Abstract
In this paper, the authors highlight several problems associated with conducting measurement validation using pooled experimental data. Beginning with a simple two-variable data set, the authors illustrate that pooling data can bias correlation and alpha coefficients, even when the data exhibit homogeneous covariance structures across treatment cells. They then introduce a data set that includes multiple measures for three latent constructs and extend the examination to include the effect of pooling bias on exploratory and confirmatory factor analysis. Three alternative approaches to conducting measurement validation that control for pooling bias are examined.
