Search tips
Search criteria 


Logo of wtpaEurope PMCEurope PMC Funders GroupSubmit a Manuscript
AIDS. Author manuscript; available in PMC 2010 March 30.
Published in final edited form as:
PMCID: PMC2847257

The socioeconomic determinants of HIV incidence: evidence from a longitudinal, population-based study in rural South Africa



Knowledge of the effect of socioeconomic status (SES) on HIV infection in Africa stems largely from cross-sectional studies. Cross-sectional studies suffer from two important limitations: two-way causality between SES and HIV serostatus and simultaneous effects of SES on HIV incidence and on HIV-positive survival time. Both problems are avoided in longitudinal cohort studies.


We used data from a longitudinal HIV surveillance and a linked demographic surveillance in a poor rural community in KwaZulu-Natal, South Africa, to investigate the effect of three measures of SES – educational attainment, household wealth categories (based on a ranking of households on an assets index scale) and per capita household expenditure – on HIV incidence. Our sample comprised of 3325 individuals who tested HIV-negative at baseline and either HIV-negative or HIV-positive on a second test (on average 1.3 years after the first).


In multivariable survival analysis, one additional year of education reduced the hazard of acquiring HIV by 7% (p = 0.017) net of sex, age, wealth, household expenditure, rural vs. urban/periurban residence, migration status and partnership status. Holding other factors equal, members of households that fell into the middle 40% of relative wealth had a 72% higher hazard of HIV acquisition than members of the 40% poorest households (p = 0.012). Per capita household expenditure did not significantly affect HIV incidence (p = 0.669).


While poverty reduction is important for obvious reasons, it may not be as effective as anticipated in reducing the spread of HIV in rural South Africa. In contrast, our results suggest that increasing educational attainment in the general population may lower HIV incidence.

Keywords: HIV incidence, socioeconomic factors, Africa, economics, epidemiology, longitudinal study, surveillance


The HIV epidemic remains one of the greatest health and development challenges facing sub-Saharan Africa. South Africa bears a substantial brunt of the HIV epidemic, with 5.5 million adults and children living with HIV and an estimated 320,000 deaths due to HIV/AIDS in 2005 [1]. From the beginning of the HIV epidemic, researchers have tried to understand its relationship with socioeconomic status (SES). Reviews of studies in Africa show a wide variety of relationships between HIV and SES [2-5]. However, most of the African studies examine cross-sectional associations between HIV serostatus and SES and thus suffer from two important limitations.

First, cross-sectional studies are usually unable to distinguish between the effect of SES on HIV infection and the effect of HIV infection on SES [6-8]. Several pathways have been suggested through which a decrease in SES may increase the risk of HIV infection. Malnutrition and micronutrient deficiencies, which are more common in the poor, can disrupt the integrity of the vaginal epithelium, increasing its permeability to HIV [9]. Ulcerative genital diseases, which are associated with low SES, increase the transmission probability of HIV [7]. Women of low SES may be economically dependent on male partners, limiting their ability to negotiate condom use in relationships, or forcing them to sell sex for money [2, 10]. On the other hand, an increase in SES may increase the risk of HIV infection because wealthy or educated people have more resources with which to attract and maintain multiple partners [11].

HIV status may also be an important determinant of SES. AIDS-related illness limits people’s ability to work, which may decrease SES, especially among people who work in economic sectors where income is closely linked to productivity such as agriculture or the informal sector [12]. In addition, HIV and its corresponding diseases may lead to substantial expenditure for health care, decreasing household wealth [12-14].

A second problem of cross-sectional studies is that they usually cannot distinguish between the effect of SES on HIV incidence and the effect of SES on survival with HIV. Prior studies have shown that HIV survival time increases with SES [15], possibly because diet or access to highly active antiretroviral therapy improve with SES [4, 16]. In a cross-sectional study, it is thus possible to find a positive association between SES and HIV infection, even if higher SES protects individuals from acquiring HIV.

Both problems – two-way causality between SES and HIV serostatus and simultaneous effects of SES on HIV incidence and on HIV-positive survival time – are avoided when a cohort of HIV-negative individuals is followed over time and the hazards of HIV seroconversion are compared across people of different SES in survival analyses. To date no such longitudinal study on SES and HIV incidence in South Africa has been published and there have only been a few longitudinal studies reporting on the effect of SES on the hazard of HIV seroconversion in other African countries[17-21]. With one exception, these longitudinal studies were conducted in cohorts of sexually active young urban women attending antenatal care (ANC) or family planning services [17, 18, 20], or in cohorts of urban factory workers [19, 21]. Relationships between SES and HIV incidence in these subpopulations may differ from those in the general population, or may not be detectable because SES varies too little [22].

