# Bayesian assumptions?

I understand (as previously discussed in the forum) that the standard parametric BF assume the same assumptions as those NHST do, but doesn't the Bayesian frame work incorporate any deviation from assumptions into the BF?

What I mean is, when comparing in a Bayesian t-test an alternative model to the null model, we usually think only how they differ - in the null the parameter is assumed to be 0, and in the alternative the parameter is assumed to be different than 0.

But in actuality, we also assume normality and homogeneity of variance. i.e., the null model assumes that data is normally distributed and with an expected value of 0.

So in theory (although this is probably not advisable?), even when assumptions are violated, we can still compute these BFs, our model of describing the data we will simply be (more?) incorrect to some degree. Right?

## Comments

Yes you can still compute the BF, that is, assess the relative predictive adequacy of the hypotheses. But what it means when the model is wildly misspecified is anybody's guess. See for instance http://www.thefunctionalart.com/2016/08/download-datasaurus-never-trust-summary.html

E.J.

Sure, if I my data is dino-distributed that would be a problem.

I was thinking more along the lines of skewed data and the like - where deviations from the norm are crucial for error-rate in NHST, but I would suspect less so for BFs.