Aging in Sub-Saharan Africa: Recommendations for Furthering Research (2006)

Barthélémy Kuate-Defo

The most influential factor historically in the aging of human populations has been fertility reductions. All populations with declining fertility become older, and the speed of aging increases as mortality declines. For the first time in history, many societies in Africa now have the opportunity to age. Accompanying this broad demographic process, however, are other changes—shifting disease patterns and emerging health threats, macroeconomic strains, emergent technologies, changing work patterns and social norms, and cultural practices within and between societies. These dramatic changes in fertility responses and unprecedented mortality reductions in Africa since the 1950s ensure that the population of this continent is bound to age quite rapidly half a century later.

The secular decline in fertility rates historically has been shown to be the most important factor of population aging, via a sustained increase in the ratio of older to younger people. The fall of mortality rates from a combination of advances in public health (e.g., immunization campaigns), medical technology, and standard of living (e.g., better nutrition) has resulted in improvements in life expectancy. Recent estimates suggest that the aggregate proportions of the elderly population in sub-Saharan Africa will grow rather modestly as a result of continued high fertility in many countries, but the size of the elderly population is expected to increase by 50 percent, from 19.3 to 28.9 million people from 2000 to 2015 (National Research Council, 2001).

These fertility and unprecedented mortality reductions, along with de-

clines in deaths from infections and parasitic diseases, have resulted in increases in life expectancy throughout the world, leading to increasing numbers and proportions of elderly people. This demo-epidemiological transition has been attributed to public health measures, advances in medical science, and health care. Irrespective of reasons, people are living longer and many of them are having more years of healthy, active, and independent life, especially in developed countries.

In contrast, there is a dearth of information and research on the health status and functional limitations of the older populations of many developing countries in general (Gorman, 2002; National Research Council, 2001; Palloni, Pinto-Aguirre, and Pelaez, 2002; Restrepo and Rozental, 1994) and notably in African countries. In this region of the world, population aging coincides with increasing social inequalities, poverty, unemployment, violence, malnutrition, and the devastating and differentiated effects of the rampant HIV/AIDS epidemic on individuals, families, communities, and nations.

The causes and consequences of aging in this region within and between countries are complex, multifactorial, and intertwined. Their study is difficult and demands an interdisciplinary approach, given the complexity of the interactions among social, economic, and environmental variables and their effect on health status and functional limitations. The projected increase in the number of older people poses new challenges to researchers, policy makers, and planners. This paper addresses the following questions: Is the population in Africa living healthier, longer lives or are added years accompanied by disabilities and generally poor health? How do changing family structures and socioeconomic conditions affect the prevalence of poor health and limited activity among the elderly?

Since current and prospective policy responses are likely to differ among countries in Africa, a number of natural experiments are needed to enable countries to learn from each other’s experiences. This study examines self-reported health and physical functional status among older people in a transitional environment—the rural and semirural societies of Cameroon—and compares their determinants in men and women. Such an investigation is important as a contrast to the general tendency to focus on urban areas of less developed countries and sub-Saharan Africa. Although differences in health between the richest and the poorest segments of the populations in many societies are clearly identifiable, differences among the rural, semirural, and urban areas of Africa may not be so obvious. In addition to variations in life expectancy, population health, and adult mortality, self-ratings of poor health and disability are likely to be lower in semirural areas than in most urban settings, and the average rate of self-reporting of good health and functional status is usually lower in rural than in urban areas. Furthermore, some differences in social influences (e.g., education, social

norms and practices) on perceived health among urban, semirural, and rural dwellers are most likely to be present.

With no study to my knowledge in Africa that has focused simultaneously on the health and functional status of rural and semirural segments of the elderly, research in this area is needed. There is a growing consensus among public health researchers and policy makers that more information is required on the mechanisms that produce gender differences in health outcomes in different social settings (Arber and Ginn, 1993; MacIntyre, Hunt, and Sweeting, 1996). This study explores the health differences between older women and men living in 75 urban and rural localities of Cameroon, with the ultimate aim of extending the information and knowledge base to prevent social exclusion and promote the health of older women and men in Africa. This paper seeks to examine the extent and nature of gender inequalities in health in later life and the extent to which these inequalities can be explained by differences in socioeconomic characteristics and the living arrangements of older women and men. The ultimate goal is to provide local health researchers, professionals, and community organizations with information needed to plan gender-sensitive interventions to promote the health and quality of life of older women and men. More specifically, I consider the following research questions:

1. What explains gender differences in health and functional status among older people?

2. By what mechanisms and to what extent do high socioeconomic status and better living arrangements lead to better health and functional status?

3. To what extent and by what mechanisms are the effects of socioeconomic status and living arrangements on health and functional status of older people dependent on gender?

There are several plausible ways in which certain aspects of gender inequality, socioeconomic status, and living arrangements may influence health and functional status at older ages. For many of these influences, however, empirical studies are lacking that can confirm the importance of particular intermediate variables. The socioeconomic status and living arrangements of older women and men and their possible connections with health and disability may be understood only in particular sociocultural contexts, given the relative position of women and men in different societies (Anker, Buvinic, and Youssef, 1983). This is because indicators of socioeconomic status and living arrangements tend to be heavily context-dependent and also because particular aspects of female versus male status may have contradictory effects on health and disability in different sociocultural contexts.

This paper is devoted to the centrality of health and functional status by focusing on gender, socioeconomic status, and living arrangement differentials in informing health, social, and economic policy formulation. The first section presents the data and methods used for analysis. The next section presents the results. The final section summarizes the main findings and discusses their implications.

