A First Difference Approach to Studying Sequential Panel Data

Historically, Cross-Lagged Panel Models have been used to try and assess causal influences of one variable on another in longitudinal studies involving panel data. However, this model has faced widespread criticism and many alternative methods of analysis have been proposed.

An econometric model proposed in the 80s attempted to use difference scores from timepoints $t_{i-1}, t_i$ to predict difference scores from timepoints $t_i, t_{i+1}$. However, this approach proves problematic with missing data, as one singular missing value can invalidate an entire row of data, if the missing value happens to be the score at time $t_1$. Our goal is to fit this model in a structural equation modeling framework so that latent difference scores can be freely estimated even in the presence of missing data.

Project is ongoing!