DE LAS LEYES QUE FORMAN LA LIBERTAD POLITICA EN SUS RELACIONES CON
DE LA CONSTITUCION DE INGLATERRA (7 )
6.1 Introduction
Over the last ten years there has been a gradual shift in the literature
towards a life-course perspective that emphasizes both social causation and social selection. For instance, Bartley (1991) pointed out the importance of social and economic influences on the relationship between chronic illness and employment status; Waldron and colleagues (1982) concluded that middle-aged women display a strong 'healthy worker' effect; Moen and colleagues (1992) showed that health problems earlier in life are associated with poorer health status in old age; Makosky (1982) demonstrated that chronic stressors, major life events and mental health are intertwined, with chronic strains dominating the stress process; and a few researchers believe that social careers during early adulthood should substantially contribute to subsequent health outcomes (Haynes and Feinleib, 1980; Elder and Liker, 1982; Moen et al., 1992).
The literature has yet to apply an adequate approach to determine which social-role careers are related to health status. Indicators have been used that probably mask many significant associations. For example, Haynes and
Feinleib (1980) investigated the effect on heart disease onset of time spent employed during women's entire adult lives, ignoring most of the social causation and selection processes that occur over such a long period. As an alternative, the present study offers a life-course approach that relies on the interrelatedness of early and later life circumstances (Spilerman, 1977). Social careers are first 'anchored' in early adulthood, and the amount of time spent within combinations of employment, marital and parental
statuses in the absence of chronic disease over age-defined periods is used to build up a picture of the social paths that lead to various health outcomes. Then the influence of chronic disease on social-role participation is
examined to emphasize the dynamic nature of the process (see Model 2.5.1, p.29).
After a review of the literature and consideration of the likely effect of recent social changes, a number of general hypotheses were generated at the end of chapter two. In this chapter a set of more specific hypotheses are tested. Furthermore, owing to the innovative nature of the study's approach some exploratory work is carried out that leads to the development of hypotheses.
The analysis is split into two sections: the first part investigates the health consequences of social-role careers anchored in early adulthood, and the second focuses on social selection, highlighting the effects of chronic disease on marital status, childbearing and employment status. Both sections use regression techniques, and a nesting procedure is carried out in each
regression analysis to determine which dummy variables from a set of
k factors can be collapsed together without significantly weakening the fit of
the models. For example, it may be found that two or more dummy variables within the early-adulthood social-role set can be combined into one dummy variable without significantly changing the nature of the relationship between early-adulthood roles and the dependent measure. The membership of a dummy variable therefore may vary between regression models that have different dependent measures. Such an approach conserves degrees of freedom and increases group sizes, and consequently is advantageous because of the small sample. In addition, in cases where a group size is too small, members will be either collapsed with the most appropriate alternative group or excluded from the analysis,
depending on the results. Finally, given the size of the present study's sample some caution is needed in the interpretation of the findings,
especially since the power of a test to correctly reject the null hypothesis may be weak in some instances.
6.2 Evidence of social-role careers influencing health status
Based on the study's theoretical model, this section tests hypotheses about the influence of the social process on health outcomes. The three main health indicators examined were retrospective dates of diagnosed chronic physical disease; scores on a measure derived from the General Health Questionnaire (GHQ); and self-rated health.^ In relation to social structures, the life-course methodology outlined in chapter four was
employed: women's social-role careers were 'anchored' in early adulthood, with group membership depending on the amount of time between ages 20 and 29 years spent working in a job outside the home (including studying), or not working while married (including de facto unions), or single
1 Self-rated health (SRH) was examined both as a current-status measure and as a change in SRH determined by a regression equation, SRH1990 = B'l * SRH1986-87 + B'2 * Controls + B'3 * Explanatory variables + error.
(including never married, separated or divorced) and childless or with children. Also, among older women (those born 1926-46) an analysis was carried out to examine the effect of subsequent social-role careers within the anchoring groups on the health outcomes of the study. The sample size limits the number and types of additional paths that could be investigated. Other more conventional approaches were examined, such as the total time spent employed or married over the life course, different 'anchoring'
periods, and cross-sectional measures. These approaches, however,
produced substantially worse-fitting models, as well as conceptually weaker evidence of social causation given that social selection and causation
increase in complexity over the life course.
