Sample Size and Model Fit Indices for Structural Equation Modelling (SEM): The Case of Construction Management Research

Structural equation modelling (SEM) has been accepted as a useful technique for testing and modelling the inherently complex and multi-dimensional relationships among a number of variables that cannot be directly observed. SEM requires "sufficient" sample size to be able to produce plausible results. Despite the prescription of sample sizes between 100 and 400 as being large enough, there exists some level of inconsistency between the concomitant sample sizes used in construction management research. Available evidence from the literature shows that most construction management research used less sample sizes than the recommended minimum. The aim of this study is to understand the reasons for the inconsistencies in sample size and how they affect the validity of the results obtained in construction management research. This study, based on extensive literature review, obtained information on sample sizes in relation to model parameters, and model fit indices. Key among the findings is that some of the fit indices are directly proportional to sample size. Therefore, there is the risk of accepting invalid models with very "small" sample sizes. Similarly, there is the risk of rejecting valid models with very "large" sample sizes. The paper surmises that the impact of sample sizes on the model fit indices depends on the complexity of the specified theoretical model. It is, therefore, recommended that researchers should pay more attention to relevant theory and model complexity than to the prescriptive numbers in determining the minimum and maximum sample sizes required for their research.