The study uses data from the first round of the Cameroon Family Life and Health Survey (CFHS), conducted in 1996-1997. This is a survey of representative and randomly selected individuals age 10 and older in 75 urban and rural localities in Cameroon. Each locality uses probability samples in which all households with individuals age 10 and older have a nonzero chance of inclusion, designed to produce comparable locality-level estimates for the population under study. The CFHS employs a self-weighted proportional sampling design, with the proportions of randomly sampled households in all 75 localities forming the Bandjoun region in the sample equal to the same proportions in the general population. The sample was drawn so as to be representative in each of the following age- and sex-specific groups: adolescent boys ages 10-19, adolescent girls ages 10-19, men ages 20-49, women ages 20-49, men age 50 and above, and women age 50 and above. After a household has been selected, one individual among all respondents in that household was randomly selected and interviewed until the sample size required for a given locality was attained. A total sample of 2,381 individuals was interviewed, of whom 631 were age 50 and older. Further details regarding sampling methodology and the survey have been published elsewhere (Kuate-Defo, 1998, 2005; Kuate-Defo and Lepage, 1997). The postulated risk and protective factors used in this study are presented in Table 9-1.

The survey was carried out in the prefecture of Bandjoun, in the western part of Cameroon. This area is representative of the system of beliefs, customs, and social structure of the population of Cameroon. In an area of approximately 274 sq km, this region combines the features of a highly modernized environment with a typical traditional Cameroonian society. The urban and semiurban localities of Bandjoun have one of the country’s universities, three public hospitals, two private hospitals in operation since the early 1950s, about a dozen public health centers, several traditional healers attracting people from various social strata, several high schools and professional schools, infrastructures for communication and transportation, and entertainment sites. In the rural areas, there are over 70 tradi-

tional chiefdoms with traditional authorities and practices, an extensive practice of polygamy and other gender-related practices, agricultural production, and extensive farming. The geographical distribution of the population reflects one of the highest population densities (a population of over 120,000 inhabitants, thus a density of about 438 inhabitants per sq km) and the highest fertility levels (a total fertility rate close to 7) in the country. The entire region is accessible in all seasons of the year.

In this study, self-rated health was assessed among 631 individuals residing in 75 urban and rural localities in Cameroon, through a question asked about perceived general health: “Would you say that in general your health is: Excellent, Very Good, Good, Fair, or Poor?” From this item, the study used a dichotomous outcome measure coded 1 if poor and 0 if excellent, very good, good, or fair. Thus, health status for all persons age 50 or older is based on an overall assessment of individual health on a 5-point scale (5 = very good, 4 = quite good, 3 = average, 2 = quite poor, 1= poor).

This measure is one of the most frequently used health status measures in population-based epidemiological research and has been a powerful predictor of morbidity and mortality. It has been demonstrated in previous studies that the reliability of self-rated health has been as good as or even better than that of the multiitem health scales. A review of 27 community studies concluded that even such a simple global assessment appears to have high predictive validity for mortality, independent of other medical, behavioral, or psychosocial risk factors (Idler and Benyamini, 1997). Several subsequent studies have also demonstrated the usefulness of capturing the health status of the elderly persons and their determinants by focusing on such a simple operational measure of health. For most studies, odds ratios for subsequent mortality ranged from 1.5 to 3.0 among individuals reporting poor health compared with those reporting excellent health.

Self-reported health has been demonstrated in longitudinal studies to predict the onset of physical disability and functional or activity limitations (Farmer and Ferraro, 1997; Ferraro, Farmer, and Wybraniec, 1997; Idler and Benyamini, 1997; Idler and Kasl, 1995; Mor et al., 1989; Wilcox, Kasl, and Idler, 1996). While the majority of the elderly are capable of maintaining their autonomy, a sizeable proportion increasing at each age becomes frail and in need of support and care at home or in institutions.

I consider a dependent variable measuring the coexistence of poor health and disability. I do so because not all of those who are ill have functional limitations, and vice versa. What is important is the ability to cope with daily life in spite of chronic morbidity and the degree to which the

elderly may need assistance to continue to do this even at decreased levels of activity.

Such functional activity is widely measured by indices of activities of daily living. In this study, functional status was measured among 631 individuals age 50 and over by asking the following questions: “Do you have any activity limitation in fulfilling your activities of daily life? If yes, what is the nature of that limitation? Is the limitation in activity at home, at work, elsewhere, and during leisure/travel?” From answers to these three questions, the researchers created a dichotomous outcome measure coded 1 if any functional limitation was reported and 0 otherwise. We also created a third outcome variable measuring the simultaneous reporting of poor health and functional limitations coded 1 if an elderly person reported as being in poor health also had a limitation in activity and 0 otherwise.

The methods of data analysis in this paper include the description of variables, followed by an examination of the association between each risk or protective factor and the three outcome variables (bivariate analyses), as well as the study of the interrelationships between the different risk or protective factors in predicting the outcomes (multivariate analyses). Since the primary goal is to evaluate the effect of a postulated risk or protective factor on each outcome, I investigate what this effect is before and after controlling for other factors so as to determine whether such an effect is direct or mediated through other postulated risk or protective factors. A mediating factor is a link in the causal chain leading from the postulated risk factor to the outcome and is partly determined by that risk factor.

To illustrate the general strategy, consider three postulated risk factors (V1, V2, and V3) of poor health or disability (or both) among older people. The overall effect of each risk factor (or group of risk factors) V1 is evaluated first (Model A). In the second step, another putative factor or a set of factors V2 is added (Model B) and its effect assessed in the presence of V1, which would then constitute a proper confounding factor. The unconfounded effect of V2 would then be obtained from this equation. The magnitude of the remaining effect of V1 in Model B would reflect only the part that is not mediated through V2. Model B is then extended to add another postulated risk factor V3 in Model C and its effect assessed in the presence of both confounding variables, V1 and V2. Any residual effect of V1 would be the part that is not mediated through either V2 or V3. Similarly, any residual effect of V2 would not be mediated through V3. In interpreting the results, it is worth noting that some if not most of the effect of V1 will be captured by the other two factors (V2 and V3). It would be incorrect to interpret that V1 has no effect after adjustment for confound-

ing variables, since in Model C the overall effect of V1 will be underestimated due to the presence of mediating factors. The strategy may be extended to situations with several variables in each hierarchical level of model building.

