Capturing Nonlinearity in Growth Curve Models: A Monte Carlo Comparison of Generalized Additive and Generalized Linear Models

• In Event: Methodological Considerations in Modeling Latent Growth

Abstract

Longitudinal data do not always follow a linear trajectory. Modeling nonlinearity using general linear models (GLM) includes adding higher-order polynomials to the model, either by knowing which polynomials to add a priori, or by running multiple models to find the highest significant polynomial. An alternative method is a generalized additive model (GAM). GAMs estimate the functional form of trend directly from the data, potentially capturing the true functional form better than GLMs. Properly capturing true nonlinearity should result in less biased estimates and better model fit. This paper examines the performance of GAMs compared to GLM via simulation. Results indicate that GAMs fit data better than GLMs—especially when the functional form of the data trend was underestimated by GLM.

Authors

• Kristynn J. Sullivan, University of California - Merced

• Sarah Depaoli, University of California - Merced