Bayesian analysis of single case experimental design count data in trauma research: A tutorial

Single case experimental designs (SCEDs) are suited to psychological trauma research that involves individualized treatments and where randomization might be impossible or inappropriate. However, SCED data present challenges such as autocorrelations, short time-series, nonnormal distributions (in count data), and the effect sizes being comparable across phase changes and participants. The present study illustrates the Bayesian rate ratio (BRR) that can be used when the phase lengths vary across phases to quantify treatment effect and can be aggregated across participants.

Bayesian estimation allows flexible modeling of autocorrelations and distributions, works well for small samples, and provides easily interpretable estimates of uncertainty in the form of credible intervals. BRR estimates from a published dataset from trauma research are compared with commonly used nonoverlap of all pairs (NAP) effect sizes.

Bayesian estimates show superior performance over NAP estimates because they reflect the patterns in the data and account for distance between observations, unlike NAP. All the R programs are shared with the readers and are annotated for their ease.

Through tutorials such as these, current and future researchers can be educated about Bayesian methodology, its appropriateness to address the idiosyncrasies posed by SCED count data, and aid further development of models for other SCEDs. Implications of the study are also discussed. (PsycInfo Database Record (c) 2023 APA, all rights reserved).