I use general health status and physical functional status and not more refined cause-specific or level of severity measures because differences in reporting of such fine-tuned measures can obscure underlying events by misclassifying outcomes. Empirically, we measure socioeconomic status using two indicators given their pertinence in the study context: level of educational attainment and economic activity status. Living arrangements are empirically assessed using three indicators: marital status by type of union, timing of first marriage in the life cycle, and family size support network. The effects of gender, socioeconomic status, and living arrangements on self-reported health and physical functional limitations are estimated with logistic regression models. The general logistic regression model used can be described as follows:

where Y_i is the value of individual i on the outcome Y. Since all three outcomes considered in this study are dichotomous (i.e., poor health, functional limitations, poor health with functional limitations), Y_i equals the logit (or log-odds), α the overall constant or intercept, X_i the value of the gender of individual i and β the vector of gender effects, Z_i the vector of socioeconomic status indicators and υ the vector of associated parameters, W_i the vector of living arrangements factors and λ its vector of parameters, and C_i the vector of control variables and η the vector of their corresponding parameters. These models estimate the odds ratios of poor health (versus excellent/very good/good/fair health), functional limitations, and poor health with functional limitations, according to gender and various indicators of socioeconomic status and living arrangements. I use this common statistical technique for studying dichotomous outcomes, which assumes that, for all individuals in the sample, these outcomes are independent. This procedure is appropriate because in the 1996-1997 CFHS survey, only one person per household was included in the sample, so that outcomes for family or household members are not correlated, and there is no room for the health or functional status of one unit in the sample to be dependent on attributes shared by other members, such as sanitary conditions in the household. Thus, the single-level regression models fitted here correctly ignore any hierarchical data structure and produce correct standard errors, so that the effects of factors associated with poor health and disability in the elderly cannot appear significant when in fact they are not.

Thus, one attractive feature of the CFHS-1996-1997 is that the survey data are self-weighted.

The sample consisted of 631 individuals age 50 years and over (44 percent men, 56 percent women) and 319 individuals age 65 and over (45.8 percent men, 54.2 percent women). Perceived health was missed in just 16 (2.5 percent) individuals age 50 or older and 6 (1.9 percent) individuals age 65 or older, and functional limitations were missed in 14 (2.2 percent) individuals age 50 or older and 5 (1.6 percent) individuals age 65 or older. The rest of the analyses focuses on individuals age 50 years and over, unless otherwise stated. Gender-specific sample characteristics are shown in Table 9-1.

Overall, 22.7 percent of individuals reported their health as being poor, 49.6 percent reported limitations in activity, and 17.5 percent reported having both poor health and functional limitations. Of the 304 individuals who reported limitations, 164 (26 percent) reported being limited in activity at home, 231 (36.6 percent) reported being limited in activity at work, 46 (7.3 percent) reported being limited in activity outside the home, and 194 (30.7 percent) reported having functional limitations for leisure or travel activities.

The two socioeconomic status variables considered in the analyses show important gender disparities. Overall, the level of education of this semirural population is quite low, and 73.7 percent of the sample is illiterate. These data indicate that the numbers and proportions of people who are unemployed or retired are highest among the elderly. The female disadvantage among the elderly is sizeable, since fully 93 percent of older women have no education compared with almost half of older men. Similarly, unemployment is substantially higher for women than for men, and fully 60 percent of men age 50 and over are still working outside the home. But one must keep in mind that most women who reported being “at home” are actually farmers, notably in typical rural and periurban areas of Africa and western Cameroon. Hence, they are probably responsible for providing food and material resources to their family through the output of their agricultural labor.

The three variables capturing living arrangements in the elderly also depict important differences by gender. About one-third of the sample is married monogamous, married polygamous, or single/widowed/divorced, respectively. When analyses are separated by gender, the proportion of older women who are widowed (the main category in the single/widowed/di-

vorced group) is almost five times higher than the proportion of their male counterparts. In contrast, the proportion of older women who are married monogamous (15.7 percent) is only about one-third the proportion of older men in monogamous marriages (48.1 percent). The pattern of age at marriage behaves as expected, with older women being married quite young compared with older men. Irrespective of gender, there is stability in the sample proportions for kinship size (roughly 42-44 percent of the sample having at least six persons in their kinship network).

The age distribution shows that almost half of the sample analyzed is between ages 50 and 65, and only 19 percent of respondents are age 75 and over; the oldest person in the sample was 96 years old. The sample proportions by gender show no differences worthy of notice.

The distribution of respondents by region of residence indicates that about one-third of the sample lives in the urban or semiurban areas of Djaa, Pete, and Yom. The urban regions are considered as the reference category in the analyses, and there are slightly more older men (39.4 percent) than women (28.9 percent) in urban centers.

Table 9-2 displays the significance of the differences in the percentages of respondents reporting poor health, functional limitations, and both poor health and functional limitations. Notwithstanding a few nonsignificant differences in kinship size and region of residence, all other differences in postulated risk and protective factors of poor health and activity limitations are statistically significant.

In particular, these bivariate analyses show that significantly higher proportions of female respondents report being in poor health, having functional limitations, or both than male respondents. Higher percentages of older respondents with no education report poor health, functional limitations, or both, than their educated counterparts. These differences are statistically significant for all outcomes, except that the results for older women with functional limitations do not vary with educational attainment. In general, older persons who are retired or who stay at home and to some extent those who are unemployed report significantly more poor health, activity limitations, or both than their counterparts who are in the labor force.

Single, widowed, or divorced elderly people tend to report significantly more poor health and functional limitations than those in monogamous marriages, and women tend to be at a disadvantage compared with men. Significantly higher proportions of older persons who are in polygamous marriages report poor health (20.4 percent), functional limitations (51.7 percent), or both (16.6 percent) than older people married in monogamous unions (13.6, 38.6, and 10.3 percent, respectively). A similar pattern is de-

tected along gender lines, although the general tendency is that older women in all marital states report more poor health and functional limitations than older men. Kinship size of six or more is an advantage for older men, who tend to report lower prevalence of poor health, activity limitations, or both; in contrast, no significant differences in health or functional status by kinship size are found among older women.

