[R] combining P values from different tests

[Duncan Murdoch]

On 08/07/2010 2:45 PM, Jonathan Baron wrote:
> On 07/08/10 19:25, Knut Krueger wrote:
> > Hi to all,
> > is there any function available to combine those p values;
> > 
> > 
> > Kind Regards Knut
>
> Here's a couple of functions I've used for one thing that you might
> mean.  The idea is that you have two p-values for different tests of
> the same hypothesis, and you want to get an overall p-value for the
> hypothesis.  The first method, for combining two p-values, was
> invented by Fisher a very long time ago.  The second method I do not
> remember a thing about, but I found it somewhere.  You need to look
> these up before trusting them.
>
> # combine and output one-tailed p's.  Enter 2-tailed p's.
> combine <- function(x,y)
> return(pchisq(-2*log(x/2)-2*log(y/2),4,low=F)) 
>   
This assumes the two p values are independent, but you don't want to 
divide x and y by 2:  that biases things towards smaller p-values.  (I'm 
not sure what you mean by "2-tailed p's" though, so I might be wrong.)
> # Stouffer's method for vector x, gives 2-tailed result
>  pnorm(sum(qnorm(x))/sqrt(length(x)))
>   
This also assumes independence between the tests.

Duncan Murdoch