■ 
 
 TN295 
 
 No. 9178 
 
 Ural 
 
 mm 
 
 
 LIBRARY OF CONGRESS 
 
 0000110^2^7 
 
 «^» *»■«&* *' 
 

 ° / 
 
 • "<r** -•jqp/ jf\ °°wm : * X m' ,/v -.™t , 
 
 
 
 
 ,4°* 
 
 «S 
 
 . V,** .'^V MM. **i .'M V/ -to- \ rf v '».\ .^.S 
 
 
 .v v y> • * • * <v> *->, o • » 
 
 
 
 
 ^ <"& 
 
 bV" 
 
 
 -4<& 
 
 ""';% /-»\ AM\ /^" 
 
 
 
 o «o 
 
 
 
 
 
 
 
 
 
 
 
 
 ^2 
 
 ,^ A 
 
 ^ 
 
 ^w^^. .^^«fe-V y.:»A ^-3«^,V ..# 
 
 
 
 
vSPv \l|Pv viP/ vSPV %^?v vS^v 
 
 •J-. \> J>>X&&% 4^«:ffifcX 4*'*i&% ♦^•tffo.X ^tfak-** / 
 
 , * » • . 
 
 
 of % 
 
 *•«- *6 
 
 &\. 
 
 
 *»»i 
 
 * 4/ >£ * 
 
 -"V 
 
 SirX A Aii3%X ^otfte.** V^iii'&X ^oSifc,^ .«>\^i.X ^ 
 
 
 -<**.4Kfc^ ^4mkS» j^JSSS^ J?titite.\> «* .*Sfc * /**m&* 
 
 lo -o vSHK' <>**% : XH3" ^°<\> va«: *°% -JOB* i\ v^R*' /^ -Ji|^ 
 
 ***% 
 
 
 
 
 
Bureau of Mines Information Circular/1988 
 
 A Catastrophe-Theory Model for 
 Simulating Behavioral Accidents 
 
 By William E. Souder 
 
 UNITED STATES DEPARTMENT OF THE INTERIOR 
 
Information Circular 9178 
 
 li 
 
 A Catastrophe-Theory Model for 
 Simulating Behavioral Accidents 
 
 By William E. Souder 
 
 UNITED STATES DEPARTMENT OF THE INTERIOR 
 Donald Paul Hodel, Secretary 
 
 BUREAU OF MINES 
 
 David S. Brown, Acting Director 
 
L 
 
 Library of Congress Cataloging-in-Publication Data 
 
 Souder, William E. 
 
 A catastrophe-theory model for simulating behavioral accidents. 
 
 (Information circular / United States Department of the Interior, Bureau of Mines ; 9178 ) 
 Bibliography: p. 18-19 
 Supt. of Docs, no.: 128.27: 
 
 1. Mine accidents— Psychological aspects. I. Title. II. Series: Information circular (United 
 States. Bureau of Mines) ; 9178 . 
 
 TN295.U4 
 
 [TN311] 
 
 622 s 
 
 [622'.8] 
 
 87-600371 
 
CONTENTS 
 
 Page 
 
 Abstract 1 
 
 Introduction 2 
 
 Background 2 
 
 Focus and scope of this study 3 
 
 Methodology 3 
 
 Accident sample 3 
 
 Variables and rating scales 3 
 
 Content analyses measurements 3 
 
 Content analyses results 4 
 
 Data reduction 4 
 
 Univariate analyses 5 
 
 Contributing factors 5 
 
 Statistical analyses of significance 6 
 
 Multivariate analyses 6 
 
 Path analyses 7 
 
 Systems model of behavioral accidents 7 
 
 Behavioral accidents 7 
 
 Relative importance of the variables 7 
 
 Cascading network effects: an example 8 
 
 Some implications 9 
 
 Simulating accident causes 9 
 
 Network preprocessors 10 
 
 Psychological conditions (PC) 11 
 
 Behavioral conditions (BC) 11 
 
 Supervisor ability (FA) and management concern 
 
 for safety (MC) 12 
 
 Environmental conditions (EC) 12 
 
 Adjustive behaviors (JB) 12 
 
 Cusp catastrophe model 12 
 
 Theory and example 12 
 
 Application to behavioral mine accidents 13 
 
 Behavioral dynamics 13 
 
 Behavioral accident simulator (BAS) 14 
 
 Illustrative application of the BAS 14 
 
 Frank: a hypothetical case 14 
 
 Coding this case into the BAS 14 
 
 Results from the BAS 15 
 
 Philosophy of using the BAS 15 
 
 Laboratory tests of the BAS 16 
 
 Field tests of the BAS 17 
 
 Summary and conclusions 18 
 
 Recommendations for further research 18 
 
 References 18 
 
 ILLUSTRATIONS 
 
 1. Behavioral accident model 2 
 
 2. Path analysis network 7 
 
 3. Results of network calculations 10 
 
 4. Network with preprocessors 11 
 
 5. Cusp catastrophe model 12 
 
 6. Effects of various BAS scenarios 15 
 
 7. Effects of stress (SLE) 16 
 
 8. Effects of supervisor abilities (FA) 16 
 
 9. Effects of various behavioral profiles (BC) 16 
 
 TABLES 
 
 1. Summary list of variables 4 
 
 2. Example of a rating scale for the norm variable 4 
 
 3. Statistically significant variables 5 
 
 4. Nonsignificant variables 5 
 
 5. Contributing factors 6 
 
 6. Pairwise data matrix 6 
 
 7. Numbers of entering and exiting paths for each variable in figure 1 8 
 
 8. VP,, data 9 
 
 9. Network calculations 10 
 
 10. Preprocessor variable definitions 11 
 
 11. Selected results from BAS experience questionnaires 17 
 
 12. Selected results from pre- and post-BAS questionnaires for BAS users who did not believe in training 17 
 

A CATASTROPHE-THEORY MODEL FOR 
 SIMULATING BEHAVIORAL ACCIDENTS 
 
 By William E. Souder ' 
 
 ABSTRACT 
 
 Behavioral accidents are a particular type of accident. They are caused by inappro- 
 priate individual behaviors and faulty reactions. Catastrophe theory is a means for 
 mathematically modeling the dynamic processes that underlie behavioral accidents. 
 Based on a comprehensive data base of mining accidents, a computerized catastrophe 
 model has been developed by the Bureau of Mines. This model systematically links 
 individual psychological, group behavioral, and mine environmental variables with 
 other accident causing factors. It answers several longstanding questions about why 
 some normally safe behaving persons may spontaneously engage in unsafe acts that 
 have high risks of serious injury. Field tests with the model indicate that it has three 
 important uses: It can be used as an effective training aid for increasing employee 
 safety consciousness; it can be used as a management laboratory for testing decision 
 alternatives and policies; and it can be used to help design the most effective work 
 teams. 
 
 1 Operations research analyst, Pittsburgh Research Center, Bureau of Mines, Pittsburgh, PA (now professor of industrial engineering and director of the 
 technology management studies institute, The University of Pittsburgh). 
 
INTRODUCTION 
 
 The following are four examples of normally safe- 
 behaving persons who suddenly stepped out of character 
 and knowingly committed unsafe acts. Their acts had dis- 
 astrous consequences for them. 
 
 1. Three experienced divers and life-saving instructors 
 ignored normal safety procedures to go on a night dive, 
 incompletely equipped, in an unexplored underwater cave. 
 All three drowned. 
 
 2. An experienced 45-yr-old supervisor assisting a 
 work crew suddenly turned and walked into the path of the 
 crew's bulldozer and was crushed before anyone could stop 
 him. 
 
 3. A 40-yr-old senior electrician who was completing 
 his work shift suddenly swung an iron wrecking bar over 
 his shoulder and abruptly turned to return it to the 
 toolcrib. The end of the bar struck the 400,000-V overhead 
 cable he had just installed, killing him instantly. 
 
 4. An experienced mine employee knowingly walked 
 under a bad roof, pointing out various roof flaws to his 
 companion. He was crushed by a sudden fall of that roof. 
 
 What caused these reckless behaviors? These were not 
 rational acts: there were few rewards and enormous per- 
 sonal risks in these acts. These were not inexperienced 
 
 and untrained personnel. They were mature, intelligent, 
 and responsible individuals. They behaved safely all their 
 lives, espoused safe behaviors, and served as role models 
 for their peers. The puzzling, unanswered question re- 
 mains: why did they do these things? 
 
 These are examples of a particular type of accident: the 
 behavioral accident. In a behavioral accident, the primary 
 cause is an inappropriate reaction or maladjustive behav- 
 ior of the individual to external stimuli. Behavioral acci- 
 dents involve complex interactions among individual 
 perceptions, attitudes, personalities, values, tolerances, 
 prior experiences, and work environments. As the four 
 cases suggest, individual and group phenomena may con- 
 tribute to behavioral accidents. Elements of carelessness, 
 inattention, thoughtlessness, poor habits, machoism, 
 bandwagon effects, and thrill seeking are suggested 
 within these cases. Because of their many causal factors, 
 the diagnosis and prevention of behavioral accidents is 
 often elusive and difficult. 
 
 This Bureau of Mines report describes research to de- 
 fine, empirically measure, and model behavioral acci- 
 dents. Perhaps a greater degree of understanding of the 
 phenomenon can lead to its prevention. 
 
 BACKGROUND 
 
 Human error often results from a mismatch between 
 individual capacities and workloads. Overloading employ- 
 ees beyond their capacities can set up stresses that cause 
 them to make errors. Conversely, underloading employees 
 may not arouse them sufficiently, causing them to make 
 errors as a result of boredom and inattention. 
 
 However, real-life situations are much more complex 
 than these simple statements might imply. Everyone's ca- 
 pacity for work differs. Moreover, people's capacities may 
 change as a result of their most recent experiences, daily 
 variations, and other factors. One of these factors is the 
 individual's perception of an overload. For example, it may 
 not matter what the ergonomic standards say: if a person 
 perceives he or she is overloaded then these perceptions 
 will guide his or her behavior. Since individual percep- 
 tions can be highly variable, the capacity-workload equi- 
 librium is likely to be correspondingly variable. To further 
 complicate matters, some persons can adjust to work over- 
 loads while others cannot. For example, some experienced 
 automobile drivers automatically adjust to fatigue and ad- 
 verse road conditions by increasing their intensity of con- 
 centration and alertness. But not everyone can so easily 
 adjust their behaviors in this fashion. Thus, the capacity- 
 workload equilibrium is a complex, dynamic, individual 
 phenomenon (14, 18). 2 
 
 Figure 1 depicts the system of factors that have been 
 found to relate to human errors (14, 18). If the capacity- 
 workload equilibrium is disturbed by some combination of 
 the factors shown in figure 1, a human error may occur. 
 Whether or not an error does in fact occur is a function of 
 the individual's adjustive behaviors. Individuals who ac- 
 
 2 Italic numbers in parentheses refer to items in the list of references at 
 the end of this report. 
 
 cordingly adjust their behaviors to changes in supervision, 
 work groups, external stimuli, and the other factors shown 
 in figure 1 may moderate the effects of these changes, 
 thereby avoiding a human error. 
 
 Note that even if a human error does occur it will not 
 necessarily lead to an accident, as illustrated in figure 1. 
 The external conditions may not be right for an accident to 
 happen. For example, adjustive behaviors may enable a 
 driver to compensate for a slick spot on the highway, thus 
 avoiding skidding into an oncoming automobile. But if 
 there is no oncoming automobile, then conditions are not 
 right for a collision and adjustive behaviors are relatively 
 less important. 
 
 Factors related 
 
 to the 
 work environment 
 
 Perceptions 
 
 External stimuli 
 
 External conditions . 
 
 Factors related Factors related 
 
 to the to the organization 
 
 individual ^ and workgroup 
 
 Capacity- workload 
 equilibrium "^v 
 
 It 
 
 Adjustive 
 behaviors 
 
 "■"» Factors related to the 
 management and supervision 
 
 Human error- 
 
 -■-No accident 
 
 "Confluence ■ 
 
 I 
 
 Accident 
 
 Figure 1. — Behavioral accident model. 
 
FOCUS AND SCOPE OF THIS STUDY 
 
 The focus of this study was on individual adjustive 
 behaviors and their relationships with the variables sys- 
 tem shown in figure 1. Why, how, and when individuals 
 successfully adjust and the consequences of failures to ad- 
 just were the topics of this research. 
 
 The adjustive behaviors of underground miners were 
 studied by analyzing a sample of fatal-accident reports. 
 
 Based on these analyses, and the use of catastrophe theory 
 concepts (20, 22), a computer model was constructed that 
 simulates behavioral accident systems. The model was 
 tested and evaluated by a sample of mine employees. Rec- 
 ommendations were made for routinely using the model 
 within mine operating firms. 
 