No clear pattern of the relationships between SES and HIV seroconversion risk emerges from these studies. In urban women attending ANC or family planning services, the risk of HIV seroconversion has been found not to be associated with education [18], to decrease with the educational attainment of women’s partners [20], and to decrease with women’s incomes [17]. In studies in urban factory workers, the risk of HIV seroconversion was positively associated with occupational status [19] and educational attainment [21]. The only published study using longitudinal, general population data to examine the association between SES and HIV incidence in Africa found that in rural Uganda educational attainment is not a significant predictor of HIV incidence [23].

Wojcicki (2005) concludes from a review of studies on SES and HIV infection in sub-Saharan Africa that “further studies should use multiple measures of SES“ because different dimensions of SES influence HIV infection risk in different ways [4]. We report findings about the effect of three different measures of SES (educational attainment, household wealth categories and household expenditure) on HIV incidence from a longitudinal HIV surveillance of the general population in a rural community in KwaZulu-Natal, South Africa.


Study area

We used data from the longitudinal population-based HIV surveillance conducted by the Africa Centre for Health and Population Studies, University of KwaZulu-Natal, and from the Africa Centre Demographic Information System (ACDIS) to investigate the relationship between SES and the hazard of HIV seroconversion. ACDIS has been collecting demographic data since January 2000 and socioeconomic data since February 2001 [24]. The ACDIS demographic surveillance area (DSA) is located in the rural district of Umkhanyakude in northern KwaZulu-Natal, South Africa. It covers 435 square kilometres and a total resident population of about 86,000 Zulu-speaking people (June 2003). While the study took place in an overall rural community, the area includes an urban township and periurban areas (informal settlements with a population density of more than 400 people per km2). In 2001, 61% of households in the ACDIS area had a toilet and only 38% had access to piped water. The unemployment rate in the same year was 30% [25].

Data collection

Teams of two trained fieldworkers visited each eligible individual in his or her household. Fieldworkers revisited the households up to four times to contact absentees. If a subject no longer lived at the household, the field worker handed the case to a specially trained tracking team that made up to ten attempts to find the individual in his or her new residence. After written informed consent, the field workers collected blood by finger stick and prepared dried blood spots for HIV testing according to the Joint United Nations Programme on HIV/AIDS (UNAIDS) and World Health Organization (WHO) Guidelines for Using HIV Testing Technologies in Surveillance [26]. HIV status was determined by antibody testing with a broad based HIV-1/HIV-2 ELISA (Vironostika, Organon Teknika, Boxtel, The Netherlands) followed by a confirmatory ELISA (GAC-ELISA, Abbott, Abbott Park, Illinois, USA). All covariates used in this study were collected by the ACDIS demographic surveillance system conducted between January 2003 and September 2004 [25].


Our sample includes all individuals who met the following criteria: they were age eligible for inclusion in the HIV surveillance during the first round from June 2003 to December 2004 and during the second round from January to December 2005 (women between 15 and 49 years of age and men between 15 and 54 years of age); they were residents in the ACDIS DSA during the first round of the HIV surveillance and either residents in the DSA or non-resident household members during the second round; they tested HIV antibody-negative during the first round and tested either HIV-negative or HIV-positive during the second round; and data on all independent variables used in this analysis was available at the time of the negative test in the first round. On average the time between the first and the second HIV test was 1.3 years.

Of the 8952 participants in the population-based HIV surveillance who were negative during the first round of testing and still eligible to be tested during the second round, information on at least one of the independent variables used in this study was missing for 1597 individuals and information on HIV status during the second round of testing was missing for 4030 (either because individuals could not be contacted (n = 1479) or were contacted and refused to consent to an HIV test (n = 2551)). In comparison with the individuals included in the sample, individuals for whom information on the independent variables used in this study was available, but a second HIV test was not, were not significantly different at the 5% confidence level with regard to per capita household expenditure, rural vs. urban/periurban place of residence or the probability of having a partner at baseline. However, they were slightly younger (25.4 vs. 26.4 years, p < 0.001), slightly more educated (8.2 vs. 7.7 average educational grade attained, p < 0.001), wealthier (−0.073 vs. −0.352 household assets index, p < 0.001), more likely to be male (49% vs. 40%, p < 0.001), more likely to have migrated out of the DSA between the two rounds (14% vs. 7%, p < 0.001) and less likely to be married (9% vs. 11%, p = 0.001) than the individuals in the sample. Table 1 shows the characteristics of the 3325 individuals included in the sample. Even though the study took place in an overall rural community, a substantial proportion of participants (28%) lived in either an urban or a periurban area (Table 1).

Table 1
Sample characteristics (N = 3325)

While in comparison to South Africa as whole the average indicators of SES in this community are low [25], their dispersion within the community is quite wide (see Table 1). For instance, in our sample the 10th percentile of educational attainment was 2nd grade, while the 90th percentile was 12th grade; and the 10th percentile of daily total per capita household expenditure was 1.2 ZAR, while 90th percentile was 7.2 ZAR, suggesting that there is sufficient variation in these two measures of SES to warrant investigating their effects on HIV incidence.

