Correlates of Work Injury Frequenc anyd Duration Among ... · fore, ther ies a economin c incentiv fo workerr e tso return to work earlier Th. opportunite y o cosf ret-maining out

Journal of Occupational Health Psychology2001, Vol. 6, No. 3, 229-242

Copyright 2001 by the Educational Publishing Foundation1076-8998/01/S5.00 DOI: 10.1037//1076-8998.6.3.229

Correlates of Work Injury Frequency and DurationAmong Firefighters

Hui Liao, Richard D. Arvey,and Richard J. ButlerUniversity of Minnesota

Steven M. NuttingCity of Minneapolis, Minnesota

This study examined demographic, personality, and economic incentive correlates of workplaceinjuries suffered by 171 firefighters over a 12-year period. Results showed that female firefightersexperienced more injuries than male firefighters. Several Minnesota Multiphasic PersonalityInventory (MMPI) scales (Conversion Hysteria, Psychopathic Deviate, and Social Introversion)were positively related to injury frequency. Regression analyses revealed that age, tenure, gender,marital status, type of injury, and wage variables were significant when predicting the duration ofinjuries as well as an interaction between marital status and gender. Two MMPI scales (Psycho-pathic Deviate and Schizophrenia) were also significantly related to injury duration. Indemnitycost estimates were calculated. The results underscore the importance of distinguishing theduration of injury from the occurrence of injury.

National statistics show that firefighters are at highrisk for injuries while on the job ("1996 U.S. Fire-fighter Injuries," 1997) because of their exposure toenvironmental hazards. The risk of a fatal incident forfirefighters is also three times greater than for allworkers (Clarke & Zak, 1999). Given this relativelyhigh risk of injury, identifying true correlates ofinjury risk for this occupational group is particularlyimportant. Although enormous effort has been di-rected to countermeasures ranging from improvedprotective clothing and equipment to more effectivehealth and safety regulations, we know relativelylittle about how individual difference factors influ-ence firefighter injuries. Knowledge of personal dif-ferences associated with accident occurrences andseverity can be applied to job assignment, safetytraining, supervision, and retum-to-work intervention(Hansen, 1991).

Prone (1998) noted several different categories ofpotential predictors of injuries in his discussion ofadolescent workers, including demographics, person-ality, substance use, health, and employment factors.

In the present study, we expanded his model in sev-eral ways. First, we differentiated between injuryfrequency and injury duration as two different out-come variables. Injury frequency refers to the numberof injuries suffered by a firefighter within a certaintime period; injury duration is time elapsed betweendate of injury and first-return-to-work date and iscommonly used as a proxy for injury severity in theworkers' compensation literature (see Butler, 1994;Johnson & Ondrich, 1990). Although these two out-comes of injury are closely related in the sense thatduration is only relevant after an injury occurs andthat both of them have important cost implications(Butler & Worrall, 1985), theoretically they mayhave different antecedents. Second, we examined abroader set of demographic, personality, and eco-nomic incentive variables affecting workplace inju-ries among firefighters while controlling for otherfactors that are also potential correlates of injuries.Third, personality data were gathered at the time ofhire and correlated with subsequent injuries over a12-year period, thus making this a predictive design.

Hui Liao, Richard D. Arvey, and Richard J. Butler, In-dustrial Relations Center, Carlson School of Management,University of Minnesota; Steven M. Nutting, Human Re-source Department, City of Minneapolis, Minnesota.

We wish to thank Jim Clack, Ellen Velasco Thompson,Jon Walstead, and Cindy Krepela-Gargulak for their help onthis project.

Correspondence concerning this article should be ad-dressed to Hui Liao, Industrial Relations Center, 3-300Carlson School of Management Building, 321 19th AvenueSouth, Minneapolis, Minnesota 55455. Electronic mail maybe sent to [email protected].

Correlates of Workplace Injury Frequencyand Duration

Demographics

Gender, tenure, and age have been found to beassociated with the frequency of workplace injuries.Frone (1998) reported that adolescent boys had morework injuries than did adolescent girls and suggestedthat such observed differences were a result of dif-ferential occupational exposure to risk and on-the-job

229

230 LIAO, ARVEY, BUTLER, AND NUTTING

substance use among boys and girls. However, adultmen typically file fewer claims for workplace injuriesthan women when controlling for occupation (Butler,Hartwig, & Gardner, 1997; DeBobes, 1986; Hirsch,Macpherson, & DuMond, 1997). For example, De-Bobes (1986) noted that women had a 45% greateraccident rate than men. This was attributed to thedisproportionate amount of external responsibilitiesthat women have to contend with before and afterwork. In the present study, we examined the injuriesof both men and women of the same occupation, inwhich presumably the exposure to injury risk is thesame for both groups. We predicted, on the basis ofprevious literature, that female firefighters will expe-rience more workplace injuries than male firefighters.

Age is generally related to tenure. As a result, it isinaccurate to discuss tenure's effects on injury with-out controlling age, and vice versa. In this study, weincluded age and tenure simultaneously to examinetheir independent impact on injuries. Frone (1998)presented alternative models in which tenure is re-lated to injuries. One model assumes that tenure is aproxy for experience, and that with more experience,workers are less likely to injure themselves. Alterna-tively, more experienced and longer tenured employ-ees might be assigned more hazardous tasks andduties, and therefore be more likely to be injured.Previous empirical research shows mixed findings onthe nature of the relationship: Cellier, Eyrolle, andBertrand (1995) reported a negative relationship,whereas Frone (1998) reported a positive relation-ship. Iverson and Erwin (1997) failed to detect asignificant relationship between tenure and injuries.Given the ambiguity of the literature, we made nodirectional prediction regarding the tenure and injuryfrequency relationship.

Prior research has shown age to be negativelyrelated to the frequency of work injuries (e.g.,Kingma, 1994; Kraus, 1985). A possible reason isthat with increasing age, employees maintain a morecautious attitude toward risky behaviors, hence fewerinjuries occur (Hirsch et al., 1997). Age, however,was shown to be unrelated to injuries among adoles-cents for whom the age range was restricted (Frone,1998). Because the age range represented among theparticipants in our sample was wide (19-49 years),we predicted a negative relationship between age andworkplace injury frequency.

We expected these demographics variables also tobe related to the duration of an injury. Prior estimatesof workers' compensation duration for workforcepopulations (not restricted to firefighters) indicatethat younger workers have shorter claim duration

than do other age groups (Butler et al., 1997; Butler& Worrall, 1985; Johnson & Ondrich, 1990). Thebiology of aging makes it clear that age is associatedwith a decline in physical hardiness or recoveryspeed from tissue damage (Butler et al., 1997); as aresult, older workers need more time to recover fromthe injuries experienced on the job. Therefore, weexpected age to be positively related with injuryduration. We also tested the impact of tenure oninjury duration while controlling for age, but wemade no specific prediction of the tenure-durationrelationship.