 METHODOLOGY 
 
 ACCIDENT SAMPLE 
 
 VARIABLES AND RATING SCALES 
 
 Mine Safety and Health Administration (MSHA) re- 
 ports of fatal underground bituminous coal mine accidents 
 from April 1979 to March 1985 were selected as the target 
 population for the study of behavioral accidents. The 
 choice of this target population was a compromise. Acci- 
 dent reports for earlier periods frequently lacked the nec- 
 essary detailed information. On the other hand, a long 
 enough time span was needed to cover a range of economic 
 conditions. And it was desirable to include the most recent 
 time periods in order to capture the latest conditions 
 within the industry. 
 
 This target population was refined by selectively re- 
 moving various reports. Reports with inadequate writeups 
 or incomplete information were removed. Since the focus 
 was on individual psychological and behavioral phenom- 
 ena, reports involving multiple fatalities, equipment fail- 
 ures, and inadequate training or inexperience of the 
 victim were removed. Reports from mines with injury 
 rates over 50 pet above the industry average were also 
 removed. The objective was to obtain a population where 
 many of the traditional accident causes were absent, yet 
 fatalities still occurred because of inappropriate individ- 
 ual adjustive behaviors. 
 
 These procedures resulted in a population of 358 fatal 
 accident cases. Since this population was too large to be 
 thoroughly studied within the available staffing and time 
 constraints, sampling was used. The population was strati- 
 fied by accident type, victim skill class, geographic loca- 
 tion of the mine, and mine size. Accident reports were then 
 randomly sampled from each stratum in numerical pro- 
 portion to their occurrence in the 358 accident case popu- 
 lation. 
 
 These procedures resulted in 60 fatal-accident cases 
 for study. The stratification and random sampling insured 
 that a range of important phenomena were present within 
 a representative sample of 60 cases. As noted later in this 
 report, these careful sampling procedures permitted the 
 behavioral accident phenomena to be distinguished from 
 the multitude of other confounding causes and factors. Be- 
 havioral accident phenomena are likely to be an impor- 
 tant component of most accidents. However, their presence 
 may remain undected because they are obscured by a mul- 
 titude of other causes. 
 
 Based on an in-depth analysis of selected literature (5, 
 12, 14, 18), the 20 variables listed in table 1 were chosen 
 for this study. Each of these variables relates to accident 
 phenomena with individuals, and each is reported to be 
 empirically measurable (5, 12, 14, 18, 21). 
 
 Some of the variables in table 1 are objectively mea- 
 surable; examples are age, experience, and size. Others 
 are highly subjective, such as carefulness, alertness, and 
 confidence. Many of these variables can be measured by 
 observation, supervisor's ratings, peer ratings, or person- 
 nel records. Others require carefully standardized rating 
 scales, for example, field dependency. The so-called field- 
 dependent individual is unable to extract salient informa- 
 tion from a complex background (12, 21). For example, an 
 individual who is unable to distinguish a zebra that is 
 standing in front of a striped background evidences field 
 dependency Standard tests have been developed for mea- 
 suring degrees of field dependency (21 ). 
 
 Rating scales were devised for measuring each varia- 
 ble, as illustrated in table 2. In pilot tests of interrater 
 reliability, a panel of 10 qualified judges applied these 
 scales in nine different exercises. After a brief training 
 and learning period, the judges ratings showed no statisti- 
 cally significant differences (using Cochran Q, binomial, 
 and kappa statistical tests of agreement (11, 17)). More- 
 over, in various trials of the scales, the author correctly 
 repeated the results with 94 pet accuracy. Thus, it appears 
 that the rating scales are highly reliable and repeatable 
 (11, 17-18). 
 
 CONTENT ANALYSES MEASUREMENTS 
 
 Each of the 60 MSHA reports was carefully read and 
 summarized to highlight its contents. In following estab- 
 lished procedures (1, 8), each report was then reread and 
 rated by the author on each of the variables listed in table 
 1, using scales like the one illustrated in table 2. While 
 this approach of reading and rating text based on the read- 
 er's impressions may be open to some arbitrariness, it is a 
 seriously accepted methodology. Two other readers who 
 were trained in content analyses methods reproduced the 
 author's ratings with 90- to 96-pct accuracy, using random 
 
Table 1.— Summary list of variables 
 
 Name 
 
 Definition 
 
 Number 
 
 INDIVIDUAL FACTORS 
 
 Age 
 
 Safety 
 
 Carefulness 
 Initiative . . . 
 Alertness . . 
 
 Evasiveness 
 
 Training 
 
 Field dependency 
 
 Self-control 
 Impulsivity 
 Experience 
 
 Victim's age 
 
 Percentage of safe behaviors demonstrated by victim 
 
 Extent to which victim showed carefulness in task behaviors 
 
 Extent to which victim demonstrated safety initiatives 
 
 Extent to which victim correctly observed danger signals that preceded 
 
 accident. 
 Extent to which victim acted to avoid or evade a potential accident situation 
 
 Recentness of training received by victim 
 
 Extent to which victim was psychologically field dependent for perceptual 
 
 information processing. 
 
 Extent to which victim maintained restraint over emotions 
 
 Extent to which victim demonstrated impulsive behaviors 
 
 Victim's level of experience 
 
 'Identification number assigned for subsequent discussion and analysis. 
 
 Table 2.— Example of a rating scale for the norm ' variable 
 
 Indicators Rating 
 
 Supervisor cautioned crew members to be aware of poor roof High safety norm or + . 
 
 Supervisor admonished crew to constantly check to see that cables were neatly Do. 
 
 stowed. 
 
 Supervisor frequently held informal safety meetings Do. 
 
 Supervisor stopped the work to remove a possible hazard Do. 
 
 Supervisor did not hold any regular safety meetings Low safety norm or - . 
 
 Supervisor seldom stressed carefulness and safety Do. 
 
 Supervisor often took chances and behaved carelessly Do. 
 
 Supervisor permitted crew to take shortcuts Do. 
 
 Inadequate information provided about supervisor's safety norm. Inadequate data or 0. 
 1 See table 1 for definition. 
 
 3 
 5 
 6 
 
 7 
 8 
 
 9 
 12 
 14 
 
 15 
 16 
 19 
 
 WORK ENVIRONMENT 
 
 Commitment 
 
 Extent to which firm demonstrated commitment to safety 
 
 10 
 
 Size 
 
 Size of firm 
 
 13 
 
 Rate 
 
 Injury rate at firm 
 
 20 
 
 Policy 
 
 Number of safety policies promulgated by firm 
 
 17 
 
 ORGANIZATION AND WORK GROUP 
 
 Integration 
 
 Confidence 
 
 Degree to which victim was integrated with work group 
 
 Degree of confidence in crew members shown by supervisor 
 
 4 
 18 
 
 
 
 
 MANAGEMENT AND SUPERVISION 
 
 Attitude 
 
 Top management attitude toward safety 
 
 1 
 
 Norm 
 
 Enforcement 
 
 Safety norm of immediate supervisor 
 
 Extent to which safety policies were enforced 
 
 2 
 11 
 
 samplings of text from the reports. This is a relatively 
 high interrater statistic that lends more confidence to the 
 results presented here. Additional standard precautions 
 were also taken to increase the validity of the results (1, 8, 
 19). 
 
 In addition to the content rating data, various contrib- 
 uting factors, such as failure to comply with safe operating 
 procedures and failure of management, were often cited in 
 the MSHA reports by the investigating teams. These 
 items were carefully noted and recorded for further analy- 
 ses. 
 
 CONTENT ANALYSES RESULTS 
 
 DATA REDUCTION 
 
 The content analyses produced a string of 60 + , - , or 
 scores (one for each accident case) for each variable listed 
 in table 1. Five of the twenty variables were then elimi- 
 nated from further consideration because their degree of 
 causal involvement (DCI) was too low. The ith variable was 
 eliminated when DCI„ defined as 
 
 was less than or equal to 0.60. Here, N,(0) is the number of 
 times the ith variable was rated for inadequate data in 
 the content analyses (see table 2). Equation 1 effectively 
 eliminates any variables that could not clearly be scored 
 either + or - in at least 60 pet of the accident cases in the 
 content analyses. Although this was a rather intuitive ap- 
 proach to data reduction, it was effective. The five varia- 
 bles thus eliminated were variables 1, 4, 5, 17, and 18 from 
 table 1. 
 
 [60 - N,(0)]/60 
 
 (1) 
 
UNIVARIATE ANALYSES 
 
 Of the 15 variables that survived the degree of causal 
 involvement test, 9 occurred often enough among the 60 
 accident cases to be statistically significant. These results 
 are summarized in table 3. Thus, the accident cases exam- 
 ined were characterized by some problem or some defi- 
 ciency in these nine aspects. That is, low supervisor safety 
 norms, carelessness, low safety initiatives, lack of alert- 
 ness, poor evasiveness, poor enforcement, high field depen- 
 dency, poor self-control, and impulsivity characterized the 
 cases. These results are consistent with the conventional 
 wisdoms about accident causation (14, 18). 
 
 On the other hand, table 4 presents results that are 
 not consistent with the conventional wisdoms. The con- 
 ventional wisdoms hold that youthful employees, weak 
 safety commitments by the firm, lack of employee train- 
 ing, lack of employee experience, large firm size, and an 
 environment of high injury rates are primary causes of 
 fatalities (14, 18). As the results in table 4 show, the sam- 
 ple of fatal accidents examined was not characterized by 
 these attributes. The sample was purposely selected in 
 such a way that these attributes were removed from it (see 
 "Accident Sample" section). The victims were not young, 
 inexperienced, and deficient in training. Over half of the 
 firms were large, accident rates at the firms were not 
 above average, and management commitments to safety 
 were strong. Nevertheless, fatal accidents occurred. 
 Clearly some other factors must have caused the accidents 
 studied. 
 
 These results thus support the central thesis of this 
 study: the victim's own inadequate adjustive behaviors 
 can be the primary cause of an accident. Such behavioral 
 accidents can occur in spite of the fact that other variables 
 and factors all point to a generally safe, potentially 
 accident-free environment. 
 
 CONTRIBUTING FACTORS 
 
 Table 5 shows the incidence, \, and statistical signifi- 
 cance of the jth contributing factors cited in the reports by 
 the investigating teams. The incidence is given by 
 
 I, = (NC/60), 
 
 (2) 
 
 where NC, is the number of times the jth contributing 
 factor was cited. As with equation 1, this is a rather intui- 
 tive approach to reducing the data. 
 
 As table 5 shows, the incidence of failures of manage- 
 ment (failure to enforce safety commitments made by the 
 firm, failure to eliminate known hazards, etc.) and failures 
 to comply with approved safe operating procedures were 
 statistically significant. That is, the investigators cited 
 these factors a significant number of times. Similarly, as 
 table 5 shows, the investigating teams cited faulty em- 
 ployee judgments and lax supervisors, who permitted un- 
 safe practices, as significant contributing factors. These 
 results are consistent with the conventional wisdoms. 
 These are precisely the factors that research has repeat- 
 
 Table 3.— Statistically significant variables ' 
 
 Name 
 
 Comments 
 
 Number 2 
 
 Norm 
 
 Carefulness 
 
 Initiative 
 
 In 76 pet of fatalities, immediate supervisor evidenced a low safety 
 norm. 
 
 In 80 pet of fatalities, victims did not show careful behaviors in performing 
 various tasks. 
 Safety initiatives were absent in 80 pet of fatalities 
 
 2 
 
 6 
 
 7 
 
 Alertness 
 
 Evasiveness 
 
 Enforcement 
 
 Field dependency 
 
 In 86 pet of fatalities, victims apparently failed to correctly observe 
 pertinent danger signals that preceded the accident. 
 
 In 98 pet of fatalities, victims failed to act to avoid or evade pending 
 danger. 
 
 In 81 pet of fatalities, supervisor did not enforce established safety rules 
 and practices. 
 
 In 95 pet of fatalities, victims appeared unable to extract salient informa- 
 tion from a complex background (high field dependency) 
 
 In 89 pet of fatalities, victims evidenced low self-control 
 
 8 
 9 
 
 11 
 14 
 
 Self-control 
 
 15 
 
 Impulsivity 
 
 In 79 pet of fatalities, victims demonstrated impulsive behaviors, with 
 little foresight into consequences and little regard for personal safety. 
 
 16 
 
 'Binomial statistical test, with level of significance for rejection set at 0.10. See reference 17 (pp. 36-42) for binomial test used. 
 Note that in this test, N = N, ( + ) + N,( - ), where N, ( + ) and N,( - ) are the number of times the \th variable was scored + and - , 
 respectively. 
 
