This article was downloaded by: [University of Sydney] On: 01 May 2015, At: 02:57 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Women & Health Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/wwah20
A Causal Analysis of Employment and Health in Midlife Women a
b
Pamela K. Adelmann PhD , Toni C. Antonucci PhD , Susan c
E. Crohan PhD & Lerita M. Coleman PhD
d
a
School of Social Work, University of Michigan
b
University of Michigan
c
University of Wisconsin
d
University of Colorado Published online: 02 Nov 2010.
To cite this article: Pamela K. Adelmann PhD , Toni C. Antonucci PhD , Susan E. Crohan PhD & Lerita M. Coleman PhD (1990) A Causal Analysis of Employment and Health in Midlife Women, Women & Health, 16:1, 5-20, DOI: 10.1300/J013v16n01_02 To link to this article: http://dx.doi.org/10.1300/J013v16n01_02
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.
Downloaded by [University of Sydney] at 02:57 01 May 2015
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
A Causal Analysis of Employment and Health in Midlife Women
Downloaded by [University of Sydney] at 02:57 01 May 2015
Pamela K. Adelmann, PhD Toni C. Antonucci, P h D S u s a n E. Crohan, PhD Lerita M. Coleman, PhD
ABSTRACT. According to the "healthy worker" hypothesis, good health selects women into the employment role; conversely,. the social causation hypothesis argues that important social roles such as employment can contribute to health. The relationship between these two variables may be of special concern to women at midlife, as both issues become increasingly salient. In this paper a model hypothesizing a nonrecursive (reciprocal) causal relationship between employment (hours employed) and health was presented. Because of the importance of both employment and health for women at midlife, the model was tested in a sample of 463 women ages 4064 from a national cross-sectional dataset. Two-stage least squares regression supported the hypothesis that employment and health have a positive reciprocal relationship for women at midlife. In addition, married women and those with a child had better health but were likely to work fewer hours for pay. The same model tested in 21 to 39-year-old women was not significant. The results serve to emphasize the importance of examining assumptions about direction of causality in research on work and well-being at midlife. Pamela K. Adelmann and Toni C. Antonucci are affiliated with the University of Michigan. Susan E. Crohan is with the University of Wisconsin and Lerita M. Coleman is affiliated with the University of Colorado. Address correspondence to Pamela K. Adelmann, School of Social Work, 330 E. Liberty, Suite 4B, University of Michigan, Ann Arbor, MI 48104-9964. This research was supported by grants from the National Institute of Aging (1R3 AG04285-01) and the National Institute of Mental Health (MH 38722-01). We wish to thank J. Veroff, E. Douvan, and R. Kulka for the use of their data, and J. Jackson, J. Krosnick, P. Niedenthal, A. Reifman, and M. Torres for helpful comments on an earlier version of this manuscript. Women & Health, Vol. 16(1) 1990 O 1990 by The Haworth Press, Inc. All rights reserved. 5
Downloaded by [University of Sydney] at 02:57 01 May 2015
6
WOMEN & HEALTH
The positive association between employment and health among women has been well documented. Employed women in general have better perceived health status (Verbrugge, 1982; Verbrugge & Madans, 1985), fewer health symptoms (Kessler & McRae, 1982; Verbrugge & Madans, 1985; Woods & Hulka, 1979), fewer days of restricted activity due to illness (Nathanson, 1980; Rivkin, cited in Nathanson, 1975; Verbrugge, 1982; Verbrugge & Madans, 1985) and less limitation from chronic conditions (Verbrugge, 1982; Verbrugge & Madans, 1985). At midlife, employment is also related to better health (Coleman & Antonucci, 1983; Jennings et al., 1984; Nathanson, 1980), and this association may even be stronger among middle-aged women than in younger women (Arber et al., 1985). The issues of health and employment are of particular interest among midlife women (Nolan, 1986). Personal achievement is a major concern of women at midlife, and women may turn to the paid work role as an important source of self-definition at this time (Guttman, 1975; Neugarten, 1968; Rossi, 1980). Flexibility of movement into and out of the paid labor force is likely to be at a maximum for women at midlife as the responsibilities of child-rearing lessen. Health is also a salient issue for midlife women. The hormonal changes of menopause may introduce or exacerbate a number of physical symptoms (Soules & Bremner, 1982), age-related health problems become more likely (Shock, 1977; Timiras, 1972), and the role changes and self-reflection characteristic of midlife may also have consequences for women's physical health. Two major arguments have been advanced to explain differences in health between employed and non-employed women. According to the selection hypothesis or "healthy worker effect," women in better health are more likely to seek, obtain, continue in or return to paid employment than their less healthy counterparts. In contrast, the social causation hypothesis implies that employment, as a valued social role, can have beneficial consequences for physical wellbeing. However, it is also possible that both causal hypotheses are valid. In this paper, it was hypothesized that employment (hours worked per week) and health are causally related in a reciprocal fashion among midlife women. The central hypothesis of the study was that while health can
Downloaded by [University of Sydney] at 02:57 01 May 2015
Adelmann et al.
