Determining an Appropriate Sample Size and Power of Test with Strength Degree of the Relationship between Variables in the Structural Equation Modeling

Abstract

When researchers apply structural equation modeling (SEM) technique, sample size issue is involved. The researchers are uncertain about how to calculate sample size and how many subjects for high confidence in the accuracy of the research. This paper aims to highlight the key points that should be considered when calculating sample size applying structural equation modeling techniques. The paper demonstrates the results of testing population parameters fewer than 7 different sample sizes conditions and determines the 3 strength degrees of correlation between variables to find accuracy sample size that power of test exceeded 0.80. The study founded that (1) an appropriate sample size for the measurement models with moderate and high correlation between latent variables was at least 150 samples, (2) the structural equation modeling without mediate latent variables and the relationship between exogenous variables was moderate; the appropriate sample size was not less than k(k+3)/2, where k was the number of observable variables, and (3) the structural equation modeling with mediate latent variables and the relationship between exogenous latent variables were at moderate and high level, the appropriate sample size was not less than 20 subjects per observed variables.