babebi

Bayesian Estimation and Validation for Small-N Designs with Rater Bias

Approximate Bayesian inference and Monte Carlo validation for small-N repeated-measures designs with two time points and two raters. The package is intended for applications in which sample size is limited and the observed outcome may be affected by rater-specific bias. User-supplied data are standardised into a common long-format structure. Pre-post effects are analysed using difference scores in a linear model with a rater indicator as covariate. Posterior summaries for the regression coefficients are obtained from a large-sample normal approximation centred at the least-squares estimate with plug-in covariance under a flat improper prior. Evidence for a non-zero pre-post effect, adjusted for rater differences, is summarised using a BIC-based approximation to the Bayes factor for comparison between models with and without the pre-post effect. Monte Carlo validation uses design quantities estimated from the observed data, including sample size, mean pre-post change, and second-rater additive discrepancy, and summarises inferential performance in terms of bias, root mean squared error, credible interval coverage, posterior tail probabilities, and mean Bayes factor values. For background on the BIC approximation and Bayes factors, see Schwarz (1978) <doi:10.1214/aos/1176344136> and Kass and Raftery (1995) <doi:10.1080/01621459.1995.10476572>.

Versions across snapshots

Dependencies (latest)