THE EFFECT OF COLLAPSING ORDERED POLYTOMOUS SCALES ON PARAMETER ESTIMATES IN STRUCTURAL EQUATION MEASUREMENT MODELS

This study explores the influence collapsing ordered polytomous variable scales has on structural equation measurement model parameter estimates. Four parameter estimation procedures (i.e., NTGLS, ADF, POLYCHORIC, and TOBIT) were investigated for their efficacy in recovering simulated structural equation measurement model parameters at three different scale collapse levels. Results from a Monte Carlo study indicate that both NTGLS and ADF procedures performed poorly in the recovery of the estimates, whereas the latent projection procedures (i.e., POLYCHORIC and TOBIT) performed well. Although a slight improvement was noticed in the estimates for the NTGLS and ADF procedures, while collapsing the scales, this result was considered trivial. The latent projection procedures were not affected by changes in scale level.