Comparing Two Independent Populations Using a Test Based on Empirical Likelihood and Trimmed Means

We develop a new robust empirical likelihood-based test for comparing the trimmed means of two independent populations. The simulation results indicate that the test has asymptotically correct level under various data distributions and controls the Type I error adequately for medium-size samples. For nonnormal data distributions, the power of the test is comparable to robust alternatives like Yuen's test for the trimmed means and considerably exceeds that of the tests based on the means. In small sample settings the test version for the difference of 10% trimmed means exhibits robustness to the combined presence of nonnormality and heterogeneity.