Permutation F test - what is the correct F statistic to use?

#1
Hi all,

I'm currently using an ANOVA to analyse some data sets, recently it occurred to me that the data may not be normally distributed - at all. So I decided to 'upgrade' to using a permutation F test which is apparently much more robust in the face of not-normally distributed data.

The test essentially requires you to run an ANOVA on your data, you then run it a number of times (i.e. 500 or 1000) on the same data, but with the data entries shuffled. Significance is determined by counting the number of F values obtained from the shuffled data that are higher than the F value obtained from the original data. To put it another way:

p = #F's > Fobs / number of permutations

My question is simply, when I'm reporting this test result what F value or degrees of freedom do I use? The ones calculated in the original ANOVA? Or do they have to be calculated in some similar fashion? I found it hard to find out how to run the test, never mind info on how to report the results etc.

Thanks in advance, as always any help is greatly appreciated.

Roddy.