Bayesian Convergent Validity
Bayesian Convergent Validity Assessment · Also known as: Bayesian convergent validity analysis, Bayesian MTMM convergent validity, Bayesian multitrait convergent validity, BCV
Bayesian convergent validity applies Bayesian statistical inference to assess whether different measures of the same construct converge as theory predicts. Rather than a single-point correlation estimate, it yields a full posterior distribution over the convergent correlation, enabling probability statements about the magnitude of shared variance between theoretically related measures.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Bayesian convergent validity when you want probability-based conclusions about construct-measure relationships rather than binary significance decisions, when sample size is small or moderate (n < 200) making frequentist standard errors unstable, or when prior validation data from comparable samples can be formally incorporated. It is particularly suited to cross-cultural scale adaptation where prior population-specific loadings can serve as informative priors. Avoid it when reviewers or journals require classical NHST-only reporting, when you lack the computational expertise to diagnose MCMC convergence (trace plots, R-hat statistics), or when the construct space is poorly defined and prior specification would be essentially arbitrary.
Strengths & limitations
- Produces full posterior distributions over convergent correlations, enabling direct probability statements about validity magnitude.
- Naturally incorporates prior evidence from previous validation studies, reducing reliance on a single study.
- Performs better than frequentist alternatives in small samples where ML-based CFA may produce inadmissible solutions.
- Credible intervals have the intuitive interpretation that practitioners often mistakenly assign to frequentist confidence intervals.
- Simultaneously estimates the entire measurement model — loadings, residuals, and construct correlations — in one coherent framework.
- Handles missing data naturally through data augmentation without listwise deletion.
- Requires specification of prior distributions; poorly chosen priors can distort posteriors, especially in small samples.
- Computationally intensive — MCMC chains may require hours to converge on complex models.
- MCMC diagnostics (R-hat, effective sample size, trace plots) add an extra layer of technical expertise beyond standard SEM.
- Results depend on prior choice, which introduces a degree of subjectivity that critics of Bayesian methods may question.
- Software options (Stan, JAGS, brms) have steeper learning curves than point-and-click CFA tools.
Frequently asked
How is Bayesian convergent validity different from classical convergent validity with CFA?
Classical CFA uses maximum likelihood estimation and reports point estimates with standard errors and p-values. Bayesian convergent validity uses MCMC to sample the full posterior distribution over model parameters, producing credible intervals and direct probability statements — for example 'there is a 92% posterior probability that the convergent correlation exceeds 0.50.' The inferential logic and outputs differ substantially even though both fit the same underlying factor model.
What software is typically used?
Stan (via the RStan or CmdStanR interfaces) and JAGS are the most common MCMC backends. The R package brms provides a higher-level interface to Stan and can fit Bayesian factor models with relatively concise syntax. Mplus also offers Bayesian CFA estimation since version 7.
How do I choose priors for the factor loadings?
A common weakly informative choice is a normal prior centered at 0.5 with a standard deviation of 0.2 for standardized loadings, reflecting the expectation of moderate-to-strong positive indicator relationships. If previous validation studies are available, empirical Bayes or formal elicitation can yield more informative priors. Always perform a prior sensitivity analysis by re-running the model with slightly different priors to check whether conclusions change.
How many MCMC samples do I need?
A minimum of 4 chains with 2,000 warm-up and 2,000 sampling iterations each is a standard starting point. Check that R-hat is below 1.01 and effective sample size exceeds 400 for all key parameters. Complex models or strong correlations among parameters may require more iterations.
Does Bayesian convergent validity also assess discriminant validity?
Not automatically. The same Bayesian CFA model that yields convergent correlations also yields between-construct correlations (discriminant correlations), but you must explicitly contrast them — for instance by computing the posterior probability that the convergent correlation exceeds the discriminant correlation. Doing both analyses together is strongly recommended.
Sources
- Levy, R. & Mislevy, R. J. (2016). Bayesian Psychometric Modeling. CRC Press. ISBN: 978-1466500952
- Van de Schoot, R., Depaoli, S., King, R., Kramer, B., Märtens, K., Tadesse, M. G., Vannucci, M., Gelman, A., Veen, D., Willemsen, J. & Yau, C. (2021). Bayesian statistics and modelling. Nature Reviews Methods Primers, 1(1), 1. DOI: 10.1038/s43586-020-00001-2 ↗
How to cite this page
ScholarGate. (2026, June 3). Bayesian Convergent Validity Assessment. ScholarGate. https://scholargate.app/en/psychometrics/bayesian-convergent-validity
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Bayesian Confirmatory Factor AnalysisPsychometrics↔ compare
- Bayesian Measurement InvariancePsychometrics↔ compare
- Confirmatory factor analysisPsychometrics↔ compare
- Construct ValidityPsychometrics↔ compare
- Convergent ValidityPsychometrics↔ compare
- Discriminant ValidityPsychometrics↔ compare