Mixture regression for observational data, with application to functional regression models (1307.0170v1)

Abstract: In a regression analysis, suppose we suspect that there are several heterogeneous groups in the population that a sample represents. Mixture regression models have been applied to address such problems. By modeling the conditional distribution of the response given the covariate as a mixture, the sample can be clustered into groups and the individual regression models for the groups can be estimated simultaneously. This approach treats the covariate as deterministic so that the covariate carries no information as to which group the subject is likely to belong to. Although this assumption may be reasonable in experiments where the covariate is completely determined by the experimenter, in observational data the covariate may behave differently across the groups. Thus the model should also incorporate the heterogeneity of the covariate, which allows us to estimate the membership of the subject from the covariate. In this paper, we consider a mixture regression model where the joint distribution of the response and the covariate is modeled as a mixture. Given a new observation of the covariate, this approach allows us to compute the posterior probabilities that the subject belongs to each group. Using these posterior probabilities, the prediction of the response can adaptively use the covariate. We introduce an inference procedure for this approach and show its properties concerning estimation and prediction. The model is explored for the functional covariate as well as the multivariate covariate. We present a real-data example where our approach outperforms the traditional approach, using the well-analyzed Berkeley growth study data.