Selection effects of source of contraceptive supply in an analysis of discontinuation of contraception: multilevel modelling when random effects are correlated with an explanatory variable

Conventional multilevel models assume that the explanatory variables are uncorrelated with the random effects. In some situations, this assumption may be invalid. One such example is the evaluation of a health or social programme that is non-randomly placed and/or in which participation is voluntary. In this case, there may be unobserved factors influencing the placement of the programme and the decision to participate that are correlated with the unobserved factors that influence the outcome of interest. The paper presents an application of a multiprocess multilevel model to assess the difference in rates of discontinuation of contraception between private and Government family planning providers, while accounting for the possibility that there may be unobserved individual and community level factors that influence both a couple's choice of provider and their probability of discontinuation. Copyright 2003 Royal Statistical Society.