 -Identification number assigned for subsequent discussion and analysis. 
 
 Table 4. — Nonsignificant variables ' 
 
 Name 
 
 Comments 
 
 Number 2 
 
 Age 
 
 Commitment 
 
 Training 
 
 Size 
 
 Experience 
 
 Rate 
 
 About half (45 pet) of victims were over 35 yr old, thus youthfulness was 
 
 not a significant factor in the fatalities studied. 
 Over half (52 pet) of cases were characterized by a strong safety 
 
 commitment by firm. 
 In over half ;51 pet) of the fatalities, victims had received formal training 
 
 for job within prior 3 months. Thus, recentness of training did 
 
 not deter the fatalities studied. 
 Over half (52 pet) of firms studied were large size (annual outputs above 
 
 industry mean). 
 Over half (59 pet) of victims had more than 10 yr experience 
 
 3 
 10 
 12 
 
 13 
 19 
 
 Over half (53 pet) of sites had injury rates below industry mean 
 
 20 
 
 'Binomial statistical test, with level of significance for rejection set at 0.10. See reference 17 (pp. 36-42) for binomial test used. 
 Note that in this test, N = N, ( + ) + N,( - ), where N, ( + ) and N,( - ) are the number of times the \th variable was scored + and - , 
 respectively. 
 
 identification number assigned for subsequent disoussion and analysis. 
 
Table 5.— Contributing factors 
 
 Factor 
 
 Number 2 
 
 I, 
 
 Significance 3 
 
 Failure of management 
 
 Failure to comply with approved safe operating procedures 
 
 Employees used unsafe judgment 
 
 Supervisors permitted unsafe practices 
 
 Chance events 
 
 Equipment failures 
 
 Faulty equipment designs 
 
 Desire for output at the expense of safety 
 
 21 
 22 
 27 
 28 
 23 
 24 
 25 
 26 
 
 0.500 
 .783 
 .800 
 .600 
 .133 
 .250 
 .317 
 .233 
 
 0.01 
 
 < .01 
 
 < .01 
 .01 
 NS 
 NS 
 NS 
 NS 
 
 NS Not statistically significant. 
 
 1 Each accident may have more than 1 contributing factor, 
 identification number assigned for subsequent discussion and analysis. 
 3 Binomial statistical test, with level of significance for rejection set at 0.10; x 
 
 NC, N = 60 in the binomial test. 
 
 edly shown to be accident contributors. When these items 
 are present, they foster other events and behaviors that 
 cause accidents (18). 
 
 The last four factors listed in table 5 were not cited a 
 statistically significant number of times. One of these fac- 
 tors was chance events. Chance events or acts of nature 
 were cited in only about one-sixth of the cases. Two other 
 factors, equipment failures and faulty equipment designs, 
 were cited in fewer than one-third of the cases examined. 
 The fourth factor, emphasis on output at the expense of 
 safety, was cited in less than one-fourth of the cases. Thus, 
 
 neither chance events, equipment failures, faulty equip- 
 ment design, nor output pressures were significant con- 
 tributors to the fatal accidents studied. This result runs 
 counter to some prevalent beliefs that these factors are 
 common causes of accidents (13, 18). However, they did not 
 cause the accidents studied. Rather, the accidents studied 
 appear to have been caused by something else. In this 
 sense, these results provide further support for the thesis 
 that at least some serious accidents are the result of poor 
 adjustive behaviors of individuals. 
 
 STATISTICAL ANALYSES OF SIGNIFICANCE 
 
 MULTIVARIATE ANALYSES 
 
 The preceding procedures resulted in a total of 13 var- 
 iables and factors that characterized the accident cases. 
 How do these variables and factors interrelate as a system 
 of accident causes? 
 
 In answer to this question, table 6 presents the results 
 from a binomial test of the pairwise interrelatedness of the 
 statistically significant items from tables 3 and 5. The 
 statistical significance of a relationship between any pair 
 of variables was determined from the binomial equation 
 for p(x): 
 
 p(x) =(N)P»Qn-». (3) 
 
 Here P = Q = 1/2, x is the frequency of content score 
 mismatches between that pair of variables, N is the total 
 number of pairs of nonzero scores for that variable pair 
 from the content analyses, and p(x) is the probability of 
 occurrence of x. A content score mismatch is the case of a 
 + score in one variable and a - score in the other variable 
 from the content analyses. Only values of p(x) less than or 
 equal to 0.10 are acceptable. For more details on the bino- 
 mial equation and the binomial statistical test see Siegel 
 (17). 
 
 As may be seen from table 6, some of the variables 
 were found to be unrelated (blank cell), some were not 
 statistically significantly related, and others were statisti- 
 cally significantly related (values of 0.10 or less). Many of 
 
 
 
 
 
 Table 6.- 
 
 -Pairwise data matrix' 
 
 
 
 
 
 
 Variable 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 or 
 
 2 
 
 6 
 
 7 
 
 8 
 
 9 
 
 11 
 
 14 
 
 15 
 
 16 
 
 21 
 
 22 
 
 27 
 
 28 
 
 factor 2 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 2 
 
 
 9/47 
 
 10/39 
 
 
 
 10/44 
 
 
 
 14/36 
 
 13/50 
 
 19/44 
 
 15/66 
 
 5/28 
 
 6 
 
 0.001 
 
 
 7/39 
 
 6/40 
 
 8/40 
 
 7/44 
 
 8/34 
 
 13/35 
 
 10/36 
 
 10/25 
 
 7/30 
 
 
 8/28 
 
 7 
 
 .001 
 
 0.001 
 
 
 8/36 
 
 6/36 
 
 
 8/31 
 
 11/35 
 
 3/22 
 
 4/31 
 
 10/39 
 
 5/28 
 
 
 8 
 
 
 .001 
 
 
 
 
 
 4/34 
 
 7/34 
 
 7/32 
 
 5/31 
 
 5/32 
 
 6/39 
 
 8/21 
 
 9 
 
 
 .001 
 
 0.001 
 
 
 
 
 6/37 
 
 4/29 
 
 7/34 
 
 6/19 
 
 2/29 
 
 4/35 
 
 3/21 
 
 11 
 
 .001 
 
 .001 
 
 .001 
 
 
 
 
 
 
 13/34 
 
 3/20 
 
 
 
 
 14 
 
 
 .001 
 
 
 0.001 
 
 0.001 
 
 
 
 8/36 
 
 5/35 
 
 
 
 3/38 
 
 
 15 .... 
 
 
 .058 
 
 .001 
 
 .017 
 
 .001 
 
 
 0.001 
 
 
 13/35 
 
 
 
 5/34 
 
 
 16 .... 
 
 NS 
 
 .002 
 
 .001 
 
 .001 
 
 .001 
 
 0.036 
 
 .001 
 
 NS 
 
 
 
 
 8/38 
 
 
 21 .... 
 
 .001 
 
 .096 
 
 .001 
 
 .001 
 
 NS 
 
 .001 
 
 
 
 
 
 21/40 
 
 19/44 
 
 13/29 
 
 22 .... 
 
 NS 
 
 .001 
 
 .001 
 
 .001 
 
 .001 
 
 
 
 
 
 NS 
 
 
 
 17/50 
 
 27 
 
 .001 
 
 
 .001 
 
 .001 
 
 .001 
 
 
 .001 
 
 0.001 
 
 0.001 
 
 NS 
 
 
 
 18/52 
 
 28 
 
 .001 
 
 .002 
 
 .001 
 
 NS 
 
 .001 
 
 
 
 
 
 NS 
 
 NS 
 
 0.002 
 
 
 NS Not statistically significant (p(x) > 0.10). 
 
 1 Data above the diagonal are x/N, where x = frequency of score mismatches and N = total number of pairs of nonzero scores 
 from the content analyses for that variable pair. A score mismatch is the case of + in 1 variable and - in the other. A blank cell 
 indicates either insufficient data to suggest a relationship or no logical reason for computing x/N statistic. Insufficient data were 
 considered to exist where either x > 24 or N < 1 8. Data below the diagonal are p(x) computed from equation 3. See Siegel (1 7) 
 for additional information. 
 
 2 See tables 1 and 5 for definitions of variables and factors. 
 
the variables were related at the 0.001 level of statistical 
 significance. 
 
 It was decided to eliminate from further consideration 
 any relationships that were not significant at the 0.002 
 level or less. This rule eliminated 4 of the 44 relationships 
 shown in table 6, while maintaining a highly desirable 
 level of confidence that the remaining relationships were 
 in fact the important ones. 
 
 PATH ANALYSES 
 
 By using a path analysis procedure and the 0.002 
 level of significance rule, causal chains were deduced from 
 the data in table 6 (2-3, 17, 19). Figure 2 was then con- 
 structed from these results. 
 
 The numbered nodes in figure 2 represent the varia- 
 bles and factors from tables 3 and 5 that survived the 
 content ratings and statistical analyses. Three types of 
 variables survived, as noted in figure 2; psychological vari- 
 ables, behavioral variables, and management variables. 
 The arrows in figure 2 depict the causal pathways that 
 were deduced from the path analysis procedures. For ex- 
 ample, the arrow running from node 6 (carefulness) to 
 node 7 (initiative), near the center of figure 2, indicates 
 that carefulness influences the extent to which individ- 
 uals demonstrate safety initiatives in their behaviors. 
 Moreover, this relationship holds true on the average in 
 998 or more cases out of 1,000, because of the 0.002 level of 
 significance rule. The path analysis procedures deduced 
 that the causality runs from node 6 to node 7 and not from 
 node 7 to node 6 (1-3, 17). This happens to be the most 
 intuitively logical causality for these two variables. In 
 other cases, the causality could have gone either way and 
 the path analysis procedure was used to make the final 
 determination of the direction (2-3, 8, 18). 
 
 Psychological 
 variables 
 
 Behavioral 
 variables 
 
 / Management 
 
 variables 
 
 KEY 
 
 (T) — »-{T) Variable i leads to variable j 
 
 Figure 2. — Path analysis network (see tables 1 and 5 for 
 definitions of nodes). 
 
 The path analyses resulted in node 27 as the terminal 
 node in the network, as shown in figure 2. That is, node 27 
 does not lead to any other subsequent nodes. Therefore, 
 the descriptor potential fatality, variable PF, was attached 
 directly to this node in figure 2. Thus, figure 2 represents 
 the system of variables that were found to lead to the 
 potential for a fatal accident. 
 
 SYSTEMS MODEL OF BEHAVIORAL ACCIDENTS 
 
 BEHAVIORAL ACCIDENTS 
 
 As the preceding results have shown, serious acci- 
 dents can and do happen to well-trained, highly experi- 
 enced and mature personnel working in mines that have 
 good safety records and strong safety commitments. 
 Chance events, equipment failures, and pressures to pro- 
 duce were not the causes. Rather, these accidents appeared 
 to be caused by the inappropriate decisions and adjustive 
 behaviors of the victims themselves in response to hazard- 
 ous conditions. In short, these were behavioral accidents. 3 
 
 These results suggest that behavioral accidents may 
 represent a hardcore type of accident that is highly resist- 
 ant to conventional treatments. Enormous amounts of 
 time and money are being spent on conventional safety 
 and training programs. Yet serious accidents continue to 
 occur and accident rates stubbornly resist falling below 
 some threshold levels. It may be that the bulk of accidents 
 
 3 As previously noted, 60 behavioral accident cases were found from a total 
 population of 358 fatal accident cases. Though this is not a relatively high 
 rate of occurrence, it represents a great cost in terms of human life. 
 Moreover, since all accidents would seem to have some behavioral content, 
 the importance of behavioral accident phenomena may be much greater 
 than this statistic might indicate. 
 
 that could be eliminated through conventional treatments 
 has now been eliminated. What remains may be the hard 
 core of behavioral accidents. More powerful, innovative 
 means may be needed to eliminate them. The systems 
 model shown in figure 2 is a basis for these more innova- 
 tive approaches. 
 
 RELATIVE IMPORTANCE OF THE VARIABLES 
 
 Figure 2 can be used to determine the relative impor- 
 tance of various variables with respect to the potential for 
 a fatality. Knowledge of the relative importance of the 
 variables is the first step in taking actions to control them. 
 
 Table 7 presents an analysis of the numbers of arrows 
 (paths) entering and exiting each variable (node) in figure 
 2. The total number of exiting paths from the psychologi- 
 cal variables set is 12. The management set has the same 
 number of exiting paths. This is a relatively large number 
 of exiting paths for a network of this size, thus indicating 
 that these variable sets are relatively important. That is, 
 both the psychological and management variables affect 
 several others. They are important affecting variables. 
 
