Mirror bootstrap method for testing hypotheses of one mean

Abstract

The general philosophy for bootstrap or permutation methods for testing hypotheses is to simulate the variation of the test statistic by generating the sampling distribution which assumes both that the null hypothesis is true, and that the data in the sample is somehow representative of the population. This philosophy is inapplicable for testing hypotheses for a single parameter like the population mean, since the two assumptions are contradictory (e.g., how can we assume both that the mean of the population is $\mu_0,$ and that the individuals in the sample with a mean $M \ne \mu_0$ are representative of the population?). The Mirror Bootstrap resolves that conundrum. The philosophy of the Mirror Bootstrap method for testing hypotheses regarding one population parameter is that we assume both that the null hypothesis is true, and that the individuals in our sample are as representative as they could be without assuming more extreme cases than observed. For example, the Mirror Bootstrap method for testing hypotheses of one mean uses a generated symmetric distribution constructed by reflecting the original sample around the hypothesized population mean $\mu_0$. Simulations of the performance of the Mirror Bootstrap for testing hypotheses of one mean show that, while the method is slightly on the conservative side for very small samples, its validity and power quickly approach that of the widely used t-test. The philosophy of the Mirror Bootstrap is sufficiently general to be adapted for testing hypotheses about other parameters; this exploration is left for future research.