Independent variables

The focus of our analyses is the socioeconomic determinants of HIV incidence. We investigated the relationship of HIV infection with three measures of SES: educational attainment, household wealth and expenditure. Increased educational attainment (the highest education grade that an individual has completed within a country’s educational system) has been hypothesized to lead to a lower risk of HIV infection, because it improves the ability to understand and act on health promotion messages or because it is associated with increased exposure to school-based HIV prevention programmes or increased access to health services [27].

Wealth and expenditure are likely to capture different financial aspects of social status. Households generate wealth through saving of income after spending money on consumption. There is commonly greater variation in wealth than in expenditure because wealth is accumulated and because some expenditure on basic items such as food and clothing are indispensable for human survival. Thus, wealth may be a more sensitive measure of the long-term social position of a household and may capture influences of social status on the risk of HIV infection better than expenditure.

We used a household assets index as a measure of wealth. As shown by Morris and colleagues, household assets indices are valid proxies for wealth in health surveys in rural Africa [28]. Following Filmer and Pritchett [29], we used the first principal component obtained in a principal component analysis of information on house ownership, water source, energy, toilet type, electricity and 27 household assets as an assets index. The assets included items that can be used for consumption, production or both, such as beds, bicycles, tables, telephones, television sets, sewing machines, block makers, wheelbarrows, tractors, cattle, and other livestock. We categorised households as either belonging to the poorest 40%, the middle 40% or the wealthiest 20% on the assets index scale. We chose these three categories of relative wealth, because they have been found to capture wealth effects well in a number of studies in poor provinces of South Africa [30, 31], including studies investigating the effect of wealth on health [32, 33].

Household expenditure captures the short-term financial liquidity of the members of a household and should thus be a better measure of the influences of current consumption of costly services on HIV infection than wealth. For example, in a cash-strapped period, an individual may not have sufficient funds to seek treatment for sexually transmitted diseases (STD) whose presence increases the risk of HIV transmission or may not be able to travel to a place where condoms are available. We measured total household expenditure by summing spending across all categories about which expenditure information is available in the ACDIS, including spending for household members (on shopping, rent, cloths, water, fuel, electricity, health, transport, religious activities, telephones, cell phones, payments for goods bought by hire-purchase or lay-bye, funerals, life insurance, and school), as well as spending for people outside the household (money, goods and food). We used daily total household expenditure and divide it by the number of members in each household to adjust for household size [34]. The resulting variable daily total per capita household expenditure is henceforth referred to as household expenditure. We logarithmically transformed the household expenditure variable to reduce skewness.

As expected, educational attainment, the household assets index, and household expenditure were positively correlated. However, the three measures of SES were not very highly correlated (with Pearson’s correlation coefficient at 0.256 between educational attainment and the household assets index; at 0.174 between educational attainment and the log transform of household expenditure; and at 0.380 between the household assets index and the log transform of household expenditure), reinforcing that each captures a different aspect of SES and cannot be used in place of one of the other two in multiple regression analysis.

In addition to the three measures of SES, we control for a number of variables that have been found to be associated with HIV infection in cross-sectional surveys in South Africa (sex [5, 35-37], age [38-41], rural vs. urban/periurban residence [36], migration status [5, 39, 41] and partnership status [5, 37, 39, 41-43]). All independent variables were measured at baseline and assumed to be time-invariant.

Statistical analysis

In order to investigate the effects of explanatory variables on the time to HIV seroconversion, we used semiparametric and parametric survival models in the following proportional hazards specification [44].


where h0(t) is the baseline hazard function that is obtained when all explanatory variables are equal to 0. A unit change of an independent variable in this model leads to a constant parallel shift of the baseline hazard function. If the baseline hazard function is left unspecified, the model is the semiparametric Cox Proportional Hazards Model (CPHM).

While parametric models that specify the functional form of the baseline hazard function have the disadvantage that they may lead to inconsistent estimates if the baseline function is misspecified, they will be more efficient than the semiparametric model if the baseline hazard is correctly specified. For the multivariate analyses with all explanatory variables we thus estimate parametric models in addition to the CPHM. First, we use the exponential model


which assumes that the hazard function is constant over time. Next, we estimate the Weibull model


which allows the hazard function to monotonically increase (p > 1) or decrease (p < 1) over time. The Weibull model includes the exponential model as a special case (p = 1).

Finally, we estimated random effects generalizations to the proportional hazards models, i.e. ‘frailty models’, that allow for variability in unobserved individual-level factors that is unaccounted for by the independent variables included in the models [45]. The unobservable individual effect (‘frailty’), v, is considered a random variable over the population that multiplicatively enters the hazard function in the above proportional hazards specification, i.e.


