Opposite results for Contingency table one-sided with ordinal data

Mirjam

Hi,

I want to compare two groups B,C (different teaching method) and test if they differ in performance on an exam. The results on the exam are failed or resit/passed. We have a poisson sample. We predict Group B to perform better than group C. Results on a two sided Bayesian contingency table shows groups differ. The contingency table shows that group B performs better. However when doing a one sided test the Bayes factor for B>C is much smaller than B<C while group B was doing better. Can anyone explain those results?

BF10 B>C BF+0= 0,000

BF10 B<C BF-0=7145.288

(see attachment)

Thank you so much!!

Mirjam

Result example one-sided contingency table.docx

EJ

Hi Mirjam,

1. I prefer to use the Bayesian A/B test in this case (see for instance https://onlinelibrary.wiley.com/doi/10.1002/sim.9278). This is a comparison of two proportions.
2. I suspect it has labeled your "failures" as successes. If you examine the posterior distribution you will see the direction of the effect.

EJ