
readily availacle to them, and they do 
not wish to wait long for those services. 

Finally, only one significant difference 
was found when the respondents were 

May1994 

stratified by the amount of their library 
use. Those who used the library heavily 
(i.e., three or more times per week), gave 
a very low preference for a possible an-
swer when compared to those who used 
the library very few times (i.e., less than 
once per week). Frequent users of the 
library may be more savvy in obtaining 
the information they need and may be 
more aware of the variety of sources in 
which to find their answers. Overall, the 
results of this study indicate a fairly sta-
ble pattern of preferences for reference 
services in the academic library. 

CONCLUSION 

Conjoint analysis provides a method · 
for determining patron preference to 
guide librarians in structuring reference 
services in the academic library. While 
individual students will vary in their 
preferences for specific aspects of serv-
ice, the means of the utility scores de-
rived from the stated preferences of 
these individuals reveal which aspects 
of existing services should be empha-
sized or deemphasized. Similar research 
can be done to determine the preferences 
of other members of the academic com-
munity who may have different service 
preferences than undergraduates, espe-
cially faculty, administration, and 
graduate students. 

Librarians should adopt and adapt 
techniques from other fields such as mar-
keting in order to improve the effective-
ness of libraries. This study shows that · 
conjoint analysis can be useful and en-
courages us to search for other such tools. 

REFERENCES AND NOTES 
1. Susan S. DiMattia, "Dogs-or Stars?" RQ 31 (1992): 307-9. 
2. R. Duncan Luce and John W. Tukey, "Simultaneous Conjoint Measurement: A New Type of 

Fundamental Measurement," Journal of Mathematical Psychologtj 1 (1964): 1-27. 
3. Paul E. Green, "On the Design of Choice Experiments Involving Multifactor Alternatives," 

Journal of Consumer Research 1(1974): 61-68; Paul E. Green, "Hybrid Models for Conjoint 
Analysis: An Expository Review," Journal of Marketing Research 21 (1984): 155-69; Paul E. 
Green and Yoram Wind, "New Way to Measure Consumers' Judgments," Harvard Business 
Review 53 (July I Aug. 1974): 107-17; Paul E. Green and Abba M. Krieger, "Segmenting 
Markets with Conjoint Analysis," Journal of Marketing 55 (Oct. 1991): 20-31; Paul E. Green 
and V. Srinivasan, "Conjoint Analysis in Consumer Research: Issues and Outlook," Journal 
of Consumer Research 5 (1978): 103-23; and Paul E. Green and V. Srinivasan, "Conjoint 



A Conjoint Analysis of Reference Services 267 

Analysis in Marketing: New Developments with Implications for Research and Practice," 
Journal of Marketing 54 (Oct. 1990): 3-19. 

4. Paul E. Green and Vithala R. Rao, "Conjoint Measurement for Quantifying Judgmental 
Data," Journal of Marketing Research 8 (1971): 355-63. 

5. Paul E. Green, Frank J. Carmone, and Yoram Wind, "Subjective Evaluation Models and 
Conjoint Measurement," Behavioral Science 17 (1972): 288-99. 

6. Green and Srinivasan, "Conjoint Analysis in Consumer Research" and Green and Srini-
vasan, "Conjoint Analysis in Marketing." 

7. Philippe Cattin and Dick R. Wittink, "Commercial Use of Conjoint Analysis: A Survey," 
Journal of Marketing 46 (Summer 1982): 44-53; and Dick R. Wittink and Philippe Cattin, 
"Commercial Use of Conjoint Analysis: An Update," Journal of Marketing 53 (1989): 91-96. 

8. Green, Carmone, and Wind, 1972; Paul E. Green and D. S. Thll, Research for Marketing 
Decisions, 4th ed. (Englewood Cliffs, N.J.: Prentice-Hall, 1978); and Green and Wind, ''New 
Way to Measure Consumers' Judgements." 

9. Richard M. Johnson, "Trade-off Analysis of Consumer Values," Journal of Marketing Re-
search 11 (1974): 121-27. 

10. Ibid., 121. 
11. Michael Halperin and Maureen Strazdon, ''Measuring Students' Preferences for Reference 

Service: A Conjoint Analysis," Library Quarterly 50 (1980): 208-24. 
12. Ibid., 223. 
13. Michael Halperin, "Determining User Preferences for Information Services," Drexel Library 

Quarterly 17 (Spring 1982): 88-98. 
14. Kenneth D. Ramsing and John R. Wish, ''What Do Library Users Want? A Conjoint Meas-

urement Technique May Yield the Answer," Information Processing and Management 18 
(1982): 237-42. 

15. See Green, "On the Design of Choice Experiments." 
16. Scott Smith, PC-MDS Multidimensional Statistics Package [Computer program] (Provo, 

Utah: Scott Smith, 1990). 

APPENDIX A 
Sample Trade-off Matrices from Conjoint Instrument 

"Cost of Service" means the actual monetary charges that you will have to pay for 
interlibrary loan (ILL) and the searching of online databases (db searching). "In Line 
Wait Time" means how long you must stand in line or wait for a librarian to assist 
you. 

Cost of Service 

Less than $5 for ILL $5 and over for ILL 
All services free and db searching and db searching 

Less than 5 minutes 

5 to 15 minutes 

More than 15 
minutes