With covariance-based SEM, one of the major assumptions is that this technique requires a larger sample size than other statistical techniques. SEM relies on tests which are sensitive to sample size as well as to the magnitude of differences in covariance matrices. There are a litany of suggestions in regards to necessary sample size using SEM. One of the most com- mon suggestions for sample size is Nunnally and Bernstein’s (1994) rule of 10.The rule of 10 states that you should have 10 observations for each indicator in your model. Another rule of thumb, based on Stevens (1996), is to have at least 15 cases per indicator. Bentler and Chou (1987) argue that a more accurate calculation should be based on free parameters of your model where you should have at least 5 cases for each parameter estimate (including error terms as well as path coefficients). Schreiber et al. (2006) made the argument that it should be 10 participants for every parameter estimated. There is no shortage of cites and methods for suggesting how large a sample is needed. More recently, a simple rule of thumb is that a “critical sample size” of 200 (Garver and Mentzer 1999; Hoelter 1983) provides stable parameter estimates and has sufficient power to test a model. While this critical sample size of 200 simplifies things, it fails to address the idea of power. Your sample should be deter- mined by the effect size you desire to capture, or put another way, the ability to capture the smallest correlation between latent variables that you wish to detect.Your sample size should be based on the complexity of your model rather than on the bare minimum sample neces- sary to run the analysis. If you want to understand how to calculate the specific sample size for a desired level of power, see Kim (2005) or McQuitty (2004).
Source: Thakkar, J.J. (2020). “Procedural Steps in Structural Equation Modelling”. In: Structural Equation Modelling. Studies in Systems, Decision and Control, vol 285. Springer, Singapore.