In the last years, the relationship between age and the parameters estimated in cognitive process models such as the drift-diffusion model has found increasing attention. In the numerous studies published on the subject, there were convergent but also divergent results. Most importantly, the parameters representing decision caution and non-decisional processes often showed very clear linear age-related patterns, while things were more complicated for the parameters representing speed of evidence accumulation. In this research project, we present a systematic multiverse analysis to better understand both the robustness and the heterogeneity found in previous studies. Specifically, we analyse how different cognitive model architectures (standard diffusion model, Levy diffusion model, linear ballistic accumulator model), different tasks studied (18 different cognitive tasks), different estimation procedures (maximum likelihood, Bayesian hierarchical, Bayesian neural-network based), and different data cleaning procedures influence the results found. Our findings thus help us make sense of the underlying dynamics of choices in modeling when utilizing cognitive process models to better understand experimental setups and individual differences.