7
influence involvement in the employment role, employment also influences self-reported health. There are some arguments in favor of both causal paths, particularly at midlife. I11 health is given by many women as a reason for not seeking employment or for having left a former job (Flaim, 1968; U.S. Department of Labor, 1975), especially among midlife women (Jennings et al., 1984; Rosenfeld & Perrella, 1965), and women in poor health are less likely to say that they plan to work in the future (Kessler & McRae, 1982). Employed women frequently give poor health as a reason that they believe would cause them to cease working in the future (Rosenfeld & Perrella, 1965). On the other hand, there are several possible mechanisms by which employment may influence health. Employed women, especially those employed full-time, may have better access to health services because of employee health insurance benefits or because of higher income levels. Psychological benefits accruing to the employment role may also have positive consequences for health. Employment provides an important source of self-esteem and identity (Warr, 1982), particularly in contrast to the low social esteem of the homemaker role (Brown & Harris, 1978; Coleman & Antonucci, 1983), and is a major source of personal satisfaction (Rosenfeld & Perrella, 1965; Weiss & Samelso,n, 1958). Women also receive social support through employment at midlife (Jacobson, 1974), which may benefit physical health (Berkman & Syrne, 1979; Gore, 1978; Seeman et al., 1985). .Waldron et al. (1982) tested both hypotheses using longitudinal data and found evidence only of the healthy worker effect. Although they found no significant effects of employment on health, their study was limited by the use of health and employment measures that were dichotomous rather than continuous, and their test of the social causation hypothesis was restricted to married women with good initial health. A secondary hypothesis was that marital status influences both health and employment involvement at midlife. Married women have better health than the unmarried (Renne, 1971; U.S. Department of Health, Education & Welfare, 1976; Verbrugge, 1979, 1982; Verbrugge & Madans, 1985), even at midlife (Nathanson, 1975), perhaps because of higher levels of social support or higher
Downloaded by [University of Sydney] at 02:57 01 May 2015
8
WOMEN & HEALTH
family income. Married women are also less likely to work for pay than unmarried women (U.S. Department of Labor, 1975). Married women may be less financially compelled to work, and may even be subject to pressure from a spouse to stay at home rather than seek paid employment (Molm, 1978). An additional secondary hypothesis was that midlife women who have children are less likely to be employed, but are more likely to be in good health. Women with no children are more likely to have a paid job (Sweet, 1973), and are more likely to work full-time (U.S. Department of Labor, 1975); these women are likely to have more available time for paid work, more desire for employment as a source qf personal gratification, and fewer social pressures not to work for pay. Women with children also tend to have better health than women without children (Marcus & Seeman, 1981). Social esteem and personal satisfactions of motherhood are possible mechanisms by which this role might benefit physical health; older children may also begin to provide their parents with social support, with beneficial consequences to health. Age was posited to have a direct relationship to the health of midlife women. Cellular and "wear and tear" theories of aging suggest that inevitable biological changes and environmental stresses associated with aging have detrimental consequences for health (Shock, 1977; Timiras, 1972). Age was hypothesized to be indirectly related to employment via its relationship with health; because older women are more likely to have health problems, they are less likely to be employed. The possibility of a direct effect of age on employment was thought to be minimal in this age group. In one study, for example, among midlife women who wanted jobs, only 6 percent said they were not seeking them because employers would think they were too old. In contrast, nearly a quarter of this group gave ill health or disability as the reason for not seeking employment (Flaim, 1968). In the present case, only a small percentage of midlife women gave age as the main reason they were not employed (either because they had experienced age discrimination or believed they were too old-less than ten percent), and they were excluded from the sample. Finally, a direct relationship of education to employment was hypothesized. Women with more education are more likely to
Downloaded by [University of Sydney] at 02:57 01 May 2015
Adelmann et al.