Table 7.— Numbers of entering and exiting paths 
 
 for each variable in figure 1 
 
 
 
 Number' 
 
 Number of paths 2 
 
 Variable set and variable or factor 
 
 Entering from — 
 
 Exiting to— 
 
 
 P 
 
 B 
 
 M 
 
 Total 
 
 P 
 
 B 
 
 M 
 
 Total 
 
 Psychological: 
 Field dependency 
 
 14 
 9 
 8 
 
 15 
 NAp 
 
 16 
 
 22 
 6 
 
 7 
 
 27 
 NAp 
 
 21 
 2 
 
 28 
 
 11 
 
 NAp 
 
 
 1 
 1 
 2 
 
 
 1 
 1 
 
 
 
 
 1 
 1 
 
 
 
 3 
 3 
 2 
 
 3 
 1 
 
 
 
 2 
 
 3 
 2 
 1 
 
 
 
 
 
 
 5 
 
 Evasiveness 
 
 Alertness 
 
 Self-control 
 
 4 
 2 
 1 
 
 Sums 
 
 4 
 
 2 
 
 2 
 
 8 
 
 4 
 
 8 
 
 
 
 12 
 
 Behavioral: 
 
 Impulsivity 
 
 Failure to comply with approved 
 
 safe operating procedures . . . 
 
 Carefulness 
 
 Initiative 
 
 Employees used unsafe 
 
 judgment 
 
 3 
 
 1 
 2 
 
 
 3 
 
 
 
 
 2 
 3 
 
 2 
 
 
 
 
 2 
 2 
 
 2 
 
 3 
 
 1 
 6 
 5 
 
 7 
 
 
 
 2 
 
 
 
 
 
 
 3 
 
 2 
 1 
 1 
 
 
 
 
 
 
 
 
 
 
 
 3 
 
 4 
 1 
 1 
 
 
 
 Sums 
 
 6 
 
 7 
 
 6 
 
 22 
 
 2 
 
 4 
 
 
 
 9 
 
 Management: 
 
 Failure of management 
 
 Norm 
 
 Supervisor permitted unsafe 
 
 practices 
 
 Enforcement 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 1 
 
 1 
 2 
 
 
 1 
 
 1 
 2 
 
 1 
 
 
 1 
 
 
 1 
 1 
 
 3 
 1 
 
 2 
 1 
 
 1 
 
 
 4 
 2 
 
 5 
 1 
 
 Sums 
 
 
 
 
 
 4 
 
 4 
 
 2 
 
 6 
 
 4 
 
 12 
 
 NAp Not applicable. 
 
 1 See tables 1 and 5 for definitions of variables and factors. 
 
 2 P, psychological variable; B, behavioral variable; M, management variable. Variable 14 and factor 27 are the origination and 
 termination, respectively, for this network. 
 
 Similarly, the total number of paths entering the behav- 
 ioral variables set is 22, as noted in table 7. Thus, the 
 behavioral variables set is a relatively important affected 
 set of variables. 
 
 Similarly, table 7 shows that the behavioral variables 
 carefulness, initiative, and employee judgment are the 
 most important affected variables. The most important 
 affecting variables are the psychological variables field 
 dependency and evasiveness, the behavioral variable fail- 
 ure to comply, and the management variables failure of 
 management and supervisor permitted. 
 
 These aspects determine the way in which the psycho- 
 logical, behavioral, and management variables dynami- 
 cally interact to culminate in the potential for a fatality. 
 Starting from any node (variable) in figure 2, the effects of 
 a change in that variable can be tracked as it cascades its 
 way through the network like a ball in a pinball machine, 
 impacting other variables, compounding their effects, and 
 building a potential for an accident. 
 
 CASCADING NETWORK EFFECTS: 
 AN EXAMPLE 
 
 Let us examine the effects of two management factors: 
 the failure of management to implement strong safety pol- 
 icies (21) and the failure of supervisors to eliminate unsafe 
 practices (28). As figure 2 shows, these two factors directly 
 affect two important psychological variables: alertness 
 and evasiveness (variables 8 and 9). This result says that 
 lax safety management and supervision can directly cause 
 employees to fail to recognize, interpret, and act upon sub- 
 tle cues and danger signs of an imminent accident. This is 
 a very reasonable finding. Employees who are not sensi- 
 tized to accident cues by their management are likely to 
 either not see those cues or fail to act on them, or both. 
 
 Note that this finding has some very significant impli- 
 
 cations. It says that lax management affects the deep- 
 seated psychological variables that control an individual's 
 long term behaviors, not just the individual's situational 
 safety behaviors. This may have some very serious and far- 
 reaching consequences. For example, once an individual's 
 alertness and evasive mechanisms have been dulled by lax 
 management practices, this may spill over into other as- 
 pects of his or her behavior. The individual may tend to 
 behave unsafely in all aspects of his or her life. Retraining 
 may thus not be effective under these circumstances. In 
 fact, conventional training, which is directed at behavior 
 modification, may have any little effect on a person whose 
 alertness and evasiveness have been dulled (14, 18-19). 
 
 These effects may be seen from the model in figure 2. 
 The impacts of low alertness (variable 8) and impaired 
 evasive abilities (variable 9) are shown to continue to cas- 
 cade throughout the network, building and amplifying 
 themselves as they go. Impaired evasive abilities lead the 
 individual to impulsive behaviors and poor judgments 
 (variable 16 and factor 27). If the individual involved hap- 
 pens to be highly field dependent (variable 14— unable to 
 extract salient information from a complex environment), 
 then the effects of the low alertness and impaired evasive 
 abilities are further magnified. The individual's self- 
 control (variable 15) becomes diminished and the individ- 
 ual's judgment (factor 27) is further impaired. Under these 
 conditions, the individual becomes less likely to comply 
 with approved safe operating procedures (factor 22). Thus, 
 his or her awareness of dangers and his or her ability to 
 avoid them (variables 8 and 9) are further dulled in a kind 
 of feedback-reinforcement cycle. Careful behaviors (varia- 
 ble 6) thus rise and fall in proportion to the strengths of 
 these influences. As a result, the individual is further 
 pushed toward a fatal accident. Recall that all of this be- 
 gan by seemingly trivial failures by management and by 
 the supervisor. 
 
SOME IMPLICATIONS 
 
 What do these results tell about how to decrease un- 
 derground mine fatalities? Since management actions and 
 styles were found to directly influence employee careful- 
 ness, safety initiatives, and judgments, it is clear that 
 management quality is a key leverage point for reducing 
 accidents. This is not new: the effects of management qual- 
 ity on employees are well known. What is new here is the 
 finding that management quality directly influences two 
 important individual psychological factors: employee 
 awareness of dangerous conditions and employee actions 
 to avoid them. 
 
 Thus, there are three major implications. First of all, 
 the preeminence of management as a factor in preventing 
 accidents has been reaffirmed. Fatalities can be substan- 
 tially reduced by improving the general quality of mine 
 management and first-line supervision. Second, miner 
 training should be more concentrated on improving miner 
 pattern recognition skills and perceptual information 
 processing abilities. The ability of miners to perceive am- 
 biguous cues of pending accidents, e.g., poor roof condi- 
 tions, is a key to accident prevention. Third, training mine 
 supervisors to emphasize and reinforce alertness skills 
 
 among their subordinates is an essential ingredient. This 
 is important to cementing the linkage between manage- 
 ment quality and individual behaviors. 
 
 Figure 2 suggests yet another way to decrease fatali- 
 ties: change the psychological variables that lead an indi- 
 vidual into accidents. At first, this may sound like some 
 sort of recommendation aimed at psychologically altering 
 the individual. But it is nothing quite so radical. Rather, 
 adjustive behavior modification is the suggested approach. 
 Employees can be trained to have greater self-control, to 
 become less impulsive in crises, to be more alert, and to 
 react more appropriately to accident situations. Though 
 this is not a new idea, it is somewhat novel for the mining 
 industry. Most training courses emphasize unitary behav- 
 iors and the one best way to do the job, rather than alter- 
 native ways and multiple roles in the work environment. 
 The ability to flex to meet the situation and to select the 
 most suitable response from a repertoire of many re- 
 sponses is essential in a complex environment like an un- 
 derground mine. 
 
 These ideas all come directly from figure 2. They are 
 innovative ideas. They are only a sampling of the kinds of 
 recommendations and innovative approaches that an anal- 
 ysis of figure 2 can yield. 
 
 SIMULATING ACCIDENT CAUSES 
 
 Figure 2 may be viewed as a system flow model in 
 which activities at any node (variable) create influences 
 that flow along the paths (arrows) into other nodes. The 
 volume or amount of any flow, VP y , along any path from 
 node i to node j is given by 
 
 VP U = (N - x)/N. (4) 
 
 The cumulative flow, VNj, at any node j is given by 
 
 VN, = E, [(VN,) (VP,)]: (5) 
 
 Here N and x are the corresponding items from table 6 for 
 the path from node i to node j. The potential for a fatality, 
 PF, is given by 
 
 PF = [VN m /(VNJ m J x 100, 
 
 (6) 
 
 where (VN m ) max is the maximum value of the mth node for 
 the system being considered. 
 
 The VP^ data for the network in figure 2 are shown in 
 table 8. Note that the VP y data are computed only for 
 those paths that resulted from the path analyses, as shown 
 in figure 2. For example, table 6 shows x/N data for the 
 relationship between node 2 and node 6. But the path 
 analyses indicated that these data resulted from the indi- 
 rect relationship of node 2 to node 6 that runs through 
 nodes 28 and 11. There is no direct path between node 2 
 and node 6. Hence, there is no corresponding VP 26 datum 
 in table 8. 
 
 Table 9 presents the results of repetitively applying 
 equations 4 and 5, starting with initial unitary activities 
 in variable 14 and factors 22 and 21. For example, case 1 
 assumes that there is initially one act of field dependency 
 on the part of some one individual. That is, case 1 assumes 
 that the individual is unable to extract salient informa- 
 
 tion from a complex or ambiguous environment in one 
 instance. To illustrate, suppose the individual is standing 
 in the pathway of an oncoming shuttle car, but he or she is 
 unable to determine whether the shuttle car is approach- 
 ing rapidly or is stopped. Case 1 shows what happens as a 
 result of this one failure. This one act grows and com- 
 pounds to an ultimate value at node 27 that is 8.23 times 
 its initial unitary value. Using equation 6, this translates 
 to a 54.9 pet potential for a fatality, PF, as noted in the last 
 line of table 9. 
 
 The effects of other single acts are shown as cases 2, 3, 
 and 4 in table 9. For example, in case 2, the effects of a one- 
 time failure of the employee to comply with established 
 safety policies is shown to compound to a 12.2-pct poten- 
 tial for a fatality. Case 3 shows that a one-time failure of 
 management compounds to a 42.5-pct potential. The maxi- 
 mum potential for an accident when starting from varia- 
 ble 14 and factors 22 and 21 is 100 pet, as shown in case 4. 
 Thus, for the variables set examined here, problems in the 
 psychological variable (field dependency) are the most 
 
 
 
 
 
 Table 8.— VP„ data ' 
 
 
 
 
 
 
 Vari- 
 
 Variable j 
 
 able 
 
 2 
 
 6 
 
 7 
 
 8 
 
 9 
 
 11 
 
 14 
 
 15 
 
 16 
 
 21 
 
 22 
 
 27 
 
 28 
 
 2 . . 
 
 6 . . 
 
 7 . . 
 
 8 .. 
 
 9 .. 
 11 . . 
 
 14 . 
 
 15 . 
 
 16 . 
 
 21 . 
 
 22 . 
 
 27 . 
 
 28 . 
 
 
 
 
 
 
 
 
 
 
 .74 
 
 
 
 
 
 
 
 
 .85 
 
 
 .84 
 
 .76 
 
 
 .73 
 
 
 .77 
 
 
 .71 
 
 
 .82 
 
 
 
 
 
 
 
 
 
 
 .69 
 .87 
 .88 
 
 
 .83 
 
 
 
 
 
 
 
 
 .89 
 
 
 
 
 .84 
 .85 
 
 
 
 
 
 
 
 
 
 
 
 
 .84 
 
 
 
 
 
 
 .86 
 
 
 
 
 
 
 
 
 
 
 
 .85 
 
 
 1.00 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 .76 
 
 
 .88 
 
 
 
 
 
 
 
 
 
 
 
 
 .81 
 
 
 .86 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 .93 
 
 
 
 
 
 
 
 
 
 0.73 
 
 
 .75 
 .85 
 .89 
 
 
 
 
 .85 
 .79 
 
 
 
 
 
 .65 
 
 0.82 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 1 See tables 1 and 5 for definitions of variables and factors. 
 