In general, the age-specific percentages of older people reporting poor health, limitations in activity, or both consistently show significant dose-response patterns: the older the respondent, the higher the proportions reporting these conditions for the total sample as well as for the male and female samples.

In this section, I regress individual health and functional status on gender, socioeconomic status, living arrangements, and biodemographic putative factors.

Table 9-3a presents the estimated odds ratios for the effects of postulated covariates on the probability of reporting poor health, functional limitations, or both. Poor perceived health is associated with low socioeconomic status (i.e., no/lower educational attainment, no labor force participation), being single/widowed/divorced or married in polygamous unions, and being older. The strongest risk factors for poor health are living as single, widowed, or divorced (e.g., odds ratio of 3.06 in Model 5 and 1.89 at the 5 percent significance level in the full model—Model 12), whereas the protective factors for poor health are younger ages (e.g., odds ratios of 0.50 in Model 6 and 0.58 in the Model 12) and working status (e.g., odds ratios of 0.49 in Model 4 and 0.57 in Model 12).

The data show the existence of significant gender differences in self-reported health: Model 1 predicts that older women are more than twice as likely to report being in poor health as older men (p < 0.01). The male health advantage persists even after controlling for level of educational attainment, although its magnitude and explanatory power is somewhat attenuated (odds ratio of 0.58, p < 0.05). Moreover, when economic activity status is included in the regression equation, the significance of the gender differences in self-reported health is eliminated (Model 7), whereas the significance of the negative association between work and poor health (odds ratios varying between 0.49 and 0.57 at the 5 percent level of significance in Models 4, 7, 10, 11, 12) remains until a control for functional limitations is introduced in the fitted model (Model 13). Put together, these results sub-

stantiate that gender differences in self-reported health among the elderly in Cameroon are entirely explained by their socioeconomic status.

The data show that there appears to be a dose-response gradient in the odds ratios for poor health across levels of socioeconomic indicators. In Model 2, the data show that older people with no education are more than twice as likely to report being in poor health as their educated counterparts (p < 0.01). Model 3 shows that gender inequality entirely explains differences in self-reported health and education, and Model 4 indicates that economic activity status fails to mediate the effects of education on self-reported health. As regards labor force activity, older persons who are work-

ing are significantly less likely to report being in poor health. This result is robust to all controls for socioeconomic status, living arrangements, age, and region of residence. However, when the activity limitations are included in the model, no statistically significant difference is found between workers and nonworkers. This finding may suggest that the observed significant advantage of workers compared with nonworkers may be a reflection of the healthy worker effect. Healthy elderly people may be more able to work than their unhealthy counterparts and may be more attractive in the job market, where they can travel and use their skills to compete successfully within the limited opportunity structures offered by the local and national

economies. In instances in which labor force participation is endogenously determined, it is obvious that there will be associations between poor health and unemployment, irrespective of the labor market environment and individual education and acquired professional skills.

The data demonstrate that older widowed, single, or divorced respondents are at least three times as likely to report being in poor health as those in monogamous marriages, while those in polygamous marriages are more

than 1.5 times as likely to assess their health as poor compared with those in monogamous marriages (model 5). The health disadvantage of elderly widows is robust to all controls, implying that this group is one of the most vulnerable segments of the elderly population. A comparison of Models 5 and 8 in Table 9-3a shows that gender inequality explains the effect of polygamous marriage on self-rated health. It is possible, as discussed above,

that women in polygamous marriages are more likely to have low self-esteem and to report self-assessments of nonphysical health as poor.

The age at first marriage appears significant only in the model including functional status as a covariate. The result indicates that the younger the age at marriage, the higher the odds of self-assessing one’s health as poor once functional limitations are taken into account. In contrast, differences by kinship size are trivial.

While there is an apparent and positive dose-response effect of age on poor health, the significance of that effect is eliminated with a control for activity limitation (Model 13), which is itself significantly associated with poor health (p < 0.01). Regional differences in self-reported health are also found, even after all postulated risk and protective factors are controlled (Models 6, 9, and 12), but they are eliminated when functional status is taken into account. Therefore, a number of influences on self-reported health are mediated by the functional status of the respondents, including labor force participation, age cohort, and region of residence.

A useful way of looking at the interactions of gender with socioeconomic status and living arrangements in predicting health and functional status is to carry out separate analyses by gender for those postulated factors. Table 9-3b presents the gender-specific estimated odds ratios for the effects of the postulated covariates on the probability of reporting poor health, functional limitation, or both.

There are no gender differences in the effects of education, and working status has protective effects on older men but not women (Models 1 and 4). This gender difference in the effect of employment is explained entirely by kinship size (Model 6). It is possible that within the social setting typical of most rural and semirural areas like western Cameroon, the influential power of older men and the control that they have over family resources within the kinship allows them to seek assistance and use labor from the family network to ease their health burden and its consequences. In fact, the data substantiate that larger kinship size has protective effects on the health of older men only and that these effects are indeed robust (Models 2 and 5-8).

The analysis also reveals that the health disadvantage associated with widowhood noted earlier is restricted to older women, even after all measured factors are controlled, including functional limitations (Models 2 and 5-8). Similarly, the younger the age at which older women entered into marriage, the higher their odds of reporting being in poor health, even after accounting for activity limitations (Model 8). Put together, these findings demonstrate that older women who married young or who are widowed are the most vulnerable to poor health status.

The age effects tend to be stronger for older women than older men, with younger generations reporting less poor health than older generations. As conjectured for both men and women, functional status entirely explains

these age effects on self-rated health. Regional differences are also trivial in most instances, except for women (see Model 3).

