Do women prefer women\'s work?

This article was downloaded by: [54.207.11.183] On: 25 March 2014, At: 06:25 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Applied Economics Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/raec20

Do women prefer women's work? a

W. Robert Reed & Julie Dahlquist a

b

Department of Economics , University of Oklahoma , Norman, Oklahoma, 73019

b

College of Business Administration , St Mary's University , San Antonio, Texas, 78228, USA Published online: 28 Jul 2006.

To cite this article: W. Robert Reed & Julie Dahlquist (1994) Do women prefer women's work?, Applied Economics, 26:12, 1133-1144, DOI: 10.1080/00036849400000111 To link to this article: http://dx.doi.org/10.1080/00036849400000111

PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions

Applied Economics, 1994, 26, 1133-1 144

Do women prefer women's work? W . R O B E R T R E E D and J U L I E D A H L Q U I S T '

Downloaded by [54.207.11.183] at 06:25 25 March 2014

Department of Economics, University of Oklahoma, Norman, OK 73019 and 'College of Business Administration, St Mary's University, Sun Antonio, T X 78228, U S A

A new methodology is implemented to determine whether job characteristics can explain why women are concentrated in low-paying, female-dominated occupations. Extensive information on jobs and labour market histories are collected from the 1982 National Longitudinal Survey, Youth Cohort, for women and men characterized by substantial labour market attachment. Significant differences in nonpecuniary job characteristics between the male and female samples are found to exist. Nevertheless, we find no evidence that women differentially favour those job characteristics commonly associated with 'women's work'.

I. I N T R O D U C T I O N In summary, the puzzle is not so much that there is a gap in the pay between men and women, but rather, why there are differences in occupational employment patterns by sex. Why are women in lower-paying jobs and lower-paying establishments even when they have relatively high education [US Bureau of the Census, 1986]? . . .women voluntarily choose low-paying jobs because they like the cushy indoor working climate better than the dirty men's jobs (Phyllis Schlafly, quoted in Cameron, 1986).

T o a large extent the male-female earnings gap can be 'explained' by the twin facts that women are located in different occupations than men and that female-dominated occupations pay lower wages than male-dominated occupations (Malkiel and Malkiel, 1973; Nadeau et al., 1993).' The degree to which men and women occupy different occupational categories is striking. For example, Rytina and Bianchi (1984) find that about half of all women work in occupations that are at least 80% female. Over 70% of men are employed in occupations that are a t least 80% male. Occupational sex concentration is prominent even for younger workers (Beller, 1984; Reskin and Hartmann, 1986). As noted above, the real puzzle in the male-female wage gap is why women work in lower-paying, female-dominated occupations. A number of researchers have suggested gender differences in preferences over job characteristics as a potential

explanation for this puzzle (Filer, 1985, 1986, 1989; Cameron, 1986; Palme and Wright, 1992). A representative statement of this hypothesis is the following: Men and women may make rational choices in the job market based on differences in utility functions that create differing preferences for certain types of work and other duties. For example, some evidence indicates that women attach greater importance to various forms of attractive working conditions and that men place relatively greater emphasis on incomes (Murray and Atkinson, 1981; Forgionne and Peters, 1982; Harvey, 1986).Such a difference in preferences, coupled with the fact that the market forces employers to pay compensating differentials to those workers who fill jobs with relatively unattractive working conditions, will, even given equal productivity, result in women being concentrated in lower paying but otherwise more attractive jobs (Filer, 1989, p. 154). In fact, there exists little evidence connecting women's preferences to those job characteristics that most characterize the class ofjobs that have come t o be known as 'women's work': jobs that offer safe and healthy work conditions, jobs that are people-oriented, that have pleasant work environments, and that d o not involve a lot of career-advancement. The big question is, are women in these jobs because they choose these kinds ofjobs, or because they are denied access to alternative 'men's work'? This study addresses this question by employing a new methodology to identify the occupational preferences of workers.

' Attempts to explain the earnings gap using worker-specific characteristics have been largely unsuccessful. When productivity-related variables such as age, education, general work experience and work interruptions are included, at best only about half of the earnings difference. between men and women can be accounted for (Seiling, 1984; Cain, 1986; US Bureau of the Census, 1986). A (very interesting) exception is Barron et al. (1993). 0003-6846 0 1994 Chapman & Hall

1133

W. R. Reed and J . Dnhlquist

1134

week or more, who began their jobs on or after their sixteenth birthday, and who were employed at least half of each year between 1982 and 1987 are included in this ample.^ In concentrating on fulltime male and female workers with significant (future) employment experience, we attempt to avoid problems associated with school-going behaviour that is a concern given the young age of the sample. We also hope to avoid spurious correlations that can arise when workers are relatively indifferent between work and its alternative^.^ Selected sample characteristics for the men and women are presented in Table I. (Variable definitions are given in the respective tables and the Appendix.) The female and male samples are compared on three sets of characteristics: ( I ) labour market characteristics, (2) personal characteristics, and (3) job characteristics. Beginning with labour market characteristics, we see that the male and

11. D A T A BASE A N D S A M P L E

CHARACTERISTICS

Downloaded by [54.207.11.183] at 06:25 25 March 2014

We begin our analysis by constructing separate samples of male and female workers who are characterized by significant labour market attachment. The data base for this analysis is taken from the National Longitudinal Survey, Youth Cohort, 1979- 1987 (NLSY). The full sample population of 12 686 consists of a nationally representative sample of young men and young women, aged 17-25 in 1982, with oversampling of blacks, Hispanics, economically disadvantaged whites, and military personnel. The NLSY is particularly attractive for our purposes because of the extensive battery of job-related questions asked of respondents. The sample used in this analysis consists of 922 white men and 61 5 white women who were holding jobs at the time of the 1982 survey.* Only those individuals working 35 hours a Table 1. Mean ~ a l u e soj'selected sample characteristics -