9
work, regardless of marital status and age, and are likely to work more hours than the less educated (U.S.Department of Labor, 1975). Education was posited to have an indirect effect on health through employment; because better educated women are more likely to be employed, they reap more of the practical and psychological benefits to health at midlife than less educated women. In sum, the theoretical and empirical evidence contribute to the plausibility of the hypothesized causal model. The central causal paths of interest are the reciprocal links between hours employed and health. The exogenous variables of education, marital status, parental status, and age, complete the specification of the hypothesized model. To verify that these relationships are of particular importance at midlife, the model was tested in a sample of younger women as well. METHOD Data
The data were drawn from the Americans View Their Mental Health (AVTMH) dataset (Veroff et al., 1981). This nationally representative survey of 2,264 adults ages 21 and over was conducted in 1976 (for sampling details, see Kish & Hess, 1965). The sample for this study included'463 midlife women between the ages of 40 and 64. The comparison sample contained 566 women aged 21 to 39.
Measures Employment was defined as number of hours worked per week and ranged from zero to eighty-four hours. Using this variable rather than a dichotomy (employed/not employed) both improved the level of measurement over previous studies (cf. Waldron, 1982) and allowed us to assume a linear relationship between employment and health (where women employed part-time should fall between non-employed women and women employed full-time). In addition, interpretation of a path coefficient leading to a continuous variable is more straightforward than of one leading to a dichotomous variable (in the former case, the coefficient would be inter-
Downloaded by [University of Sydney] at 02:57 01 May 2015
10
WOMEN & HEALTH
pretable in hours employed per week, while in the latter it would be roughly comparable to the probability of being employed). Because preliminary analysis revealed parallel results using the two employment measures, the more easily interpretable "hours employed" was chosen for the final model. The health measure is a summary score of six items derived in an orthogonal factor analysis from'a symptoms checklist by Veroff et al. (1981). This physical health factor was statistically independent of the more psychologically-oriented symptoms factors (anxiety and immobilization).' The index reflects respondents' general selfreported health status and is conceptually similar to measures in common use in studies of employment and health (e.g., Coleman & Antonucci, 1983; Nathanson, 1980; Waldron et al., 1982). The items included "Do you have any particular physical or health trouble?," "Do you feel you are bothered by all sorts of pains and ailments in different parts of your body?," coded 1 = yes, 4 = no; "For the most part, do you feel healthy enough to carry out the things that you would like to do?" was coded 1 = no, 4 = yes. The remaining three items ("Have you ever been bothered by your heart beating hard?," "Have you ever been bothered by shortness of breath when you were not exercising or working hard?," and "Has any ill health affected the amount of work you do?") were coded 1 = many times, 2 = sometimes, 3 = hardly ever, 4 = never. Scores on each item were summed and could range from 6 = poor health to 24 = good health (Cronbach's alpha = .77) Marital status was coded as a dichotomous variable (0 = not married, 1 = married), as was parental status (0 = not a parent, 1 = parent). Age of respondent was coded continuously in years, while education was coded for the highest grade of school completed (less than one grade to seventeen grades and beyond). The intercorrelations for all variables in the model appear in Table 1. PROCEDURE
We used the simultaneous equations technique of two-stage least squares regression (2SLS) to test the hypothesized nonrecursive causal model (Berry, 1984; Blalock, 1971). In this model, the two
Adelmann el al. Table 1. lntercorrelatlons among variables In the model.
