Does anyone have good resources on actually implementing Bayesian #statistics in #neuroimaging research?
I'm currently reading Intuitive Biostatistics by Harvey Motulsky and Bernoulli's Fallacy by Aubrey Clayton.
I know that Bayesian stats make more sense to use. I have just been trained in using P-values for so long that I'm not sure how to do it any other way. How do I determine prior probabilities? Just by feeling it out?
I do try to use CI and effect size instead of P values, but that's not enough. I don't think...
@AmpBenzScientist amazing thank you!
@weberam2
Not in neuroimaging but generally Bayesian inference is straightforward. The difficulty isn't usually with the prior probabilities, its with the choice of generative model. One way to specify priors is by pushing them through the generative model and then seeing what it implies about data. If some parameter values result in ludicrous data predictions those parameter values should have ludicrously small probability associated to them for example.
@weberam2
Its in fact possible to create a set of plausible data values and then insert them into the model using a ultra broad prior on the parameters to get an effective prior, then add in the observed measured data to the dataset and get the full posterior.
@weberam2 I hope this helps.
https://pypi.org/project/pymc3/
It's available with pip.
pip install pymc3
I believe there's a GitHub repo too.