Comparing empirical likelihood and bootstrap hypothesis tests (Q1340294)

The aim of this paper is to compare the powers of empirical likelihood and bootstrap tests by developing expansions for their powers. In Section 2 we give definitions for empirical likelihood and bootstrap tests. After developing higher order expansions for the power functions in Section 3, we propose in Section 4 two rules for practically choosing between empirical likelihood and bootstrap tests for univariate and bivariate cases. In the univariate case, the rule says that the empirical likelihood test is more powerful than the corresponding bootstrap test when \(\tau\alpha_ 3 >0\), and vice versa when \(\tau\alpha_ 3 <0\), where \(\alpha_ 3\) is the population skewness parameter. For higher dimensional cases, similar rules may be developed. In Section 5 we present simulation studies.