Title: An independence test for a nonparametric random effects meta-regression model
Authors: Daniel Gaigall - Leibniz University Hannover (Germany) [presenting]
Abstract: A nonparametric generalization of a random effects meta-regression model frequently used in life sciences is considered. For testing goodness-of-fit for the regression function or testing independence of input and between study variation noise, We apply the Hoeffding-Blum-Kiefer-Rosenblatt independence criterion to pairs of input and residuals. It turns out that the test statistic has the well-known distribution free limiting null distribution of the classical criterion. Related quantiles are available as critical values. Properties of the test statistic under alternatives are pointed out as well. A permutation procedure is a second option to obtain critical values. Simulations investigate size and power of both tests for small and moderate sample sizes. Application to real data from clinical trials illustrates how the tests work in practice.