--

-

-

Characteristic

Females

Males

(t-statistics)

(1)

(2)

(3)

(4)

Lubour market characteristics URBAN"

URATE Personal characteristics EDUCATION (years) DEPENDENTS (number) WORKPLANAa W O R K PLANE" MARRIEDa AGE (years) L M E X P E R week^)^ JOBEX PER (weeks)' Job characteristics WAGES (centslhour) UNIONa FLXHOURS" Work commute (in minutes)* COMMUTE81 COMMUTE80 COMMUTE79

0.793

0.764

1.39

0.097

0.100

- 1.64

12.74 0.145 0.647 0.161 0.23 1 22.05 110.1 85.85

12.07 0.250 0.862 0.0 14 0.23 1 2 1.98 1 14.7 84.98

510.4 0.140 0.106

618.0 0.1 85 0.129

-9.01* -2.41* - 1.41

16.85 16.52 16.32

18.1 1 17.37

- 1.43

16.59

6.79* - 3.27*

9.61* 9.58* - 0.0 I 0.7 1 - 1.07 0.22

-0.88 - 0.24

*An asterisk indicates that the difference between the means of the two samples is significant at the 5%) level (two-tailed test). " U R B A N takes the value 1 if worker's residence is located in an urban area; M A R R I E D takes the value I if currently married; W O R K P L A N A takes the value 1 if respondent plans to be working at age 35: W O R K P L A N B takes the value 1 if respondent plans to be raising a family at age 35 and also wants to work outside the home: U N I O N takes the value 1 if wages are determined by a collective bargaining agreement; and F L X H O U R S takes the value 1 if hours are flexible. L M E X P E R is weeks of labour market experience after sixteenth birthday or date last enrolled in school (whichever is latest) minus J O B E X P E R . 'J O B E X P E R is weeks in current job at time of 1982 survey date. Commuting times are not available for the 1982 sample. These (one-way)commute values are calculated for earlier years using samples with similar inclusion criteria. *Unfortunately, of the nine years included in the survey period, only the 1979 and 1982 surveys provide a large universe of questions concerning workers' jobs, including their nonpecuniary job attributes. Only 1982 is included in the analysis here because inclusion criteria significantly reduced the size of the 1979 sample. Also excluded from the sample are all military personnel, government employees, self-employed workers and those working without pay. Another example of a study which omits temporary and part-time workers is Nadeau et ul. (1993).

Downloaded by [54.207.11.183] at 06:25 25 March 2014

Do women prefer women's work? female samples differ little with respect to residence in urban areas (URBAN) and local unemployment rates (URATE). In contrast, there are some group differences between the men and women with respect to personal characteristics. In general, the female workers in our sample are likely to be better educated (EDUCATION) and have fewer dependents (DEPENDENTS). Further, their career plans differ significantly from men in the expected direction. When asked, 'What would you like to be doing when you are 35 years old?' 86.2% of the men, but only 64.7% of the women, responded 'working in present job' or 'working in same occupation'; as opposed to 'being marriedlraising a family' and 'other' (cf. WORKPLANA). Even so, of the female respondents who identified 'being married/raising a family', approximately two-thirds expressed a desire to work outside the home in addition to their family pursuits (cf. WORKPLANB). If we add the percentage of women who plan to be working at age 35 with those who report wanting to work outside the home along with raising a family, the difference in working plans for the two groups narrows substantially: WORKPLANA plus WORKPLANB equals 81% for women, compared to approximately 88% for men. The remainder of the personal characteristics variables show no significant differences between the male and female samples. Approximately 23% of both the women and men samples are married (MARRIED), and the mean age for each group is about 22 (AGE). O n average, both males and females report a little more than two years of accumulated work experience prior to beginning their jobs (LMEXPER). Similarly, the average time spent with their current employer up to the time of the survey date is approximately one and two third years (JOBEXPER). Turning now to job characteristics, we see the all too familiar result that average women's wages lie below those of men. For our two samples, the hourly wages (WAGES) for the women average approximately $1.08 less than their male counterparts, measured in 1982 dollars. This difference amounts to a little more than 20% of the average female wage. Female workers also have somewhat lower rates of unionization (UNION). We find no evidence that the

1135 women have more flexible schedules than the men (FLXHOURS) or that they face different commuting lengths to work (COMMU TE81, COMMUTE80, C O M M U TE79).' In summary, Table 1 includes a number of variables that have long been hypothesized as contributing to the wage gap between men and women. In this light, the NLSY proves to be a particularly attractive data set. Its focus on young workers early in their labour market careers allows us to avoid some of the problems attending older-aged samples in which many of the women have interrupted work histories (i.e., d o women work less because of low wages, or d o they receive low wages because they work l e s ~ ? ) Indeed, .~ the extensive labour market history information available on the NLSY allows us to determine that there are essentially no differences in either labour market experience or current job tenures between the men and women in our samples. This suggests that the gender wage gap in our data is not attributable to productivity differences, at least as measured by these variables. Further, the women in our samples d o not differ significantly from the men in either their job locations or the flexibility of their work schedules: two reasons often mentioned by human capital proponents as possible contributors to the wage gap. A few significant differences d o exist, however, and these comprise potential explanations for the $1.08 difference in hourly wages between men and women. In particular, women have somewhat lower rates of union representation and assign a greater priority to family pursuits in assessing their future occupational aspirations. They also report a number of work-related differences. Table 2 reports sixteen additional job characteristics for comparison between men and women. The first three variables indicate whether workers received paid vacations (PAIDVAC), health insurance (HLTHINS) and life insurance (LIFEINS) at their 1982 jobs. In each case, more women than men report receiving these fringe benefits. In addition, workers were asked to rate their jobs across a number of nonpecuniary dimensions. The questions are all of one general type: respondents are asked to indicate which one of four or five provided categories most clearly describe

