p values for confidence intervals vs. credible intervals

[lizrichardson]

Hi,

I'm new to MCMC models but am finding runmlwin a very helpful and flexible way to use MLwiN. Also this forum is a wealth of great advice for a newbie. Thanks for all the contributions.

My query is about p values. A p value is given by default when I run my model, but I'm not sure how to interpret it. I started to suspect that this isn't like a frequentist p-value when I got a p=0.049 for a credible interval that included zero. If I add the zratio option I get a pa value that make more sense to me for the same CI: p=0.106 rather than p=0.049. I know that confidence intervals and credible intervals aren't the same thing, and that p values are frowned upon by Bayesians, so I'm wondering how to proceed with the interpretation.

Thanks,

Liz

[ChrisCharlton]

The Bayesian p-values are calculated as the proportion of the chain where the values are of the opposite sign to that of the corresponding estimate (usually the mean). This is one-tailed, which is why it is approximately half the value you obtained via the zratio option.

[lizrichardson]

Many thanks for the reply Chris.

Liz