PLS-SEM Measurement Model Lab.
Evaluate reflective measurement quality on a seeded synthetic dataset (n = 180 responses, 2 reflective constructs x 4 indicators). Toggle individual indicators on or off to watch standardised outer loadings, Cronbach's alpha, composite reliability (rho_c) and AVE update live against Hair et al. PLS-SEM primer thresholds (0.70 / 0.70 / 0.50).
Loading interactive measurement laboratory (requires JavaScript ES modules).
Plain-English guide to Hair et al. thresholds.
Reference: Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2022). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) (3rd ed.). Sage.
-
Cronbach's alpha (threshold ≥ 0.70)
Estimates internal consistency reliability assuming all indicators carry equal loadings (tau-equivalence). Because it treats every item equally, a single noisy indicator pulls Cronbach's alpha down sharply, making it a conservative lower bound.
-
Composite reliability rho_c (threshold ≥ 0.70)
Weights each indicator by its actual outer loading (Joreskog's rho). Hair et al. note that true internal consistency typically lies between Cronbach's alpha (lower bound) and composite reliability rho_c (upper bound). Values above 0.95 can warn of redundant items.
-
Average Variance Extracted, AVE (threshold ≥ 0.50)
The mean of the squared outer loadings across all active indicators in a construct. An AVE of 0.50 or higher shows convergent validity: the latent construct explains at least half of the variance in its indicators rather than measurement error.
-
Outer loadings lambda_i (rule of thumb ≥ 0.708)
Because 0.708 squared equals 0.501, an outer loading above 0.708 means the construct explains more than 50% of that indicator's variance. Indicators between 0.40 and 0.70 should only be dropped if removal raises AVE or composite reliability above threshold without harming content validity.
Classroom reflection questions.
- With all four indicators active in Perceived Credibility (CR1 to CR4), why is composite reliability rho_c noticeably higher than Cronbach's alpha?
- Uncheck CR4 and PI4. By how much does AVE improve in each construct, and why should a researcher check the wording of CR4 before deleting it in a real study?
- What happens to reliability if you leave only two strong indicators versus three strong indicators? Why do Hair et al. caution against shrinking a reflective scale to a single item?