Statistical Models for Sexual Networks on Likoma Island, Malawi: Implications for Sexual Behavior and HIV Control

James H. Jones, Stanford University
Stephane Helleringer, University of Pennsylvania

Despite its importance, little research has addressed structural factors of sexual networks contributing to the generalized epidemic in Sub-Saharan Africa. We use sociocentric network data derived from a population-based survey of sexual partnerships conducted in Likoma Island (Malawi) and fit exponential random graph models (ERGMs) to a one-year sexual network, comprising 1463 individuals, from Likoma Island. The network is sparse, with a network density of 0.001. A simple structural ERGM incorporating edge effects and dyad-wise shared partnerships (DSPs) indicates that the log-odds of observing a new tie in the network are approximately -5.675, reflecting the sparseness of the observed network. Also significant is the coefficient on the DSP(1) term, with a log-odds of a new tie creating a DSP(1) of -0.519, indicating that a tie added to the sexual network is less likely to create a shared partnership than tie added to a random graph. This has important implications for the formation of sexual networks on Likoma.