With respect to the fact we find no evidence that women have more flexible work schedules than men, one might suspect that this is an anomaly produced by restricting the data sets to full-time workers. In fact, the same results obtained when one conlpares all employed workers in every year from 1979 to 1982. While data on length ofjob commute is difficult to come by, the few studies that do report commute times find that women choose jobs closer to home than men (Rees and Schultz, 1970; Filer, 1985). Unfortunately, no information on commuting times was requested on the 1982 survey. As a means of ascertaining whether male and female workers differ systematically in this dimension, we constructed samples of workers from the 1979,1980,and 1981 surveys using similar inclusion criteria to those used for the 1982 sample. We then calculated average commute times for these sets of workers. As Table 1 reports, in each case the difference in commute times was negligibly small and statistically insignificant.Ofcourse, it might be the case that as more women have dependents for which they are the primary caretakers, the difference between commute times for men and women will increase. Enlightening in this regard is the paper by Goldin and Polachek (1987) and the subsequent comments by Sherwin Rosen. Note that we do not mean to imply that there may not be issues of self-selection here, though in this case they work to make the wage gap smaller than it might otherwise be. Further, there may also be issues of statistical 'discrimination' (Aigner and Cain, 1977), if employers expect the two groups to have different average rates ofjob withdrawal. However, we argue below that this is not likely to be the case, since job leaving behaviour is, if anything, more prevalent among males than females, ceteris paribus.

W. R. Reed and J . Dahlquist

Downloaded by [54.207.11.183] at 06:25 25 March 2014

Table 2. Mean values

of

additional lob charucreristics

Characteristic

Females

Males

(t-statistics)

(1)

(2)

(3)

(4)

Fringe benejits PAID VACa HLTHINSa LIFEINSa

0.8 1 8 0.748 0.607

0.745 0.696 0.534

Nonpecuniury job characteristicsb 3.410 SAFETY PLSNTSR 3.359 3.852 PEOPLE HLTH Y 3.472 OPPROMO 2.676 FEDBACK 3.873 JOBSIG 3.476 C H DOBS T 3.187 3.504 FRSHIPS VARIETY 3.182 AUTNOM Y 3.276 VALXPNS 3.257 3.836 CMPTASK

2.716 3.155 3.476 3.08 1 2.875 3.744 3.377 3.121 3.42 1 3.152 3.287 3.208 3.879

* An asterisk indicates that the difference between the means of the two samples is significant at the 5% level (two-tailed test). " P A I D V A C , H L T H I N S and L I F E I N S each take the value 1 if

these fringe benefits are received by t h e worker. variables are defined in the Appendix.

I,These

a given job attribute. For example, the survey question which underlies the variable SAFETY asks, 'How well or poorly does the following statement describe your job? 'The job is dangerous'. The response categories are '1 -very true'; '2-somewhat true'; '3-not too true'; and '4-not at all true'.' It is worth noting that, together, these variables represent most of the nonpecuniary job characteristics cited in the job satisfaction literature as being highly valued by workers (Warr and Wall, 1978; Mobley et al., 1979). As column (4) demonstrates, six of these variables differ significantly between the female and male samples at the 5 % level of significance (two-tailed test). In our opinion, the striking result here is that the self-reported responses from the workers closely conform to common assertions about male-female job differences; namely, that women are more likely to have jobs that offer safe and healthy work conditions (SAFETY and HLTHY), that are people-oriented (PEOPLE), that have pleasant work surroundings (PLSNTSR), and that d o not offer much opportunity for promotion (OPPROMO). In addition, the women in our

sample are more likely to report having jobs in which they are kept informed of their job performance (FEDBACK). These results suggest that differing preferences towards work characteristics may be responsible for the gender wage gap. 111. A N O T E O N T H E U S E O F T H E CONVENTIONAL DECOMPOSITION APPROACH T O EXPLAIN GENDER WAGE DIFFERENCES If men and women ofequal productivity face identical wageattribute choices, but women favour attractive job dimensions relatively more than men, then the theory of equalizing differences predicts that women will choose jobs characterized by more attractive job characteristics and lower wages. A number of researchers have explored this possibility by using the standard Blinder decomposition methodology to determine the extent to which attractive job characteristics 'explain' the male-female wage gap (Filer, 1985; Palme and Wright, 1992).8 Unfortunately, there are a number of reasons for being wary of this approach. First, the equalizing differences approach assumes that the different endowments of job characteristics are due to the differing choices of male and female workers. This need not be the case. For example, suppose dangerous jobs are high-paying (perhaps because of equalizing differences) and traditionally male. Suppose women are shut out of these occupations by prejudicial employers and ended up in low-paying, relatively safe occupations. A researcher might find that a decomposition of the wage gap revealed that differences in work conditions 'explained' all of the wage difference between men and women. However, it would clearly be a mistake in this case to attribute the job locations of women, and the resulting gender wage gap, to the preferences of women for safe working conditions. This underlines the fact that the equalizing differences/Blinder wage decomposition approach is unable in principle to distinguish choice-based (supply-side) theories of the wage gap from their discrimination (demandside) alternatives. The second deficiency concerns the static framework which underlies the theory of equalizing differences. In that framework, workers are assumed to face a 'budget constraint' of job choices, represented by the equilibrium hedonic wage function. It is crucial for the conventional