Gender is another commonly studied correlate ofinjury duration. Johnson, Baldwin, and Butler (1999)found longer indemnity duration for injured femaleworkers. We therefore expected that female firefight-ers will have longer injury durations. As to maritalstatus, Butler and Worrall (1985) and Worrall, Appel,and Butler (1987) found that married workers hadlonger recovery periods than nonmarried workers,controlling for gender. Generally, lost wages of in-jured workers are not fully replaced under workers'compensation (most states have a replacement rate of67%, i.e., two thirds of the work-time wage loss canbe compensated up to a statutory maximum); there-fore, there is an economic incentive for workers toreturn to work earlier. The opportunity cost of re-maining out of work is most likely lower, however,for married workers who have a working spouse.Thus, married workers may have less financial incen-tive to return to work and might tend to stay longeron disability. Consequently, we expect married fire-fighters to have longer duration periods than unmar-ried firefighters.

There is also some reason to expect an interactionbetween marital status and gender on injury duration.Hirsch et al. (1997) showed that workers' compen-sation claim rates were different according to themarital status and gender of the employee. Althoughthe pattern of results for marital status was similar formen and women, the differences observed amongmen were small and insignificant, suggesting an in-teraction. Hirsch et al. (1997) suggested that suchdifferences might occur because of differential in-come "needs"—those employees with greater needs(e.g., having fewer alternative sources of income)take fewer risks and have a greater incentive not toforgo earnings in the event of minor injuries. Thus,unmarried women are expected to demonstrate alower injury duration rate than married women,whose spouses are likely to be working. Hence, wepredicted an interaction between gender and maritalstatus, specifically that lower injury duration rates

WORK INJURY FREQUENCY AND DURATION 231

will be observed for unmarried women than for mar-ried women but no differences between married andunmarried men.

Personality Constructs

A number of personality constructs have been hy-pothesized to be related to the frequency of work-place injury, and there is some empirical support forthese propositions. For example, Hansen (1991)found considerable support for the proposition thatmeasures of general social maladjustment (e.g., de-linquency, sociopathic attitudes, drinking, and au-thority problems) and measures of neuroticism wererelated to accidents. Thus, there is reason to believethat the constructs assessed using the Minnesota Mul-tiphasic Personality Inventory (MMPI) may be re-lated to on-the-job injuries. MMPI is the psychomet-ric test most frequently used for screening policeofficers (Burbeck & Furnham, 1985) and has alsobeen widely used in other high-risk or sensitive jobssuch as nuclear power plant workers and firefighters(Kelley, Jacobs, & Fair, 1994). Past studies show thatworkers' compensation claim samples differ fromnorm groups on several MMPI scale scores (e.g.,Gandolfo, 1995; Hersch & Alexander, 1990; Repko& Cooper, 1983). These studies suffer, however,because they are conditional on employees havingalready reported a workers' compensation claim, andthey fail to establish a direct correlation betweenMMPI scales and workers' compensation claims. Al-though no study has directly examined the predictiveefficiency of MMPI scale scores with injuries, par-ticular MMPI scales seem logically related to work-place injuries, or at least to the reporting of injuries.

The Hypochondriasis (Hs) scale was developed ona group of patients with neuroses who showed anexcessive concern about their health, presented avariety of somatic complaints with little or no organicbasis, and rejected assurances that there was nothingwrong with them. High scorers typically see them-selves as physically ill and seek medical explanationsand treatment for symptoms (Graham, 1993). Wor-ries over their physical health dominate their life andoften seriously restrict the range of their activitiesand interpersonal relations, possibly increasing therisk of injury. Past research supports a positive rela-tion between Hs and workplace injuries (Gandolfo,1995; Repko & Cooper, 1983).

The Depression (D) scale was developed on psy-chiatric patients with various depressive reactions orwith a depressive episode of a manic-depressive dis-order. High scorers tend to feel sad, blue, unhappy,

and pessimistic about the future. Several studies havereported that depression was implicated in accidentcausation (e.g., Craske, 1968; Selzer, Rogers, &Kern, 1968). Iverson and Erwin (1997) suggested thechronic experience of negative emotional statusmight lead to lapse in attention or higher levels ofdistractibility, thereby increasing the likelihood ofinjury at work. High depression scores also indicateindividuals who report physical complaints, sleepdisturbances, weakness, fatigue, loss of energy, poorconcentration, and so on (Graham, 1993). Thus, fire-fighters who are depressed may be more vulnerableto injuries when fighting fires, which is an extremelyphysically demanding and dangerous job.

The Conversion Hysteria (Hy) scale was con-structed on patients who exhibited some form ofsensory or motor disorder for which no organic basiscould be established. People with elevated Hy scalescores often feel overwhelmed and tend to developsymptoms when under stress (Graham, 1993). Wetherefore expected firefighters with higher Hy scoresto report more injuries. Past research provides sup-port for this assertion (Gandolfo, 1995; Hersch &Alexander, 1990).

The Psychopathic Deviate (Pd) scale was devel-oped on patients diagnosed as having psychopathic,asocial, or amoral type personalities. High scores onthis scale indicate people who have difficulty incor-porating the values and standards of society, arerebellious toward authority, tend to act without con-sidering the consequences of their actions, show poorjudgment and take risks, and tend to be aggressive(Graham, 1993). Logically then, individuals scoringon the high end of the Pd scale seem more likely toignore rules and regulations regarding safety andhealth, resulting in an increased risk of injury. Driv-ers with various social maladjustment problems suchas irresponsibility and authority problems are foundto have higher accident rates (McGuire, 1972). Fur-thermore, high scorers on the Pd scale may rush tocomplete a task without adequate consideration of theconsequences, putting themselves (and others) at ahigher risk for error and accidents. Hansen's (1991)review of the literature relating impulsivity to acci-dents also supports the notion. Additional evidence isprovided by Bigos et al. (1991), who found a positiverelationship between the Pd scale and workplaceinjuries.

On the basis of the literature cited above, weexpected that the MMPI scales of D, Hs, Hy, and Pdwould be positively related to injury frequency. Wealso examined the relationships of other major clin-


ical MMPI scales against injury frequency but madeno specific predictions.

Personality may be more directly related to dura-tion of injuries rather than the number of injuriessuffered among firefighters. Once someone leaves theworkplace because of injury, work norms are lessimportant. Therefore, the decision to return to work ismore likely to be a function of personality factorsthan the decision to miss work. The absenteeismliterature makes just this argument: Attitudes andindividual differences are more important when nor-mative constraints on attendance are reduced (e.g.,Johns, 1997). Hence, firefighters with elevatedMMPI profiles might be more likely to stay off thejob longer after an injury. Given the paucity of ex-isting research along this line, we tested the relation-ships between various MMPI scales and duration ofinjuries but made no hypotheses regarding the direc-tion for any particular scale and the duration ofworkplace injuries.

