Abstract:

The empirical likelihood inference for the parameter of interest in linear model with missing covariates is considered. Using the inverse probability-weighted method, the estimating function for the unknown parameter is constructed. The empirical log-likelihood ratio based on our estimating function is proved, under some suitable conditions, to be a standard chisquare distribution asymptotically. With this result,  the confidence region for the parameter can be constructed. At last,  our result is further verified through some numerical simulations.

Key words:  linear model; estimating function; empirical likelihood; confidence region