Latent vs. Manifest Variable Analyses to Test Group Differences in a Two-Group Randomized Pretest-Posttest Design

Two-group randomized pretest-posttest designs commonly measure multiple items believed to reflect the same underlying latent construct. To analyze such data, modern methods like the Latent Change Score Model (LCSM) utilize latent variables, while traditional methods analyze manifest composite scores. In a Monte Carlo experiment, we compared the LCSM on item scores to Welch’s <i>t</i>-test applied to composite gain scores. We simulated multiple congeneric items at pretest and posttest in a control group (no average change) and a treatment group (positive treatment effect). We examined bias, efficiency, Type I error rates, and power to detect group differences in mean change. Although the LCSM showed slightly higher Type I error rates and reduced efficiency in a few conditions, it consistently produced unbiased estimates. In contrast, the Welch’s <i>t</i>-test estimates became more biased as effect size increased, and this test at times demonstrated lower power. Based on our findings, we offer practical recommendations.