Over 30 years into the HIV epidemic there is still little consensus as to what drives the generalized HIV epidemics in sub-Saharan Africa 
. The large differences in HIV prevalence between the various races in South Africa offer a useful standpoint from which to investigate putative risk factors. South Africa has conducted three nationally representative HIV serosurveys that include 15–49 years olds. In 2004, the HIV prevalence in 15–49 year olds was 19.9% (95% CI, 18.1–21.4) in blacks, 3.2% (95% CI, 2.1–4.3) in coloureds, 0.9% (95% CI, 0.08–1.7) in Indians and 0.5% (95% CI, 0.1–0.9) in whites. HIV prevalences by race vary to a similar degree in the other two surveys conducted in 2001 
and 2007 
as well as in a nationally representative sample of 15–24 year olds 
, a national survey of tertiary students 
, a survey of company employees 
and the country's annual antenatal surveys 
. Controlling for various socioeconomic variables makes little or no difference to the differences in HIV prevalence by race 
. An example is provided by a multivariate analysis of the 2004 HIV survey. When education and socioeconomic status are controlled for, being black remains the strongest factor associated with testing HIV positive – the odds ratios varying from 7.9 (95% CI, 4.3–14.5) to 8.7 (95% CI, 5.1–14.8) in the men and women only models respectively 
There has only been one published study that has attempted to systematically explore the risk factors which co-vary with HIV prevalence by race in South Africa 
. This study found that the individual-level risk factors such as number of sex partners in the last 12 months, condom usage and circumcision prevalence did not co-vary with HIV prevalence by race. The prevalences of partner and respondent concurrency, both network-level properties, were however found to differ considerably between the different racial groups and to do so in a way which mirrored the differences in HIV prevalence. This study was limited to 14–22 year olds in the city of Cape Town. The current study extends this analysis to include five nationally representative samples of 15–49 year olds.
Its findings concur to some extent with the Cape Town study. The Indian and white groups are both numerically small and have similarly low HIV prevalences. If we consider them together as the low HIV prevalence groups, then the risk factors which co-vary with HIV prevalence by race in the six surveys are age of sexual debut (in five out of five surveys for men and three out of six surveys for women), age gap (zero surveys in men and three in women), mean number of sex partners in the previous year (five surveys in men and one in women) and respondent concurrency (five surveys in men and one in women). Condom usage and circumcision were both more prevalent in the high HIV prevalence groups. There was little evidence of difference in the prevalence of those who had had sex. The survey which demonstrated the largest difference in this variable was the CAPS. This was likely due to the fact that it was the only survey which was limited to younger persons. In four of the five surveys where people up to the age of 49 were included, there were no differences in sexual experience.
Which of the co-varying risk factors could be responsible for the large differential HIV spread by race? Age of sexual debut, by itself, is an unlikely candidate. This is for a number of reasons, including the fact that the age of sexual debut in the highest HIV prevalence groups is higher than that in the very low prevalence countries of the USA 
and Western Europe 
A number of publications have argued that age-mixing plays a significant role in HIV spread in sub-Saharan Africa 
. Age-mixing, whilst of likely importance, cannot without an interconnected sexual network result in a generalized HIV epidemic. This is evident if we consider a hypothetical population where there is an age gap of 10 years in all couples but the couples practice exclusive lifetime monogamy. Purely sexually transmitted infections cannot spread in this population despite extreme age-mixing since STI spread depends on an interconnected sexual network 
. Factors such as age-mixing are, however, likely to influence transmission across an interconnected network, particularly to new cohorts of younger persons 
. The fact that, in three out of five surveys of women, the prevalence of age-discordant coupling co-varied with HIV prevalence may be indicative of age-mixing having an influence on HIV prevalence.
This analysis finds evidence of a covariance between concurrency and HIV prevalence. Higher prevalences of sexual partner concurrency have been shown to lead to exponential increases in the degree of network connectivity and thereby the potential for HIV transmission 
. Although certain studies have not found an association between HIV and concurrency 
, a number of good modeling-based and empirical studies have shown that concurrency prevalence covaries closely with HIV prevalence inter- and intra-nationally 
and that it is a key driver of incident HIV at a partnership level in sub-Saharan Africa 
. In particular declines in concurrency have been shown to be important in the rapid decline of HIV incidence in Zimbabwe, Uganda and elsewhere 
. Amongst the women, the prevalence of concurrency was only higher in the blacks in one of five surveys – the CAPS survey. Finding lower prevalence of concurrency in women compared to men is a common result of surveys in Southern Africa and further afield 
. This may reflect a combination of a lower prevalence of concurrency 
and a differential male-female courtesy bias induced by the fact that concurrent partnering is considerably more stigmatized for women than men in many communities 
. The importance of a courtesy bias in this regard is suggested by studies in Southern Africa that found that changes in the ways that surveys are conducted and the ways questions are asked, can lead to a significant increase in the measured prevalence of concurrency in women 
. Even in the likely scenario that women are less likely to have concurrent partners than men, concurrency can still lead to extensive HIV spread in women. This is for two main reasons. Firstly, at an individual level, concurrency acts to increase the risk of HIV to the partners of the individual engaging in concurrency rather than to the individual him or herself 
. Secondly, and most importantly, concurrency's major impact on HIV transmission operates by connecting together a large proportion of the population into a transmission pathway for HIV 
. This is a network level effect and thus would be experienced by all members of the connected-network (both men and women).
The total number of sexual partners, though important, is less likely to be a crucial factor for a number of reasons. Firstly, the lifetime number of partners is no higher in sub-Saharan African countries with generalized HIV epidemics than the USA 
and other low HIV prevalence countries 
. Secondly, the available evidence suggests that much of the higher number of sexual partners in the past year amongst the black group represents long-term concurrent partnering 
. This is supported by the fact that this analysis could find no evidence of a difference in the total life-time number of partners between the races.
Other comparative studies of sexual behavior in sub-Saharan Africa have reached different conclusions. In a comparative study of sexual behavior of 18–24 year olds in the USA and South Africa, Pettifor et al, found that three out of four risk factors assessed (age of sex-debut, lifetime number of sex partners and lack of condom usage) were more prevalent in the USA (HIV prevalence <1%) than South Africa (HIV prevalence 10.2%) 
. Only age-mixing was more prevalent among South African women. Pettifor et al, reached the conclusion that “unique biological forces” must be important factors in explaining the more extensive spread of HIV in South Africa. Of note, this study did not assess differences in concurrency prevalence