The use of these self-reported, Likert-scale variables introduces a number of problems into the empirical analysis. Different workers may rate identical jobs differently. This introduces an errors-in-variable problem. This problem is exacerbated by the fact that these variables are discrete and not continuous. One implication of this for statistical analysis is that standard errors will be inflated, reducing the significance of these variables. See Duncan and Holmlund (1983) for a nice discussion of this point. A more serious problem concerns the question of whether these variables represent interval scales (Stevens, 1960). If not, then classical statistical analysis is inappropriate. Although there are means of addressing this problem (Maddala, 1983), these are impractical given the purpose of this study and the large number of variables used in the empirical analysis. The assumption of equality of intervals is made throughout the subsequent analysis. Unfortunately, this assumption cannot be tested. Illuminating discussions of the Blinder decomposition methodology may be found in Jackson and Lindley (1989) and Nadeau et al. (1993).

Do women prefer women's work?

Downloaded by [54.207.11.183] at 06:25 25 March 2014

0.30 -fl

Fig. 1. Number of new jobs held by workers from 1982-87: (m) Male, (B) Female approach that the estimated wage equation correctly identifies the implicit prices of the respective job characteristics. However, recent theoretical contributions have established that the relationship between the estimated wage line and the underlying costs and benefits to firms and workers is distorted when labour markets are characterized by job search (Hwang et ul., 1993; Gronberg and Reed, 1994). Hwang et al. (1993) demonstrate that in an equilibrium search environment, estimated wage equations will yield biased estimates of attribute prices, causing the researcher t o misestimate not only the size but sometimes even the sign of the job characteristics' price^.^ This leads to inconsistent estimates of the monetary valuations of gender differences in job characteristics.

The severity of this last deficiency is related to the extent that search is a pervasive element of the labour market. In support of a dynamic view of the labour market, we consider the job changing behaviour of the workers in our sample. Figure 1 plots the number of new jobs held by male and female workers during the five year period immediately following the 1982 survey.'0 Approximately 75% of the women and men in the sample made at least one job change during this period. Over half of the women and men made two job changes. Furthermore, the majority of these job changes were voluntary." For this reason, we adopt a job search framework in developing our approach to the question, d o women prefer women's work?I2

9The intuition behind this result is as follows. It is well known that estimated wage equations can yield wrong-signed estimates of the true implicit prices of job characteristics when workers' job productivities are only partially observed. The same effect results when unobserved productivity differences exist across firms. More productive firms may offer higher wages and more attractive job characteristics in order to attract workers to their firms. 'OActually, this represents a slight undercount of new jobs, because some of the people working at the time of the 1982 interview had dropped out of the sample by 1987. ' I While reasons for quits were not collected during this later period (only two categories for quits exist in the 1984-87 surveys: 'quit for pregnancy or family reasons'; and 'quit for other reasons') we note that in earlier surveys, when this information was collected, 'found better job', 'bad working conditions', and 'pay too low' were all frequently cited reasons for leaving one's job. l Z In fact, we did perform the conventional wage decomposition approach and found that 13.3 cents (using female wage equation coefficients) or 22.4 cents (using male wage equation coefficients) of the $1.08 wage gap could be 'explained' by gender differences in nonpecuniary job characteristics.

W. R. Reed and J . Duhlquist

1138 IV. I M P L E M E N T A T I O N O F A N E W METHODOLOGY FOR DETERMINING WORKERS' PREFERENCES TOWARDS JOB ATTRIBUTES Theory

Downloaded by [54.207.11.183] at 06:25 25 March 2014

This paper employs the hazard model approach developed by Gronberg and Reed (1994) for determining workers' preferences towards job attributes. The hazard model approach works on the simple idea that workers will remain longer at jobs having more desirable characteristics. Let us suppose that each job may be characterized by a given vector Z of n + 1 dimensions, consisting of the wage and n nonwage job dimensions.

where the w subscript designates the wage dimension. Let u ( Z ) measure the utility that a worker receives from having a job characterized by Z. Following Burdett (1978), we presume an environment where worker turnover is due to on-the-job search. Define h ( - ) as the hazard function, and let it represent the probability that a worker voluntarily leaves (quits) his current employment in the next instant of time. Let t represent the elapsed time since the worker began employment at the current job; 6 the exogenous (involuntary) separation rate of workers from firms, say due to firm firings; q(u(Z), t) the rate at which job offers arrive, related to the utility the worker receives at his current job because he is assumed to choose his intensity of search; and F(-) the known distribution of feasible job-specific utility values. Then the hazard function can be written as follows. h(u(Z), t)=6+r](u(Z), t). [ I - F(u(Z), [)I.

(2)

Equation 2 can be used to analyse the effect on h(.) of an increase in a given job characteristic, Z i . [ I -F(u(Z), +r]

d [ l - F(u(Z), r)] du

I

01 (3)

that there are two lerms within the large parentheses i n Equation 3, and both have negative signs when workers favourably evaluate an increase in Zi. The first term captures the effect that workers are less likely to leave jobs with higher values of Z i because they will choose to d o less searching at such jobs, thus locating fewer alternative job offers. The second term represents the fact that there is a smaller probability that a n alternative job offer - once received - will dominate the worker's current job. It follows then that

The inequality of Equation 4 expresses the underlying notion that workers are less likely to quit jobs having more attractive job characteristics. Consider now a popular specification of the hazard function. It is common in thc duration literature to assume that the true hazard function can be specified as a proportional hazards modcl. h(Z, t) = j,,(t) exp (ZIP)

(5)

where I.,([) is the baseline hazard function, and /I is a vector of unknown coeficients. Note that the portion of the hazard function that is determined by job characteristics is separate from, but interacts multiplicatively with, the portion of the hazard function determined by tenure on the job (hence the name proportional hazards). As a result, the effect of a change in Z i on A(.) is given by r3h for i = rv, 1, . . . , n. CiZ, /liAo(t) exp (ZIP) = pi/?,

(6)

--=

It is casily seen that dl1 --30

azi

iff P i P o , i = w , 1, ..., n.