Women bom between 1956 and 1966 (aged between 20 and 29 years at the baseline survey) had not yet lived long enough to describe their social careers adequately or to demonstrate the effect of career participation on health, and therefore were excluded from the analysis. Women who developed a chronic disease before age 30 years were also omitted, which reduces the effect of poor health on early-adulthood role participation (social selection), and increases the probability of excluding those who have health conditions caused by factors not associated with social influences.
Cox proportional hazards models (Cox, 1972) were used to determine the relationships between social-role careers and the risk of chronic disease. The formal model based on the hazard rate at time t is: X{t;z) = exp(z(3)>^(t),
where (3 is a p x 1 vector of unknown parameters, and ^o(t) is an unknown fimction giving the hazard function for the standard set of conditions z = 0. The results from analyses using this regression technique show parameter estimates and their standard errors; and exponentiated parameter estimates (hazard ratios) which give the associated risk of developing a chronic disease relative to an unknown (arbitrary) baseline hazard, although since the
simplest Cox model assumes that the hazard function and underlying hazard are proportional, inferences can be made about the risk multiplier (exp zP) without involving the baseline hazard.^ Also, approximate 95 per
2 A range of time-varying covariates were examined to test the proportionality assumption. AH time-varying covariates were found to be small and non-significant. Also, similar models were found when splitting the period of disease onset into two (30-39 years and 40-63 years). Finally, two parametric survival models (Weibull and Exponential) were examined as
cent confidence intervals associated with the hazard ratios are given, computed using exp((3' ± 1.96'^S.E.). At the bottom of each table is the final Log-Likelihood estimate for the model, its associated x^ statistic, and the statistic related to the change in the Log-Likelihood as a result of social-role careers.3 Only the onset of the first chronic disease was considered, and a range of control variables were included in the models to reduce
confounding effects: for instance, current age (controlling for cohort
differences); birthplace; home ownership in 1986-87 (adjusting for sampling biases); occupational class,-^ personality (introversion-extroversion);
number of children before the onset of chronic disease; and health-related childbearing difficulties (further reducing the effect of social selection). In relation to the other health outcomes, ordinary least-squares regression models, adjusted for the control variables and the presence of chronic disease, were employed to investigate the relationship between
early-adulthood social-role careers and mental health (GHQ) ^ Ordered probit models (McKelvey and Zavoina, 1975), adjusted for the control variables, chronic disease and mental health, were used to examine the effects of role careers on self-rated health (see Appendix 5.E for a description of ordered probit models). Given that these two health indicators are likely to be influenced by recent life events (Makosky, 1982; Avis et al., 1991), there was an analysis of the relative importance of some major life events in the models. The questionnaires did not include a life-event check list, but instead asked respondents about key turning points in their lives: for
alternatives. Both models produced substantially poorer fitting models than the Cox. Consequently, there is strong evidence to suggest that the proportionality assumption is met. 3 To obtain the Log-LikeHhood when the slopes are restricted to zero use the formula
(X^f/-2) + the Log-Likelihoodf; or the Log-LikeUhood with only the control variables in the model can be obtained from (X^c/-2) + the Log-Likelihoodf, where f represents the value corresponding to the final model, and x^c represents the x^ value for the change in the model resulting from social-role participation.
4 In the main analysis occupational class and home ownership were chosen as indicators of socio-economic status over educational attainment and income because income data had too much missing information; and it was thought that educational attainment was less representative of women's social conditions than occupational class: this hypothesis was supported by an analysis that showed occupational class to have a stronger relationship with health outcomes than educational attainment.
5 The distribution of GHQ scores is positively skewed. Analyses employing transformed scores were compared to results using the raw scores: the findings were similar. In addition, tests for curvilinear trends were carried out, with the findings suggesting that the tiends were substantially linear. Preference, therefore, was given to the untiransformed data using an ordinary least squares technique.
example, dates of marriage, migration, childbirth, when parents and
children died, and starting and stopping work. It was not possible, therefore to obtain a life-event score. Furthermore, the sample size limited the types of life events that could be examined.
Finally, the measurement of mental health and self-evaluated health
precluded comparing cohorts when they were at similar ages, and as a result, the study's hypotheses focus on relationships within cohorts and not
between cohorts. The stratification of the analysis does reduce the power of the statistical tests because of the smaller sample sizes. Consequently, caution is needed when interpreting the results from these analyses.