Geneticists seeking to understand HIV-1 evolution among human hosts generally assume that hosts represent a panmictic population. Social science research demonstrates that the network patterns over which HIV-1 spreads are highly nonrandom, but the effect of these patterns on the genetic diversity of HIV-1 and other sexually transmitted pathogens has yet to be thoroughly examined. In addition, interhost phylogenetic models rarely account explicitly for genetic diversity arising from intrahost dynamics. This study outlines a graph-theoretic framework (exponential random graph modeling, ERGM) for the estimation, inference, and simulation of dynamic partnership networks. This approach is used to simulate HIV-1 transmission and evolution under eight mixing patterns resembling those observed in empirical human populations, while simultaneously incorporating intrahost viral diversity. Models of parametric growth fit panmictic populations well, yielding estimates of total viral effective population on the order of the product of infected host size and intrahost effective viral population size. Populations exhibiting patterns of nonrandom mixing differ more widely in estimates of effective population size they yield, however, and reconstructions of population dynamics can exhibit severe errors if panmixis is assumed. I discuss implications for HIV-1 phylogenetics and the potential for ERGM to provide a general framework for addressing these issues.