Structural Equation Modeling (SEM) Discussion > Sample Size and Power Analysis
This is a great question, Kristin. You can estimate power for each prediction in your model by simply using a power calculator, such as G*Power (http://www.psycho.uni-duesseldorf.de/abteilungen/aap/gpower3/).
Alternatively, if you want to try to estimate power for the entire SEM model, you will need to count the parameters and then multiply by a "rule of thumb" (I've heard anywhere from 5 to 30, depending on how conservative the source). Examples of what is considered a parameter include: regression pathways, error terms, covariances...etc.
I recently found a really useful resource for calculating power in SEM, based on which fit statistic you will be examining:
Jeremy Taylor
Hi Jeremy,
You say that you can use G*Power to estimate the power of each prediction, when you are in G*Power which 'Test Family' and 'Statistical Test' would you use? I cannot see anything related to SEM.
Many thanks
Alan
Hi
G*Power and Hair et al. 10 times rule are no longer valid for estimating sample size requirements
Please see
https://www.scribd.com/document/370278672/Kock-2018-Sample-Size
Jim
Hello!-I know that the recommended minimum has been said to be 200. I am running an SEM model with my total sample of 656 people which I know is a good size. However I am also looking at gender differences between men and women, but my men sample is only 150. Are there references out there that say that this would be an adequate sample size?
Also, how can you do a power analysis in SEM--perhaps this would help with defending my sample size?
Thanks in advance for any help with this!