10 
 
 Table 9. — Network calculations 
 
 VN 14 
 VNz, 
 VN 21 
 
 VN 2 
 VN 2e 
 VN„ 
 VN 9 
 VN,, 
 VN 8 
 VN 15 
 VN 16 
 VN 6 
 VN 7 
 VN 27 
 
 Case 1 Case 2 Case 3 Case 4 
 
 INITIAL VALUES 
 
 1.0 
 .0 
 .0 
 
 0.0 
 
 1.0 
 
 .0 
 
 0.0 
 
 .0 
 
 1.0 
 
 COMPUTED VALUES 
 
 Fatality potential (PF) pet 
 
 0.0 
 
 .0 
 
 .0 
 
 .84 
 
 .78 
 
 1.55 
 
 1.52 
 
 1.54 
 
 1 3.80 
 
 4.87 
 
 8.23 
 
 '54.9 
 
 0.0 
 
 .0 
 
 .0 
 
 .0 
 1.0 
 
 .85 
 
 .0 
 
 .0 
 
 .72 
 1.47 
 1.83 
 
 12.2 
 
 0.74 
 
 .61 
 
 .85 
 
 .52 
 
 .48 
 
 .84 
 
 .39 
 
 .42 
 
 3.05 
 
 4.58 
 
 6.37 
 
 42.5 
 
 1.0 
 1.0 
 1.0 
 
 0.74 
 .61 
 1.46 
 1.36 
 2.26 
 2.58 
 1.91 
 1.96 
 7.78 
 9.98 
 14.99 
 
 100 
 
 'Sample calculation: 
 
 VN 6 = (VN*) (VP^ 
 
 + (VNJ (VP m6 ) + (VN„) (VP 86 ) + (VNJ (VP l66 ) 
 
 = (0.78) (0.77) + (0.0) (0.84) + (0.0) (0.71) + (1.0) (0.76) 
 + (1.55) (0.85) + (1.54) (0.73) = 3.80. 
 
 PF, case 1 = (VN 27 for case 1)/ (VN 27 for case 4) 
 = (8.23/14.99) x 100 = 54.9 pet. 
 
 likely to result in a fatality. Management failure is the 
 next most likely cause of a fatality, and an individual fail- 
 ure to comply with an established safety procedure is the 
 least likely to result in a fatality. Note that these results 
 could be different for a different set of initial values and 
 nodes. 
 
 The data from table 9 are plotted in figure 3 in such a 
 way as to display the tracking of the contributions that 
 each key variable makes to the potential for a fatality. 
 Figure 3 clearly demonstrates the impact of the behavioral 
 variables. In each of the four cases, the node flows, VN j; 
 are all relatively low until the behavioral variables are 
 reached within the system. At that point, the node flows 
 increase dramatically This increase is greatest in case 4, 
 where multiple interactions of the management, psycho- 
 logical, and behavioral variables occur. 
 
 Thus, it becomes apparent from these few sample cal- 
 culations that the network model depicted in figure 2 and 
 table 7 has many potentials for simulating accident situa- 
 tions. That model is further developed, in the following 
 section, in order to make it more realistic and relevant for 
 mining applications. 
 
 15 r- 
 
 10 
 
 5 — 
 
 Node j = 2 
 
 Norm 
 
 11 9 8 15 16 6 7 27 
 
 Evasiveness Self-control Carefulness Employee 
 
 Enforcement Alertness Impulsivity Initiative judgment 
 
 KEY VARIABLES 
 
 Figure 3.— Results of network calculations (table 9). 
 
 NETWORK PREPROCESSORS 
 
 In the simulation exercises, initial values of some key 
 variables were simply assumed in order to determine the 
 ultimate potential for a fatality under several conditions. 
 In reality, several psychological, behavioral, managerial, 
 and environmental conditions that naturally surround the 
 network in figure 2 often determine the initial values of 
 these key variables (4, 6, 9, 15). These conditions are natu- 
 ral preprocessors for the network. These preprocessors can 
 
 be used to flexibly model particular real situations. These 
 situations will then drive the fixed data base of relation- 
 ships shown in table 8, which are the flow rates for the 
 paths depicted in figure 2. 
 
 Seven preprocessors are used: psychological condi- 
 tions (PC), behavioral conditions (BC), supervisor ability 
 (FA), management concern for safety (MC), environmental 
 conditions (EC), physiological state (PS), and adjustive be- 
 
11 
 
 haviors (JB). These preprocessor variables are defined in 
 table 10 and are presented as part of the network shown in 
 figure 4. Note that the discussion here focuses on the de- 
 velopment of these preprocessors, a subsequent section of 
 this report discusses their use. 
 
 PSYCHOLOGICAL CONDITIONS (PC) 
 
 Research (4, 6, 9, 19) indicates that the three psycho- 
 logical conditions listed in table 10— pattern recognition 
 skills, alertness, and discriminatory abilities— strongly 
 impact a person's likelihood of acting to avoid an accident. 
 For example, pattern recognition training teaches visual 
 differences between faulty and safe roofs in a mine. Alert 
 miners will observe these differences. Discriminating min- 
 ers will know the appropriate actions to take based on 
 these observations. 
 
 Research (1-2) suggests that pattern recognition 
 skills (Xi), alertness (X 2 ), and discriminatory abilities (X 3 ) 
 combine according to 
 
 PC = [(X 2 /1.67) + (X 3 /2.5)] - (X./2.0), 
 
 (7) 
 
 where PC gives their combined effects. The parameter PC 
 is an input to node 14 of the network in figure 4. The value 
 of PC is the initial value of VN 14 in calculations like those 
 in table 9. Thus, a large value for PC will have a large 
 impact on node 14; a small value will have a small impact. 
 Recall that node 14 is the variable field dependency (psy- 
 chological ability to distinguish salient cues from an am- 
 biguous environment). When the value of PC is high, the 
 flow into node 14 is high and field dependency is high. 
 Similarly, when the value of PC is low, field dependency is 
 necessarily low. 
 
 BEHAVIORAL CONDITIONS (BC) 
 
 The four behavioral conditions shown in table 10— 
 volatility, machoism, consistency, and influence— can bias 
 individual reactions to emergencies (4, 6, 9). For example, 
 persons who characteristically act in a macho fashion may 
 become involved with more accidents than those who do 
 not. This may be especially true where others emulate the 
 macho individual. 
 
 KEY 
 
 (T) — *"{7) Variable i leads to variable j 
 |~K~] User inputs that define 
 
 preprocessors 
 
 «■ Feedback I oops 
 
 (T) Preprocessors, calculated from 
 
 user inputs 
 
 Figure 4. — Network with preprocessors. 
 
 Table 10 
 
 Preprocessor 
 
 Psychological conditions (PC): 
 
 Pattern recognition skills (X,) 
 
 Alertness (XJ 
 
 Discriminatory abilities (X 3 ) 
 
 Behavioral conditions (BC): 
 
 Volatility (Y,) " 
 
 Machoism (Y^ 
 
 Consistency (Y 3 ) 
 
 Influence (Y„) 
 
 Supervisor ability (FA): 
 
 Leadership skills (LS) 
 
 Interpecsonal abilities (IA) 
 
 Technical proficiency (TP) 
 
 Planning skills (PP) 
 
 Communication skills (CS) 
 
 Directing abilities (DA) 
 
 Management concern for safety (MC) 
 
 Environmental conditions (EC): 
 
 Perceived ambiguity of job-role (PJA) 
 
 Production pressure and fatigue (PPF) 
 
 Job physical annoyances (PA) 
 
 Perceived economic climate (PEC) 
 
 Stressful personal life events (SLE) 
 
 Physiological state (PS) 
 
 Adjustive behaviors (JB): 
 
 Aggression 
 
 Projection 
 
 Withdrawal 
 
 Sublimation 
 
 Adaptation 
 
 Consultation 
 
 .—Preprocessor variable definitions 
 
 Definition 
 
 Extent to which individual- 
 Has received formal training in pattern recognition. 
 
 Is an accurate observer. 
 
 Knows appropriate actions to take in an emergency. 
 Extent to which individual — 
 
 Reacts irrationally to crises. 
 
 Has compelling need to demonstrate his "manliness." 
 
 Behaviors are consistent day to day. 
 
 Is emulated by others. 
 
 Ability to lead others. 
 
 Ability to get along with others. 
 
 Proficiency in technical aspects of work directed. 
 
 Ability to plan work of others. 
 
 Ability to communicate with others. 
 
 Ability to direct work of others. 
 
 Extent to which top management evidences concern for safety. 
 
 Degree of understanding or lack of clarity about nature of the job. 
 
 Amount of perceived pressures to get out the product or produce at a high level. 
 
 Extent to which employee feels job is dirty, noisy, or otherwise physically annoying. 
 
 Extent to which employee feels trapped in an underpaid and/or lowly esteemed job. 
 
 Extent to which employee appears to be affected on job by traumatic personal or family events. 
 
 Extent to which individual is physiologically qualified for particular job. 
 
 Misplaced direction of pent-up emotions, stubborn and persistent nonadjustive reaction in face of 
 
 evidence that this is inappropriate or ineffectual behavior. 
 Attributing one's own failures to another person or thing, blaming others for own shortcomings. 
 Use of fantasies, emotional flight, repression, or regression to infantile behaviors. 
 Replacing urge to vent one's anger and frustrations with higher level substitutes without ever fully 
 
 resolving basic conflicts or issues. 
 Adjustments that accommodate the stimuli but fail to make permanent and complete internal 
 
 reconciliation, so that recidivism is likely. 
 Complete and thorough resolution of issues through consultative sessions between employee 
 
 and supervisor or other affected parties. 
 
12 
 
 Research (18-19) suggests that these four conditions 
 combine according to 
 
 BC =t(Y x + Y 2 )/2.0] (Y 3 ) (Y 4 ), 
 
 (8) 
 
 where BC is their envelope. The parameter BC is the input 
 to node 22, as shown in figure 4. The value of BC is the 
 initial value of VN 22 in computations like those illustrated 
 in table 9. 
 
 the product, feels trapped in an underpaid job, or finds the 
 physical environment annoying is more likely to have an 
 accident than other employees who are not under these 
 pressures. In addition, personal financial problems, family 
 trauma, or other stressful life events can promote careless 
 behaviors on the job. Holmes and Rahe (7) have devised a 
 scale for measuring such events. 
 
 The envelope, EC, of these five environmental condi- 
 tions is 
 
 SUPERVISOR ABILITY (FA) AND 
 MANAGEMENT CONCERN FOR SAFETY (MC) 
 
 The six supervisor abilities listed in table 10 are gen- 
 erally felt to be important influences on employee accident 
 behaviors (13, 15). Their envelope, FA, is 
 
 FA = LS + IA + TP + PS + CS + DA. 
 
 (9) 
 
 The value of FA is part of the input to node 28 in figure 4. 
 Management concern for safety (MC), as defined in 
 table 10, is generally believed to affect employee accident 
 behaviors. Research (18-19) suggests that the effect is an 
 inverse geometric one, i.e., 
 
 SC = (1.0 - MC) + [(1.0 - MOV2.0], 
 
 (10) 
 
 where SC is simply an envelope. 
 
 The value of the parameter SC directly affects node 21 
 in figure 4. It is the initial value of VN 21 in calculations 
 like those in table 9. 
 
 ENVIRONMENTAL CONDITIONS (EC) 
 
 The five environmental conditions listed in table 10 
 have repeatedly been found to seriously affect employee 
 work behaviors (13-15). An employee who perceives that 
 his or her job is ambiguous, feels under pressure to get out 
 
 EC = (PJA + PPF + PA + PEC + SLE) (PS), (11) 
 
 where PS is the individual physiological abilities state, as 
 defined in table 10. The value of the parameter EC is part 
 of the input to node 9 in figure 4. 
 
 ADJUSTIVE BEHAVIORS (JB) 
 
 Individuals may react to their environments in any of 
 the six adjustive behaviors (JB) listed in table 10. Which of 
 these behaviors actually occurs in any situation will de- 
 pend on the individual's personality. But it will also de- 
 pend in part on the supervisor's ability to successfully 
 intervene. A highly skilled supervisor, i.e., one with a 
 large FA value from table 10, may be able to intervene 
 early enough in the sequence of events to neutralize dan- 
 gerous adjustive behaviors such as aggression (4, 10, 14- 
 15). 
 
 Let IF be an intervention factor such that IF = 1 if 
 the supervisor successfully intervenes, and otherwise IF 
 = 0. Then let 
 
 FA* = F, (IF, FA) (12) 
 
 and let 
 
 EC* = F 2 (JB, EC), (13) 
 
 where F t and F 2 are functions. Then FA* and EC* will 
 respectively affect nodes 28 and 9 as shown in figure 4. 
 The exact impacts of EC* and FA* are governed by the 
 cusp catastrophe model described in the following section. 
 