The random variable v is assumed to be positive, to be distributed independently of t and X, and to follow a gamma distribution with unit mean (required for identification) and finite variance (θ). If θ is not significantly different from zero, individual heterogeneity is not important and it is appropriate to estimate the non-frailty models.


During 4352 person-years at risk, 131 of the 3325 individuals in our sample became seropositive. The overall incidence of HIV infection was 3.0 per 100 person-years (95% confidence interval 2.5 to 3.6). Table 2 shows the unadjusted hazard ratios when we examine separately the effects on time to seroconversion of sex; age, age2 and age3; educational attainment; wealth categories; household expenditure; place of residence; migration status; and partnership status in CPHM. In these separate regressions, female sex, age, belonging to the middle wealth category, urban/periurban place of residence, having migrated out of the DSA between the first and the second round of the HIV surveillance, and not being married but having a partner were positively associated with the hazard of HIV seroconversion (all p < 0.010). Educational attainment and household expenditure were not significantly associated with the hazard of HIV seroconversion (both p > 0.556).

Table 2
Unadjusted hazard ratios and hazard ratios adjusted for sex and age

By experimentation we found that a third-order polynomial age specification provides a good fit for the relationship between age and time to HIV seroconversion. In order to reduce multicollinearity we expressed age as its deviation from its mean [46]. When we adjusted for sex and age (in the third-order polynomial specification) two of the relationships from the individual regressions that d not adjust for any other factor changed significantly (Table 2, Adjusted HR). First, the hazard ratio of educational attainment changed from 0.99 (p = 0.785) to 0.93 (p = 0.022). Holding age and sex constant, each additional year of educational attainment reduced the hazard of HIV seroconversion by 7%. Second, the hazard ratio for the group of unmarried people with a partner changed from 2.13 (p < 0.001) to 1.60 (p = 0.041).

Table 3 shows estimation results of the semiparametric CPHM and the parametric exponential and Weibull regression models in their proportional hazard specification (expressions 1, 2 and 3, respectively). We tested the proportional hazards assumption for all variables jointly and for each variable individually using the tests proposed by Grambsch and Therneau [47]. The null hypothesis that the hazard rates are proportional could be rejected at the 10% confidence level in any of the tests. A unit change in one of the independent variables leads to a proportional shift of the hazard rate.

Table 3
Multiple regression models of the hazard of HIV seroconversion

The sizes and significance levels of the adjusted hazard ratios (aHR) were similar in all three estimations. While the CPHM leads to consistent estimates and is more flexible than the parametric models, it is less efficient than the appropriate parametric model. In the Weibull regression, the null hypothesis that p = 1 could not be rejected (Table 3), i.e. the hazard function is neither increasing nor decreasing over time. Thus, the exponential estimation is preferred over the Weibull estimation. In order to check for unobserved individual heterogeneity, we estimated the exponential model (Table 3, IIB) and the Weibull model (Table 3, IIIB) including an individual random effect, or frailty, ν (see expression 5). The null hypothesis that the individual random effect is equal to zero was not rejected at the 10% significance level. The exponential estimation without frailty is thus the preferred parametric model and was used for the description of results below.

In multiple regression analysis, belonging to the middle 40% of households as ranked by the assets index increased the hazard of HIV seroconversion by a factor of approximately two (p = 0.001) (Table 3, IIA). Controlling for place of residence, migration status and partnership status in addition to sex and age reduced the size of the hazard ratio (to 1.72) but the effect remained significant (p = 0.012) (Table 3, IIB). To test whether our finding that belonging to a household in the middle wealth category increases the risk of HIV incidence is robust to a change in the choice of household wealth categories, we repeated the regressions in Table 3 with households categorized into wealth tertiles on the assets index scale. The alternative categorization changed the sizes and significant levels of all coefficients only slightly. In particular, when we replaced the wealth variables in model IIB with variables capturing wealth tertiles, the coefficient of the middle wealth category became 1.62 (p = 0.037).

One additional grade of educational attainment reduced the hazard of HIV seroconversion by about 7% (Table 3, IIB). Household expenditure was not a significant determinant of HIV seroconversion in any of the models. Urban residence was associated with a 65% increase in the hazard of HIV seronconversion (p = 0.012) (Table 3, IIB). Individuals who remained residents in the ACDIS DSA between the two rounds of the HIV surveillance faced about half the hazard of HIV seroconversion of those who migrate out of the DSA between the two rounds (p = 0.006). Once all other variables were controlled for, the hazard ratio of unmarried people with a partner at baseline remained borderline significant (p = 0.074) and the partnership variables jointly increased the model fit with borderline significance (p = 0.099).