1 Employment
Downloaded by [University of Sydney] at 02:57 01 May 2015
2 Health
.24
3 Marltal status
-31
.09
4 Parental status
-.l I
.11
.22
5 Age
-.lo
-.2 1
-.I9
.20
.27
.09
.06
1
2
3
4
6 Education
-.I5
-.08 5
nonrecursively (reciprocally) related variables of employment and health are known as endogenous variables. The exogenous variables of marital status and parental status predict both employment and health. Age, which is directly related only to health, and education, which is directly related only to employment, are known as instrument variables. In the 2SLS program we used, in the first stage each endogenous variable was regressed on the set of exogenous variables by which it is directly caused, yielding a "purged" version of the endogenous variable; for example, health was regressed on age, parental status, and marital status. In the second stage each endogenous variable was regressed on the purged endogenous variable and the exogenous variables by which it is directly caused (e.g., employment was regressed on education, marital status, parental status, and the "purged" health variable). The regression coefficients resulting from the second-stage regressions were used as path coefficients for the model. Several assumptions must be met in using the 2SLS technique (James & Singh, 1978). Linear and additive .effects are assumed, and measures are assumed to be interval level scales with an acceptable degree of reliability. The error terms of the endogenous variables are assumed to be normally distributed with means of zero, and are uncorrelated with the exogenous variables in the model. The exogenous variables are assumed to be measured without error.
I2
WOMEN & HEALTH
Downloaded by [University of Sydney] at 02:57 01 May 2015
Finally, it is assumed that the directions of causal flow have been correctly specified. It is important to note that a 2SLS solution does not provide proof of a causal model. Hypotheses concerning causation can be rejected by a 2SLS procedure, but there are usually multiple models in addition to the one being tested that are plausible and consistent with the data. However, if predictions regarding parameters are substantiated by analysis and a sound theoretical argument is advanced, the model is corroborated and a basis for causal inference exists (James & Singh, 1978; Schmitt & Bedeian, 1982). RESULTS Description of the Sample Among this sample of midlife women, approximately 65% are married and 35% are not married; about 89% have children, while the remaining 11% do not. The average number of grades of school completed is over eleven (with a standard deviation of three grades), and the mean age of this group is nearly 52 years old (standard deviation of seven years). The average number of hours employed per week is 19 from a range of 0 to 84 hours (sad. = 19.7), and the average health score on a scale of 6 to 24 is 17, (s.d. = 3.7). In comparison, for the 21 to 39 year old women, 65% are married and 68% have children. They have completed almost 13 years of school (s.d. = 2.4), are an average of 29 years old (s.d. = 5.2), work 22 hours per week for pay (s.d. = 20.4), and have health scores of 18.8 (s.d. = 2.7). Tests of Assumptions Several of the assumptions of the 2SLS technique can be put to empirical tests. The assumption that the error or disturbance terms (Ue and Uh) are not correlated with the exogenous variables in the model was tested first by examining the zero-order correlation coefficients. The results indicate that none of the correlations between
Downloaded by [University of Sydney] at 02:57 01 May 2015
Adelmann el al.