 CUSP CATASTROPHE MODEL 
 
 THEORY AND EXAMPLE 
 
 Thorn's catastrophe theory (13, 20) enables the com- 
 bined influence of the environmental conditions, the su- 
 pervisor's ability, the intervention factor, and the 
 adjustive behaviors to be depicted as shown in figure 5. 
 The behavioral accident potential', BAP, is given by 
 
 BAP = F 3 (FA*, EC*), 
 
 (14) 
 
 where F 3 is some function. 
 
 Zeeman (22) cites a vivid illustration of the cusp catas- 
 trophe theory model shown in figure 5 with a friendly dog 
 that is teased to the point of biting. Starting at point B on 
 the behavioral surface in figure 5, where the dog is 
 friendly and affectionate, the dog is purposely teased by 
 its master. Thus, the dog's environment is depreciated and 
 a corresponding movement occurs from left to right along 
 the 1-EC* axis in figure 5, where < EC* < 1. The 
 teasing also depreciates the relationship between the dog 
 
 Behavorial 
 surface 
 
 ggression 
 
 I- EC — 
 
 Consultation 
 
 KEY 
 FA'^UFjFA) 
 EC*=F 2 (JB-,EC) 
 
 Figure 5.— Cusp catastrophe model. 
 
13 
 
 and its master, so that a corresponding movement occurs 
 toward the origin along the FA* axis in figure 5. The com- 
 bined effects of these actions is to move the dog along the 
 pathway from point B towards point C. As the teasing 
 continues, the dog becomes increasingly more agitated. 
 The dog reaches its threshold of tolerance at point C, 
 where it suddenly jump shifts its behavior to point D on 
 the upper surface of the cusp. Or, to put it bluntly, the dog 
 bites its master. 
 
 Several subsequent events are now possible. When the 
 dog's master ceases the teasing and goes off to tend his or 
 her wounds, the dog's aggressive behavior may gradually 
 decay back to point B along the decay path DB. Or the 
 dog's behavior may recede along path DA and jump shift 
 back to point B. Or the dog may remain at point D for some 
 time, as long as its master continues to maintain the cor- 
 responding EC* and FA* stimuli. Still other more complex 
 outcomes are possible if this static model is permitted to 
 undergo dynamic changes. For example, suppose the teas- 
 ing permanently changes the dog's personality. It is not 
 uncommon for a dog to be turned into a mean and vicious 
 animal through repeated teasing. This could be modeled 
 as a shift in the shape and location of the cusp. Or the cusp 
 could collapse once point D has been reached, i.e., the dog 
 only attacks once and then regresses into a totally submis- 
 sive state (19). 
 
 It is clear from these discussions that this is a static 
 model with some limitations. But it is also clear that this 
 static model may be quite relevant for modeling accident 
 behaviors. 
 
 APPLICATION TO BEHAVIORAL MINE 
 ACCIDENTS 
 
 It is easy to see how the story of the teased dog can be 
 analogous to the underground miner who becomes agi- 
 tated. At point B in figure 5, either the individual is in 
 harmony with his or her environment or the supervisor's 
 intervention factor (IF) is high enough that the individ- 
 ual's adjustive behavior is adequate for him or her to cope. 
 Change any of the variables that contribute to this equi- 
 librium, i.e., give the individual a less skilled supervisor, a 
 worse environment, a supervisor who is less able to inter- 
 vene, etc., and dramatic things may happen. The individ- 
 ual's adjustive behaviors may jump shift to aggression at 
 point D in figure 5. Once at point D, the individual could 
 remain there for relatively long periods of time, since the 
 behavioral surface is relatively flat in the vicinity of point 
 D. Large changes in the environment or the supervisor's 
 abilities may be required to move the individual away 
 from point D. All the subsequent events reviewed above 
 may now occur, i.e., the individual's behavior may decay 
 back to point B, the individual may remain aggressive, the 
 individual's behavior may jump shift to other positions, 
 etc. It may be noted that in terms of the JB scale in table 
 10, point D is aggression, point C is projection, and point B 
 is consultation. The other JB scale positions, i.e., adapta- 
 tion, etc., are points along the path BC. 
 
 The output from figure 5 is a value for BAP, calculated 
 according to equation 14. As shown in figure 4, the value 
 of the parameter BAP goes directly into node 9 of the net- 
 work. Recall that node 9 was the variable evasiveness (see 
 table 1). Thus, BAP controls the ability of the individual to 
 avoid an imminent accident. 
 
 BEHAVIORAL DYNAMICS 
 
 Even with the addition of the cusp catastrophe mecha- 
 nism, the network in figure 4 is primarily a static model. 
 However, one important dynamic aspect can easily be 
 added: behavioral reinforcement. 
 
 Real, vicarious, or social experiences can reinforce un- 
 safe behaviors. To illustrate, suppose a normally safe- 
 behaving individual knowingly performs a familiar job in 
 a careless manner and does not have an accident. This 
 experience may reinforce the individual's deviant behav- 
 ior and increase the likelihood that he or she will repeat 
 the careless act. Or, to put it more bluntly, because he or 
 she got away with it one time he or she will probably try it 
 again. This same result can occur if the individual has 
 never actually experienced the careless act, but thinks 
 about it longingly (vicarious experience) or observes oth- 
 ers doing it (social learning). 
 
 To account for such phenomena, three feedback loops 
 are included in figure 4. One loop runs from node 15 to 
 node 9, a second feedback loop runs from node 16 to node 9, 
 and a third runs from node 27 to node 22. The first loop 
 permits changes in self-control to feed into evasiveness. 
 That is, as one's self-control decreases (or increases), one's 
 evasive ability also decreases (or increases). The second 
 feedback loop permits impulsivity to feed into evasiveness. 
 The third loop permits individual judgments to feed into 
 
 compliance behaviors, i.e., once poor judgments occur they 
 can further cloud the individual's compliance behaviors. 
 Many other feedback loops are possible. However, these 
 three appeared to be the most prominent ones for the ap- 
 plications here (16, 19). 
 
 With feedbacks, the cumulative flows at any node j 
 are given by 
 
 VN* = VN, + E k (VN k ) (VP kj ), (15) 
 
 where VNj is the value at the kth node where the feedback 
 originates and VP kj is the value of the feedback path. Here 
 
 for the case where 
 
 VP,, = A (VP ik ), 
 
 < A < 1.0. 
 
 (16) 
 
 (17) 
 
 The coefficient, A, is parametrically set to reflect particu- 
 lar situations. For example, if social learning is believed to 
 influence an individual's behaviors by 20 pet, i.e., he or 
 she is affected 20 pet of the time, then A would be set at 
 0.20 for that case. Different values of A may be used in 
 each of the three feedback loops (19). 
 
14 
 
 BEHAVIORAL ACCIDENT SIMULATOR (BAS) 
 
 A review of the literature and a consideration of the 
 application here (12, 16, 22) suggested that the following 
 functional forms of equations 13 and 14 should be used. In 
 the high accident zone of figure 5, 
 
 BAP = [a 1 (l-EC*)] la 2 ,1 - EC *' 1 , 
 where 
 
 EC* > a 3 . 
 
 (18) 
 (19) 
 
 In the low accident zone, 
 
 
 BAP = [a 4 (l-EC*)] e ' 2 , 
 where 
 
 EC* < a 3 . 
 
 (20) 
 (21) 
 
 In equations 18 through 20, 
 
 
 EC* = a 5 EC - a 6 (x,JB + x 2 FA*), 
 
 (22) 
 
 for x, = or 1, when x 2 = 1 or 0, and when e = 2.718 + . 
 Note that equation 22 is simply a convenient functional 
 way to combine equations 12, 13, and 14. By using equa- 
 tion 22, FA* becomes an input to EC* instead of an input 
 to BAP as shown in figure 4. The coefficients a,, a 2 , etc., are 
 parameters set by the model builder that control the posi- 
 tioning and length of the behavioral decay path (path DB 
 in figure 5), the jump shift and return shift paths (paths 
 
 CD and AB in figure 5), and the behavioral buildup path 
 (path BC in figure 5). 
 
 The probability of a fatality (the output from node 27 
 of figure 4) is given by 
 
 CK = a(Z be > c + del 
 
 (23) 
 
 where CK is the chance that an individual experiencing 
 the conditions within the network will be killed and 
 
 Z = k [(VN 27 )/(VN 27 ) ma J, 
 
 (24) 
 
 where a, b, c, d and k are suitably chosen constants, and 
 where e = 2.718 + . 
 
 The model, as defined by equations 4 through 24, and 
 the VP,, data base shown in table 8, were programmed in 
 conversational mode using the Fortran 77 language. This 
 computer program, called the behavioral accident simula- 
 tor (BAS), can be run on VAX, DEC-1099, and IBM per- 
 sonal computer systems (19). The BAS provides many 
 different parametric options to the user, so that a variety 
 of scenarios can be run and a variety of conditions can be 
 simulated. Graphic outputs, tabular outputs, and several 
 user-selected reports are available. The conversational 
 nature of the BAS permits the user to flexibly interact 
 with the program at the computer keyboard. Users can 
 describe various situations and receive immediate print- 
 outs of the fatality probabilities and other statistics for 
 those situations. 
 
 ILLUSTRATIVE APPLICATION OF THE BAS 
 
 FRANK: A HYPOTHETICAL CASE 
 
 To illustrate the use of the BAS, the following is a 
 purely hypothetical case of Frank, an underground miner 
 at the hypothetical XYZ mine. Frank is currently as- 
 signed to one of the dirtiest and least attractive jobs at the 
 XYZ mine. He often works overtime, he believes he is gen- 
 erally underpaid, and he feels he is under a great deal of 
 pressure to get out the product. Frank's life philosophy is: 
 "you gotta fight everybody just to stay even in this world." 
 He is currently contesting his third wife's divorce suit. 
 Yesterday, his new Corvette was vandalized and over 
 $1,000 of uninsured damage was done to the interior of the 
 car. Frank is a generally aggressive person with many 
 pent-up emotions. He is a poor observer, often flies off the 
 handle, and is prone to react inappropriately to emergen- 
 cies. He has a consistent need to demonstrate his macho 
 self-image and he thinks training is for sissies. Unfortu- 
 nately, Frank often influences his peers to act just like 
 him. Frank's supervisor is technically competent, but he is 
 an inadequate planner, director, and leader. The supervi- 
 sor has poor interpersonal skills and does not communi- 
 cate well with the crew. However, top management at the 
 XYZ mine is very safety conscious. 
 
 CODING THIS CASE INTO THE BAS 
 
 It is apparent from this description that the hypotheti- 
 cal Frank has behavioral and psychological characteristics 
 that incline him toward accidents. Moreover, he is cur- 
 rently experiencing some personal life stresses, a poor 
 working environment, and an ineffective supervisor. By 
 using scales that were designed to be used with the BAS 
 (7, 18), Frank's situation was coded into the BAS on a 
 personal computer as scenario A. For example, using the 
 scales for the variables listed in tables 10, Frank's envi- 
 ronmental conditions (EC) and stressful life events (SLE) 
 were coded as follows in scenario A: PJA = 100, PPF = 
 100, PA = 100, PEC = 100, and SLE = 100. A score of 100 
 indicates the worst possible conditions. From table 10, 
 Frank's behavioral conditions (BC) were coded as follows: 
 volatility (V\) = 1, machoism (V 2 ) = 1, consistency (V 3 ) = 
 0, and influence (V 4 ) = 1. These are the worst scores possi- 
 ble for these variables. 
 
 Four other scenarios were also similarly coded into 
 the BAS. In scenario B, the impacts of changing Frank's 
 supervisor were simulated by recoding the BAS with a 
 superior-performing supervisor while leaving all the other 
 data the same. In scenario C, the codes were altered to 
 
15 
 
 simulate the effects of transferring Frank and his cowork- 
 ers to a cleaner and more attractive job environment. Sce- 
 nario D simulates the effects of transferring Frank out of 
 the work group, thus breaking up the work clique. Sce- 
 nario E simulates the effects of retraining Frank in a way 
 that would totally modify his behavior. 
 
 RESULTS FROM THE BAS 
 
 It should be noted here that the purpose of these exer- 
 cises is not to get absolute answers about how to deal with 
 Frank. Rather, the purpose is to gain insights about the 
 impacts of various management alternatives and what-if 
 conditions, as one way of learning more about the entire 
 accident system. For example, there is an implication in 
 scenario E that Frank can be totally retrained. In reality, 
 this is highly unlikely. Yet, experimenting with this alter- 
 native enables one to learn a great deal about accident 
 systems and how they operate. 
 
