The Prevalence of Sexual Partner Concurrency Is Not Correlated with Markers of Poverty or Gender Inequality: An Ecological Analysis ()
Received 21 October 2015; accepted 13 December 2015; published 16 December 2015
1. Introduction
Sexual partner concurrency is believed by many to play an important role in the genesis of the generalized HIV epidemics in sub-Saharan Africa [1] - [5] . The factors underpinning the high concurrency rates in this region have however been hotly debated [6] - [8] . While few would dispute that a combination of socioeconomic [9] [10] , gender-equity-related [10] and cultural factors [11] play a role, the emphasis placed on each differs considerably. Establishing whether or not one of these plays a dominant role has significant programmatic consequences for HIV prevention. A first step in elucidating the factors responsible for high concurrency rates should be assessing the way concurrency and its putative determinants are patterned internationally. This should tell us if the prevalence of concurrency is higher in regions with greater economic or gender inequality for example.
We present an ecological analysis of the correlation between point-concurrency prevalence and various socioeconomic and gender-equity indicators that have been argued to be important in the generation of extensive concurrency.
2. Methodology
2.1. Dependent Variable
To overcome the problems of comparability with different datasets using different methodologies and definitions of concurrency, we used what is to the best of our knowledge, the only multinational surveys which assess concurrency using the same methodology. These are the WHO/Global Programme on AIDS (GPA) sexual behavioural surveys conducted in 1989/1990. Using data from 1990 is advantageous for the purposes of this study as this period predates the marked changes in sexual behavior as a response to the HIV epidemic that have occurred in certain countries [12] - [14] . The surveys were performed in 15 countries but only the 11 countries that asked questions about concurrency are evaluated here [15] [16] . These countries were: Brazil, Central African Republic, Côte d’Ivoire, Kenya, Lesotho, Philippines, Singapore, Sri Lanka, Tanzania, Thailand and Zambia. All these surveys followed WHO/GPA protocols. These included that national probability samples of the general populations aged 15 to 49 should be utilized. In two cases, Manila and Rio de Janeiro, the samples were representative of these large cities rather than being nationally representative. All samples were selected based on probability principle with various designs depending on national factors. A two stage sampling strategy was the norm, with census enumeration areas as the first stage and households as the second. The sample sizes were typically 1000 - 3000 men and women (Table 1). Response rates were high in all cases. The questionnaire started with innocuous questions about the respondents’ backgrounds. This was to establish a rapport before exploring more sensitive issues. The variable for concurrency was derived from the question “Do you now have one or more than one spouse/regular partner?” The dependent variable we used in our analysis was the percentage of all men 15 - 49 years old who had more than one sexual partnership active at the time of the survey. The denominator for this variable was thus the total number of survey participants (Table 1). Only 7 of the countries surveyed asked the women if they had concurrent partners and so we limited our analysis to male concurrency. Concurrency prevalences were natural log transformed to create more normal distributions for regression analyses.
2.2. Independent Variables
Both absolute poverty and economic inequality are plausible drivers of concurrency [9] [17] . We used two indicators to measure the impact of absolute poverty on concurrency rates; national Gross Domestic Product (GDP) per capita from 1989 (expressed in Purchasing Power Parity 1987 US$) and the percentage of the population living below the poverty line. The Gini coefficient was used to assess the impact of economic inequality. Overall development status was measured via the United Nations Development Programme’s (UNDP) Human Development Index (HDI) as well as its three constitutive indices―the GDP, life expectancy and education indices. All these data were obtained from the UNDP’s 1990 Human Development Report [18] . In a few cases the 1990 report did not contain the information pertaining to a particular country and we were forced to obtain the data from the 1995 Human Development Report (see Table 1 for details) [19] .
Gender inequality in relationships has been associated with increased HIV transmission [20] . We compared concurrency rates with two composite indicators of gender equity obtained from the 1995 Human Development Report that focused on “Gender and Human Development”. The gender-related development index (GDI) measures achievement in the same basic 3 capabilities (life expectancy, education outcomes, standard of living) as the Human Development Index does, but takes note of and penalizes inequality in achievement between women and men [19] . The gender empowerment measure (GEM) evaluates the extent to which women are able to actively participate in economic and political life and take part in decision-making. It includes three dimensions: the share of jobs classified as professional, administrative and managerial, the share of seats in parliament and women’s income [19] .