13
the disturbance terms and age, education, marital status, and parental status are significantly different from zero in the midlife sample. Two assumptions that are important to the identification of the model and that bear on the suitability of the variables selected as instruments (Berry, 1984; Blalock, 1971; Kenny, 1979) were also tested. That each instrument has a significant association with the endogenous variable to which it is directly causally related was supported; education is significantly correlated with employment at midlife (r = .20, p
Women & Health Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/wwah20
A Causal Analysis of Employment and Health in Midlife Women a
b
Pamela K. Adelmann PhD , Toni C. Antonucci PhD , Susan c
E. Crohan PhD & Lerita M. Coleman PhD
d
a
School of Social Work, University of Michigan
b
University of Michigan
c
University of Wisconsin
d
University of Colorado Published online: 02 Nov 2010.
To cite this article: Pamela K. Adelmann PhD , Toni C. Antonucci PhD , Susan E. Crohan PhD & Lerita M. Coleman PhD (1990) A Causal Analysis of Employment and Health in Midlife Women, Women & Health, 16:1, 5-20, DOI: 10.1300/J013v16n01_02 To link to this article: http://dx.doi.org/10.1300/J013v16n01_02
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.
Downloaded by [University of Sydney] at 02:57 01 May 2015
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
A Causal Analysis of Employment and Health in Midlife Women
Downloaded by [University of Sydney] at 02:57 01 May 2015
Pamela K. Adelmann, PhD Toni C. Antonucci, P h D S u s a n E. Crohan, PhD Lerita M. Coleman, PhD
ABSTRACT. According to the "healthy worker" hypothesis, good health selects women into the employment role; conversely,. the social causation hypothesis argues that important social roles such as employment can contribute to health. The relationship between these two variables may be of special concern to women at midlife, as both issues become increasingly salient. In this paper a model hypothesizing a nonrecursive (reciprocal) causal relationship between employment (hours employed) and health was presented. Because of the importance of both employment and health for women at midlife, the model was tested in a sample of 463 women ages 4064 from a national cross-sectional dataset. Two-stage least squares regression supported the hypothesis that employment and health have a positive reciprocal relationship for women at midlife. In addition, married women and those with a child had better health but were likely to work fewer hours for pay. The same model tested in 21 to 39-year-old women was not significant. The results serve to emphasize the importance of examining assumptions about direction of causality in research on work and well-being at midlife. Pamela K. Adelmann and Toni C. Antonucci are affiliated with the University of Michigan. Susan E. Crohan is with the University of Wisconsin and Lerita M. Coleman is affiliated with the University of Colorado. Address correspondence to Pamela K. Adelmann, School of Social Work, 330 E. Liberty, Suite 4B, University of Michigan, Ann Arbor, MI 48104-9964. This research was supported by grants from the National Institute of Aging (1R3 AG04285-01) and the National Institute of Mental Health (MH 38722-01). We wish to thank J. Veroff, E. Douvan, and R. Kulka for the use of their data, and J. Jackson, J. Krosnick, P. Niedenthal, A. Reifman, and M. Torres for helpful comments on an earlier version of this manuscript. Women & Health, Vol. 16(1) 1990 O 1990 by The Haworth Press, Inc. All rights reserved. 5
Downloaded by [University of Sydney] at 02:57 01 May 2015
6
WOMEN & HEALTH
The positive association between employment and health among women has been well documented. Employed women in general have better perceived health status (Verbrugge, 1982; Verbrugge & Madans, 1985), fewer health symptoms (Kessler & McRae, 1982; Verbrugge & Madans, 1985; Woods & Hulka, 1979), fewer days of restricted activity due to illness (Nathanson, 1980; Rivkin, cited in Nathanson, 1975; Verbrugge, 1982; Verbrugge & Madans, 1985) and less limitation from chronic conditions (Verbrugge, 1982; Verbrugge & Madans, 1985). At midlife, employment is also related to better health (Coleman & Antonucci, 1983; Jennings et al., 1984; Nathanson, 1980), and this association may even be stronger among middle-aged women than in younger women (Arber et al., 1985). The issues of health and employment are of particular interest among midlife women (Nolan, 1986). Personal achievement is a major concern of women at midlife, and women may turn to the paid work role as an important source of self-definition at this time (Guttman, 1975; Neugarten, 1968; Rossi, 1980). Flexibility of movement into and out of the paid labor force is likely to be at a maximum for women at midlife as the responsibilities of child-rearing lessen. Health is also a salient issue for midlife women. The hormonal changes of menopause may introduce or exacerbate a number of physical symptoms (Soules & Bremner, 1982), age-related health problems become more likely (Shock, 1977; Timiras, 1972), and the role changes and self-reflection characteristic of midlife may also have consequences for women's physical health. Two major arguments have been advanced to explain differences in health between employed and non-employed women. According to the selection hypothesis or "healthy worker effect," women in better health are more likely to seek, obtain, continue in or return to paid employment than their less healthy counterparts. In contrast, the social causation hypothesis implies that employment, as a valued social role, can have beneficial consequences for physical wellbeing. However, it is also possible that both causal hypotheses are valid. In this paper, it was hypothesized that employment (hours worked per week) and health are causally related in a reciprocal fashion among midlife women. The central hypothesis of the study was that while health can
Downloaded by [University of Sydney] at 02:57 01 May 2015
Adelmann et al.
