[R-lang] Re: p-values from pvals.fnc

[David Reitter]

Jakke,

On Jul 29, 2011, at 3:58 PM, Jakke Tamminen wrote:

> and gave me a t-value of 2.07 for the interaction, with a pMCMC p-value of 0.4756 and a t-distribution p-value of 0.0381. Obviously I like one of these better than the other! I know that the latter p-value is anticonservative, but the magnitude of the discrepancy is nonetheless surprising to me, given the t-value. I'd be very grateful for any advice on how to proceed in cases like this. I'm using lme4 version 0.99875-6.

Are your x and y well-behaved, that is, are they normally distributed?
The usual caveats of t-tests apply, which is why people (read reviewers) often demand MCMC sampling to derive p-values.  Rightfully so, as your real-world example shows (provided the effect is bogus, that is).

If you know the distribution of your x and y variables, transforming them appropriately might give you more power.  I don't see right now why this shouldn't be allowed, but others here may weigh in on that question.