As computers have improved, so has the prevalence of simulation studies to explore implications for assumption violations and explore statistical power. The simglm package allows for flexible simulation of general(ized) linear mixed models (multilevel models) under cross-sectional or longitudinal frameworks. In addition, the package allows for different distributional assumptions to be made such as non-normal residuals and random effects, missing data, and serial correlation. A power analysis by simulation can also be conducted by specifying a model to be simulated and the number of replications. This package can be useful for instructors or students for courses involving the general(ized) linear mixed model, as well as researchers looking to conduct simulations exploring the impact of assumption violations prior to data collection or grant submission.
The focus of the presentation will be on a discussion of the benefits of a Monte Carlo framework for power analysis and how the simglm package can facilitate this framework for robust power analyses across a wide range of data conditions.
|