7
influence involvement in the employment role, employment also influences self-reported health. There are some arguments in favor of both causal paths, particularly at midlife. I11 health is given by many women as a reason for not seeking employment or for having left a former job (Flaim, 1968; U.S. Department of Labor, 1975), especially among midlife women (Jennings et al., 1984; Rosenfeld & Perrella, 1965), and women in poor health are less likely to say that they plan to work in the future (Kessler & McRae, 1982). Employed women frequently give poor health as a reason that they believe would cause them to cease working in the future (Rosenfeld & Perrella, 1965). On the other hand, there are several possible mechanisms by which employment may influence health. Employed women, especially those employed full-time, may have better access to health services because of employee health insurance benefits or because of higher income levels. Psychological benefits accruing to the employment role may also have positive consequences for health. Employment provides an important source of self-esteem and identity (Warr, 1982), particularly in contrast to the low social esteem of the homemaker role (Brown & Harris, 1978; Coleman & Antonucci, 1983), and is a major source of personal satisfaction (Rosenfeld & Perrella, 1965; Weiss & Samelso,n, 1958). Women also receive social support through employment at midlife (Jacobson, 1974), which may benefit physical health (Berkman & Syrne, 1979; Gore, 1978; Seeman et al., 1985). .Waldron et al. (1982) tested both hypotheses using longitudinal data and found evidence only of the healthy worker effect. Although they found no significant effects of employment on health, their study was limited by the use of health and employment measures that were dichotomous rather than continuous, and their test of the social causation hypothesis was restricted to married women with good initial health. A secondary hypothesis was that marital status influences both health and employment involvement at midlife. Married women have better health than the unmarried (Renne, 1971; U.S. Department of Health, Education & Welfare, 1976; Verbrugge, 1979, 1982; Verbrugge & Madans, 1985), even at midlife (Nathanson, 1975), perhaps because of higher levels of social support or higher
Downloaded by [University of Sydney] at 02:57 01 May 2015
8
WOMEN & HEALTH
family income. Married women are also less likely to work for pay than unmarried women (U.S. Department of Labor, 1975). Married women may be less financially compelled to work, and may even be subject to pressure from a spouse to stay at home rather than seek paid employment (Molm, 1978). An additional secondary hypothesis was that midlife women who have children are less likely to be employed, but are more likely to be in good health. Women with no children are more likely to have a paid job (Sweet, 1973), and are more likely to work full-time (U.S. Department of Labor, 1975); these women are likely to have more available time for paid work, more desire for employment as a source qf personal gratification, and fewer social pressures not to work for pay. Women with children also tend to have better health than women without children (Marcus & Seeman, 1981). Social esteem and personal satisfactions of motherhood are possible mechanisms by which this role might benefit physical health; older children may also begin to provide their parents with social support, with beneficial consequences to health. Age was posited to have a direct relationship to the health of midlife women. Cellular and "wear and tear" theories of aging suggest that inevitable biological changes and environmental stresses associated with aging have detrimental consequences for health (Shock, 1977; Timiras, 1972). Age was hypothesized to be indirectly related to employment via its relationship with health; because older women are more likely to have health problems, they are less likely to be employed. The possibility of a direct effect of age on employment was thought to be minimal in this age group. In one study, for example, among midlife women who wanted jobs, only 6 percent said they were not seeking them because employers would think they were too old. In contrast, nearly a quarter of this group gave ill health or disability as the reason for not seeking employment (Flaim, 1968). In the present case, only a small percentage of midlife women gave age as the main reason they were not employed (either because they had experienced age discrimination or believed they were too old-less than ten percent), and they were excluded from the sample. Finally, a direct relationship of education to employment was hypothesized. Women with more education are more likely to
Downloaded by [University of Sydney] at 02:57 01 May 2015
Adelmann et al.