Overall, there is a strong association between poor health and low socioeconomic status (measured by no education and nonparticipation in the labor force), unfavorable living arrangements (captured by being single/ widowed/divorced or married in polygamous unions), and being older and female. The data confirm the existence of gender differences in self-reported health, functional limitations, and the combination of the two. These gender differences in self-reported health among the elderly in Cameroon are entirely explained by the superior socioeconomic position of men versus women, and that gender inequality explains differences in self-rated health by level of educational attainment. Older people who are working are significantly better off than those who are not working, controlling for socioeconomic status, living arrangements, age, and region of residence. But this work status difference disappears in the presence of a control for activity limitations. Age at first marriage is also found to predict health status: the younger the age at marriage, the higher the odds of poor health. Regional differences in self-rated health exist, but they appear to be determined by activity limitations. Kinship size appears to predict the health status of older men, which may reflect the power differentials in access to family resources (human and material) in their favor. Similarly, the health disadvantage associated with widowhood is limited to older women who also tend to report poor health associated with early marriage. Since no interaction terms are present in the models, the estimated odds ratios represent adjusted estimates that control for all other measured covariates. I refer to these estimates as the “gold standard estimates” of effects, because I consider them to be the best estimates that can be obtained that control for all the potential confounders in the models.

Table 9-4a presents the odds ratios of covariates of functional limitations. As in the case with self-rated health, gender differences exist in functional limitations, with older women at least twice as likely to have activity limitations as older men. Unlike the case with self-rated health, the effects of gender remain robust to all controls, so that the most complete model (Model 12) shows an odds ratio of 0.44 for older men (p < 0.01). This finding implies that there are inherent gender differences in functional limitations, which may be explained by the cumulative consequences of reproductive activities and other harsh farming-related activities that women engage in for family survival.

The effects of socioeconomic status are also robust. Older persons with no education are about twice as likely to report activity limitations as their

educated peers. Similarly, older people who are working are less likely to report having functional limitations; in contrast, those “at home” or retired are more likely to have functional limitations. As the gender-specific models show, the disadvantage associated with being “at home” is mainly a female disadvantage (see Table 9-4b, Models 1, 4, 6, and 7).

Living arrangements are also important predictors of activity limitations. Again, the groups at a disadvantage in terms of marital status are widowed/single/divorced older people and those living in polygamous mar-

riages (Models 3 and 6). However, as Model 8 shows, gender inequality in many domains of family and social organization and practices tends to be unfavorable to widowed/single/divorced respondents (especially women), and polygamy is less likely to promote women’s status. As in Tables 9-3a and 9-3b, one finds that gender explains at least in part the effect of polygamy on the odds of reporting activity limitations (see Models 3 and 8). Any residual effect of polygamy on the probability of reporting functional limitations is captured by socioeconomic status (see Models 8 and 11).

As in the case of self-rated health, age at first marriage is a significant predictor of functional limitations at older ages. As expected, the inverse relationships between age cohort and functional limitations are consistently strong in all models (p < 0.01) and show a direct connection between growing older and developing functional limitations. Finally, the effects of region of residence emerge strongly, with respondents from all rural regions reporting fewer functional limitations than urban and semiurban dwellers from Pete, Djaa, and Yom.

Table 9-4b shows the gender-specific odds ratios of covariates of functional limitations, so their interactions with gender can be assessed. The robustness of the protective effects of education on functional status played

out to the advantage of older men only (Models 1, 4, 6, and 7). Working status is advantageous for older men, while being “at home” or retired is a risk factor of activity limitations among older women (Models 1, 4, 6, and 7). Again, these findings are robust across a range of specifications of models.

As before, age at first marriage is negatively associated with activity limitations for older women, but not for older men. Age effects are also present across gender categories, which shows that the link between disability and age is independent of gender. Regional differences in functional limitations are statistically noticeable only among elderly women. Overall, there are significant interactions among gender, socioeconomic status, and

living arrangements in predicting functional status, just as in the case of self-reported health.

In sum, gender differences in activity limitations are apparent, with older women being at least twice as likely to report functional limitations as older men. Here, these differences remain robust to controls to all postulated risk and protective factors. The robustness of this finding suggests that the association between gender and functional limitations does not reflect a methodological artifact but rather suggests a depletion process and the trajectory of health underlying the relationship for women but not for

men. The effects of socioeconomic status are also quite robust, showing that low socioeconomic status is associated with functional limitations, just like substandard living arrangements in the study context (being widowed/ single/divorced elderly or living in polygamous marriages), especially for women. As for self-assessed health, age at first marriage is a robust predictor of functional limitations at older ages. The strong protective effects of education on functional status are advantageous to older men only, and similarly for working status. Here again, young age at first marriage is associated with activity limitations among older women.

Tables 9-5a and 9-5b show the odds ratios of influences on the combination of self-rated health and functional limitations.

Gender differences in poor health with activity limitations are expected, since this outcome results from the previous ones, which have shown that, at least in the unadjusted models, gender effects do exist. As in the case of self-reported health, these gender differences are entirely explained by socioeconomic status (Model 7, Table 9-5a) and not by living arrangement

covariates (Model 8, Table 9-5a) or age cohort and the socioeconomic contextual variable, which is the region of residence (Model 9, Table 9-5a).

The influences of socioeconomic status on poor health with functional limitations are present (Model 2, Table 9-5a) as before and show the protective effects of education and work on health and functional status. But those effects are somewhat attenuated after controlling for age and region of residence (p < 0.10, Model 5, Table 9-5a; Model 4, Table 9-5b for men only) and are eliminated in the most complete model (Model 12,

Table 9-5a). This implies that some effects of socioeconomic status operate at least in part through demographic attributes and the socioeconomic context of the communities in which older people live.

The health disadvantage of widowhood remains strong, and to some extent older people in polygamous marriages tend to report being in poor health with functional limitations. As in Tables 9-3a and 9-3b, gender explains the effects of being in polygamous marriages on the odds of reporting poor health with activity limitations (see Models 3 and 8, Table 9-5a). Age at first marriage is also significantly associated with poor health with functional limitations. The age cohort effects also remain strong and unaltered

by various controls, but the regional disparities noted in previous models are somewhat attenuated.

