Rank Regression Inference via Empirical Likelihood
Bishop, E. (2005, August). Rank Regression Inference via Empirical Likelihood. Presented at Joint Statistical Meetings, Minneapolis, MN.
Rank regression has been developed as an alternative semi-parametric method for statistical analysis when the assumptions of parametric methods are not sufficiently met, such as normality of the residuals or constant variance. This paper will explore estimating beta by minimizing the dispersion function. This paper develops confidence intervals using an empirical likelihood (EL) ratio method and presents coverage probabilities. EL has the advantage of not requiring variance estimation which is required for the normal approximation method. Simulation studies are used to compare and evaluate normal approximation versus EL inference methods for various conditions such as sample size or error distribution. Simulation results reveal conditions when the EL method results in a coverage probability closer to the true significance level than the normal approximation method. An application of stability analysis also shows the EL method to result in shorter confidence intervals for real life data.