Speaker: Hao Sun
Abstract: Many large-scale surveys collect response variables including both discrete and continuous variables. The joint model of the variables can be viewed as a graphical model where each node denotes one of the variables and each edge shows the correlation between the pair of corresponding nodes. We call the model as survey graphical model (SGM) because the sample data are sampled under a survey sample design. We fit a survey regression model with lasso to reduce the number of edges in the graph. Our simulation study shows that the algorithm performs well to detect the true edges based on ROC curves. Besides, we show that the inclusion probability is important in SGM and neglecting it could result in a bad performance.