As in previous models, the younger the age at first marriage, the higher the odds of reporting poor health with functional limitations (Models 8 and 11, Table 9-5a) but the significance of such a relationship is largely restricted to older women (Models 2 and 5-7, Table 9-5b). Since early marriage is associated with high fertility, which in turn is strongly correlated with short birth intervals in highly fertile populations, such as those of western Cameroon, it is likely that the effects of age at marriage on poor health and functional limitations at older ages operate through the parity

effect and maternal depletion syndrome, which has been well documented in the literature on women’s health in Cameroon and elsewhere (Kuate-Defo, 1997).

Kinship size appears to have a protective effect on older men but not on older women. The larger the kinship size, the lower the odds of reporting being in poor health with activity limitations among older men: those with kinship size of six or more are at least three times less likely to report these conditions than those with kinship size under six (Models 2 and 5-7, Table 9-5b).

The odds ratios of poor health with activity limitations increase with age, even after all other postulated risk and protective factors are taken into account. Irrespective of gender, there is strong evidence that the younger the generation, the lower the odds of being in poor health with functional limitations, even after all controls are introduced in models for men and women (p < 0.05, Model 7, Table 9-5b). Therefore, functional limitations explain the age-health relationship found above, but age appears to have a direct relationship with the combination of poor health and functional limitations for the elderly in western Cameroon.

Finally, when they are present, regional differences in poor health with functional limitations appear to show a disadvantage for urban and semiurban residents (Models 4-6 and 9, Table 9-5a), but after all controls are introduced, only residents of Tsela, Famla II, and Bagang Fodji have some significant advantage vis-à-vis those from Pete, Djaa, and Yom. When the analyses are separated by gender, only older women from Tsela, Famla II, and Bagang Fodji are at an advantage (Models 3 and 5, Table 9-5b), which is purged by controls for socioeconomic status. Hence, it is likely that regional differences in poor health with activity limitations are largely explained by differences in the socioeconomic context of each region.

There is evidence that gender differences in poor health with activity limitations are present but are fully captured by differences in the socioeconomic standing of older men versus women. The influences of socioeconomic status on poor health and functional limitations are mediated at least partly through age and the socioeconomic context of communities of residence. The disadvantage associated with widowhood is unaltered, and gender inequality explains the deleterious effects of living in polygamous unions on self-assessed health and functional status. Age at first marriage is strongly associated with poor health with functional limitations among older women, and the age effects remain unaffected by controls for other covariates. Older men who belong to a larger kinship are less likely to report poor health with functional limitations. Poor health with functional limitations increases with age, even after controlling for all postulated risk and protective factors, and this finding applies to both older men and women. Finally, only older women who are residents of Tsela, Famla II,

and Bagang Fodji have some significant advantage vis-à-vis those from Pete, Djaa, and Yom, and socioeconomic standing explains their relative advantage. The data also show that residing in the Tsela, Famla II, and Bagang Fodji neighborhoods is associated with decreased odds of poor health of 45 to 55 percent, compared with living in the highest development localities of Djaa, Pete, and Yom. Similar findings emerge concerning functional limitations. Together, these results suggest that living in urban areas may not be sufficient to provide better health status in contexts similar to the study sites investigated.

This study has demonstrated that it is possible to lengthen life expectancy while maintaining the quality of life in Africa. The results extend previous findings on the health advantages stemming from socioeconomic status and living arrangements to semirural areas of Africa. These factors appear to exert independent effects on self-rated health and functional limitations in most instances. Overall, there are significant interactions among gender, socioeconomic status, and living arrangements in predicting poor health, functional limitations, or both. A number of interactions were tested in models for the total sample as well as gender-specific models, but none reached significance levels. However, after adjusting for all other measured covariates, the models fail to show significant interactions of several measures of socioeconomic status and living arrangements in predicting poor health and disability, but that does not mean that such links are nonexistent. For example, it is likely that decisions about work, retirement, and unemployment are not independent of considerations of living arrangements for the elderly, just as it is the case for younger age groups.

The effects of socioeconomic status on perceived health were quite similar among men and women. Although a robust relationship between education and health status has been demonstrated in previous research (for a review, see Robert, 1999), the processes that explain the link are not well understood, especially in developing countries (Zimmer, Liu, Hermalin, and Chuang, 1998). This study advances such understanding in two notable ways. First, the study found differences in the relationship of education and self-rated health and education and functional limitations among older people. Second, it found that gender inequality entirely explains the education differences in self-rated health.

The importance of socioeconomic differences in health status is well documented in the health literature, and this study provides empirical evidence on the robustness of these differences in the African context. Related to diversity is the fact that the health of elderly populations in all countries varies according to socioeconomic position (National Research Council,

2001). The magnitude of these differences, as well as their causes, varies over time within and among societies.

To develop policies effectively, one must have an understanding of these causes, which in turn requires a fuller understanding of the determinants of socioeconomic differences in health and functioning. Policy responses to such differences will ideally cover a wide range of determinants, including the provision of, access to, and response to medical care and social services. Therefore, an underlying need in any research agenda is to include a link to socioeconomic position in the collection of population and health data on elderly individuals. Economic activity status captures the extent to which an older person is unemployed, working, retired, or housewife. In the multivariate analyses, work was measured broadly to include both paid and unpaid work.

As life expectancy increases, the postretirement life period is expected to get longer; greater health and other needs demand appropriate income for the elderly, who may increasingly be forced to compete in the labor market against younger, more skilled, and highly educated people in order to secure an income. This study confirms that the aging population that is unemployed or retired is numerically and proportionally more important as it grows older, as previous studies have found (e.g., Møller and Devey, 2003). Statistically significant differences in self-rated health by level of education are entirely explained by gender inequality, which is likely to operate in turn through the income or wages advantage of men compared with women.