9
work, regardless of marital status and age, and are likely to work more hours than the less educated (U.S.Department of Labor, 1975). Education was posited to have an indirect effect on health through employment; because better educated women are more likely to be employed, they reap more of the practical and psychological benefits to health at midlife than less educated women. In sum, the theoretical and empirical evidence contribute to the plausibility of the hypothesized causal model. The central causal paths of interest are the reciprocal links between hours employed and health. The exogenous variables of education, marital status, parental status, and age, complete the specification of the hypothesized model. To verify that these relationships are of particular importance at midlife, the model was tested in a sample of younger women as well. METHOD Data
The data were drawn from the Americans View Their Mental Health (AVTMH) dataset (Veroff et al., 1981). This nationally representative survey of 2,264 adults ages 21 and over was conducted in 1976 (for sampling details, see Kish & Hess, 1965). The sample for this study included'463 midlife women between the ages of 40 and 64. The comparison sample contained 566 women aged 21 to 39.
Measures Employment was defined as number of hours worked per week and ranged from zero to eighty-four hours. Using this variable rather than a dichotomy (employed/not employed) both improved the level of measurement over previous studies (cf. Waldron, 1982) and allowed us to assume a linear relationship between employment and health (where women employed part-time should fall between non-employed women and women employed full-time). In addition, interpretation of a path coefficient leading to a continuous variable is more straightforward than of one leading to a dichotomous variable (in the former case, the coefficient would be inter-
Downloaded by [University of Sydney] at 02:57 01 May 2015
10
WOMEN & HEALTH
pretable in hours employed per week, while in the latter it would be roughly comparable to the probability of being employed). Because preliminary analysis revealed parallel results using the two employment measures, the more easily interpretable "hours employed" was chosen for the final model. The health measure is a summary score of six items derived in an orthogonal factor analysis from'a symptoms checklist by Veroff et al. (1981). This physical health factor was statistically independent of the more psychologically-oriented symptoms factors (anxiety and immobilization).' The index reflects respondents' general selfreported health status and is conceptually similar to measures in common use in studies of employment and health (e.g., Coleman & Antonucci, 1983; Nathanson, 1980; Waldron et al., 1982). The items included "Do you have any particular physical or health trouble?," "Do you feel you are bothered by all sorts of pains and ailments in different parts of your body?," coded 1 = yes, 4 = no; "For the most part, do you feel healthy enough to carry out the things that you would like to do?" was coded 1 = no, 4 = yes. The remaining three items ("Have you ever been bothered by your heart beating hard?," "Have you ever been bothered by shortness of breath when you were not exercising or working hard?," and "Has any ill health affected the amount of work you do?") were coded 1 = many times, 2 = sometimes, 3 = hardly ever, 4 = never. Scores on each item were summed and could range from 6 = poor health to 24 = good health (Cronbach's alpha = .77) Marital status was coded as a dichotomous variable (0 = not married, 1 = married), as was parental status (0 = not a parent, 1 = parent). Age of respondent was coded continuously in years, while education was coded for the highest grade of school completed (less than one grade to seventeen grades and beyond). The intercorrelations for all variables in the model appear in Table 1. PROCEDURE
We used the simultaneous equations technique of two-stage least squares regression (2SLS) to test the hypothesized nonrecursive causal model (Berry, 1984; Blalock, 1971). In this model, the two
Adelmann el al. Table 1. lntercorrelatlons among variables In the model.
