Path Analysis With Nonnormal Continuous Data: A Monte Carlo Simulation

• In Event: When Sharp Statistics Meet Complex Data

Abstract

Our study evaluated the impact of nonnormality, estimation method, and sample size on the fit indices, power, and standard errors of parameter estimates in path model with continuous data. Data were simulated from a five-variable path model with different degrees of skewness and kurtosis. We found that standard error bias was the smallest when all observed variables were normal. For nonnormal data, the robust maximum likelihood (MLR) and maximum likelihood with mean-adjusted (MLM) estimators produced smaller standard error bias. Sample size and data configuration appeared to be significant drivers of the standardized root mean squared residual (SRMR) and root mean square error of approximation (RMSEA) fit indices. Finally, statistical power was higher for configurations with normal or moderately nonnormal data.

Authors

• Nnamdi Chika Ezike, University of Arkansas at Fayetteville

• Chunhua Cao, The University of Alabama

• Xinya Liang, University of Arkansas

• Wen-Juo Lo, University of Arkansas

• Paula Castro Abbott, University of Arkansas at Fayetteville

• Ji Li, University of Arkansas for Medical Sciences