(7)

Note the similarity between the relationships in Equations 4 and 7. This leads us to the first of two results. Let be a vector of estimates of the individual Pi's. Then serves as indicator of the desirilbility of job dimension in particular,

Bi

pi 3

0 implies that

ciu(Z) $0, dZi

-

i = I , ..., 11.

B

(8)

This rcsult will prove useful as we investigate whether it is true that women positively value those job characteristics that most distinguish the kinds of jobs found in femaledominated occupations. The second result concerns the relative responsiveness of male and female quitting behaviour towards individual job dimensions. Suppose it is true that equally productive men and women have equal access to all jobs. Suppose it is also true that women particularly favour jobs that are safe and have pleasant work that are people-oriented, and d o not put a great deal of career demands on them. Then it follows that women should be observed to be less likely than men to le;lve jobs that are by these attributes. We measure this reponsiveness to a change in a given job characteristic Ziby the elasticity of the hazard function, chi. From Equation 6 it is easily shown thatI3 chi= B i z i .

(9)

l 3 This formulation, which is a direct consequence of the proportional hazards model, implies that the elasticity of the hazard is increasing in the value of the job attribute. Note that it does not imply that the elasticity of the quit rate is increasing in the job attribute.

1139

Do women prefer women's work? This yields the second result of the paper.

l ~ L l > < l iff~ ~ ll ~ / l > < l ~ i ~ l .

(10)

Thus, a direct empirical test of whether women are more responsive to changes in certain kinds of job attributes is provided by a comparison of the estimated 8;s for men and women. The remainder of this paper is concerned with an empirical investigation of the quitting behaviour of men and women.

Downloaded by [54.207.11.183] at 06:25 25 March 2014

Estimution of the hazard model This study estimates two specifications of proportional hazards models, Cox's proportional hazards model and the Weibull model (Cox, 1972; Kalbfleisch and Prentice, 1980).14 Table 3 reports the results of this analysis. Since an increase in the hazard rate implies an increase in workers' quit rates, positive coefficients indicate a positive association between quits and the respective variables. The following discussion focuses on those variables with significant coefficients. We first consider the labour market and personal characteristics variables. The coefficient for the local unemployment rate (URATE) is negative for both men and women,

though it is significant only for men. This is consistent with a model of on-the-job search in which workers are less likely to quit their jobs when alternative jobs are difficult to find. For both male and female workers, being married (MARRIED) is associated with decreased job mobility. This result agrees with previous studies which report that married workers generally display greater job stability. In contrast, older workers (AGE) are more likely to leave their current employment.ls Most studies find that older workers are less likely to quit their jobs. We suspect that our result is due to the fact that the workers in our samples are just beginning their labour market careers. When queried about their future working plans, most workers responded by saying that they planned on working for some other employer at age 35. Thus, the fact that older workers in this sample are more likely to quit may reflect that, ceteris purihus, older workers are better able to generate more attractive alternative employment opportunities. We turn next to the job characteristic variables. Recall that these are categorized in four groups. First is the category of wage and hours, in which the unionization variable is also included. As expected, increases in the log of the wage rate (LN WAGES) are associated with decreased quits for both men and women. Even so, there does appear

Table 3. Female and male hazard equations Cox model Variable (1)

Female (2)

Weibull model Male

Female

(3)

(4)

Male (5)

CONSTANT" URBAN URATE EDUC2

EDUC3 EDUC4 EDUCS DEPENDENTS WORKPLANA WORKPLANB MARRIED

l 4 Self-reported 'quits' were treated as completed spells. Self-reported 'fireds' were treated as censored spells. The BMDP utilities on SAS were used to estimate the hazard functions. Descriptions of the respective likelihood functions can be found in the corresponding SAS documentation. l 5 The age variable in the hazard equation is the age of the worker measured at the time he began his current employment spell. This contrasts with the age variable reported in Table 1, which is the worker's age as measured at the time of the 1982 survey date.

W. R. Reed and J . Dahlquist

1140 Table 3. Continued Cox model Female (2)

Variablc (1)

Male (3)

Weibull model Female (4)

Male (5)

AGE LMEXPER LN WAGE UNION FLXHOURS PAIDVAC

Downloaded by [54.207.11.183] at 06:25 25 March 2014

HLTHINS LIFEINS SAFETY PLSNTSR PEOPLE HLTHY OPPROMO FEDBACK JOBSIG CHDOBST FRSHIPS VARIETY AUTNOMY VALXPNS CMPTASK P"

Observations Log likelihood'

-

Test of restrictionsd Note: Asymptotic t-statistics are in parentheses below coefficient estimates. *Indicates coefficient estimate is significant at the 5% level (two-tailed test). a Note that no constant term for the Cox model is reported since there are one less as many constant terms as there are failure times. "The t-test on the p parameter tests whether it is significantly different from one, where one represents the special case when the Weibull model simplifies to the exponential model. 'The log likelihood values between the Cox and Weibull models are not directly comparable. *'This row reports the x2 statistics from the joint hypothesis that the coefficients for S A F E T Y , P L S N T S R , P E O P L E , H L T H Y and O P P R O M O are all equal to zero. None of the statistics is significant at the 5% level, where the corresponding critical value with five degrees of freedom is 1 1.07.

