06 / RESEARCH PLAN FREE BROWSER TOOL
SEM sample size calculator
Estimate a lower-bound sample size for a structural equation model from your effect size, power, significance level and model dimensions.
Adjust the inputs
Replace the sample figures with your own assumptions.
Your assumptions
Calculations run in this tab. Your entries are not sent to a server.
Live estimate
LOCALSEM sample size lower bound
Method lower bound only. It does not establish model identification, convergence, or adequate power for your estimator and data.
Calculation notes
What this number means
The result takes the larger of two lower bounds from Westland (2010): a model-structure bound based on the observed-to-latent variable ratio, and an effect-detection bound based on the Gini-correlation estimator variance for paired latent variables. For the a-priori calculation, the entered minimum absolute effect is used as the modeled correlation; the significance level is Sidak-adjusted across k(k−1)/2 possible latent-variable pairs. These bounds are necessary conditions, not a guarantee of adequate power or model convergence. Estimator, identification, missing data, non-normality and study design can require more observations; use an estimator-specific power analysis or Monte Carlo study for a final design.
Method references: Westland (2010), Lower bounds on sample size in structural equation modeling · equation documentation.
Useful context
Before you plan a study
What does the recommended minimum mean?
It is the larger of a model-structure lower bound and an effect-detection lower bound from the Westland method. It is a planning threshold, not proof that a specific model will fit or converge.
How should I choose the effect size?
Enter the smallest absolute correlation you want the study to detect. A smaller target effect generally requires more observations. Use prior research or a justified design assumption rather than choosing a value to reach a preferred sample size.
Does this guarantee enough participants?
No. The method states necessary lower bounds, and estimator choice, model identification, data distribution, missingness and other design details can increase the sample needed. Consider estimator-specific simulation or Monte Carlo power analysis.
Are my inputs uploaded or saved?
The calculation runs in this browser tab. The tool does not send entered numbers to a calculation server or save them to an account.