Table shows descriptive statistics for all variables included in the analysis. On average, HIV prevalence declined by 13% between 2001 and 2009, and HIV incidence declined by 26%. These averages, of course, obscure a good deal of variation, with change in HIV prevalence ranging from a decline of 48% (Côte d’lvoire) to an increase of 25% (Guinea-Bissau), and change in HIV incidence varying from an 81% decline (Namibia) to a 4% increase (Uganda). In terms of the other two dependent variables, on average 36% of people needing ARV therapy in 2009 received it, while 42% of HIV-positive pregnant women on average received PMTCT interventions.
In terms of independent variables, almost 80% of countries have a population policy, and slightly less than two-thirds of countries have an IPPF affiliate founded before 1986. On average, countries had a GDP per capita of slightly more than $800 per year. Cultural diversity is relatively high, with an average fractionalization score of 0.42, and 40% of countries are former British colonies. In terms of the HIV-related controls, a third of countries are PEPFAR focus countries, and countries received on average slightly less than $10 per person from the Global Fund between 2002 and 2009. Finally, in 2006, only slightly more than a quarter of people in need of antiretroviral therapy were receiving it (based on the old, WHO guidelines), and the epidemic peaked before 1999 in approximately one-third of countries.
Figure depicts the bivariate relationships between the key population interventions (early IPPF affiliates and the existence of a population policy) and the dependent variables. All of these figures show effects in the expected direction: having an early IPPF affiliate or a population policy is associated with better HIV outcomes. The difference is statistically significant at the p <0.05 level for the change in HIV prevalence, and at the p <0.10 level for ARV and PMTCT coverage.
Table presents the results from the multivariate analysis, which includes four different dependent variables and similar independent variables. Models 1 and 2 present standardized coefficients from ordinary least squares (OLS) regressions predicting change in HIV prevalence and change in HIV incidence, respectively, between 2001 and 2009. A decrease in the dependent variable is a good thing, so negative coefficients indicate a more favorable outcome (a larger decline in prevalence or incidence). Models 3 and 4 present standardized coefficients from OLS regressions predicting levels of ARV and PMTCT coverage in 2009. In these models, positive coefficients indicate a more favourable outcome (greater coverage).
Standardized coefficients from ordinary least squares regressions predicting HIV outcomes, sub-Saharan Africa, 2001-2009
Model 1 shows that after controlling for whether a country has an “old” epidemic (one that peaked prior to 1999), as well as antiretroviral coverage, the best predictors of declines in HIV prevalence are having an early IPPF affiliate and being a PEPFAR focus country. Holding all variables constant at their means, a country with an early IPPF affiliate is predicted to experience a 20.6% decline in HIV prevalence between 2001 and 2009, while a country without an affiliate is expected to have only a 3.0% decline. The positive sign on the PEPFAR variable indicates that focus countries experienced smaller declines in prevalence than non-focus countries. This finding is most likely not reflective of the impact of PEPFAR’s activities, but rather the result of PEPFAR generally targeting hard-hit countries.
Model 2 shows that many more factors are predictive of change in HIV incidence than of change in HIV prevalence. Specifically, having a population policy, as well as more GDP per capita is associated with greater declines in incidence. Specifically, holding all variables constant at their means, a country with a population policy would be predicted to experience a 32.7% decline in HIV incidence between 2001 and 2009, while a country without such a policy would be predicted to experience only a 3.8% decline. Furthermore, in this model, Global Fund disbursements play a positive role (leading to greater declines). Intriguingly, the coefficient for former British colonies is significant but positive, indicating that former British colonies experienced smaller declines in incidence. While it is possible that this outcome reflects some difference in these countries’ institutional capacity resulting from colonialism, more likely it is the result of the fact that most former British colonies are located in southern and eastern Africa, the areas hardest hit by the HIV epidemic.
Models 3 and 4, predicting different types of antiretroviral coverage, are largely consistent with one another. Countries with population policies do a better job of providing such services to their citizens. Indeed, a country with a population policy is predicted to have an ARV coverage rate 13.3 absolute percentage points higher, and a PMTCT coverage rate 18.7 absolute percentage points higher, than a country without such a policy, holding all other variables constant at their means. In addition, wealthier countries also more effectively provide antiretroviral coverage. Interestingly, in the case of overall ARV coverage specifically, high levels of cultural fractionalization are associated with lower levels of coverage, which echoes the findings of Lieberman [37
]. The effect of international funding for HIV-related activities clearly comes to bear in these models, as PEPFAR and Global Fund funds are both positive predictors of antiretroviral coverage.
The four models are designed to capture two types of success in addressing HIV. Models 1 and 2 analyze factors that may be associated with successful prevention efforts, while Models 3 and 4 are solely about treatment. Looking at the models in this way suggests two key observations. First, it seems to be more difficult to predict changes in HIV prevalence and incidence than it is to predict treatment success: the R2 values are somewhat lower for the first two models than for the second two models. This means that there are additional factors, most likely difficult-to-measure ones, driving differential success in prevention efforts.
Second, resources (broadly construed) are clearly very important to both prevention and treatment. Greater amounts of GDP per capita are associated with better prevention and treatment options. This may indicate that there are actually more resources to be put towards interventions, that there exist other social institutions that similarly facilitate interventions, or may reflect lower levels of poverty, which can drive HIV outcomes through numerous pathways. And while there is some evidence that greater resources in the form of more foreign aid directly targeting HIV is associated with greater prevention success, funds from PEPFAR and the Global Fund are strongly associated with treatment success, indicating the challenges of prevention interventions.
Omitted variables are an important consideration in any statistical analysis. Two hypothetically important variables, democracy and conflict, were knowingly omitted from the regression analysis in the name of parsimony because they showed no correlation with any of the dependent variables in either bivariate or multivariate contexts. There are, however, other variables that may influence the transmission of HIV and so may also particularly influence Models 1 and 2, but that were not entered into the regressions. These include variables related to the prevalence of parasitic infections, including malaria, schistosomiasis and various intestinal helminths [10
], as well as variables measuring exposure to HIV through unsafe injection practices [75
While rigorous debate continues about the relative role of these factors in explaining variation in HIV prevalence [76
], their potential impact is partially reflected by the inclusion of GDP per capita. Inclusion of alternative measures for GDP per capita that relate to different theories about the transmission of HIV (the percentage of the population living on less than two dollars per day, the percentage of the population with access to clean water, and the percentage of the population under-nourished) yielded substantively similar results for the analysis of change in HIV prevalence, and were much less predictive than GDP per capita in the analysis of change in HIV incidence (results not shown).
Taken together, these models show a significant impact of organizational and political factors resulting from population interventions – specifically, IPPF affiliates and population policies – on HIV outcomes. They also indicate that other factors are important. Level of resources, both in the form of GDP per capita and in the form of funds from major HIV donors, are significantly associated with positive HIV outcomes, particularly those related to provision of antiretroviral coverage. Cultural fractionalization plays a role in overall ARV coverage, while the role of being a former British colony seems to be mixed. In the case of change in HIV incidence, it is associated with worse outcomes (perhaps because this variable also captures southern African countries), while in the case of PMTCT coverage, it is associated with better outcomes (perhaps because of better institutional structures).