In order to test whether the effects of education and wealth on HIV incidence are modified by sex, we added in turn education-sex and wealth category-sex interaction terms to all the models reported in Table 3. None of the interaction terms was significant at the 5% confidence level. Further, we added the square of the educational attainment variable to the regression models IB, IIB and IIIB in order to investigate if the effect of education on HIV incidence is non-linear. In none of the models was the added term significant; we thus did not include it in the final regression equations. Similarly, we replaced the log transform of the household expenditure variable in models IB, IIB and IIIB with alternative functional forms (linear, linear and square). In none of the alternative regression equations was any of the household expenditure terms significant at the 5% confidence level.


We show that educational attainment significantly reduces the hazard of becoming infected with HIV in a poor rural community in South Africa when controlling for sex, age, wealth, household expenditure, place of residence, migration status and partnership status. The protective effect of education shown in this study differs from the findings of previous studies that suggest that educational attainment is not significantly associated, or positively associated, with the risk of HIV infection [27].

The differences between our results and the findings of previous studies may be due to methodological issues such as how they control for confounding. We find that educational attainment is not correlated with the risk of HIV seronconversion in univariable analysis but that its protective effect against HIV seroconversion becomes apparent once sex and age are controlled for. In the case of South Africa, the relationship between educational attainment and time to HIV seronversion is likely to be confounded by sex and age. For instance, HIV seroconversion risk decreases with age above certain peak ages in women and men [48], while average educational attainment decreases with age in older age groups. The age-specific pattern of education reflects secular changes in South African education policy, such as the South African Schools Act of 1996 that abolished racial segregation in schools [49]. For instance, the matric pass rate (i.e. attainment of grade 12) increased from 40% in the late 1990s to 68% in 2005 [50].

Alternatively, education effects may differ by stage of the HIV epidemic. Most of the published studies that have examined the relationship between education and HIV infection were conducted in early stages of the epidemic [27], when educational attainment may have been positively associated with HIV infection, for instance because the more educated had more partners in any given period of time than the less educated. In contrast, as the epidemic matured, the more educated may have adopted HIV risk-reducing behaviours more quickly than the less educated because they were more exposed to health promotion messages or more empowered to negotiate protective behaviours with sexual partners [27].

We also show that in this overall poor community it is not the members of the asset-poorest households who are at highest risk of HIV acquisition but people who live in households belonging to the middle category of relative wealth. Recent analyses of cross-sectional surveys of HIV serostatus in Africa have shown that the poor do not have the highest HIV prevalence [11, 22]. Our longitudinal study provides evidence about the causal effect of relative wealth on the risk of HIV acquisition. First, our results obtain while important other determinants of acquisition of HIV that may be correlated with wealth are controlled for, particularly urban residence and migration status. Second, unlike analyses of cross-sectional surveys, we can rule out the possibility that positive HIV status of study participants caused his or her household to fall into poverty because of loss of employment or increased expenses related to disease. Third, unlike results from cross-sectional surveys, our findings cannot have been caused by a wealth gradient in the survival with HIV.

Our third main finding is that household expenditure does not seem to influence the hazard of HIV seroconversion in this population. In the ACDIS DSA, government clinics distribute condoms for free and provide basic health services, including STD treatment, for free. One plausible explanation of our result is thus that access to services that can help prevent HIV transmission does not depend on households’ short-term ability to pay. Alternatively, it is possible that financial liquidity does improve access (for instance, because transport to government clinics is costly or because private health care providers offer some services that are effective in reducing HIV transmission that are not available at government clinics), but that access does not translate into actual utilisation of such services (for instance, because people who could access them do not believe that they are effective).

Finally, we find that other covariates (sex, age, place of residence, migration status and partnership status) influence the hazard of HIV seroconversion as expected based on previous studies [35-38, 40, 41]. Studies of risky sexual behaviour in Africa have shown striking differences between women and men [51-54]. In as far as education and wealth effects on HIV incidence are conveyed by sexual behaviour we expected to find – but did not – that the effects of education and wealth on HIV incidence are modified by sex. It is possible that pathways from education and wealth to HIV acquisition that are not sex-specific (e.g. malnutrition) are relatively more important in explaining our findings than sex-specific pathways, or that – after controlling for sex and other factors – different pathways in women and men have similar effects on HIV incidence. However, given that our sample includes fewer men than women, it is also possible that our study lacks power to detect a sex differential in the effects of education and wealth on HIV incidence.