Downloaded by [54.207.11.183] at 06:25 25 March 2014

Do women prefer women's work? to be a substantial difference in the sizes of the coefficients for men and women. This is discussed in further detail below. Comparing the coefficients for the UNION variable, we see that unionization is associated with decreased quits for men, and insignificantly related to quits for women. This is consistent with the view that unions serve to address workers' grievances, diminishing conflicts that can result in voluntary job separations (Freeman, 1978). The second category is fringe benefits. For women, both paid vacations (PAIDVAC) and life insurance (LIFEINS) are significantly related to decreased workers' quits. For males, only the coefficient for paid vacations is significant. According to Equation 8, the negative coefficients here indicate that workers positively value these fringe benefits, which is of course what one would expect. We now consider the set of 13 nonpecuniary variables included in the analysis. Only two of these 13 are significant in any of the equations, CHDOBST for women and F R S H I P S for men. Both of these have the expected negative coefficients, with the opportunity to 'do the things you d o

1141 best' (CHDOBST) and 'develop close friendships in you job' (FRSHIPS) both reducing quits significantly. None of the other 13 nonpecuniary characteristics are significant, and that includes the five variables of particular interest to this study: SAFETY, PLSNTSR, PEOPLE, HLTHY, and OPPROMO. Further, for both the male and female samples, we cannot reject the hypotheses that all of the coefficients for these variables are equal to zero (see the X 2 values reported on the bottom of Table 3). While we d o not suggest that this finding indicates that workers d o not care about safe and attractive work conditions, it does indicate that they are not of sufficient importance to have an impact on the job mobility of the workers in our samples. Further evidence against voluntary sorting as a n explanation of occupational segregation is provided by a test of differences in male and female quitting behaviour. As indicated by the relationship (Equation 10) above, if women particularly favour 'women's work', then they should be less likely than men to leave jobs that are distinguished by traditionally female characteristics.

Table 4 . A comparison of male-female differences in the hazard equations

Hypotheses tested (1)

A. Labour market characteristics (H,: D U R B A N = D U R A T E = 0 ) B. Personal characteristics (H,: D = D E D U C 2 = DEDUC3 = DEDUC4= DEDUCS

=DDEPENDENTS= D WORKPLANA =DWORKPLANB=DMARRIED=DAGE =DLMEXPER =0)

C . Job characteristics: wage and hours ( H , : DLN W A G E S = D U N I O N =DFLXHOURS=O) D . Job characteristics: fringe benefits (H,: D P A I D V A C = D H L T H I N S =DLIFElNS=O) E. Job characteristics: nonpecuniary characteristics (first set) (H,: D S A F E T Y = D P L S N T S R =DPEOPLE =DHLTH Y =DOPPROMO=O)

F. Job characteristics: nonpecuniary characteristics (second set) (H,: DFEDBACK =DJOBSIG = DCHDOBST= DFRSHIPS

=DVARlETY=DAUTNOM Y =DVALXPNS=DCMPTASK=O)

G. Job characteristics: (E)-(F)

H. Job characteristics: (C)-(F)b All characteristics: (A)-(F)b

I.

Note: Values in parentheses are zZ statistics associated with the likelihood ratio test. Asterisks identify statistics significant at the 5% level. "Specification A consists of the basic hazard equation with sex dummies ( D equals 1 for females, 0 for male) interacted with all variables. Specification B consists of the basic hazard equation with sex dummies for L N WAGES, U N I O N and F L X H O U R S . bNote that these omit coefficients in (C) for Specification B.

Downloaded by [54.207.11.183] at 06:25 25 March 2014

1142 Table 4 presents the results of combining the male and female samples and testing for differences in coefficients for various subsets of variables.16 Aside from the wage and hours variables (see row (C) of Table 4), no statistically significant differences between female and male hazard elasticities can be identified. Taken on their own, the coefficients for LN WAGE d o suggest that women are less elastic in their quit probabilities than men (see Table 3). However, given the large number of variables included in this analysis, we cannot reject the possibility that this difference is a random deviation. Finally, we note that when the samples are expanded to include part-time workers, a larger number of the coefficients for the nonpecuniary job characteristics are significant in the respective hazard equations (results not reported). This suggests that nonpecuniary dimensions of work may be more important for those workers with less labour market attachment. Unfortunately, the signs of the coefficients are not consistent with the hypothesis that women prefer the job characteristics associated with 'women's work'."

V. S U M M A R Y A N D C 0 N C : L U S I O N

This paper investigates the hypothesis that occupational sex concentration is the result of female workers choosing to forego higher paying jobs in order to obtain more attractive job characteristics. Standard wage decompositions which attempt to 'explain' the gender wage differentials suffer from a number of problems. Most importantly, they are unable to identify whether the job characteristics of men and women represent voluntary choices (preferences) or differing degrees of access (discrimination) to certain kinds of jobs. This deficiency is addressed by implementing a new methodology for identifying worker preferences towards job

W. R. Reed und J . Duhlquist characteristics based o n quit behaviour (Gronberg and Reed, 1994). The intuition underlying this analysis derives from the notion that if women prefer particular types ofjobs, then they should be observed quitting those jobs less frequently. Summarizing the results of this analysis, we find that the full-time women workers in our sample are more likely than the full-time men workers to be employed in jobs that are safer, have greater people contact, more pleasant work environments, and involve lower rates of promotion. These job characteristics are ones commonly associated with 'women's work'. Even so, we find no evidence that the women workers disproportionately favour these job characteristics. While the present findings d o not prove that gender discrimination is responsible for observed occupation,'I 1 sex concentration, they are consistent with this h y p o t h e ~ i s . ' ~ In contrast, we d o find some evidence that women's quit rates are less responsive than men's to wage increases. As Viscusi (1980) notes, if women are less wage elastic in their quit behaviour, thcn, ceteris pcrribus, the optimizing firm would respond by paying women a lower wage rate, even if female workers displayed the same rate of quits as males when given the male wage rate.'"his is a topic for future research.