Economic Incentive Factors

We also expected the duration of workplace inju-ries to be related to economic incentive factors suchas workers' wage and potential workers' compensa-tion benefits. In a study of injury duration for workerswho reported low-back injuries in Illinois, Butler andWorrall (1985) found that the higher the preinjurywages, the shorter the expected duration of the claim.The model adopted to explain this result was thathigher preinjury wage provided more economic in-centive for injured workers to return to work. On thebasis of this theory and previous literature, we ex-pected the weekly wages of the firefighters prior totheir injury to be negatively related to the duration oftheir injuries.

Although all workplace injuries filed in our samplestate had an identical 67% wage replacement rate andwere subject to the same statewide maximum pay-ment, the firefighters were sometimes entitled to In-jury-on-Duty (IOD) benefits. For firefighters, onlylost-workday injuries occurring during a fire emer-gency run were to be considered as IOD claims andwere reimbursed at 100% of lost wages. Other validinjuries, such as those occurred while administeringemergency medical aid, are filed as an ordinary work-ers' compensation claim and lost time is compen-sated at two thirds of the preinjury wage rate. Butlerand Worral (1985) found that the greater the workers'compensation benefits, the longer the duration of theinjury. On the basis of this literature, we expected

lost-time IOD claims to be longer than lost-timenon-IOD claims.

Method

Sample Description

The study was conducted on the basis of data drawn fromthe fire department of a major midwestern U.S. city. Injurydata, personality scores, wage, demographic, and other rel-evant information were collected from the employees' riskmanagement, medical, and personnel files maintained by thefire department. Workers' compensation data were availablestarting from 1987. Therefore, we chose January 1, 1987 toDecember 31, 1998 as the studying period. A total of 194nonclerical employees were hired during this period, and allof them took the MMPI as part of the psychological eval-uation during the application process. Few individuals werescreened out because of elevated MMPI profile scores.Complete MMPI scores were available for 171 employees.We based our model estimates on these 171 employees.Statistical tests showed that the sample did not differ fromthose 23 firefighters with missing MMPI scores in terms ofgender, age, and ethnicity. All of the firefighters in thesample worked full time. The majority of the firefighters inthis sample were male (83%) and Caucasian (68%). Theminority racial breakdowns were as follows: 32 AfricanAmericans (19%), 6 Hispanic (4%), 4 Asian Pacific Island-ers (2%), and 12 American Indian or Alaskan Native (7%).Compared with the total of 402 employees (including the208 employees hired before 1987) serving in the fire de-partment on December 31, 1998, the sample containedgreater number of female and minority employees, reflect-ing the recent efforts on the part of the city to employ amore diverse group of employees. Firefighters in this sam-ple occupied a variety of positions: firefighter, fire motoroperator, fire captain, battalion chief, and district chief,among others.

The workers' compensation claim records provided in-formation on all injury claims filed by fire departmentemployees from 1987 to 1998. During the 12-year period(1987-1998), 765 claims were filed, of which 1.4% weredenied as valid claims. Only those claims that were judgedas valid were used in our analyses to represent an injuryoccurrence. Records with obvious coding errors (e.g., re-turn-to-work date was earlier than injury date) were alsodeleted, leading to 678 lost-time injuries used for the dura-tion analysis. The per capita injuries experienced for fire-fighters over the whole sample period ranged from 0 to 15.

Measures

Dependent variables. The two dependent variables ofmajor interest in this article were injury frequency andinjury duration. Injury frequency was measured by a simplecount of the number of injuries filed for a firefighter for eachparticular calendar year in which the firefighter was em-ployed. We determined the duration of each injury by cal-culating the elapse time between an injury and a firefighter'sreturn to work. Almost all claims had been closed by the endof 1998, so most insurance payments had been made. How-ever, four injury claims were still open by the end of 1998,


indicating that the injured workers had not completed theirspells on disability and were continuing to receive workers'compensation benefits. These continuing status recipientsare usually referred to as right-censored (e.g., Butler &Worrall, 1985).

Demographics. Gender was coded as 1 if the firefighterwas female and 0 if male. Age and tenure were coded inyears. Marital status was coded as 1 if the firefighter wasmarried as of the date of injury and 0 if not married. Data onmarital status were available for firefighters who were in-jured but not for those individuals for whom no injuryclaims were made. Thus, it was only included in analysespredicting injury duration.

MMPI scales. The MMPI is one of the mandatoryscreening tests for firefighters in this particular city and hasbeen used for more than 20 years. Until 1991, the firstversion of the MMPI was administered. Since 1993 (no onewas hired in 1992), the MMPI-2 has been administered.Although these two versions differ, the 10 clinical scales arecomparable using T scores (Graham, 1993). These T scoresfor 9 of the clinical scales were used for our analyses. TheMasculinity-Femininity scale was not used because it hasopposite interpretation for female and male participants.Longitudinal studies (Costa & McCrae, 1986) demonstratethat individual differences in personality are highly stable inadulthood. Pancoast and Archer (1988) found stability in theMMPI scale scores of normal adult samples across 40 years.R. L. Greene (1990) also found that the validity and clinicalscale scores were very stable over 40 years in samples ofpsychiatric patients. Thus, the fact that the MMPI data forthe present study were collected over a 12-year period isunlikely to have much effect on the reported scores.

Economic incentive variables. Data on the weeklywage for each firefighter who filed an injury claim werecoded as a continuous variable. This variable was includedonly in the duration model because the information wasonly available for individuals who were injured. A dummyvariable, IOD, was included in the duration model and wascoded as 1 if the injury was approved as an IOD claim, and0 if otherwise.

Other controls. Race was represented by two dummyvariables: White, coded as 1 if the firefighter was Caucasian;and Black, coded as 1 if the firefighter was African Amer-ican. The omitted ethnic groups include Hispanic, Asian,American Indian, and so on. In the injury duration model,two dummy variables, fire motor operator and fighter, werecreated to represent the type of position held by the indi-vidual firefighter at the time of an injury. The omittedcategories include fire captain, chief, trainee, and so on.They served as control for different levels of risk exposureassociated with different positions. Six dummy variableswere created to represent different types of injuries in du-ration model. These variables represent injuries due to backsprain, other strains and sprains, burns or exposure to chem-icals, fracture or laceration, contusion, and potential expo-sure to contagious diseases. For both the frequency andduration models, year dummy variables were created foreach of the years from 1987 to 1997 (year 1998 was omittedand used as the comparison year) and served as controlvariables to capture any changes in the state's administra-tion of workers' compensation as well as secular trends inthe injury rates and duration that would be not explained bythe demographic changes in the workforce.

Estimation Models

The annual injury count per person was used as thedependent variable in the injury frequency model. Becausethis count variable is clearly discrete and contains a sub-stantial number of zeros and other small values, the negativebinomial regression model was used to account for theseattributes. The negative binomial regression has been usedextensively in statistics and actuarial practice to model thenumber of insurance claims filed in a year (W. H. Greene,1997). The estimated regression coefficient for an explana-tory variable indicates the percentage change in the ex-pected injury frequency, given a unit change in that explan-atory variable. Stata (Release 6.0) was used to estimate thenegative binomial regression. Because the data involvednonindependent data points (data were recorded for thesame individuals over time), the robust and cluster optionswere used to adjust the covariance matrix to account forrepeated measures across individuals (W. H. Greene, 1997;Stata Reference Manual, 1999; Stata User's Guide, 1999).

