There is no universal sample size that makes every SEM defensible. Required sample size varies with model complexity, indicator quality, effect sizes, missing data, estimator, distribution and the performance criterion being protected.
Build the planning record
- Specify the model. Record latent variables, observed indicators, paths, groups, repeated measures and free parameters.
- State the smallest effect that matters. Sample-size planning should reflect the effects or parameters the study needs to estimate or detect.
- Describe measurement quality. Factor loadings, indicator reliability and model identification affect estimation performance.
- Allow for data conditions. Missingness, non-normality, categorical indicators and attrition can increase information requirements.
- Choose the criterion. Power, bias, coverage, convergence and improper solutions are different planning targets.
- Use model-specific analysis or simulation where possible. Preserve assumptions and sensitivity ranges rather than reporting one unexplained number.
Why fixed thresholds fail
Simulation research shows that sample-size requirements change substantially as model characteristics change. Lower-bound approaches also depend on model structure, anticipated effects, power and significance assumptions. A threshold such as 100, 200 or ten cases per parameter may be a rough screening reference, but it is not a general design justification.
Worked use case
Two studies both propose N=200. One has a simple, well-measured two-factor model; the other has weak loadings, many parameters, missing data and a small target path. The same N does not imply the same power, bias or convergence risk. Record the assumptions and evaluate each model separately.
Evidence boundary
A planning calculator is not a guarantee that the fitted model will converge or be correct. Final adequacy still depends on data quality, model specification, assumptions, estimator behaviour and transparent reporting of uncertainty.
Primary research
- Wolf, E. J., Harrington, K. M., Clark, S. L., & Miller, M. W. (2013). Sample Size Requirements for Structural Equation Models: An Evaluation of Power, Bias, and Solution Propriety.
- Westland, J. C. (2010). Lower bounds on sample size in structural equation modeling.