ACKNOWLEDGEMENTS We wish to thank Donald Deere, Tim Gronberg, John Lott, Tom Saving, and scminar participants at Texas A&M University and the University of Texas for helpful comments on earlier versions of this paper. A number of staff at the Center for Human Resources, Ohio State University supplied valuable advice relating to the data set; special thanks go to Laura Brandon, Steve McClaskey, and Karima Nagi. Robert Maness provided capable research assistance.

l 6 Table 4 reports likelihood ratio tests using Cox's proportional hazards model. Though not reported here, the likelihood ratio tests based on the Weibull model of the hazard function yield similar outcomes. "While we do find evidence that women prefer jobs with pleasant work surroundings ( P L S N T S R ) ,jobs requiring substantial peoplecontact ( P E O P L E ) appear to be negatively valued by women. l 8 If this were the case, then we would expect to see women suing to gain access to certain occupations. While the extent of this phenomenon is difficult to gauge, its existence is easily demonstrated. For example, women have taken legal measures to gain access to jobs as police officers (Costa v. Markejr, 706 F.2d. 1; Dal;is v. City ofDallas, 483 F. Supp. 54, 51 -61; Blake v. City of Los Angeles, 595 F. 2d. 1367);prison guards (Dothard v. Rawlinson, 433 US 321, 15 FEP Cases 10; Smith v. Fairmun, 678 F.2d. 52); construction workers (Local Union No. 3.5 of Intern. Broth. of Elec. Workers v. City of Hartford, 625 F.2d. 416); meat cutters (Babrocky v. Jewel Food C o . and Retail Meatcutters, 645 F . Supp. 1396, 1423);sheet metal workers (Reynolds v. Sheet Metal Workers Local 102); railroad brakemen (Coleman v. Missouri Pac. R. Co., 622 F. 2d. 408); and to jobs in work environments in which they were excluded because the work involved 'laborious tasks and long hours of work'(Chrapi1iwig v. Uniroyal, Inc., 458 F. Supp. 252) or because the unhealthy nature of the workplace threatened their offspring (Wright v. Olin Corp., 697 F.2d. 1172; Harper v. Thiokol Chetnical Corp., 619 F.2d. 489). Lastly, we note that we also observe cases in which men sue to gain access to traditional women's occupations, such as nursing (Mississippi University.for Women v. Hogan, 102 S. Ct. 3331) and flight attending (Diaz v. Pan American World Airways, Itlc., 442 F.2d. 385, 3 FEP Cases 337). l 9 There is another, related implication involvingjob attributes. Suppose, as the evidence suggests, that safer, more attractive jobs pay lower wages. Suppose further that men and women are more likely to enter the labour market at these jobs. If men are more responsive to wage changes than women, then men would be more likely to leave. The result would be a concentration of women in lower-paying, safer, and more attractive jobs. Not because they particularly valued those characteristics, but because they were less likely to leave those jobs once there.

Do women prefer women's work?

Downloaded by [54.207.11.183] at 06:25 25 March 2014

REFERENCES Aigner, D. and Cain, G. (1977) Statistical theories of discrimination in the labor market, Industrial and Lahor Relations Review, 30, 175-87. Barron, J., Black, D. and Loewenstein, M. (1993) Gender differences in training, capital, and wages, Journal of Human Resources, 28, 343-64. Beller, A. (1984) Trends in occupational segregation by sex and race, 1960- 1981, in Sex Segregation in the Workplace: Trends, Explanations, Remedies, (ed) B. Reskin, (National Academy Press, Washington, DC). Blinder, A. (1973) Wage discrimination: reduced form and structural variables, Journal of Human Resources, 8, 436-55. Burdett, K. (1978) Employee search and quits, American Economic Review, 68, 212-20. Cain, G. (1 986) The economic analysis of labor market discrimination: a survey, in Handbook of Labor Economics, Volume I , (eds) 0 . Aschenfelter and R. Layard, (North-Holland, Amsterdam). Cameron, T. (1986) Some reflections on comparable worth, Contemporary Policy Issues, 4, 33-9. Cox, D. (1972) Regression models and life tables, Journal of the Royal Statistical Society B, 34, 187-200. Duncan, G. and Holmlund, B. (1983) Was Adam Smith right after all? another test of the theory of compensating wage differentials, Journal of Labor Economics, 1 , 366-79. Filer, R. (1985) Male-female wage differences: the importance of compensating differentials, lndustrial and Labour Relations Review, 38, 426-37. Filer, R. (1986) The role of personality and tastes in determining occupational structure, lndustrial and Labor Relations Review, 39,412-24. Filer, R. (1989) Occupational segregation, compensating differentials, and comparable worth, in Pay Equity: Empirical Inquiries, (eds) R. Michael, H. Hartmann, and B. O'Farrell, (National Academy Press, Washington, DC). Freeman, R. (1978) Job satisfaction as an economic variable, American Economic Review, 68, 135-41. Goldin, C. and Polachek, S. (1987) Residual differences by sex: perspectives on the gender gap in earnings, Papers and Proceedings of the American Economic Association, 77, 143-51. Gronberg, T. and Reed, W. (1994) Estimating workers' marginal willingness to pay for job attributes using duration data, Journal (4'Human Resources, 29, 9 1 1-3 1. Hwang, H., Mortensen, D. and Reed, W. (1993) Hedonic wages and labor market search, mimeo, Northwestern University. Jackson, J. and Lindley, J. (1989) Measuring the extent of wage discrimination: a statistical test and caveat, Applled Economics, 21, 5 15-40. Kalblleisch, J. and Prentice, R. (1980) The Statistical Analysis of Failure Time Data, (Wiley, New York). Maddala, G . (1983) Limited-Dependant and Qualitative Variables in Econometrics, (Cambridge University Press, Cambridge). Malkiel, B. and Malkiel, J. (1973) Male-female pay differentials in professional employment, American Economic Review, 63, 693-705. Mobley, W., Griffeth, R., Hand, H. and Meglino, B. (1979) Review and conceptual analysis of the employee turnover process, Psychological Bulletin, 86, 493-522. Nadeau, S., Walsh, W. and Wetton, C. (1993) Gender wage discrimination: methodological issues and empirical results for a Canadian public sector employer, Applied Economics, 25, 227-41. Palme, M. and Wright, R. (1992) Gender discrimination and