Evidence of gender differences in self-rated health has been inconsistent, with a male advantage reported in some studies (Gijsbers van Wijk, van Vliet, Kolk, and Everaerd, 1991; Rahman, Strauss, Gertler, Ashley, and Fox, 1994; Zimmer, Natividad, Lin, and Chayovan, 2000), but no advantage in others (Jylhä, Guralnik, Ferrucci, Jokela, and Heikkinen, 1998; Leinonen, Heikkinen, and Jyhhä, 1998; McDonough and Walters, 2001; Zimmer et al., 2000). The evidence from developing measures to test these conjectures remains very limited. Even if measures of self-rated health were reliable and comparable across populations, empirical evidence is likely to be greatly influenced by the cultural and social norms and practices as well as power relations in specific socioeconomic environments.

This study has found that older men are indeed at an advantage compared with older women (odds ratio of 0.49, p < 0.05), but that these gender differences in self-rated health in favor of men in Cameroon are entirely explained by the health advantage conferred by their labor force and economic activity status (see Models 1, 3, and 7, Table 9-3a). Similarly, older women tend to report being in poor health with functional limitations more than men, but the female disadvantage again is entirely explained by differences in their socioeconomic status relative to older men. It has been sug-

gested that respondents draw on a number of sources when they make their self-assessment of health, including family history, severity of current illness, possible symptoms of diseases not yet diagnosed, trajectory of health status over time, as well as the availability of external resources (e.g., social support) and internal resources (e.g., perceived control) (Idler and Benjamini, 1997). The findings indicate that irrespective of such sources, gender inequality in health status necessarily operates through a rise in socioeconomic status in Cameroon.

One of the most important findings of this study is that the younger the age at first marriage of an older woman, the higher her odds of reporting being in poor health (Models 2, 5, 6, and 7, Table 9-3b) and having functional limitations (Models 2, 5, 6, and 7, Table 9-4b) as well as the combination of poor health with functional limitations (Models 2, 5, 6, and 7, Table 9-5b), even after all postulated covariates are included in the models. In contrast, age at marriage has a trivial effect on older men’s health and functional status. Since early marriage is associated with high fertility, which in turn is strongly correlated with short birth intervals, it is likely that the effects of age at marriage on poor health and functional limitations at older ages operate through the maternal depletion syndrome, which has been well documented in the literature on women’s health in Cameroon (Kuate-Defo, 1997).

This finding is also consistent with previous studies, which have demonstrated a relationship between early life conditions and later health and survival (Alter, Oris, Neven, and Broström, 2002; Conde-Agudelo and Belizan, 2000; King, 2003; Lundberg, 1993; Mosley and Gray, 1993). Because early marriage is strongly associated with high fertility and reproductive health problems, these robust findings suggest that women’s issues in the areas of child marriage and childbearing are of paramount importance. This concerns the promotion of reproductive health in young girls and their successful transition to adulthood, as well as throughout the life cycle and especially in considering social and health policies for the elderly population.

Indeed, Bledsoe (2002) shows that, in rural Gambia, women view aging as contingent on the cumulative physical, social, and spiritual hardships of personal history, especially obstetric trauma. It is likely that such ill health and disability during old age among women are a result of exacerbated risks across the life course as they assume their reproductive and productive roles. The fact that early marriage is a robust risk factor for poor health and functional limitations among older women suggests that delaying marriage will have large payoffs not only in the short term, to ensure a successful transition to adulthood for girls, but also in the long term, in protecting the health of women and enhancing their quality of life at older ages. Therefore, there is an urgent need for the international community to address

more vigorously the problem of early marriage in developing countries and to go beyond rhetoric, following repeated calls for action made by such organizations as UNICEF (2004) to prevent children from bearing children.

The finding that kinship size has a protective effect on the self-rated health of older people (especially men) is also consistent with the finding from a recent study showing that the risk of poor health and mortality is decreased by membership in large patriarchal kin groups (Hammel and Gullickson, 2004). I consider kinship size as a nurturing and enhancement factor in residential patterns that promote health and prevent disability, at least when the older person has some control over the social relations within the kinship. In most African societies the older the men, the stronger their control over familial matters and living arrangements and therefore the bigger their advantage with a larger kinship. The impact of social relations within the kinship is likely to interact with age in predicting health status. Specifically, the type and nature of social relations within the family connections as well as their usefulness will also vary across the lifespan. In contrast to the situation for men, social relations for women are not necessarily positive and may not always contribute to improved coping with illness and disability, especially for those who are caregivers and give social support to family members, in the addition to their traditional roles of reproduction and production in most societies in Africa.

Many researchers have examined the relationship between age and self-reported health, but the evidence is inconsistent (Helweg-Larsen, Kjoller, and Thoning, 2003). About one-third of studies show that older people evaluate their health more positively, another one-third show that the elderly evaluate their health more negatively, and one-third show no relationship between self-reported health and age. Idler (1993) found in a sample of elderly age 65 and older in the United States that older participants rated their health as better than younger participants at any given level of health status. Idler attributed this result to both a cohort effect (i.e., older cohorts may have different perceptions about what constitutes good health), an age effect (i.e., people evaluate their health differently as they age), and a survival effect (i.e., individuals who evaluate their health positively are more likely to survive).

This study shows that younger elderly (under age 65) are significantly less likely to report poor health, before controlling for functional limitations (odds ratio of 0.58, p < 0.05). However, when functional limitations are taken into account in the model fitting, the age differences are trivial. Indeed, functional ability is typically highly correlated with self-reported health and declines with age, as expected from biological theory. It has been suggested that if functional ability is not controlled, then self-reported health may appear to decline with age, when in fact this decline can be entirely

accounted for by functional ability (Bjorner et al., 1996). The analysis confirms this conjecture (see Model 13, Table 9-3a, and Model 8, Table 9-3b).

