Sensitivity Analysis for Pearson-Lawley Corrections in the Context ofNonignorable Missingness

摘要

We consider a model sensitivity problem of a dependent variable on severalexogenous variables while the dependent variable has some missing data. Under certain assumptions on the model of selected sample and on the selection mechanism, a mixture model is derived and some statistical properties are discussed. This model gives a way to derive Pearson Lawley (PL) correction formula for the covariance matrix and leads to a modification when the missingness is not ignorable. A sensitivity analysis is then discussed for the PL method. Finally, this modified PL method is applied to a real data set from Project A of Office of Naval Research. The results show some difference from that of using Pearson-Lawley method or of using listwise deletion.