The visualization shows a Bayesian two-sample t test, for simplicity the variance is assumed to be known. It illustrates both Bayesian estimation via the posterior distribution for the effect, and Bayesian hypothesis testing via Bayes factor. The frequentist p-value is also shown. The null hypothesis, H0 is that the effect δ = 0, and the alternative H1: δ ≠ 0, just like a two-tailed t test. You can use the sliders to vary the observed effect (Cohen's d), sample size (n per group) and the prior on δ.
The prior on the effect is a scaled unit-information prior. The black, and red circle on the curves represents the likelihood of 0 under the prior and posterior. Their likelihood ratio is the Savage-Dickey density ratio, which I use here to compute the Bayes factor. The p-value is the traditional p-value for a two-sample t test with known variance (i.e. a Z test). HDI is the posterior highest density interval, which in this case is analogous a credible interval. And CI is the traditional frequentist confidence interval.
Have any suggestion? Or found any bugs? Send them to me, my contact info can be found here.
Kristoffer's LinkedIn
profile
