This is an example of a power analysis for a simple mediation model with observed variables (no latent variables). This example is for simsem. The Mplus counterpart can be found here:
If you are unsure of what the population regression parameter estimates should be (that is the beta values), and prior work in the literature is not proving helpful, you might consider using Cohen’s (1988) effect size guidelines of 0.10 = small, 0.30 = medium, and 0.50 = large.
For this example you will need to use the package simsem which you can install by using:
Line 3 is where the population model is defined. In this example, the means of the observed variables x, m, and y are fixed to 0 while their variances are fixed to 1.
Starting at line 15, the standardized beta coefficient y ~ m is set to .18, y ~ x is .22 and m ~ x is .20.
The sample model starts at line 19. If you studying power of your model and not the effects of model misspecification then you would likely want to make the sample model the same as the population model. The parameters in the model have been labelled (e.g.: b*m) so that the indirect and total effects can be calculated.
Line 26 is where the simulation is run. This line is requesting 1,000 replications of a sample of n = 150. The summary command then provides you with the results of the power analysis.