The findings of this study allow one to identify opportunities and priorities for further research and appropriate interventions. It has been argued that a sounder theoretical basis for socioeconomic classification would yield better understanding of the determinants involved. One approach to this end is to conceive of three different modes of social stratification: one based on degree of material deprivation, one based on social power relations, and one based on general social standing (National Research Council, 2001). Measures of social deprivation are appropriate for assessing health differences among those living in absolute poverty. Such measures are less appropriate when health follows a social gradient. In such cases, there are clear social inequalities in health among people who are not materially deprived. Other concepts must therefore come into play.

A second approach that does not potentially relate to the whole social gradient is based on power relations in the workplace. Occupations are defined in terms of power and autonomy, a perspective that has its origin in the Marxist concept of class. Such a measure is appropriate for social classification among people of working age and less appropriate for those beyond working age, especially in settings in which aging starts early and the retirement age is 55, as in most sectors of employment in Cameroon until recently. The degree to which occupation continues to provide a reliable indicator of socioeconomic position beyond working age will vary, and additional measures of socioeconomic classification will be needed. This is especially the case for older women, particularly those who are single, widowed, or divorced, which is why I explicitly consider marital status as a predictor in the analyses.

A third approach—general social standing—has features in common with the concept of status based on patterns of consumption and lifestyle. The status group shares the same level of prestige or esteem and, in addition to common forms of consumption and lifestyle, limits its interactions with members of other groups. This approach fits the typical way of life in semirural Cameroon, where the study’s data come from, given the social organization of the society and interactions by membership groups in terms of adult roles and responsibilities, sociotraditional ranking in the social hierarchy, and so forth.

Among the most important policy concerns relevant to health and longevity in modern economies are the future fiscal viability of pension, health, and social insurance systems, if any, both public and private, and the implications of these systems for savings and investment rates. How long people continue working, paying taxes, and saving will feature prominently in the consequences of population aging. Many people already work less than half a lifetime because of extended periods of schooling and training in early

life, earlier retirement, and enhanced longevity, posing a challenge to the sustainability of systems designed to support older people. In all age groups, but particularly among older people, there is a substantial amount of selfcare, as well as varying levels of alternative and complementary health care practices, including self-medication with herbs and the use of alternative practitioners, that may have an impact on health outcomes. These issues also deserve consideration in the context of health research among the elderly, but they were beyond the scope of this study.

This study suggests that a fuller understanding of the appropriate determinants of socioeconomic differences in health and functioning generally requires longitudinal, representative population surveys. Such surveys are essential for establishing causal associations and assessing the magnitude of causes operating in all directions. In other words, longitudinal data are important for determining the degree to which levels of health and functioning determine social and economic position, as well as for assessing the magnitude and nature of these determinants of health and functional status. Both personal behaviors and many public health measures bear on health status. Health promotional activities aimed at older persons may or may not involve direct contact with the formal health care system; examples of such activities include education programs and provision of good preventive nutrition, safe transportation to enhance mobility, and adequate housing. Effective national and regional policies for health promotion among older people require that important deficits in these areas be identified. Community-based population and health surveys may be the only means of acquiring accurate information on such issues as cigarette and alcohol consumption, perceived health status, levels of mobility, and social interaction. Coordination of public and clinic policies relevant to health promotion and disease prevention among the elderly is essential if these policies are to have the desired positive effects on the health status of older people. Again, the most effective means of obtaining the information necessary for such cross-national research is representative household surveys of older people, like the CFHS panel surveys, which so far have been fielded in 140 localities in western and northwestern Cameroon. Because of the higher rates of morbidity and disability that occur with increasing age, older people make substantial use of formal health services. Such services consume an enormous amount of resources. Again, cross-national comparative research is one important avenue for addressing this issue by examining international variations in organization, financing, delivery, and evaluation of elder health services.

Given the current state of knowledge in African countries, one cannot prevent the majority of the diseases and impairments of old age, but to make a start it is necessary to study the epidemiology of these conditions and measure their risk and protective factors. A further extension of this

study will involve the epidemiology of the specific diseases with increasing prevalence in the elderly populations of Africa, including but not limiting to hypertension. In addition, it would be necessary to study the aggravating factors that change disease to impairments and impairments to handicaps in the elderly.

This work was supported in part by the Rockefeller Foundation’s Intervention Research grant RF 97045 no. 90 to the author; supplemental support was provided by the National Research Council of the National Academy of Sciences of the United States and the PRONUSTIC Research Laboratory at the University of Montreal. I thank two anonymous referees for their suggestions and comments.

Alter, G., Oris, M., Neven, M., and Broström, G. (2002). Maternal depletion and mortality after the childbearing years in the nineteenth century East Belgium. (Early Life Conditions, Social Mobility, and Longevity Working Paper Series 7). Presented at the Social Science History Association, October 24-27, St. Louis, MO.

Anker, R., Buvinic, M., and Youssef, N.H. (Eds.). (1983). Women’s roles and population trends in the third world. London, England: Croom Helm.

Arber, S., and Ginn, J. (1993). Gender and inequalities in health in later life. Social Science and Medicine, 36, 33-46.

Bledsoe, C. (2002). Contingent lives: Fertility, time and aging in West Africa. (with contributions by Fatoumatta Banja). Chicago, IL: University of Chicago Press.

Bjorner, J.B., Kristensen, T.S., Orth-Gomer, K., Tibblin, G., Sullivan, M., and Westerholm, P. (1996). Self-rated health: A useful concept in research, prevention, and clinical medicine. Stockholm, Sweden: Forskningsrädsnämnden.

Conde-Agudelo, A., and Belizan, J. (2000). Maternal morbidity and mortality associated with interpregnancy interval: Cross sectional study. British Medical Journal, 321, 1255-1259.

Farmer, M.M., and Ferraro, K.F. (1997). Distress and perceived health: Mechanisms of health decline. Journal of Health and Social Behavior, 39, 298-311.

Ferraro, K.F., Farmer, M.M., and Wybraniec, J.A. (1997). Health trajectories: Long-term dynamics among black and white adults. Journal of Health and Social Behavior, 38, 38-54.