We also estimated injury duration for all firefighters whotook time off work because of an injury. Survival analysis,which has its origin in medical statistics, has been widelyapplied to the study of unemployment spell (Kiefer, 1988)and the duration of nonwork spell due to injury (Johnson &Ondrich, 1990), and has recently been introduced to orga-nizational psychology analyses of turnover and absenteeism(Johns, 1997; Morita, Lee, & Mowday, 1989). Survivalanalysis is well suited for identifying determinants of injuryor nonwork spells. Although most of the injury claims hadbeen closed by 1998, some were still open or right-cen-sored. Right-censored observations represent a problem forordinary regression models (e.g., Butler & Worrall, 1985)but can be readily modeled by survival analysis. Specifi-cally, we used a Weibull Survival model. This is a com-monly used model in the unemployment and workers' com-pensation literature (e.g., Butler & Worrall, 1985; Heckman& Singer, 1984). The Stata survival analysis program wasused, again with the robust and cluster options to deal withthe repeated measures aspect of the data. The resultingestimated regression coefficient for an explanatory variableis interpreted similarly to that in the negative binomialmodel: the percentage change in the expected durationgiven a unit change in that explanatory variable.

Results

Predicting Injury Frequency

Table 1 reports the means, standard deviations, andintercorrelations for the data used in the negativebinomial regression for the frequency of injury data.As can be seen in Table 1, there was an average of0.62 injuries per individual per year for the 171firefighters. Inspection of the intercorrelations amongthese variables and the frequency of injuries suggeststhe presence of relatively low relationships betweenthese variables and injuries in this sample—only fiveof the single-order correlations were significant at the.05 level: gender, tenure, the D scale, the Psychas-thenia scale, and the Schizophrenia (Sc) scale.


i

a

c

.2

I

3 I

r- CN ONen »n oor-' CN i-<

o r- en >n —'

>n NO CN i-i

t~* en en o I~H O ON <nen NO en en >n i~ *-*

I oo (S »n o «n ON en' O fN O •—i i—i O •—'

^-.^,^-,^-ifvjOi-H'—(i-«OOr r r r ' r r r r

> < N Q N < N e n o J C N O O N r t - N O O

i r r

;$:S O N t ^ - e N i n O N e n ^ e n ^ b o O i — '> O p C > c n ' - ' r 4 p p e N O p pr r r r r r r r ' r

m —" ON O —( O *i s s s s s s ;

) en vo <N fN "

*«II3 9

-H" o rf-o- .S 3S § ^o. S sii «>S " 1

iT'3 u

I | |.fi 4) f-5y *o o.

ll 11Sill

v

in 'g

*~i oa~i ">

, -e -B ^!.3 s o

1! ° S.El|,l §

g '1 £ "


Table 2 exhibits the results for the negative bino-mial model in which the natural log of the count ofinjuries per year per individual was the dependentvariable. The impact of demographic and personalityvariables was estimated. The resulting chi-square was64.05, indicating that the full model was significant atthe .01 level. The pseudo R2 statistics (see Sheridan,1992) is .05; thus, a large portion of variance wentunexplained.

In support of our hypothesis, female firefightersexperienced a greater number of injuries comparedwith male firefighters (p < .05), with female fire-fighters reporting 33% more injuries than male fire-fighters. Contrary to our prediction, tenure and agewere not significantly related to the frequency ofinjury in this sample.

Consistent with our predictions, the MMPI scalesHy (p < .05) and Pd (p < .05) were positively relatedwith injury frequency. However, the MMPI scales of

Table 2Estimates of Negative Binomial Regression forFrequency Model (N = 1,286)

Variable

FemaleTenureAgeHsDHyPdPaPtScMaSio"/(25)Prob > x2

Pseudo R2

Log likelihood

Unstandardized ft

.330**-.046-.009

.003-.013

.022*

.012*-.008-.012-.011

.004

.020*

.13364.050.0000.05

-1,324.5465

SE

.133

.026

.014

.016

.010

.013

.008

.008

.014

.013

.006

.010

.073

Note. One-tailed tests were performed for female, age, Hs,D, Hy, and Pd. Two-tailed tests were performed for theremaining independent variables. Control variables, includ-ing 11 year dummies and two racial dummies, were omittedfrom the table. Hs = hypochondriasis; D = depression;Hy = conversion hysteria; Pd = psychopathic deviate;Pa = paranoia; Pt = psychasthenia; Sc = schizophrenia;Ma = hypomania; Si = social introversion.a a is the overdispersion parameter. When a = 0, thenegative binomial model reduces to Poisson model. A like-lihood ratio test showed that a was larger than zero andstatistically significant in the sample, indicating the Poissonmodel would have underestimated the variance.*p < .05. **p < .01.

Hs and D failed to show significant relationships tothe frequency of injuries. Another scale, Social In-troversion (Si), was significantly related (p < .05) tothe frequency of injuries when a two-tailed test wasapplied.

Among the year dummy variables, the years 1989and 1993 were significant; there was a tendency in1989 to report fewer injuries after controlling for thedemographic changes across years, whereas therewas a tendency in 1993 to report more injuries. Noevidence was found for racial differences in theirexpected injury rates.

Predicting Injury Duration

More information about individual claims wasavailable in the duration model sample than in thefrequency model sample. In addition to basic demo-graphics and MMPI scores, the duration model sam-ple also contained data for each record of injuryabout the injured firefighters' job title, marital status,wage rate, injury types, whether approved for IOD,and so on. Table 3 presents means, standard devia-tions, and intercorrelations for these variables. Table4 presents the results of the survival analyses. Theresulting chi-square for the overall model was529.07 and was significant at p < .01. The pseudoR2 statistics was .45, indicating 45% of the variancein injury duration was explained by the model wespecified.

Inspection of Table 3 shows that the averagelength of injury duration for this sample was 10.04days. Again, the zero-order correlations between theduration of injury and other variables were consis-tently small—all were under .10, except IOD, whichhad a correlation of .18 with duration.

Table 4 shows that female firefighters and marriedfirefighters had significantly longer claim durations,consistent with our hypotheses. The interaction betweengender and marital status was also statistically signifi-cant, but the nature of the interaction was not consistentwith our hypothesis. Compared with unmarried malefirefighters, married female firefighters had a 38%shorter injury duration, unmarried female firefightershad a 49% longer injury duration, and married malefirefighters had a 52% longer injury duration. Tenureshowed a significant negative relationship with dura-tion. For each additional year of tenure, injury durationdecreased by 27%. Consistent with our hypothesis, agewas positively related to injury duration, indicating thatolder firefighters took more time recovering from inju-ries. A 7% increase in injury duration accompaniedeach 1-year increase in age. Also consistent with our


Table 3Means, Standard Deviations, and Intercorrelations for Duration Models (N = 678)

1.2.3.4.5.6.7.8.9.

