Many investigations into marital status differentials in health or mortality at older ages have considered only current status or transitions over a short period, here we investigated associations between an indicator of marital history spanning twenty or thirty years and mortality and long-term illness in older age groups.
Results showed that relative to men in long-term first marriages, men in all non-married groups and those in long-term remarriages had raised mortality 1991-2001, although for the small proportions divorced before 1971 or after 1981 (and not remarried) this excess was not significant once socio-economic status was controlled. Men in long-term remarriages and those divorced or widowed since 1971 also had higher odds of long-term illness in 1991; in 2001, however, the long-term remarried were the only group with significantly raised odds of long-term illness. Women in long-term remarriages also had higher odds of reporting long-term illness in 1991 and in 2001, relative to women in long-term first marriages, and remarried women who had previously been divorced had both raised odds of long-term illness and raised mortality 1991-2001, although this latter effect ceased to be significant when parity was also considered. After control for socio-economic status, some groups of divorced women had higher mortality risks 1991-2001 and raised odds of long-term illness in 1991; mortality was also raised for all widows. Relative to women in long-term first marriages, never-married women had raised mortality risks but the association with 1991 long-term illness was not significant and in 2001 never-married women had lower odds of reporting long-term illness than women in long-term first marriages. Neither the positive association with mortality nor the negative one with long-term illness in 2001 remained significant when parity was also taken into account but a limitation of this analysis is that we had no information on non-marital births prior to 1971 and so, taking into account the low levels of non-marital childbearing in the relevant period, made the assumption that all women who were never-married in 1971 were then childless.
It is interesting that for both never-married men and women, the results show a divergence between associations with mortality risks, which were raised, and odds of long-term illness which were either not raised or, in the case of women in 2001, significantly reduced relative to those in long-term first marriages. This is consistent with the literature in that several studies have found that older never-married women report similar or better health than married women although many studies have found higher mortality for never-married compared with married groups [5, 21–23, 30]. The measure of health status used in this study - self-reported illness limiting daily activities- is a subjective one and it is possible that people's perceptions of health problems or the limitations they produce may vary with marital status. It has been suggested, for example, that some of the health benefits of marriage result from monitoring of health by a spouse and also that those with fixed obligations that cannot be reassigned may be less likely to adopt the 'sick role' . Possibly, never-married older people are less aware of health changes and may have a higher threshold for reporting illness. The subjective nature of the health measure used also makes it difficult to interpret the higher age specific prevalence rates of reported limiting long-term illness in 2001 compared with 1991 referred to earlier. Although it is possible that this indicates a real deterioration in population health, it may reflect increases in people's health expectations and consequent greater propensity to report health limitations.
In terms of what consideration of marital history, rather than just current marital status, adds to our understanding and knowledge, this is clearest for the remarried. Although those in the various categories of remarried considered accounted for only small proportions of the study samples, the remarried as a group outnumbered the never-married for both men and women in 1991 and for men in 2001 and will be larger in more recent cohorts, so understanding possible implications for health is important. We found that men in long-term remarriages contracted before 1971 had higher mortality and higher odds of reporting long-term illness than men in long-term first marriages whereas men remarried since 1971 generally had better health and lower mortality, although this only reached conventional levels of statistical significance in the analysis of 1991 long-term illness.
In the comparable analyses for women, the long-term remarried had higher odds of reporting long-term illness in 1991 and 2001, relative to women in long-term first marriages, and those remarried following divorce had significantly worse health and mortality in all analyses. No such disadvantage was evident for remarried women who had previously been widowed. Those already remarried by 1971 would have experienced marital dissolution at a relatively young age, possibly the disruption to life course trajectories resulting from this may have enduring health consequences. One such effect may be reduced opportunities for having children and in the case of women who had remarried following divorce our results showed that apparent health disadvantages tended to weaken or disappear when parity was controlled. It is also possible that relatively early marital termination, particularly through divorce, and remarriage in these cohorts is associated with risk taking and unhealthy lifestyles which unfortunately we had no information on. However, we did not find any equivalent consistent disadvantage among women divorced before 1971 who had not remarried, whose characteristics are likely to be even more unfavourable than women who remarried after divorce. This group was relatively small (517 women in 1991 and 342 in 2001) and the power of the analysis consequently weaker.
Among the much larger proportions of widows and widowers, there were no clear and consistent differences between those with the differing durations of widowhood we considered. Many studies of widowhood have suggested that excess mortality is most pronounced within the first year following bereavement, with less or no excess risk at longer durations of widowhood [48, 49]. However, Manzoli and colleagues found little evidence for such an effect in their meta analyses of studies since 1994 and suggested this was mainly a feature of older studies . We did not separately identify the very recently widowed because it was not possible to identify a similar group of very recently divorced people and small numbers would have further limited power of the analysis.
