A Short Note on Interpretation in the Dual Change Score Model

Abstract

The latent change score framework allows for estimating a variety of univariate trajectory models, such as the no change, linear change, exponential forms of change, as well as multivariate trajectory models that allow for coupling between two or more constructs. A particularly attractive feature of these models is that it is easy to decompose and interpret aspects of change. One particularly flexible model, the dual change score model, has two components of change: a proportional change component that depends on scores at the previous time point, and a constant change component that is additive. We demonstrate through simulation and an empirical example that in a correctly specified model, the correlation between the proportional change parameter and the mean of the constant change component can approach either −1 or 1, thus complicating interpretation. We provide recommendations and code to aid researchers’ ability to diagnose this issue in their own data.

Keywords:

• Longitudinal
• latent change score
• latent growth

Notes

¹ This can generally be overcome by changing the control arguments of max_treedepth to 20 and adapt_delta to 0.99. However, this results in slower sampling.