10.11.12.13.14.

15.16.17.

18.19.20.21.22.23.24.25.26.

Variable

Duration (days)FemaleWhiteBlackMarriedAgeTenureFighterFire motor operatorInjury on dutyWeekly wageBack sprain/strainOther sprain/strainBurn/exposure to

chemicalsFracture/lacerationContusionExposure to

contagious diseaseHsDHyPdPaPtScMaSi

M

10.060.230.760.130.53

32.594.420.800.080.20

670.830.120.28

0.100.100.08

0.1548.9247.5453.6756.2652.0452.1852.2455.0642.44

SD

56.670.420.430.340.504.983.300.400.270.40

183.150.320.45

0.300.300.27

0.365.686.786.457.526.786.446.477.796.31

1

—.026

-.012-.001-.018

.005-.008-.012

.032

.181-.007

.020

.088

-.008.006

-.038

-.067-.018-.033-.007

.014-.003-.044-.046

.009-.042

2

— ..125

-.146-.419

.013-.010

.040-.064

.058

.015

.017-.007

-.027-.024

.133

-.021-.328-.208-.273-.041

.044-.220-.073-.006-.067

3

—-.683

.029

.198

.111-.042

.024-.037

.073

.009

.002

.028-.043-.005

-.048.036.076.218.187.129.237.192

-.013-.002

4

—.041

-.146-.137

.029

.002-.006-.080-.009-.027

-.027.034.027

.044-.172-.143-.227-.113-.169-.103-.146

.026-.003

5

—.233.235

-.123.152

-.050.156.013.029

.018-.016-.072

-.047.150.090.187.089.026.156.119.092.083

6

—.686

-.220.161.062.591.021.027

-.014-.081

.072

.032

.001-.066

.102

.226

.108

.032

.179

.102

.131

7

—-.481

.383

.108

.897

.010

.028

-.041-.059

.044

.038

.142

.155

.318

.227

.183

.223

.299

.262

.146

8

—-.582-.020-.571-.042

.048

-.018.016

-.012

-.006-.120-.133-.152-.096-.054-.108-.141-.062-.025

9

—.008.361

-.006-.009

-.023-.003-.027

.062

.056

.103

.123-.004-.049

.082

.090

.078

.140

10

—.109.179.134

.020

.022-.081

-.178-.027-.084

.019

.024

.058-.022-.007

.107-.082

Note. \r\ > .075, p < .05; |r| > .099, p < .01. Hs = hypochondriasis; D = depression; Hy = conversion hysteria;Pd = psychopathic deviate; Pa = paranoia; Pt = psychasthenia; Sc = schizophrenia; Ma = hypomania; Si = socialintroversion.

expectations, IOD injuries significantly increased theduration time, indicating that firefighters whose loss ofwork time would be fully compensated took more timeoff. Specifically, an individual who made an IOD claimhad injury duration 192% longer than one who made anon-IOD claim. Contrary to our expectation, the injuredworker's weekly wage was positively related to disabil-ity duration rather than negatively related; however, themagnitude of the impact was rather small.

Two MMPI scales were significant predictors of in-jury duration. Higher scores on the Pd scale were asso-ciated with an increase in injury duration. Interestingly,the higher value of the Sc scale was associated withshorter claim duration. Individuals who score high onthis scale are described as isolated, alienated, misunder-stood, and unaccepted by their peers, as well as expe-rience a good deal of apprehension and anxiety (Gra-ham, 1993). It is not intuitively obvious why individualswith relatively high Sc scores would experience rela-tively shorter injury durations.

The data also indicated that the type of injury wasrelated to the duration of injury. Firefighters who re-ported non-back sprain and strain stayed 68% longer ondisability than those with injuries of the types not listedin Table 3, whereas those who reported injury due toexposure to contagious disease returned to work 51%sooner. Other frequently occurring injuries, such as burns,fractures, and lacerations, did not show any significantimpact on injury duration. Race did not show significantrelationships to the duration of injuries. Finally, injuryclaims filed in 1988 were found to have 134% longerduration than claims filed in other years.

Impact on Indemnity Cost

Another way to understand the relationships be-tween these variables and injuries is to examine therelative costs of the injuries and the change in costs inrelation to the per unit change in the independentvariables. On the basis of our sample, the average


11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

.009 —

.023 -.229 —

.014 -.122 -.206 —-.038 -.121 -.205 -.109 —.051 -.110 -.186 -.099 -.098 —

.035

.104

.097

.168

.085

.083

.095

.155

.135

.039

-.155-.076-.095-.045-.027-.044-.025-.030

.008-.044

-.262.042.014.035.055.116.057.028.056.002

-.140.009.008.022

-.023-.055-.040-.028

.011-.045

-.139-.029

.021-.022-.051-.026-.033-.021-.033-.017

-.126-.029-.010-.033

.015-.031-.016-.003

.024-.023

.018

.007-.007-.029-.048-.026-.027-.087

.015

.425

.671

.404

.342

.517

.606

.125-.106

.271

.275

.233

.434

.334

.016

.407

—.510.419.567.649.344

-.195

—.484.542.648.236

-.028

—.441.504.273.037

—.749 —.286 .317 —.158 .007 -.127 —

indemnity cost (the amount paid to injured employ-ees for lost wages) per injury was $283.43. Theaverage annual indemnity cost per employee can becalculated by taking the product of the number ofclaims per employee times the average duration (indays) per claim times the average statutory dailybenefit. To calculate the impact of a particular inde-pendent variable (i.e., the MMPI Pd scale) on annualindemnity cost per employee, we take the derivativeof costs (in natural logarithm terms) with respect tothat variable:

3 ln(cos t) d In(Frequency)

dPd dPd

d ln(Duration) d \n(Benefit)

dPd dPd

On the right-hand side of the equation, the firstterm represents the percentage change in injury fre-

quency in relation to a unit change in the Pd MMPIscale and is exactly estimated by the negative bino-mial regression for the frequency model. The secondterm represents the percentage change in injury du-ration in relation to a unit change in the Pd scale andis represented by the estimated regression coefficientfrom the survival analysis for the duration model.The third term captures the influence of MMPI onstatutory benefits, which is zero (because there is noreason that change in MMPI scores, or change in anyof the other demographics, will change the legislatedbenefit level).

