Unusual notation of p-values

Question

[lvogel]

Hi. I have data from a tiling microarray experiment. The column I am interested in contains the p-values for sex bias. They are supposedly all significant (only the significant ones were included). However, instead of being in the usual decimal notation of 0 to 1 (or 0 to 0.05 in this case), they range from -300 to -1.5 and 1.5 to 300. Does anyone know how to convert this? The person who originally generated this data is not easy to get a hold of. Thanks.

[brennanpincardiff]

Are you sure it's a p-value? Could it be an F-test statistic? Maybe try to reproduce the calculation for one of the bits of data, then you could check if it's log transformed suggested by the answer?

[lvogel]

It's unlikely to be anything other than a p-value, as the column heading is "sex.P". My supervisor just remembers that the 2.0 corresponds to 0.01 and the 1.5 to 0.05. I can reproduce the first (10^-2 =0.01), but not the second.

Oh, well. It will get figured out eventually.

Answer 1

[ethan.kaufman]

Since 10^(-1.5) = 0.03, it is likely that the P-values have just been log transformed. The sign probably refers to direction of the bias (eg. +'ve for male, -'ve for female). To convert back, ignore the sign and just compute: Pvalue = 10^(-1 * score).

[karl.stamm]

This is done when data storage precision is an issue. If you only get eight characters to write each number, both 10^-100 and 10^-150 look like exactly zero when truncated.