Abstract
Power and precision in confirmatory analyses is a cornerstone for the reliability and replicability of empirical findings. Albeit important, calculating power is not necessarily a trivial task, and may pose a feasibility barrier to scientists. One of these cases is power analysis for linear and generalized linear mixed-effect models (LMMs, GLMMs), a popular and widely used tool in experimental research, including scene perception and visual search. Mixed models are a powerful tool for modelling fixed and random effects simultaneously, but do not offer a general and feasible analytic solution to estimate the probability that a test correctly rejects the null hypothesis. Therefore, a simulation-based approach is necessary. Although a mixture of tools for conducting simulation-based power analyses for mixed-effect models are available, there is a lack of structure and guidance on how to appropriately use them in different scenarios. Therefore, we have developed an openly accessible R package to assist with conducting power analyses for G/LMMs. With this package, we provide code and resources for performing such simulation-based power analyses on openly accessible data sets from cognitive research to illustrate possible practical solutions to transparent sample size and design planning where the effect of adding different units (e.g. participants or trials) can be explored and justified. We also discuss common pitfalls and theoretical constraints that need to be considered when conducting a power analysis for LMMs/GLMMs. Our resources are tailored to cognitive and vision scientists and aim at empowering researchers to set up highly powered research designs when sophisticated analysis procedures like G/LMMs are outlined as inferential procedures.