Another possible limitation of our study are uncontrolled selection effects because of selection into the baseline sample, because of missing information on independent variables, or because of attrition between the first and the second round of the HIV surveillance. While selection on observed factors that are associated with HIV seroconversion will bias estimates of HIV incidence (unless different selection biases balance each other out), coefficient estimates in multiple regression will be consistent if the observed factors that determine selection are included as independent variables in the regression equation [55, 56]. Thus, our regression equations control for selection on sex, age, education, wealth, household expenditure, rural vs. urban/periurban place of residence, migration status and partnership status. As these characteristics are among the most commonly observed correlates of HIV infection in South Africa [57], it is possible that our model adequately controls for selection effects. However, we cannot completely rule out that selection on unobserved characteristics that are associated with the risk of contracting HIV affect our findings. One possibility to adjust for selection on unobservable factors are Heckman-type selection models, which are not as well developed for survival analysis as they are, for instance, for ordinary least squares regression or probit regression, and whose performance commonly depends on the existence of a valid and relevant exclusion restriction, i.e. a variable that is a significant predictor of selection into the sample, but not independently associated with the time to HIV seroconversion [58-60]. Future studies are needed to investigate the effect of selection on unobservable factors in analysis of the socioeconomic determinants of HIV incidence in this community.

In sum, our results provide little support for the assertion that “reducing poverty will be at the core of a long-term, sustainable solution to HIV/AIDS” (Lynda Fenton, 2005) [61]. While poverty reduction is important for obvious reasons, it may not be as effective as anticipated in reducing the spread of HIV in rural South Africa. In contrast, increasing educational attainment in the general population – whatever the precise pathways of the effect – may lower HIV incidence.


We thank the fieldworkers and supervisors at the Africa Centre for Health and Population Studies for their excellent work in the HIV surveillance and the Demographic Information System. We thank the Africa Centre community for their participation in the surveys. The research reported in this paper was supported by the Wellcome Trust through grant GR065377/Z/01/B for the HIV surveillance and grant GR065377/Z/01/H for the Africa Centre Demographic Information System. Some of the methods used in this analysis were devised and tested at workshops run by the Wellcome Trust-funded ALPHA network.


