Small Sample Parameter Recovery in Sigmoidal Structured Latent Curve Models: A Simulation Study

• In Event: Marriott Roundtable Session 73
  In Roundtable Session: Precision, Power, and Small Sample Size Issues in Growth Curve Modeling Variations

Abstract

This simulation study examined sigmoidal structured latent curve models under conditions with small sample sizes and large numbers of repeated measures to provide evidence of acceptable parameter estimate bias for a future applied study. Sigmoidal models using the generalized logistic function were generated using the first order Taylor series approximation method detailed in Browne and du Toit (1991). Convergence rates, parameter estimate bias, and standard error bias were examined. Manipulated design factors for this study include sample size (25, 50, 75), the number of repeated measures (18, 36, 52), and the amount of growth measured (70%, 80%, 90%). Results and implications of the findings are discussed.

Authors

• Kevin Wells, Baylor University

• Tracey Nicole Sulak, Baylor University