1 Employment
Downloaded by [University of Sydney] at 02:57 01 May 2015
2 Health
.24
3 Marltal status
-31
.09
4 Parental status
-.l I
.11
.22
5 Age
-.lo
-.2 1
-.I9
.20
.27
.09
.06
1
2
3
4
6 Education
-.I5
-.08 5
nonrecursively (reciprocally) related variables of employment and health are known as endogenous variables. The exogenous variables of marital status and parental status predict both employment and health. Age, which is directly related only to health, and education, which is directly related only to employment, are known as instrument variables. In the 2SLS program we used, in the first stage each endogenous variable was regressed on the set of exogenous variables by which it is directly caused, yielding a "purged" version of the endogenous variable; for example, health was regressed on age, parental status, and marital status. In the second stage each endogenous variable was regressed on the purged endogenous variable and the exogenous variables by which it is directly caused (e.g., employment was regressed on education, marital status, parental status, and the "purged" health variable). The regression coefficients resulting from the second-stage regressions were used as path coefficients for the model. Several assumptions must be met in using the 2SLS technique (James & Singh, 1978). Linear and additive .effects are assumed, and measures are assumed to be interval level scales with an acceptable degree of reliability. The error terms of the endogenous variables are assumed to be normally distributed with means of zero, and are uncorrelated with the exogenous variables in the model. The exogenous variables are assumed to be measured without error.
I2
WOMEN & HEALTH
Downloaded by [University of Sydney] at 02:57 01 May 2015
Finally, it is assumed that the directions of causal flow have been correctly specified. It is important to note that a 2SLS solution does not provide proof of a causal model. Hypotheses concerning causation can be rejected by a 2SLS procedure, but there are usually multiple models in addition to the one being tested that are plausible and consistent with the data. However, if predictions regarding parameters are substantiated by analysis and a sound theoretical argument is advanced, the model is corroborated and a basis for causal inference exists (James & Singh, 1978; Schmitt & Bedeian, 1982). RESULTS Description of the Sample Among this sample of midlife women, approximately 65% are married and 35% are not married; about 89% have children, while the remaining 11% do not. The average number of grades of school completed is over eleven (with a standard deviation of three grades), and the mean age of this group is nearly 52 years old (standard deviation of seven years). The average number of hours employed per week is 19 from a range of 0 to 84 hours (sad. = 19.7), and the average health score on a scale of 6 to 24 is 17, (s.d. = 3.7). In comparison, for the 21 to 39 year old women, 65% are married and 68% have children. They have completed almost 13 years of school (s.d. = 2.4), are an average of 29 years old (s.d. = 5.2), work 22 hours per week for pay (s.d. = 20.4), and have health scores of 18.8 (s.d. = 2.7). Tests of Assumptions Several of the assumptions of the 2SLS technique can be put to empirical tests. The assumption that the error or disturbance terms (Ue and Uh) are not correlated with the exogenous variables in the model was tested first by examining the zero-order correlation coefficients. The results indicate that none of the correlations between
Downloaded by [University of Sydney] at 02:57 01 May 2015
Adelmann el al.
13
the disturbance terms and age, education, marital status, and parental status are significantly different from zero in the midlife sample. Two assumptions that are important to the identification of the model and that bear on the suitability of the variables selected as instruments (Berry, 1984; Blalock, 1971; Kenny, 1979) were also tested. That each instrument has a significant association with the endogenous variable to which it is directly causally related was supported; education is significantly correlated with employment at midlife (r = .20, p