Various demographic and personality variables in-fluenced the amount of indemnity costs through theirimpacts on either frequencies or duration of injuries,or both. Table 5 lists the demographic and personalityvariables identified as being statistically significant inthe negative binomial regression or the survival re-gression analysis. The percentage change in indem-


Table 4Estimates of Survival Analysis for Duration Model(N = 678)

Variable

FemaleMarriedAgeTenureInjury on dutyWeekly wageHsDHyPdPaPtScMaSiFemale X Married^(38)Prob > x2

Pseudo R 2

Log likelihood

Unstandardized j3

.494**

.516**

.074**-.271**1.924**.003*

-.014.001.047.029*

-.009.003

-.062**-.001-.011

-1.385**529.07

0.000.45

-1,187.2894

SE

.216

.168

.026

.089

.279

.002

.021

.016

.025

.015

.014

.024

.023

.012

.018

.346

Note. One-tailed tests were performed for female, age,married, injury on duty, wage, and Female X Married.Wage and Female X Married were significant but had anunexpected sign. Two-tailed tests were performed for theremaining independent variables. Control variables, includ-ing 11 year dummies, two racial dummies, two positiondummies, and six injury type dummies were omitted fromthe table. Hs = hypochondriasis; D = depression; Hy =conversion hysteria; Pd = psychopathic deviate; Pa = para-noia; Pt = psychasthenia; Sc = schizophrenia; Ma = hy-pomania; Si = social introversion.*p < .05. **p < .01.

nity costs with respect to a unit change in the inde-pendent variable is the sum of the relevant negativebinomial regression coefficient and the survival anal-ysis coefficient, if both coefficients were availableand at least one of them was statistically significant.If an independent variable's information was notavailable in the negative binomial regression, weused only the coefficient estimated in the survivalanalysis as a proxy for the percentage change in theindemnity cost caused by that particular variable. Theoverall change in indemnity costs can be approxi-mated by multiplying the percentage change in theindemnity costs and the average indemnity costs inthis sample ($283.43).

Table 5 exhibits the overall impact that the demo-graphic and personality variables have on indemnitycosts. As an example, to calculate the overall changein the indemnity cost of unmarried female compared

with unmarried male firefighters, we added and mul-tiplied the two regression coefficients from the firstcolumn (.330) and second column (.494) by$283.4322, to yield a value of $233.83. Therefore,unmarried female firefighters experienced on average$233.83 more indemnity costs compared with unmar-ried male firefighters. However, married female fire-fighters experienced $12.47 lower average injurycosts compared with unmarried male firefighters. A1-year increase in age increased the costs by $18.42,whereas a 1-year increase in tenure decreased costsby $90.13. The four statistically significant MMPIscales also demonstrate some cost impact. With onestandard deviation increase on the Hy scale, the costswere increased by only $126.16. If one standarddeviation increase on the Pd scale is considered, thecosts involved would be $87.38. A one standarddeviation increase on the Si scale would also increasethe costs by $16.09. But a similar one standard de-viation increase on the Sc scale would result in anaverage decrease in costs by $133.86.

Discussion

This research examined the correlates of two im-portant injury variables among firefighters: the fre-quency of injuries and the duration of injuries. Weextended previous research in predicting workplaceinjuries in several aspects. First, most of the priorresearch has either been based on cross-sectional dataor used data from individuals who were already in-jured—both these approaches do not allow for pre-dictive inferences. The present research used datagathered over a 12-year period, while the majority ofthe explanatory variables were collected before thedate of hire, thus allowing for predictive inferences.By using a longitudinal design such as this, causalinferences can be strengthened. Second, we focusedon a single occupation, namely, firefighters, henceruling out extraneous risk factors associated withdifferent occupations. Third, we used injury frequen-cies and duration of actual workers' compensationclaims as the outcome variables. Workers' compen-sation records may be biased if job-related injuries gounreported (Parker, Carl, French, & Martin, 1994).However, the advantages of using the workers' com-pensation database are dominant, which includedecreasing the potential overreporting problem thataccompanies self-reported data and avoiding prob-lems of having participants recall events in the past.Furthermore, using valid injury claim data facilitatedthe estimates of indemnity costs, because only work-


Table 5Demographic and Personality Variables' Influence on Indemnity Cost

Variable

FemaleMarriedAgeTenureHyPdScSiFemale X Married

% change ininjury frequency*

.330**NA

-.009-.046

.022*

.012*-.011

.020*NA

% change ininjury duration1"

.494*

.516**

.074**-.271**

.047

.029*-.062**-.011

-1.385**

% change inindemnity cost0

.824

.516

.065-.317

.069

.041-.073

.009-1.385

Change inindemnity costs'1

233.83146.2518.42

-90.13126.1687.38

-133.8616.09

-392.55

($)

Note. NA = information on this variable was not available in the negative binomial regression; Hy = conversion hysteria;Pd = psychopathic deviate; Sc = schizophrenia; Si = social introversion."The estimated coefficients from the negative binomial regression. bThe estimated coefficients from the survivalanalysis. c The sum of the two regression coefficients in each row if both are available; if the negative binomial regressioncoefficient is unknown for that variable, then only the value of the survival analysis coefficient is used. d Calculated bymultiplying the percentage change in indemnity cost (in previous column) with the average indemnity cost per injury in thissample ($283.4322). For Hy, Pd, Sc, and Si, it was further multiplied by each scale's standard deviation, so it was the changein indemnity cost associated with one standard deviation increase in the corresponding Minnesota Multiphasic PersonalityInventory scale.*p < .05. **p < .01.

ers' compensation claims cause direct costs to theemployers.

Main Findings

Predicting injury frequency. Female firefightersexperienced more injuries, and this relationship per-sists even after we have controlled for factors such asage, race, tenure, personality, and secular trends. Areporting bias might also help to explain the differentinjury rates between male and female firefighters.Within male firefighters, there may be a strong cul-tural norm for not reporting minor injuries (e.g., it is asign of weakness). For female firefighters, this normmight be different. Future research should attempt toidentify the mediating variables that explain the re-lation between gender and work injury frequencies.

Consistent with our predictions, the frequency ofinjuries rose with higher values for the Hy and Pdscales. We formed no hypothesis concerning the Siscale, but it demonstrated a significant positive rela-tionship to the frequency of injury variable. Onepossible explanation is that individuals who are lesslikely to engage in interpersonal interactions aremore insecure and uncomfortable when working withothers (Costa & McCrae, 1992), resulting in a hesi-tation to call for assistance. Extraversion predictssuccessful performance in jobs involving interactionswith others (Mount, Barrick, & Stewart, 1998). Fire-

fighting is essentially teamwork (Clarke & Zak,1999). Firefighters perform more safely and effec-tively if they cooperate well with each other. There-fore, those who are more reluctant to interact withteam members may seek less help from coworkersduring an emergency, thereby exposing themselvesto greater risks. Alternatively, there could be omittedvariables that are common causes of both higher Siscores and frequency of injury. Future researchshould examine why high level of introversion isassociated with more injuries.

Predicting injury duration. Among the demo-graphic predictors, older firefighters suffered longerinjury duration than did younger firefighters. Tenurewas also negatively related to injury duration, indi-cating that longer tenured firefighters suffered lesssevere injuries than shorter tenured firefighters. Therewas a significant interaction between gender andmarital status, but in a direction contrary to ourhypothesis. Results showed that married female fire-fighters returned to work the earliest, followed byunmarried male, unmarried female, and married malefirefighters. A possible explanation hinges on wom-en's traditional family role: Because of their dualresponsibility at work and at home, married femalefirefighters may be more cautious than unmarriedfemale firefighters and thereby less likely to be ex-posed to certain injury risks. Alternatively, fire-fight-ing crews may implicitly assign working mothers to


the least risky tasks when fighting any given fire.Hence, they would be less likely to experience certaintypes of severe injuries. Future research may includenumber of dependents in their analysis, which mayhelp clarify the reason why female firefighters, whenmarried, returned to work much sooner.