Controlling for socio-economic status, which was strongly associated with mortality and health, considerably modified associations found, particularly for women and particularly for the never-married. This indicates a greater co-variance of marriage and socio-economic status for women as compared with men (consistent with the idea that many benefits of marriage come from this association) and a greater effect of absence or loss of marriage on men compared with women (consistent with the idea that men are more dependent than women on the social support and social control elements of marriage). This can be seen as evidence for both protection and selection effects - marriage may bring socio-economic advantage and not getting married, or experiencing marriage termination, may be associated with characteristics that make attainment of socio-economic advantage less likely.
Effects of parity were also significant. We have previously reported results of analyses between women's reproductive histories and mortality (from age 50) and long-term illness in 1991, using the same data set but a different design and sample and taking account of factors such as length of birth intervals and timing of first and last births but not detailed marital history . Overall, results from these two analyses are similar except that in our earlier work we found greater disadvantages associated with nulliparity (here positively associated with mortality, but not with long-term illness in 1991 and 2001). Results of another study, again using a sample drawn from this database but restricted to ever-married women, also showed raised mortality among nulliparous women; but no adverse effects of high parity, however this may have been because the investigators grouped all those with three or more children together . The associations found probably reflect the influence of a range of factors, including selective influences (the childless and mothers of one child only may include women with health problems precluding successful first or second pregnancy and delivery); protective factors (social support from children); long-term sequelae of physiological challenges associated with high parity, and unmeasured characteristics of low and high parity women which may be associated with health and health related behaviors.
We did not explore interactions with age in any detail in this study but the fact that there were few significant associations between marital experience and long-standing illness in 2001 (when sample members were aged 70-89) but more indication of association in 1991 could be interpreted as a tendency towards convergence in differentials with older age. However, the smaller size of the 2001 sample and the fact that they represent a more selected group of survivors is also relevant, as are effects of earlier differential mortality more generally. We lacked information on mortality prior to 1971 (when those included in this study were aged 40-59), but mortality differences 1971-91 were explored in preliminary analysis and showed clear marital status differences. We investigated the influence of selection more formally by fitting a Heckman probit model  to presence of long-term illness in 2001 in which we included the same terms as in the logistic model reported in Table 4 and also a selection equation including marital status and socio-economic variables in 1971. This model takes account of the fact that those observed in 2001 represent only a selected sub set of the original 1971 population. Results showed that men who were never-married, widowed or divorced in 1971 were significantly less likely to be present in the 2001 sample; older age also reduced chances of inclusion in the 2001 analysis whereas owner occupation in 1971, access to a car and having an educational qualification was positively associated with survival to 2001. However, even allowing for this, coefficients for the 1971-91 marital history categories and other co-variates were very similar to those from models not including a selection term. Equivalent analysis for women showed that those who were divorced in 1971 (but not those then never-married or widowed) had significantly lower chances of survival to 2001 and that the effects of 1971 age and socio-economic status on inclusion in the 2001 sample were in the same direction as for men. Taking account of this selection had only slight effects on estimated covariates, although these were sufficient to make the reduced risk of 2001 long-term illness among women who had married for the first time since 1971 statistically significant in the Heckman probit model including a selection equation.
Strengths of this study include large sample size, availability of socio-economic indicators over several time points and in analyses for women, consideration of parity, inclusion of those in institutions (apart from the tiny proportion already resident in an institution before age 60) and low rates of non-response and loss to follow-up. However we were unable to precisely time marriages and divorces and lack information on childhood or early adult circumstances (other than an educational qualification indicator) which are likely to be associated both with marital trajectories and with later health. We also lacked data on health related behaviours and although we used several indicators of socio-economic status these were in some cases fairly crude. The measure of education available, for example, did not allow us to make any distinctions among the majority who only had lower level qualifications. It is therefore possible that some of the reported associations between marital history and status and health outcomes reflect residual confounding by socio-economic status. This may be particularly true for women as in the cohorts we consider the entwinement of female family trajectories with other aspects of the life course, including labour market participation and acquisition of wealth, makes disentangling their implications for health very difficult. As already discussed there are also limitations to the health information available. We had no information on health status in 1971 or 1981 and the data available for 1991 and 2001 were drawn from a single item question. A further limitation of this study, which is relevant more generally, is that despite the large size of the study sample, numbers in subgroups of interest were in some cases very small and power to detect differences between them consequently limited.
The cohorts we studied were born between 1912 and 1931. The earliest born within this range may have had marriage and fertility plans disrupted by war; the latest born are members of the 'marriage boom' generations with high rates of nuptiality, rising but still low rates of divorce and, on average, later widowhood than preceding generations. Cohorts born more recently have shown further changes in partnership behavior with later marriage, and more fluid partnership arrangements. Analyses of the implications of marital or partnership histories for later life health may be both more feasible and more relevant for these later born cohorts. Comparative studies of populations, such as those of the US and some Nordic countries, which already include higher proportions of older people with disrupted marital histories (because of the earlier adoption of more fluid and disrupted partnership arrangements), would also be useful. Such research is important because although changes in partnership patterns would suggest that the difference between married and unmarried is becoming less distinct, some large studies suggest that marital status differentials in mortality may be increasing [6, 52] and we need to discover to what extent the changing marital history composition of marital status groups may account for this.