![]()
Table 1. The frequency distribution of various socioeconomic and gender indicators and their correlation with the natural log of point concurrency prevalences in 11 countries.a
HDI―Human Development Index. aAll the data is from the Human Development Report 1990 except the following: 1. Gender Related Development Index, Gender Empowerment Measure and the Human Development Index 1995 are all from the 1995 Human Development Report [19] . 2. The Gini coefficients for Central African Republic (1993), Kenya (2005), Lesotho (2003), Tanzania (2007) and Zambia (2004) are all from the World Bank’s World Development Reports (year of publication in parentheses). Empty cells reflect data was not obtainable from the above sources.
The natural log of point concurrency prevalence was related to the gender and socioeconomic variables through linear regression. The analyses were repeated limited to sub-Saharan African countries (henceforth termed African countries) to assess if there were any region specific effects. Variables that were associated with the natural-log-of-point-concurrency-prevalence in univariate analysis at a significance level of P value of 0.1 or lower were entered into a multivariate linear regression model. Because of a high degree of collinearity between HDI and GDI only GDI was entered into the multivariate analysis. All analyses were performed in STATA 13.
This study involved secondary data analysis of surveys that had each received ethical committee clearance for data analyses such as the one conducted here. No extra ethics committee approval was necessary for this study.
3. Results
The point prevalence of male concurrency and the various socioeconomic indicators is detailed in Table 1. The prevalence of male concurrency varied between 2 and 55%. GDP in 1990 varied between US$405 and US$12 790. There was no evidence of a relationship between the point-prevalence of concurrency and Gini coefficient, percentage of the population living below the poverty line or the Gender Empowerment Measure (Table 1). In the all-country analysis, there was a negative relationship between concurrency prevalence and GDI (r = −0.87; P < 0.001), HDI (1995: r = −0.85; P < 0.001), women’s life expectancy (r = −0.83; P = 0.002) and women’s literacy rate (r = 0.72; P = 0.01).
Within African countries there was a positive association between GDP per capita and the prevalence of concurrency (r = 0.85; P = 0.03). There was no significant association between GDI, HDI, women’s literacy or women’s life expectancy and concurrency.
None of the four variables entered into the multivariable analysis were significantly associated with concurrency prevalence (Table 2).
4. Discussion
Although only based on a sample size of 11 populations, the results of this study do not support the thesis that levels of poverty or inequality (economic or gender-related) are dominant drivers of higher concurrency rates. Concurrency rates were higher in countries with lower GDIs and HDIs. However much of this association appears to be driven by the lower life expectancy in countries in Africa, which also had higher concurrency rates. Life expectancy constitutes one third of the weight of the HDI and GDI. The lower life expectancies in Africa could be due to the effects of HIV or the multiple other factors that have resulted in the lower life expectancy in this region [21] . The finding that concurrency rates were higher in countries with higher GDP per capitas in Africa is unexpected and requires further evaluation in larger datasets.
This analysis is weakened by a number of factors including the small sample-size of only 11 populations, the absence of a concurrency-as-normative-indicator (measuring the acceptability of concurrent partnering), the fact that concurrency prevalence was based on self-reported data from surveys and the ecological nature of the study. Nonetheless the major effect that concurrency has on enhancing HIV transmission is via increasing network connectivity―a population level attribute. Ecological studies are therefore necessary to investigate the drivers and effects of concurrency.
5. Conclusions
I draw two conclusions from this study about the patterning of concurrency at the level of countries. Firstly, there is little evidence for a relationship between concurrency and various markers of relative and absolute poverty or gender inequality. Individual-level [22] and ecological studies [23] in sub-Saharan Africa have likewise found little or no relationship between economic deprivation and concurrency. Secondly, all the countries with point-concurrency above 10% are in sub-Saharan Africa where numerous qualitative studies, particularly in Southern and Eastern Africa, have reported a widespread tolerance for main and side partners for men [11] [24] . Much more can and should be done to address the issues of economic deprivation and gender inequality. The available evidence, however, suggests that high concurrency prevalence rates in parts of Africa are not predominantly determined by socioeconomic or gender-equity related issues. More qualitative and quantitative research is needed to better delineate the socio-cultural factors underpinning the high concurrency rates in the area. In the interim it may be prudent for countries affected by generalized HIV epidemics to follow the example of Uganda where culturally appropriate campaigns to diminish the practice of having partners-on-the-side were instrumental in reducing the prevalence of concurrency and HIV [13] .
![]()
Table 2. Multivariate linear regression analysis for variables associated with log transformed concurrency prevalence.