Regression and PLS-SEM Sample Size Guide

Why sample size for regression and PLS-SEM is debated

There is no single correct answer. Researchers, supervisors and journal reviewers use different rules, and each rule makes different assumptions. This guide shows the most cited approaches together, so you can pick a defensible target and report how you reached it.

The methods compared

• Green (1991) rules: 50 + 8k for testing the overall regression model, and 104 + k for testing individual predictors, where k is the number of predictors.
• 10-times rule: a sample at least ten times the largest number of arrows pointing at any construct. It is popular in PLS-SEM but criticised as too simple.
• Inverse square root method: based on the smallest path coefficient you want to detect. At 5% significance and 80% power the minimum is (2.486 / p_min)².
• Power analysis: uses Cohen’s f² effect size and the F distribution to find the sample that gives 80% power, the same logic as G*Power.

How to choose

Take the largest value from the methods that fit your design, add a margin for non-response, and state the method in your methods section. For PLS-SEM, k is the maximum number of structural arrows pointing at one construct, not the total number of indicators. For regression, k is the number of predictors in the model.

Worked example

With five predictors, a medium effect size (f² = 0.15) and a smallest expected path coefficient of 0.21, the Green rules give 90 and 109 cases, the 10-times rule gives 50, the power analysis gives about 91, and the inverse square root method gives 141. A safe target is therefore around 141, or about 166 once you allow for 15% non-response.