and cause of death; and on census data, providing age and sex-specific ''denominator" data. Vital registration data in Africa are scanty, however. Sub-Saharan Africa lacks vital registration systems (with the exception of Mauritius and South Africa) that even approach the United Nations standards for completeness.
Another traditional source of information on mortality and morbidity is based on the health care services, but this information is of limited use in Sub-Saharan Africa. One reason for this is that the majority of deaths occur outside the health care system. Another is that health care service information is usually restricted to inpatients in tertiary facilities, and there are substantial selection biases among the populations that use these health services, reflecting, for example, differentials in access to health services according to age, gender, and geography or a higher level of risk among those utilizing the services. Access may be lower, for instance, in young teenage groups or in rural areas (Graham, 1991; Timaeus, 1991d). Only a small proportion of women at risk in Sub-Saharan Africa have access to hospitals (Graham, 1991). Much of the health care provided to the African population is through smaller, community-based primary health care services (Ewbank, 1988), and evidence suggests that deaths among patients in contact with these facilities tend to be omitted in health service statistics because of incomplete recordkeeping (Graham, 1991).
Since the 1960s, demographic data have improved substantially. Censuses and surveys, described below, have filled some of the statistical needs traditionally met through vital registration data. Most important to the current level of knowledge about African mortality, however, has been the development of survey instruments and methodologies to measure mortality through specific retrospective questions asked of a mother or other family member about births and deaths, which can be included in a census or survey. These questions, described below, result in estimates of mortality over a broad age group. Indicators most often used are the probability of survival from birth to exact age 5, and the conditional probability of survival from age 15 to age 50. The estimates thus reflect "partial lifetime" mortality over a substantial age period. The mortality measures refer to the time period over which the deaths occurred, although greater weight is generally given to the information from more recent deaths. Expert matching of these estimates with model schedules of mortality, developed on the basis of mortality experience over a broad range of countries and time periods, can be used to estimate life expectancy at birth.
The above methodology is most developed for the estimation of mortality among infants and children, which includes the majority of deaths in a developing country. Estimates are based on a mother's retrospective reporting of her fertility history, or a question on the number of children ever born and the number of children surviving to the time of the survey (UN, 1983). These measures also provide limited information on mortality over time. Information on survival between birth and age 5 within the two to three years immediately prior to the survey is generally considered the most accurate. Limitations of the method include probable understatement of birth and death events recalled from the past, inclusion of fostered and adopted children, and decreasing accuracy of reporting as overall numbers of births and childhood deaths decline. Whether there is a selective underreporting of female versus male deaths through use of these measures is uncertain.
Procedures to estimate adult mortality are still being developed and refined, spurred on by more recent recognition that child and adult mortality are not as closely associated as had been thought; thus, traditional methods of estimating adult mortality based on child mortality are no longer considered accurate. The most direct approach—to ask about deaths in a household in a fixed previous time period—has not proved very useful in developing countries, because it requires large sample sizes to yield sufficient numbers of observations. Indirect methods have been more successful. These methods include, for example, the widowhood and orphanhood methods (Timaeus, 1991a,b,c; UN, 1983) and the sisterhood method (Graham et al., 1988, 1989). Such methods are based on questions that ask the respondent about survival of close relatives. Estimates are converted to survival probabilities across an age span, often ages 15–60, based on assumptions about past age patterns and trends of fertility and mortality. Widowhood questions have not been as widely used in Africa as orphanhood questions, because the former require a greater number of respondents to obtain accurate estimates. Sisterhood methods are fairly recent developments and, as such, are still being evaluated, but early field trials are encouraging (Graham, 1991; Trussell and Rodriguez, 1990). Orphanhood questions may underestimate mortality for theoretical reasons, as well as because of adoption, where respondents may refer to surviving stepparents or foster parents, rather than to early mortality of the biological parents. The impact of the adoption effect is possibly greater on reporting about mothers than about fathers, which may underlie the relatively steep declines noted in female