Consistent with our prediction, IOD claims hadlonger injury durations than non-IOD claims. How-ever, higher wages did not seem to provide an incen-tive to return to work early in this study. Wage waspositively related to nonwork spell, but the magni-tude of this impact was not particularly strong. Fire-fighters seemed to respond more to the workers'compensation benefits than to wage loss. This isconsistent with Butler and Worrall (1985), who alsofound the benefit coefficient was about double thesize of the wage coefficient.

Among the personality predictors, the Pd scaleshowed a positive relationship with injury duration.Possibly, firefighters who tend to ignore safety rulesand regulations not only had accidents more fre-quently but also suffered more severe injuries. Alter-natively, high scorers may also ignore return-to-workrules and abuse injury leave, resulting in longer in-jury duration periods.

The evidence suggests much could be gained byexamining the duration of injury as an importantvariable distinguished from the occurrence of injury.Thus, researchers might want to differentiate thesetwo outcomes and examine their relative importanceto organizations.

Implications

Because workplace injuries are costly, it is impor-tant to identify individual risk factors predictive ofworkplace injury. These factors may help screen outhigher risk individuals, provide safety education, andoffer accommodations to those who need longer timeto recover from injuries.

The MMPI was developed to diagnose a variety ofpsychological disorders and screen out candidateswho would be ill-suited to high-risk or sensitive jobs.The present research provided evidence that severalMMPI scales are significant predictors of two dimen-sions of safety performance among firefighters: thefrequency and the duration. Specifically, if applicantswith elevated Hy and Si scales are screened out,workers' compensation costs can be cut by reducingthe occurrence of injuries. Additionally, if applicantswith high Pd scores are screened out, workers' com-pensation costs can be cut by reducing both theoccurrence and the length of injuries.

Although MMPI measures are frequently used forselecting firefighters, police officers, and nuclearpower workers (Inwald, 1988; Kelley et al., 1994),they may not be readily available to human resourcemanagers of other occupations, who may adopt someother personality measures such as the Big Five con-structs (Barrick & Mount, 1991). Costa, Busch, Zon-derman, and McCrae (1986) correlated MMPI factorscales with measures of the Big Five and concludedthat Neuroticism, Extraversion, Openness to Experi-ence, and Agreeableness were well represented in theMMPI scales. Cortina, Doherty, Schmitt, Kaufman,and Smith (1992) identified the Pd scale as a measureof the Conscientiousness factor. In the present study,scores on the Pd scale were positively associated withboth injury frequency and duration, indicating a moredirect measure of Conscientiousness may also predictthese two dimensions of safety performance. The Siscale is an obvious measure of Extraversion (Cortinaet al., 1992). Our finding that higher Si scores wereassociated with more injuries suggests that Extraver-sion predicts the injury frequency dimension ofsafety performance. Future research could further ourunderstanding of the links between Big Five person-ality constructs and the frequency and duration di-mension of injury performance.

Unlike personality variables, demographic vari-ables are not appropriate for personnel selection pur-poses; however, identifying high-risk employees pro-vides important information for organizations tomanage their employment risks and forecast trainingand accommodation needs. Adequate health andsafety education has been repeatedly shown to beeffective in preventing injuries (Hunt, Habeck, Van-tol, & Scully, 1993). More safety training and injury-prevention intervention could be provided if the em-ployee demographic composition indicates moreinjuries are likely to occur. When injuries occur andimpairments do result, accommodation might be theemployers' best solution for preventing adverse dis-ability outcomes (Hunt et al., 1993). Our study iden-tified individuals who are in most need for accom-modation (e.g., older firefighters and firefighters whosuffered from injuries of sprains or strains). Accom-modations such as light-duty assignments may helpthem return to work earlier.

Limitations and Direction of Future Research

Our model explained a substantial portion of vari-ance (45%) in the injury duration variable. However,the injury frequency model explained only a smallportion of the variance (5%) in the dependent vari-


able, indicating an obvious need for further modeldevelopment. Inclusion of other constructs, such asemployee attitudes and social support, may enhancethe explaining power of the model. For example,Sherry (1991) showed that poor person-environmentfit is an important correlate of the occurrence ofaccidents among railroad transportation workers andthat employee-perceived peer and supervisory sup-port is associated with decreased accidents. Addition-ally, some variables used in the injury duration modelwere not available for the injury frequency model(e.g., marital status and weekly wage); therefore, wecould not estimate the impact of these variables onthe frequency of injuries. Consequently, our esti-mates of these variables' overall impact on indemnitycost might be biased in an unknown direction. In oursurvival analysis, marital status and weekly wagewere significant predictors of the injury duration.Future studies should investigate whether they arealso related to injury frequency.

Although we focused on firefighters, similar re-search designs and methodologies could be applied toother occupations. It will be useful to replicate thestudy in different settings to examine whether thesame set of demographic and personality variablespredict workplace injuries for other occupations inother places. Another avenue for future research is toextend the dimensionality of the criterion space forincluding accidents and injuries as an element of thebroadly defined job performance (Campbell,McHenry, & Wise, 1990). An important question ishow safety performance might fit into the other, morefrequently studied individual outcomes: task perfor-mance quantity and quality, contextual performance,and work avoidance or withdrawal behavior.

In conclusion, there is much yet to be done inves-tigating the individual differences and on-the-job in-jury relationship. The knowledge gained in such re-search may help develop effective interventions fororganizations to improve workplace safety and re-duce injury-related costs.

References

Barrick, M. R., & Mount, M. K. (1991). The Big Fivepersonality dimensions and job performance: A meta-analysis. Personnel Psychology, 44, 1-26.

Bigos, S. J., Battle, M. C., Spengler, D. M., Fisher, L. D.,Flordyce, W. E., Hansson, T. H., Nachemson, A. L., &Wortley, M. D. (1991). A prospective study of workperceptions and psychosocial factors affecting the reportof back injury. Spine, 16, 1—6.

Burbeck, E., & Furnham, A. (1985). Police officer selection:A critical review of the literature. Journal of PoliceScience and Administration, 13, 58-69.

Butler, R. J. (1994). Economic determinants of workers'compensation trends. Journal of Risk and Insurance, 61,383-401.

Butler, R. J., Hartwig, R. P., & Gardner, H. H. (1997).HMO's, moral hazard, and cost shifting in workers'compensation. Journal of Health Economics, 16, 191-206.

Butler, R. J., & Worrall, J. D. (1985). Work injury compen-sation and the duration of nonwork spells. EconomicJournal, 95, 714-724.

