Data collection procedures have been published elsewhere . Briefly, data were collected by visiting households and conducting face-to-face interviews to obtain information on demographic characteristics, wealth, anthropometry, female genital cutting, HIV knowledge, and sexual behaviour between March and August 2003.
For the present study, 6362 never, currently, or formerly married women, all of whom who have had at least one episode of sexual intercourse in their lifetime, were drawn from the overall sample of 7620 women to examine factors associated with "high-risk" sexual behaviour. Thus, if a woman has never had sexual intercourse in her life (e.g., a "virgin"), she was not included in the analysis. "High-risk" sexual behaviour was defined as having two or more sex partners in the last 12 months.
The correlates of sexual behaviour were selected by reviewing the literature and grouped into individual- and community-level factors.
Age was categorized into three groups: 15–24 years, 25–34 years, and 35 years or older.
Education attainment was grouped into three bands: never been to school, primary, and secondary or higher education.
Wealth index a score was attributed to each household amenity and the total score constituted the wealth index score. We divided this score into three classes of wealth: poorest (below 20th quantile), middle (between 20th and 80th quantile), and richest (above 80th quantile).
Past alcohol use: past alcohol use was defined as number of days drank alcohol last three months (never, one day, two days, three or more days)
Marital status: grouped into never, currently, or formerly married.
Religion: respondents' religion was stratified into Christian, Muslim, and others
Community economic status is an average wealth index at weight index at community level. Economic inequality is measured by dividing community wealth index into three equal quintiles.
Place of residence was defined as rural or urban
Geographic region 1) North central, 2) North East, 3) North west, 4) South East, 5) South south, and 6) South west
The descriptive statistics show the distribution of respondents by the key variables. Values were expressed as absolute number (percentages) and mean (standard deviation) for categorical and continuous variables respectively.
Given the hierarchical structure of the sample and the binary outcome, a logistic multilevel modelling approach was adopted. A two-level model with a binary response (y, whether the respondent had extramarital sex in the past 12 months or not) for a woman i living in community j of the form:
The probability was related to a set of categorical predictors, X; and a random effect for each level, by a logit-link function as
) = log [π
/(1 - π
)] = β
0 + βX
The linear predictor on the right-hand side of the equation consisted of a fixed part (β
0 + β X
) estimating the conditional coefficients for the explanatory variables, and random intercept attributable to communities (u
The analysis was done in three steps. In Model 1 (empty model), no explanatory variable was included. In model 2, only individual-level factors were included. Model 3 we controlled for both individual and community-level factors. The results of fixed effects (measures of association) were shown as odds ratios (ORs) with 95% confidence intervals (CIs). The results of random effects (measures of variation) were presented as variance partition coefficient and percentage change in variance.
The MLwiN software, version 2.0.2, was used for the analyses. Parameters were estimated using the Markov Chain Monte Carlo (MCMC) procedure. The default settings in MLwiN were used for the analyses, i.e., chains of length 5000 after a burn-in of 500. The Deviance Information Criterion (DIC) was used as a measure of how well our different models fitted the data. A lower value on DIC indicates a better fit of the model.