 Figure 6 plots the results of the above five scenarios 
 from the BAS. Scenario A, the current situation, gives 
 intolerable results: Frank's chances of being killed are ex- 
 traordinarily high. Thus, it is clear that this current situa- 
 tion cannot be permitted to exist. Some changes must be 
 made. Would things be better under a new supervisor? 
 Scenario B responds to this question. As figure 6 shows, 
 though Frank's chances of being killed are lessened under 
 the new supervisor, his chances still remain intolerably 
 high. Scenario C, transferring Frank and the crew to a 
 better working environment, substantially lowers the 
 
 ■Current situation ' 
 
 New supervisor 
 
 New environment 
 
 SCENARIO 
 
 Figure 6. — Effects of various BAS scenarios. 
 
 chances of being killed. This result is shown in figure 6. 
 But major decreases in the chances of being killed do not 
 occur until scenario D, where Frank is transferred out of 
 the work group and the work clique is broken up. Still, 
 Frank's chances of being killed remain unacceptably high. 
 It is apparent from the scenario E results that in this 
 extreme example Frank himself must be changed if his 
 chances of being killed are to be lowered to acceptable 
 levels. 
 
 PHILOSOPHY OF USING THE BAS 
 
 Though the example with Frank was purely hypothet- 
 ical, it was nevertheless an informative and valuable exer- 
 cise. The BAS demonstrated the relative effectiveness of 
 several alternative management policies. The BAS showed 
 that, under the given conditions, the most potent manage- 
 ment action is to break up the work clique and form a new 
 work group. Some other common alternatives, e.g., bring- 
 ing in a new supervisor and altering the work environ- 
 ment, were found to be much less effective. These are 
 valuable results. They show that time and effort should 
 not be wasted on the other alternatives: they will not be 
 effective. This was not intuitively apparent from the de- 
 scription given for Frank's case. 
 
 Thus, the BAS can make four valuable contributions. 
 First, it encourages managers to explore a wide variety of 
 creative new alternatives that they might not otherwise 
 consider. The very nature of the BAS encourages man- 
 agers to ask what-if questions. Its sole purpose is to simu- 
 late real world conditions as a basis for generating and 
 experimenting with creative solutions. The BAS catalyzes 
 a kind of brainstorming that encourages one to generate 
 and try out a variety of conditions and scenarios. For ex- 
 ample, in the hypothetical case with Frank, a natural 
 question to ask is: "What would happen if we could make 
 the mine perfectly clean and quiet?" Of course, this is not 
 feasible. But by asking an exaggerated question of the 
 BAS, the extreme cases can be tested. The result in sce- 
 nario C, was highly revealing: even making the mine con- 
 ditions perfect would not solve the problem. The BAS has 
 
 indicated the need to look somewhere else for the solution. 
 A more creative solution, breaking up the work group, was 
 then tried in scenario D. This was found to be highly effec- 
 tive. Given these results, management can now try other 
 proposed solutions— somewhere between the impossibly 
 perfect mine conditions and the possible but impractical 
 breakup of the work group. For example, what happens if 
 we clean up the mine a little, provide some very directed 
 training, and modify the work team somewhat? Such com- 
 binations can now be tested with the BAS, in order to move 
 toward a solution that is both feasible and practical to 
 implement. 
 
 Second, the BAS is a unique management laboratory 
 where alternative actions can be tried before they are im- 
 plemented. By using the BAS, the relative effectiveness of 
 various alternatives can be studied as a basis for selecting 
 the most cost-effective choices. For example, in the hypo- 
 thetical case with Frank, bringing in a new supervisor is 
 an intuitively appealing solution. But the BAS demon- 
 strated that this would have been an ineffective and costly 
 solution in this case. The BAS gives managers the oppor- 
 tunity to try new policies and actions before implementing 
 them. This capability to test the impacts of new safety 
 policies, organization designs, and changes in mine envi- 
 ronments before developing or implementing them is 
 clearly highly valuable. 
 
 Third, the BAS emphasizes the systems approach to 
 accidents. It permits one to look at the total system of 
 accident factors and to observe the ways in which they 
 
16 
 
 interact to cause accidents. BAS users acquire an impor- 
 tant conceptual appreciation for the way various factors 
 can interact, augment each other, or cancel each other. An 
 understanding of the complex dynamics of human acci- 
 dent phenomena is essential to the selection of effective 
 accident prevention policies. Focusing on only one or two 
 factors in the hopes of eliminating accidents will not work. 
 An understanding of the entire accident system dynamics 
 is required. The BAS provides this understanding. 
 
 Fourth, the BAS is not only a management tool. It has 
 been shown to be a highly effective self-learning or self- 
 teaching tool for miners. Miners who used the BAS ac- 
 quired a significantly improved sense of safety, and an 
 appreciation for ways to avoid hazards. 
 
 One reason for much of the potency of the BAS is that 
 
 it is not simply a theoretical model, designed in a vacuum. 
 The network of variables that compose the heart of the 
 BAS (fig. 4) was derived purely empirically, from actual 
 MSHA accident cases. The preprocessor variables and 
 other relationships in the BAS are all empirically based, 
 from studies reported in the literature. Thus, everything 
 within the BAS is a reflection of reality, at least to the 
 extent that it has been captured in various accident re- 
 ports and research studies. 
 
 Another feature that makes the BAS attractive is its 
 ability to flexibly and realistically model a wide variety of 
 circumstances. Many different individual, organizational, 
 group, managerial, and mine conditions can be repre- 
 sented and tested. 
 
 LABORATORY TESTS OF THE BAS 
 
 Over 200 hypothetical cases have now been run on the 
 BAS to demonstrate its utility. Figures 7, 8, and 9 present 
 some of the more interesting results from these cases. For 
 example, as figure 7 shows, the BAS can be used to test 
 interactions between work environments and perceived 
 levels of stress by individual miners. These BAS results 
 indicate that the difference between high and low stress 
 only becomes significant as the work environment de- 
 grades. However, in very poor work environments even low 
 stresses are quite deleterious. As figure 8 demonstrates, 
 the BAS can be used to test the effectiveness of various 
 levels of supervisor competencies. As figure 9 demon- 
 strates, the influence of several types of individual behav- 
 ior profiles can also be readily tested with the BAS. 
 
 These results are informative and interesting. But the 
 real test of the BAS is a matter of whether or not it actu- 
 ally improves mine safety. 
 
 100 
 
 0.8 
 
 Bad 
 
 WORK ENVIRONMENT (EC) 
 Figure 7.— Effects of stress (SLE). 
 
 
 1UU 
 
 1 1 1 
 
 Worst < 
 
 
 75 
 
 / 
 
 o 
 a. 
 
 
 /Poori 
 
 LU 
 
 
 
 O 
 
 
 
 z 
 
 
 
 < 
 
 
 
 X 
 
 o 
 
 50 
 
 _ / / Good-, 
 
 >- 
 
 
 
 H 
 
 
 
 _l 
 
 
 
 < 
 
 
 / / / Best 
 
 < 
 
 
 
 U. 
 
 25 
 n 
 
 ► r~^"^ i i 
 
 0.2 0.4 0.6 0.8 
 
 Good Bad 
 
 WORK ENVIRONMENT (EC) 
 Figure 8.— Effects of supervisor abilities (FA) 
 
 1.0 
 
 100 
 
 < 
 i 
 o 
 
 0.2 
 
 Good 
 
 0.4 0.6 0.8 1.0 
 
 Bad 
 WORK ENVIRONMENT (EC) 
 
 Figure 9.— Effects of various behavioral profiles (BC). 
 
17 
 
 FIELD TESTS OF THE BAS 
 
 Twenty-eight persons from the mining industry exper- 
 imented with the BAS on an individual basis. They in- 
 cluded miners, mining engineers, and mine managers. 
 Each subject used the BAS to generate scenarios of inter- 
 est and observed the results. These scenarios included 
 tests of various work environments, stresses, psychologi- 
 cal conditions, behavioral conditions, management safety 
 policies, supervisor abilities, crew composition, organiza- 
 tional arrangements, and employee adjustive behaviors. 
 
 Before experimenting with the BAS, each subject com- 
 pleted an attitude questionnaire. A postexperience ques- 
 tionnaire was then administered to each subject at the 
 completion of the BAS session. Each session lasted from 2 
 to 3 h. Tables 11 and 12 present the results from the pre- 
 and post-BAS experience questionnaires. The following 
 are comments from post-BAS experience questionnaires 
 (each comment is from a different subject). 
 
 1. Too many times management does not seem to ex- 
 haust all their alternatives before they try something. 
 What I mean here is that they simply do not look at all the 
 alternative things they could do. They just do whatever 
 first comes to mind. If firing the employee comes to mind 
 first, then that is what they do. The BAS says "wait a 
 minute. Let's look at some other alternatives." And then it 
 shows you what these alternatives can do. You get to see 
 the effectiveness of each alternative before you do it. Now 
 that is worthwhile. Even if some of the relations are made 
 up in the model and not exact, they at least seem to go in 
 the right direction. That's all you need. 
 
 2. Working with the BAS can be a useful experience 
 for mine managers. It will help them do things that reduce 
 mine accidents. 
 
 3. The BAS is a good tool for new employees. But it 
 might end up as a tool for management to get rid of certain 
 individuals, instead of trying to change things within the 
 organization itself. 
 
 4. Most accidents seem to be caused by mental lapses, 
 or at least a "hurry-up to get the job done" attitude. Safety 
 meetings and films, as boring as they are, are helpful if you 
 will only remember one small item from them and practice 
 it. 
 
 5. Management and employees should look into this 
 computer program and see how they can make changes 
 based on what it might suggest to them. 
 
 6. The BAS is clearly useful to anyone dealing with 
 hazards. 
 
 7. The BAS is a good device for mine managers and 
 for foremen to learn skills that will improve their judg- 
 ments in coping with problem employees and difficult situ- 
 ations. 
 
 8. Because there may be vast differences in individ- 
 ual receptiveness to safety messages, something like the 
 BAS is a good idea as a training device. No matter how you 
 feel about training, this thing is sort of interesting and it 
 holds your attention. 
 
 9. The BAS gives management better insights into 
 reducing accidents, improving work relations, and under- 
 standing employees. 
 
 10. I can see this as a very effective working tool in the 
 mining industry. But management must be willing to take 
 the time to care about their employees, to be willing to 
 better understand human nature and to be willing to in- 
 vest dollars to improve safety. This is a good tool for learn- 
 ing a lot about employees and how they act. But, if 
 managers don't care about that, then your model is no 
 good. 
 
 11. This is the first time I have ever seen all the acci- 
 dent factors put together like this. It really shows the rela- 
 tionships and gives you a feel for the whole thing. 
 
 Table 11.— Selected results from BAS experience 
 questionnaires, percent 
 
 Agreement 
 Pre-BAS questionnaire: 
 
 High accident rates are inherent in the job 40 
 
 The miner's own behavior is seldom the cause of accidents 46 
 
 Conventional safety training is ineffective 38 
 
 Post-BAS questionnaire: 
 
 The BAS clarifies behaviors that lead to accidents 92 
 
 The BAS is a useful training device for mine supervisors . 92 
 
 The BAS is a useful training device for miners 88 
 
 The BAS is more effective than conventional training 
 
 methods 88 
 
 The BAS sharpens a person's ability to recognize and avoid 
 
 hazards 73 
 
 Use of the BAS could help reduce mine accident rates ... 76 
 
 Table 12.— Selected results from pre- and post-BAS 
 questionnaires for BAS users who did not believe in training 
 
 
 Averages' 
 
 Level of sig- 
 
 
 Pre- 
 BAS 
 
 Post- 
 BAS 
 
 nificance of 
 change, 2 pet 
 
 Individual miners can do very little to 
 reduce accident rates 
 
 1.67 
 3.89 
 3.72 
 
 4.28 
 1.33 
 2.22 
 
 >99 
 
 Mine accidents are caused by a sys- 
 tem of interacting factors 
 
 Miners often take chances and do 
 things that are not very safe 
 
 >99 
 
 >95 
 
 '1.0, strongly agree; 5.0, strongly disagree. 
 
 2 Wilcoxian 1-tailed test of statistical significance; numbers represent confi- 
 dence levels; e.g., 99 pet means there is 99-pct confidence that a real change 
 in attitudes and beliefs has occurred. 
 
 12. This is a good thing to help analyze the human 
 factors and how they influence safety in the mine. I hope 
 the possibilities exist to make this available to all mining 
 companies for training of their supervisors and employees. 
 
 13. The BAS gives good appreciation for the hazards 
 and the miner's own reactions. It shows how one's own gut 
 reactions in a time of crisis can be the wrong thing to do. 
 
 14. A very good program that should be made available 
 to the industry for training supervisors as well as workers 
 alike. 
 
