The scope of the worldwide AIDS epidemic is staggering. As of 2008, there are an estimated 32 million people living with HIV/AIDS, of whom more than 90% live in developing countries. Africa alone accounts for two-thirds of the world total and almost all of the infected children according to the UNAIDS 2008 Report on Global Aids Epidemic. It is hard not to argue that HIV/AIDS is a humanitarian crisis. However there is still debate among economists on how HIV/AIDS will impact the future development of these countries.
Drawing a parallel between AIDS and the Black Death plague in Europe, Young (2005) suggests that population declines will lead to higher per capita incomes in the affected countries. While the epidemic will have a detrimental impact on human capital accumulation, he postulates that widespread community infection will lower fertility, both directly through a reduction in willingness to engage in unprotected sex and indirectly by increasing the scarcity of labour and the value of women's time. Using household data from South Africa and relying on between-cohort variation in country-level HIV infection and number of births, he estimates a large negative effect of HIV prevalence on fertility. Fertility declines reinforce population declines and improve the future per capita income of South Africa.
In recent research (Juhn, Kalemli-Ozcan and Turan, 2008), we revisit this question using newly available micro data from population-based surveys. In the latest rounds of the Demographic Health Surveys, HIV testing was administered in 16 African countries, allowing us to link an individual woman's detailed fertility and health history to her own HIV status. Demographic and Health Surveys (DHS), which are based on nationally representative samples, are designed to gather information on fertility and child mortality. Recent waves of these surveys have sought information on HIV/AIDS status by asking a subset of women who are interviewed to provide a few drops of blood for HIV testing. The collected blood specimens and the main surveys are linked by case identification numbers. The linked data are available for 13 out of the 16 countries who conducted the testing.1
Population-based samples lead to lower estimates of HIV prevalence
One clear advantage of this newly available data is that it provides us with a more accurate estimate of HIV prevalence in the population. Previous researchers, including Young (2005) relied on estimates based on samples of pregnant women attending prenatal clinics. Since pregnant women are engaging in unprotected sex, they are at higher risk and such estimates are likely to overstate the prevalence rate in the population as a whole. Table 1 compares, by country, HIV prevalence rates among 15-49 year old women from the Demographic and Health Surveys (column 1) and other data sources that rely on surveillance data from pre-natal clinics (columns 3-5). The table shows that country-level prevalence from other data sources are indeed higher than those we estimate from the Demographic and Health Surveys.2
Table 1. Contrasting HIV prevalence estimates from various sources
|Country||Survey year||DHS||UNAIDS/WHO||US Census||US Census projections3|
Behavioural response versus physiological impact in Africa
Another clear advantage of the new data is that we can examine the impact of own HIV status separately from the impact of community-wide prevalence. Women who are HIV-positive may have lower fertility due to physiological reasons, i.e. the disease may lower fecundity or the individual may be too sick to be sexually active. By examining changes in fertility among non-infected women, we can focus on the behavioural response to increased risk of infection and death.
Young (2005) posits that the physiological effects of HIV on fertility are minimal and interprets his finding as being largely due to change in behaviour, including sexual practices, in response to increased mortality risk. Our study, in fact, points to the opposite. Women who are HIV-positive have approximately 20% lower probability of giving birth in a given year compared to women who are HIV-negative. This result is fairly robust across countries and remains even after we control for condom use and other measures of risky sexual behaviour. Our investigation of births prior to 1986 (prior to the on-set of the HIV/AIDS) using women's fertility histories suggests that unobserved heterogeneity is not driving our results.
In contrast to Young (2005), however, we do not find a negative effect of local community HIV prevalence on the fertility of non-infected women. In high HIV countries such as Kenya and Lesotho, the effect of community HIV prevalence on fertility of non-infected women is actually positive and statistically significant. However, when we pool all the countries we do not find a statistically significant effect.4
Overall our estimate of the impact of HIV on total fertility rate is considerably smaller than the estimate reported in Young (2005). His estimate suggests that a community that has 100% prevalence would have fertility that is approximately 80% lower than a community with zero prevalence. Our estimate of the impact of HIV, working entirely through the own effect, suggests that fertility would be approximately 20% lower. Aside from the difference in magnitude, is our distinction between the physiological and behavioural channels important?
The absence of a behavioural response among non-infected women is consistent with recent findings in Oster (2005), among others, who document relatively little change in sexual behaviour in response to HIV. It points to the importance of disseminating information and knowledge about the disease in these poorer and less developed countries. It may also be premature to assume that families necessarily want fewer children if they, the parents, or the children, are more likely to die. In fact, a large theoretical literature links rising life expectancy to reductions in fertility. The HIV/AIDS epidemic has produced a shock to life expectancy that should increase fertility according to these models.
How will HIV impact Africa’s future?
Will the fertility response to HIV reinforce or offset the declines in population due to mortality? Our results show that only fertility of infected women will decline and hence the total impact of HIV on the aggregate economy is much smaller than the effect implied by Young (2005). There is an extensive literature that documents substantial declines in human capital accumulation as a result of the disease. Complementary to these results, our evidence suggests that HIV/AIDS is likely to decrease rather than increase future per capita incomes in Africa.
Juhn, Kalemli-Ozcan and Turan (2008), HIV and Fertility in Africa: First Evidence from Population Based Surveys, NBER Working Paper 14208
Oster, E., (2005). Sexually Transmitted Infections, Sexual Behavior, and the HIV/AIDS Epidemic, Quarterly Journal of Economics, 120(2), 467-515.
Young, A., (2005). The Gift of the Dying: The Tragedy of AIDS and the Welfare of Future African Generations, Quarterly Journal of Economics, 120(2), 423-466.
1 Mali and Zambia have HIV data but cannot be linked to the main survey questions while Tanzanian survey does not include fertility questions.
2 Low response to the testing might cause lower estimates. However even in the case of countries such as Malawi where there was a 100% response rate to testing, the new estimates are considerably lower than the previous ones.
3 The US Census Bureau’s projections use the Estimation and Projections Package from WHO/UNAIDS to estimate HIV trends by fitting an epidemiological model to the surveillance data.
4 We employ a number of alternative estimation strategies including instrumental variables estimation where we instrument community level HIV prevalence with distance to the Democratic Republic of Congo as in Oster (2005). In another specification, we assume zero HIV prevalence before 1986 and regress change in age-specific births on the change in community-level HIV prevalence rate. None of these alternative specifications led to a significant community-level HIV effect in pooled data.