1. UNAIDS . 2006 Report on the global AIDS epidemic, Annex 2: HIV and AIDS estimates and data. Joint United Nations Programme on HIV/AIDS; Geneva: 2006.
2. Buve A, Bishikwabo-Nsarhaza K, Mutangadura G. The spread and effect of HIV-1 infection in sub-Saharan Africa. Lancet. 2002;359:2011–2017. [PubMed]
3. Nyindo M. Complementary factors contributing to the rapid spread of HIV-I in sub-Saharan Africa: a review. East Afr Med J. 2005;82:40–46. [PubMed]
4. Wojcicki JM. Socioeconomic status as a risk factor for HIV infection in women in East, Central and Southern Africa: a systematic review. J Biosoc Sci. 2005;37:1–36. [PubMed]
5. Johnson L, Budlender D. HIV risk factors: a review of demographic, socioeconomic, biomedical and behavioural determinants of HIV prevalence in South Africa. In: Centre for Actuarial Research, editor. CARE Monograph No. 8. University of Cape Town; Cape Town: 2002.
6. Braveman PA, Cubbin C, Egerter S, Chideya S, Marchi KS, Metzler M, Posner S. Socioeconomic status in health research: one size does not fit all. JAMA. 2005;294:2879–2888. [PubMed]
7. Mosley A. Does HIV or poverty cause AIDS? Biomedical and epidemiological perspectives. Theor Med Bioeth. 2004;25:399–421. [PubMed]
8. Myer L, Ehrlich RI, Susser ES. Social epidemiology in South Africa. Epidemiol Rev. 2004;26:112–123. [PubMed]
9. Stillwaggon E. HIV/AIDS in Africa: fertile terrain. Journal of Development Studies. 2002;38:1–22.
10. Booysen FL, Summerton J. Poverty, risky sexual behaviour, and vulnerability to HIV infection: evidence from South Africa. J Health Popul Nutr. 2002;20:285–288. [PubMed]
11. Shelton JD, Cassell MM, Adetunji J. Is poverty or wealth at the root of HIV? Lancet. 2005;366:1057–1058. [PubMed]
12. Greener R. The Impact of HIV/AIDS on Poverty and Inequality. In: Haacker M, editor. The Macroeconomics of HIV/AIDS. International Monetary Fund; Washington, DC: 2004. pp. 167–181.
13. Bachmann MO, Booysen FL. Health and economic impact of HIV/AIDS on South African households: a cohort study. BMC Public Health. 2003;3:14. [PMC free article] [PubMed]
14. Booysen FL, Mafereka R. The impact of HIV/AIDS on household savings in two Free State communities; Conference on Reducing Poverty and Inequality: How can Africa be included?, Oxford University; Centre for Study of African Economies, Oxford. 2006.
15. Braitstein P, Brinkhof MW, Dabis F, Schechter M, Boulle A, Miotti P, et al. Mortality of HIV-1-infected patients in the first year of antiretroviral therapy: comparison between low-income and high-income countries. Lancet. 2006;367:817–824. [PubMed]
16. Wood E, Montaner JS, Chan K, Tyndall MW, Schechter MT, Bangsberg D, et al. Socioeconomic status, access to triple therapy, and survival from HIV-disease since 1996. AIDS. 2002;16:2065–2072. [PubMed]
17. Bulterys M, Chao A, Habimana P, Dushimimana A, Nawrocki P, Saah A. Incident HIV-1 infection in a cohort of young women in Butare, Rwanda. AIDS. 1994;8:1585–1591. [PubMed]
18. Kapiga SH, Lyamuya EF, Lwihula GK, Hunter DJ. The incidence of HIV infection among women using family planning methods in Dar es Salaam, Tanzania. AIDS. 1998;12:75–84. [PubMed]
19. Mayala M, Minlangu M, Nzila N, Mama A, Jingu M, Mundele L, et al. [Prevalence and incidence of HIV-1 infection among employees of a large textile business and their wives in Kinshasa, 1991-1996] Rev Epidemiol Sante Publique. 2001;49:117–124. [PubMed]
20. Mbizvo MT, Kasule J, Mahomed K, Nathoo K. HIV-1 seroconversion incidence following pregnancy and delivery among women seronegative at recruitment in Harare, Zimbabwe. Cent Afr J Med. 2001;47:115–118. [PubMed]
21. Senkoro KP, Boerma JT, Klokke AH, Ng’weshemi JZ, Muro AS, Gabone R, Borgdorff MW. HIV incidence and HIV-associated mortality in a cohort of factory workers and their spouses in Tanzania, 1991 through 1996. J Acquir Immune Defic Syndr. 2000;23:194–202. [PubMed]
22. de Walque D. Policy Research Working Paper Series: 3844. The World Bank; 2006. Who gets AIDS and how? The determinants of HIV infection and sexual behaviors in Burkina Faso, Cameroon, Ghana, Kenya, and Tanzania.
23. Wawer MJ, Serwadda D, Gray RH, Sewankambo NK, Li C, Nalugoda F, et al. Trends in HIV-1 prevalence may not reflect trends in incidence in mature epidemics: data from the Rakai population-based cohort, Uganda. AIDS. 1997;11:1023–1030. [PubMed]
24. Hosegood V, Timaeus IM. Household Composition and Dynamics in KwaZulu-Natal, South Africa: Mirroring Social Reality in Longitudinal Data Collection. In: van der Walle E, Sharpe ME, editors. African Households: Censuses and Surveys. University of Illinois Press; Champaign, IL: 2006. pp. 58–77.
25. Case A, Ardington C. Socioeconomic Factors. In: Africa Centre Population Studies Group, editor. Africa Centre Demographic Information System: sociodemographic and health events. Africa Centre for Health and Population Studies; Mtubatuba: 2004. pp. 68–78.
26. UNAIDS/WHO . Guidelines for using HIV testing technologies in surveillance: selection, evaluation, and implementation. UNAIDS/WHO; Geneva: 2001.
27. Hargreaves JR, Glynn JR. Educational attainment and HIV-1 infection in developing countries: a systematic review. Trop Med Int Health. 2002;7:489–498. [PubMed]
28. Morris SS, Carletto C, Hoddinott J, Christiaensen LJ. Validity of rapid estimates of household wealth and income for health surveys in rural Africa. Journal of Epidemiology and Community Health. 2000;54:381–387. [PMC free article] [PubMed]