 15. The whole problem is management. Until they will 
 pay something for safety, nothing is going to help. 
 
 16. This should be required of all the employees who 
 are known to work carelessly. You should pick out the ones 
 who don't work safe and make them take this. 
 
 17. I had fun, but I would like to see how this correlates 
 with real data and situations. 
 
 18. Like any training, just doing this does not make 
 you a more safe person. 
 
 19. I doubt that this is useful in real situations. I don't 
 see how this will be very useful. Some people just work 
 safer than others, just like some people are more messy 
 than others. You can't change people. 
 
 As table 11 shows, before working with the BAS, many 
 of the subjects felt that high accident rates were inherent 
 in mining. Thus, they seldom perceived that the miner's 
 own behavior was a cause of accidents. Consistent with 
 this, many of the subjects felt that conventional training 
 was ineffective. However, as a result of using the BAS, 
 these attitudes dramatically changed. As table 11 shows, 
 over 80 pet of the subjects felt the BAS was a useful and 
 effective training aid. Most of the subjects agreed that the 
 BAS clarifies behaviors that lead to accidents and 
 sharpens a person's ability to recognize and avoid hazards. 
 Three-fourths of the subjects said that using the BAS could 
 help reduce mine accident rates. 
 
 For some subjects, these attitude shifts were rather 
 dramatic. Table 12 summarizes the attitude shifts for 
 
18 
 
 those subjects who stated in their pre-BAS experience 
 questionnaires they did not believe conventional training 
 was effective. Before their BAS experiences, these nonbe- 
 lievers felt that individual miners could do little to prevent 
 accidents. They had minimal appreciation for the ways 
 that accidents could be caused by a system of interacting 
 factors, and they generally denied that miners often take 
 chances. Their BAS experiences caused statistically sig- 
 nificant changes in these attitudes. After their BAS exer- 
 cises, these nonbelievers had a much greater appreciation 
 for mine accident systems. They became much more aware 
 of the ways that miners unwittingly take risks and the 
 ways in which individual behaviors are vitally important 
 within the entire accident system. They still did not be- 
 lieve in conventional training. However, they thought the 
 BAS was a potentially effective new approach to training. 
 
 The post-BAS experience questionnaire comments 
 summarize both the favorable and unfavorable opinions. 
 Three messages are clearly conveyed in these comments. 
 One, the BAS was generally viewed as a potentially useful 
 and effective device. It may be much more effective than 
 many existing conventional training aids. Two, the BAS is 
 a potentially useful device for managers to use in explor- 
 ing alternative safety actions, work environment designs, 
 and organization structures. Three, the actual effective- 
 ness of the BAS in any application appears to depend 
 greatly on the creativity of the users and the philosophy of 
 use behind the BAS. The BAS is at best an information 
 collection and decisionmaking aid. It can help the user 
 gain numerous valuable insights, but it cannot be ex- 
 pected to make decisions or provide the ultimate answers. 
 
 
 SUMMARY AND CONCLUSIONS 
 
 A simulation model of human accident systems was 
 developed based on an analysis of MSHA accident reports, 
 some theories about human behavior, and a mathematical 
 approach called catastrophe theory. A Fortran computer 
 program, the behavioral accident simulator (BAS), was 
 written for this model. 
 
 BAS users sit at a computer terminal or personal com- 
 puter and experiment with various scenarios of their 
 choosing. They observe the impacts of their decisions and 
 their choices on accident rates and fatalities. The BAS 
 enables the user to see the impacts of each decision and 
 each behavior on every variable in the system. Users are 
 able to vividly observe how these various factors cascade 
 together to cause fatal accidents. 
 
 The BAS has been extensively tested in over 200 labo- 
 ratory simulations and in a pilot field test with 28 coal 
 mine industry personnel. These experiences demonstrated 
 that the BAS is a highly valuable decision aid and train- 
 
 ing device. Mine managers can use the BAS as a kind of 
 management laboratory to ask what-if questions about al- 
 ternative organizational arrangements, working condi- 
 tions, and other actions as a basis for choosing the best 
 ones. Trainers can use the BAS as an innovative and effec- 
 tive way to teach mine management skills. Moreover, the 
 BAS can be used by mine employees to improve their un- 
 derstanding of how their individual behaviors can cascade 
 into serious accidents. 
 
 The BAS is a unique and innovative approach to the 
 study and alleviation of human error accidents. It empha- 
 sizes experimentation with the entire system of individ- 
 ual, group, organizational, management, and mine 
 variables to reduce accident rates. The pilot field test with 
 mine employees demonstrated that this approach can sig- 
 nificantly improve employee safety attitudes and mine 
 management decisionmaking. 
 
 RECOMMENDATIONS FOR FURTHER RESEARCH 
 
 The success of the BAS encourages the development 
 and implementation of advanced BAS-type models 
 throughout the mining community. More sophisticated, 
 precise, and powerful models should be developed, perhaps 
 using stochastic variables and nonlinear dynamics tech- 
 niques. Stochastic variables would permit more accurate 
 modeling of actual human accident behaviors. Nonlinear 
 dynamics techniques could be used to comprehensively 
 
 treat the complex and varying interactions between these 
 variables that are common in the real world. Because of its 
 deterministic nature, the BAS is limited in its capabilities 
 to handle these realities. However, now that its utility has 
 been demonstrated, a step up to more elaborate models 
 seems warranted. Because of their greater precision and 
 power, more elaborate models would be both easier to vali- 
 date and more useful to the mining community. 
 
 REFERENCES 
 
 1. Berelson, B. Content Analysis. Hafner (New York), 1971, 
 pp. 5-70. 
 
 2. Blalock, H. M. Causal Inference in Nonexperimental Re- 
 search. Univ. NC Press (Raleigh, NC), 1964, pp. 3-27. 
 
 3. Conway, J. R. Path Analysis Techniques for the Behavioral 
 Sciences. Penton (London), 1983, pp. 21-105. 
 
 4. Denton, D. K. The Unsafe Act. Prof. Safety, July 1979, pp. 
 34-37. 
 
 5. Haddon, W., E. A. Suchman, and D. Klein (eds.). Accident 
 Research. Harper & Row (New York), 1964, pp. 29-87. 
 
 6. Hale, A., and M. Hale. A Review of the Industrial Accident 
 Literature. H. M. Stationary Office (London), 1972, 75 pp. 
 
 7. Holmes, T. H, and R. H. Rahe. The Social Readjustment 
 Rating Scale. J. Psychosom. Res., No. 11, 1967, p. 216. 
 
 8. Kripendorff, K. Content Analysis: An Introduction to Its 
 Methodology. Sage Publications (London), 1980, pp. 105-210. 
 
19 
 
 9. Lykes, N. R. A Psychological Approach to Accidents. Van- 
 tage Press (New York), 1954, pp. 37-112. 
 
 10. McGlade, F. S., and L. Brody. Adjustive Behavior and Safe 
 Performance. Thomas (Springfield), 1970, pp. 51-118. 
 
 11. Meister, D. Behavioral Analysis and Measurement Meth- 
 ods. Wiley (New York), 1985, pp. 304-310. 
 
 12. Mihal, W. L. Individual Differences in Perceptual-Informa- 
 tion Processing and Their Relation to Accident Behavior. Ph. D. 
 Thesis, Univ. Rochester, Rochester, NY, 1974, 205 pp. 
 
 13. Petersen, D. The Human Error Model of Accident Causa- 
 tion. Focus (Rochester, NY), July 1983, pp. 97-100. 
 
 14. . Human Error Reduction and Safety Management. 
 
 Garland Press (New York), 1982, pp. 29-101. 
 
 15. Pfeifer, C. M., J. L. Stefanski, and C. B. Grether. Psychologi- 
 cal, Behavioral, and Organizational Factors Affecting Coal Miner 
 Safety and Health. Westinghouse Behavioral Services Center 
 Tech. Rep. BSC-5, 1972, pp. 12-99, NTIS PB 275599. 
 
 16. Poston, T, and I. Stewart. Catastrophe Theory and Its Ap- 
 plications. Pitman (London), 1978, pp. 53-186. 
 
 17. Siegel, S. Nonparametric Statistics. McGraw-Hill (New 
 York), 1956, pp. 36-44, 161-166. 
 
 18. Souder, W. E. The Behavioral Accident Phenomenon: Re- 
 view and Models. Univ. Pittsburgh TMSG Study Rep., Feb. 22, 
 1984, 68 pp. 
 
 19. . A Catastrophe Model of Behavioral Accidents. Paper 
 
 in Proceedings of the IEEE International Conference on Systems, 
 Man & Cybernetics (Tucson, AZ, Nov. 12-15, 1985), pp. 514-516. 
 
 20. Thorn, R. Structural Stability and Morphogenesis. Ben- 
 jamin, 1975. pp. 5-87. 
 
 21. Whitkin, H. A. Psychological Differentiation and Forms of 
 Pathology. J. Abnorm. Psychol., v. 70, 1965, pp. 317-336. 
 
 22. Zeeman, E. C. Catastrophe Theory. Sci. Am., v. 234, 1976, 
 pp. 56-83. 
 
 U.S. GOVERNMENT PRINTING OFFICE: 1988 — 505-016/80,011 
 
 INT.-BU.OF MINES,PGH.,PA. 28693 
 
295C 
 
U.S. Department of the Interior 
 Bureau of Mines— Prod, and Distr. 
 Cochrans Mill Road 
 P.O. Box 18070 
 Pittsburgh, Pa. 15236 
 
 OFFICIAL BUSINESS 
 PENALTY FOB PRIVATE USE. S900 
 
 ] Do not wi sh to recei ve thi s 
 material, please remove 
 from your mailing list. 
 
 "2 Address change. Please 
 correct as indicated. 
 
 AN EQUAL OPPORTUNITY EMPLOYER 
 

 
 ,o< 
 
 
 afc % ^ »£■£• ^ ^ •Ssxk*. ^ ^ •»&*• ^ /*&^ % £* »i9B* %. ** 
 
 JF**Mk% #tfsi>S ^*tite?% ^.$a&S &*. titer % *•:££& 
 
 ' SxKfoSi 4^tf£fe*« **\^i*\> ^«2ffifi»*» ^•li&V 
 
 «*v^ v^v o«>v \r^9£> v*-**> # v^v 
 
 6° 42afo% 6° -Stated 4?. titer \ <P -s£fr% ^. titer 
 
 6 <fc 
 
 
 
 ^ ♦♦ 
 
 **% -^1^ ,/\ -JP? A ^% l -W«* ,/\ -SR* y\ ?' >% 
 
 
 • V^' 
 
 
 5- ; 
 
 3W5 ^. % 
 
 % ^ 
 ^H v 
 
 
 "Gam: #<*. i) 
 
 %^SPV %v!B^^ %*3Rf^^ V^^\/ %^f"- v ^ %^ T v # 
 
A* ^ 
 
 
 
 ^0* 
 
 40. 
 
 cv * 
 
 *bV" 
 
 o5°^ .1 
 
 ^d* 
 
 
 ■4, °>. 
 
 ; . ^ •«•' v : A % s •*£&• w 
 
 ...» A <v ^TTT* A*- *o, '«.»* A x» *vT7T» .g* ^5 '0.1* a <► 
 
 tar- ^« .«l^ ^^ .v^to* -^ c ^gi^. ^ .\jf|a; ^, 
 
 l* v v *o 
 
 O „ ' .0 
 
 S\ 
 
 ^6* 
 
 r*°* - 
 
 
 A^ o^^f^^ A-* 
 
 '•-\ /,.^4. °o ,* 4 .^:-^ oo*.^.°o >*.^%^ c o> 
 
 
 s ^"^ "mww: a* >^ J .^iii^'»" ^""^ 
 
 
 l>* 
 
 
 . 5 «f . 
 
 
 
 !\°* 
 
 
 -.^ o_ 
 
 
 ^ ' ^ ^ %, \% ^ : ^'\/ 
 
 0? .°«^ia*. ^0^ 
 
 
 
 ^^ 
 
 '^S 3 
 
 
 ^d* 
 
 n 4o^ 
 
 
 ^°^ 
 
 
 I-- 
 
 % > ^V^t\ c^iSgkr °o ^-^^\ c^°- ^-^Ur^ 
 
 
 .0° *~ 
 
 
 
 
 
 
 
 
 5 • 
 
 :-%> ^s#mk°* J^^kSr Ac^>% ^-iSit\ c°**c^,% 
 
 A * - 
 
 
 
 A.V-^. 
 
 
 
 A 
 
 •• s"\ •' 
 
 A c ° " ° * <^ 
 
BBM 
 
 ■ 
 
 . 1 I 
 
 ■ 
 
 :• ■ 
 
 ■ 
 ■ 
 
 HS