Campbell, J. P., McHenry, J. J., & Wise, L. L. (1990).Modeling job performance in a population of jobs. Per-sonnel Psychology, 43, 313—333.

Cellier, J. M., Eyrolle, H., & Bertrand, A. (1995). Effects ofage and level of work experience on occurrence of acci-dents. Perceptual and Motor Skills, 80, 931-940.

Clarke, C., & Zak, M. J. (1999, Summer). Fatalities to lawenforcement officers and firefighters, 1992—97. Compen-sation and Working Conditions, 3—8.

Cortina, J. M., Doherty, M. L., Schmitt, N., Kaufman, G., &Smith, R. G. (1992). The "Big Five" personality factorsin the IPI and MMPI: Predictors of police performance.Personnel Psychology, 45, 119-141.

Costa, P. T., Busch, C. M., Zonderman, A. B., & McCrae,R. R. (1986). Correlations of MMPI factor scales withmeasures of the five factor model of personality. Journalof Personality Assessment, 50, 640-650.

Costa, P. T., & McCrae, R. R. (1986). Personality stabilityand its implications for clinical psychology. ClinicalPsychology Review, 6, 407-423.

Costa, P. T., & McCrae, R. R. (1992). The five-factor modelof personality and its relevance to personality disorders.Journal of Personality Disorders, 6, 343-359.

Craske, S. (1968). A study of the relation between person-ality and accident history. British Journal of MedicalPsychology, 41, 399-404.

DeBobes, L. (1986). The psychological factors in accidentprevention. Personnel Journal, 65, 34-38.

Prone, M. R. (1998). Predictors of work injuries amongemployed adolescents. Journal of Applied Psychology,83, 565-576.

Gandolfo, R. (1995). MMPI-2 profiles of worker's compen-sation claimants who present with complaints of harass-ment. Journal of Clinical Psychology, 51, 711-715.

Graham, J. R. (1993). MMPI-2: Assessing personality andpsychopathology. New York: Oxford University Press.

Greene, R. L. (1990). Stability of MMPI scale scores withfour code types across forty years. Journal of PersonalityAssessment, 55, 1-6.

Greene, W. H. (1997). Econometric analysis. New York:MacMillan.

Hansen, C. P. (1991). Personality characteristics of theaccident-involved employee. In J. W. Jones, B. D. Steffy,& D. W. Bray (Eds.), Applying psychology in business:The handbook for managers and human resource pro-fessionals (pp. 801-812). New York: Lexington Books.

Heckman, J. J., & Singer, B. (1984). Econometric durationanalysis. Journal of Econometrics, 24, 63-132.

Hersch, P. D., & Alexander, R. W. (1990). MMPI profilepatterns of emotional disability claimants. Journal ofClinical Psychology, 46, 793-799.

Hirsch, B. T., Macpherson, D. A., & DuMond, J. M. (1997).Workers' compensation recipiency in union and non-


union environments. Industrial Relations and Labor Re-lations Review, 50, 213-236.

Hunt, H. A., Habeck, R. V., Vantol, B., & Scully, S. M.(1993). Disability prevention among Michigan employ-ers, 1988-1993 (Upjohn Institute Technical Report No.93-004). Kalamazoo, MI: W. E. Upjohn Institute forEmployment Research.

Inwald, R. (1988). Five-year follow-up study of departmen-tal terminations as predicted by 16 preemployment psy-chological indicators. Journal of Applied Psychology, 73,703-710.

Iverson, R. D., & Erwin, P. J. (1997). Predicting occupa-tional injury: The role of affectivity. Journal of Occupa-tional and Organizational Psychology, 70, 113-128.

Johns, G. (1997). Contemporary research on absence fromwork: Correlates, causes, and consequences. Interna-tional Review of Industrial and Organizational Psychol-ogy, 12, 115-173.

Johnson, W. G., Baldwin M., & Butler, R. J. (1999). Work-ers' compensation low back claims: The costs and out-comes of chiropractic and physician care. Journal of Riskand Insurance, 66, 185—205.

Johnson, W. G., & Ondrich, J. (1990). The duration ofpost-injury absences from work. Review of Economicsand Statistics, 72, 578-586.

Kelley, P. L., Jacobs, R. R., & Farr, J. L. (1994). Effects ofmultiple administrations of the MMPI for employeescreening. Personnel Psychology, 47, 575-591.

Kiefer, N. (1988). Economic duration data and hazard func-tions. Journal of Economic Literature, 26, 646-679.

Kingma, J. (1994). Causes of occupational injuries. Percep-tual and Motor Skills, 79, 1025-1026.

Kraus, J. F. (1985). Fatal and nonfatal injuries in occupa-tional settings: A review. Annual Review of PublicHealth, 6, 403-418.

McGuire, F. L. (1972). The understanding and prediction ofaccident producing behavior. North Carolina Symposiumon Highway Safety, 1, 116-118.

Morita, J. G., Lee, T. W., & Mowday, R. T. (1989). Intro-ducing survival analysis to organizational researchers: A

selected application to turnover research. Academy ofManagement Journal, 36, 1430—1463.

Mount, M. K., Barrick, M. R., & Stewart, G. L. (1998).Personality predictors of performance in jobs involvinginteraction with others. Human Performance, 11, 145-166.

1996 U.S. firefighter injuries. (1997, November/December).National Fire Protection Association Journal, 67-77.

Parker, D. L., Carl, W. R., French, L. R., & Martin, F. B.(1994). Characteristics of adolescent work injuries re-ported to the Minnesota Department of Labor and Indus-try. American Journal of Public Health, 84, 606-611.

Pancoast, D. L., & Archer, R. P. (1988). MMPI adolescentnorms: Patterns and trends across 4 decades. Journal ofPersonality Assessment, 52, 691-706.

Repko, G. R., & Cooper, R. (1983). A study of the averageworkers' compensation case. Journal of Clinical Psy-chology, 39, 287-295.

Selzer, M. L., Rogers, J. E., & Kern, S. (1968). Fatalaccidents: The role of psychopathology, social stress, andacute disturbance. American Journal of Psychiatry, 124,8-19.

Sheridan, J. E. (1992). Organizational culture and employeeretention. Academy of Management Journal, 35, 1036—1056.

Sherry, P. (1991). Person-environment fit and accidentprediction. Journal of Business and Psychology, 5, 411-416.

Stata Reference Manual: Release 6.0. (1999). College Sta-tion, TX: Stata Corporation.

Stata User's Guide: Release 6.0. (1999). College Station,TX: Stata Corporation.

Worrall, J. D., Appel, D., & Butler, R. J. (1987). Sex,marital status, and medical utilization by injured workers.Journal of Risk and Insurance, 54, 27—44.

Received August 16, 2000Revision received December 8, 2000

Accepted February 1, 2001 •

Documents

Correlates of Work Injury Frequenc anyd Duration Among ... · fore, ther ies a economin c incentiv fo workerr e tso return to work earlier Th. opportunite y o cosf ret-maining out