Reviews: Robust Hypothesis Test for Nonlinear Effect with Gaussian Processes

The paper proposes a statistical test for particular non-linear effects in a linear mixed model (LMM). The problem of testing non-linear effects is relevant, especially in the natural sciences. The experimental validation has its flaws, but may be considered acceptable for a conference paper. The method consists of multiple parts: 1) The main new idea introduced in the paper is to introduce a kernel parameter (garotte) that interpolates between a null model and the desired alternative model and to perform a score test on this parameter. This elegant new idea is combined with several established steps to obtain the final testing procedure: 2) Defining a score statistic and deriving an approximate null distribution for the statistic based on the Satterthwaite approximation.