compensating wage differentials in Sweden, Applied Economics, 24, 751 -9. Rees, A. and Schultz, G. (1970) Workers and Wages in an Urban Labor Market, (University of Chicago Press, Chicago). Reskin, B. and Hartmann, H. (1986) (eds) Women's Work, Men's Work: Sex Segregation on the Job, (National Academy Press, Washington, DC). Rytina, N. and Bianchi, S. (1984) Occupational reclassification and changes in distribution by gender, Monthly Labor Review, CVII, 1 1 - 17. Seiling, M. (1984) Staffing patterns prominent in female-male earnings gap, Monthly Labor Review, CVII, 29-33. Stevens, S. (1960) O n the theory of scales of measurement, in Philosophy of Science, (eds) A. Danto and S. Morgenbesser, (Meridian Books, New York). US Bureau of the Census, Current Population Reports, Series P-23, No. 146, Women in the American Economy, by C. Taeuber and V. Valdisera, (US Government Printing Office, Washington, DC). US Department of Labor, Bureau of Labor Statistics, Employment and Earnings, (US Government Printing Office, Washington, DC). Viscusi, W. (1980) Sex differences in worker quitting, Review of Economics and Statistics, 62, 388-98. Warr, P. and Wall, T. (1978) Work and Well-being, (Penguin Books, New York).

APPENDIX Description of vuriubles AUTNOMY

CHDOBST

CMPTASK

EDUC2 EDUC3

EDUC4 EDUCS

FEDBACK

a measure of the independence a worker has a t current j o b (takes values between 1 a n d 5, with larger values indicating greater independence). a measure of the opportunity for the worker t o use his most valued skills a t current j o b (takes values between 1 a n d 4 , with larger values indicating greater opportunity t o use most valued skills). a measure of the opportunity t o complete assigned tasks a t current j o b (takes values between 1 a n d 5, with larger values indicating greater opportunity t o follow task from beginning t o end). takes the value I if worker started b u t did not complete high school. takes the value 1 if worker completed high school but did not continue formal schooling. takes the value 1 if worker started but did not complete college. takes t h e value 1 if worker completed college. a measure of feedback o n performance a t current j o b (takes values between 1 a n d 5, with larger values indicating m o r e feedback).

W. R. Reed and J . Dahlquist

1144 FLXHOURS FRSHIPS

HLTHINS HLTHY

JOBSIG

Downloaded by [54.207.11.183] at 06:25 25 March 2014

JOBEXPER LIFEINS LMEXPER

takes the value 1 if hours are flexible at current job. a measure of the opportunity to form close friendships at current job (takes values between 1 and 5, with larger values indicating greater friendship opportunities). takes the value 1 if worker receives health insurance benefit at current job. a measure of the healthiness of current job (takes values between 1 and 4, with larger values indicating healthier work conditions). a measure of the worker's evaluation of the significance of the current job (takes values between I and 5, with larger values indicating greater significance). weeks employed in current job measured at time of 1982 survey date. takes the value 1 if worker receives life insurance benefit at current job. weeks of labour market experience after sixteenth birthday or date last enrolled in school (whichever is latest) - measured at time of 1982 survey date minus JOBEXPER. takes the value 1 if worker is currently married. a measure of the probability of promotion at current job (takes values between 1 and 4, with larger values indicating greater likelihood of promotion).

PAID VAC PEOPLE

PLSNTSR

SAFETY

UNION URBAN VALXPNS

VARIETY

-

MARRIED OPPROMO

WORKPLANA WORKPLANB

takes the value 1 if worker receives paid vacation benefit a t current job. a measure of the degree of contact with other people at current job (takes values between I and 5, with larger values indicating greater contact). a measure of the pleasantness of physical surroundings at current job (takes values between 1 and 4, with larger values indicating more pleasant surroundings). a measure of the safety inherent in current job (takes values between 1 and 4, with larger values indicating the job is less dangerous). takes the value 1 if wages are determined by a collective bargaining agreement. takes the value I if worker's residence is located in an urban area. a measure of general, on-the-job training at current job (takes values between 1 and 4, with larger values indicating more valuable training). a measure of the variety of work at current job (takes values between 1 and 5, with larger values indicating greater variety). takes the value 1 if respondent plans to be working at age 35. takes the value 1 if respondent plans to be raising a family at age 35 and wants to work outside the home.