The article discusses the importance of specifying the random effects structure in linear mixed-effects models (LMEMs) for confirmatory hypothesis testing in psycholinguistics. The authors argue that LMEMs should include the maximal random effects structure justified by the experimental design to ensure the best generalization of results. They highlight that the generalization performance of LMEMs with data-driven random effects structures depends on modeling criteria and sample size, often yielding reasonable results on moderately-sized samples but with little or no power advantage over maximal models. The article also emphasizes that random-intercepts-only LMEMs, especially on within-subjects and within-items data, generally generalize worse than separate F1 and F2 tests. The authors conclude that maximal LMEMs should be the 'gold standard' for confirmatory hypothesis testing in psycholinguistics and related fields. They provide theoretical arguments and Monte Carlo simulations to support their claims and discuss the implications for research practice.The article discusses the importance of specifying the random effects structure in linear mixed-effects models (LMEMs) for confirmatory hypothesis testing in psycholinguistics. The authors argue that LMEMs should include the maximal random effects structure justified by the experimental design to ensure the best generalization of results. They highlight that the generalization performance of LMEMs with data-driven random effects structures depends on modeling criteria and sample size, often yielding reasonable results on moderately-sized samples but with little or no power advantage over maximal models. The article also emphasizes that random-intercepts-only LMEMs, especially on within-subjects and within-items data, generally generalize worse than separate F1 and F2 tests. The authors conclude that maximal LMEMs should be the 'gold standard' for confirmatory hypothesis testing in psycholinguistics and related fields. They provide theoretical arguments and Monte Carlo simulations to support their claims and discuss the implications for research practice.