29. Filmer D, Pritchett LH. Estimating wealth effects without expenditure data - or tears: an application to educational enrollments in States of India. Demography. 2001;38:115–132. [PubMed]
30. van der Berg S, Louw M. Changing patterns of South African income distribution: towards time series estimates of distribution and poverty. South African Journal of Economics. 2004;72:546–572.
31. Gyekye A, Akinboade O. A profile of poverty in the Limpopo province of South Africa. Eastern Africa Social Science Research Review. 2003;19:89–109.
32. Blaauw D, Penn-Kekana L. Socio-economic inequalities and maternal health in South Africa. Health Systems Development Programme. Centre for Health Policy, University of Witswatersrand; Johannesburg: 2002.
33. Booysen FL. Using demographic and health surveys to measure poverty – an application to South Africa. Journal for Studies in Economics and Econometrics. 2002;26:53–70.
34. World Bank Poverty in Africa -- standardized welfare indicators. [accessed: 12 June 2006]. 2007.
35. Auvert B, Ballard R, Campbell C, Carael M, Carton M, Fehler G, et al. HIV infection among youth in a South African mining town is associated with herpes simplex virus-2 seropositivity and sexual behaviour. AIDS. 2001;15:885–898. [PubMed]
36. Connolly C, Shisana O, Colvin M, Stoker D. Epidemiology of HIV in South Africa--results of a national, community-based survey. S Afr Med J. 2004;94:776–781. [PubMed]
37. Pettifor AE, Rees HV, Kleinschmidt I, Steffenson AE, MacPhail C, Hlongwa-Madikizela L, et al. Young people’s sexual health in South Africa: HIV prevalence and sexual behaviors from a nationally representative household survey. AIDS. 2005;19:1525–1534. [PubMed]
38. Dunkle KL, Beksinska ME, Rees VH, Ballard RC, Htun Y, Wilson ML. Risk factors for HIV infection among sex workers in Johannesburg, South Africa. Int J STD AIDS. 2005;16:256–261. [PubMed]
39. Lurie MN, Williams BG, Zuma K, Mkaya-Mwamburi D, Garnett G, Sturm AW, et al. The impact of migration on HIV-1 transmission in South Africa: a study of migrant and nonmigrant men and their partners. Sex Transm Dis. 2003;30:149–156. [PubMed]
40. Shisana O, Hall EJ, Maluleke R, Chauveau J, Schwabe C. HIV/AIDS prevalence among South African health workers. S Afr Med J. 2004;94:846–850. [PubMed]
41. Zuma K, Gouws E, Williams B, Lurie M. Risk factors for HIV infection among women in Carletonville, South Africa: migration, demography and sexually transmitted diseases. Int J STD AIDS. 2003;14:814–817. [PubMed]
42. Ahmed S, Lutalo T, Wawer M, Serwadda D, Sewankambo NK, Nalugoda F, et al. HIV incidence and sexually transmitted disease prevalence associated with condom use: a population study in Rakai, Uganda. Aids. 2001;15:2171–2179. [PubMed]
43. Shisana O, Zungu-Dirwayi N, Toefy Y, Simbayi LC, Malik S, Zuma K. Marital status and risk of HIV infection in South Africa. S Afr Med J. 2004;94:537–543. [PubMed]
44. Cox DR. Regression models and life tables. Journal of the Royal Statistical Society. 1972;34:187–220.
45. Gutierrez RG. Parametric frailty and shared frailty survival models. The Stata Journal. 2002;2:22–44.
46. Kutner MH, Nachtsheim CJ, Neter J, Li W. Applied linear statistical models. 5th ed. McGraw-Hill; Boston: 2004.
47. Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika. 1994;81:515–526.
48. Bärnighausen T, Tanser F, Mbizana C, Gqwede Z, Herbst K, Newell ML. HIV incidence in a rural community in KwaZulu-Natal, South Africa; 3rd South African AIDS Conference; Durban. 2007.
49. Maithufi IP. The South African Schools Act of 1996: a break with the past. In: Robinson JA, editor. The law of children and minors in South Africa. Butterworth; South Africa: 1997.
50. Education in South Africa. [accessed: 15. January 2007]. 2006.
51. Babalola S. Gender differences in the factors influencing consistent condom use among young people in Tanzania. Int J Adolesc Med Health. 2006;18:287–298. [PubMed]
52. Boer H, Mashamba MT. Gender power imbalance and differential psychosocial correlates of intended condom use among male and female adolescents from Venda, South Africa. Br J Health Psychol. 2007;12:51–63. [PubMed]
53. Dunkle KL, Jewkes RK, Nduna M, Levin J, Jama N, Khuzwayo N, et al. Perpetration of partner violence and HIV risk behaviour among young men in the rural Eastern Cape, South Africa. AIDS. 2006;20:2107–2114. [PubMed]
54. Jewkes R, Dunkle K, Nduna M, Levin J, Jama N, Khuzwayo N, et al. Factors associated with HIV sero-status in young rural South African women: connections between intimate partner violence and HIV. Int J Epidemiol. 2006;35:1461–1468. [PubMed]
55. Wooldridge JM. Econometric analysis of cross section and panel data. MIT Press; Cambridge, MA: 2001.
56. Vella F. Estimating models with sample selection bias: a survey. The Journal of Human Resources. 1998;33:127–169.
57. Shisana O, Rehle T, Simbayi LC, Parker W, Zuma K, Bhana A, et al. South African National HIV prevalence, HIV incidence, behaviour and communication survey. HSRC Press; Cape Town: 2005.
58. Greene WH. Econometric analysis. Prentice Hall; Upper Saddle River, NJ: 2002.
59. Heckman JJ. Sample selection bias as a specification error. Econometrica. 1979;47:153–161.
60. Ahn HT, Powell JL. Semiparametric estimation of censored selectin models with nonparametric selection mechanism. Journal of Econometrics. 1993;58:3–29.
61. Fenton L. Preventing HIV/AIDS through poverty reduction: the only sustainable solution? Lancet. 2004;364:1186–1187. [PubMed]