Hello,
I'm trying to estimate the sample size I need for my data. I'm just starting to learn Structural Equation Modeling with Amos.
I have pilot data for about ~150 participants. I want to use this to calculate the sample size I need for 80% power at the standard 5% confidence interval.
My model has 6 exogenous variables (individual differences). My hypothesized model is that these 6 would predict 3 different types of enjoyment (I don't expect any latent variable here because it's logically impossible for these 3 distinct situations to occur at the same time). Then these 3 types of enjoyment would predict overall enjoyment.
Are there any step by step guides on how to calculate a sample size from pilot data like this? How would I calculate this? I have G* Power and can calculate sample sizes for regressions using "rules of thumb" for expected effect sizes, but am having trouble fifuring out how to do it for SEM.
Hi Steve,
the most straightford approach would be a simulation study that you can easily do with R
Muthén, L. K., & Muthén, B. O. (2002). How to use a monte carlo study to decide on sample size and determine power. Structural Equation Modeling, 9(4), 599-620. doi:10.1207/S15328007SEM0904_8
Beaujean, A. A. (2014). Sample size determination for regression models using Monte Carlo methods in R. Practical Assessment, Research & Evaluation, 19(12), 1-16.
I further attached a small but nice tutorial (besides the paper by Beaujean)
Best,
Holger
Rules of thumb are problematic because they assume a certain effect size and a certain degree of collinearity. As neither of those assumptions is likely to be true for your problem the rule of the dumb is not going to be particularly useful.
Either simulate as suggested by Holger or reduce the problem to simple regression effect in GPower for the most important effect or effects in the model. Select the size of effect you'd like to be able to detect. Then adjust n based on your expected dropout/attrition and degree of collinearity. Vary the input parameters to see how sensitive n is to the inputs. The latter approach is fairly crude but is at least driven by relevant information from your study.
- Badges
- Science topic
- Similar topics
- Psychology
- Mental Processes
- Learning