How to increase p-value decimals for non parametric tests ?

[Michel Passager]

Hi everybody
I have to do a lot of Mann-Whitney and Wilcoxon tests. Stata "gives" p-values with only 4 decimal.
Beyond that, it indicates "p = .0000". However, I will have to make corrections of p-values because of the important number of tests. I will use the approach of Benjamini & Hochberg for this. So I would need more precise values, for example 6 to 8 decimal places.
Can you help me to find a solution?
Thank you very much for your help.
Michel

[Andrew Musau]

The p-values for r-class commands are usually stored as -r(p)-, but do check

return list

for variations of this. That being the case, just type

display %7.6f r(p)

to get 6 decimal places displayed.

[Michel Passager]

Many thanks for your help Andrew. It works !!!

[Mike Lacy]

I'd like to offer a side note here, namely to raise question about whether a procedure that relies on lots of decimal places in a p-value is wise. Any conventional p-value here necessarily will involve a bunch of asymptotic approximations. Consequently, while Stata will offer lots of decimal places in r(p), I would not believe that those p-values can be trusted: Stata might report 0.00004, but the "true" value might actually be 0.007. If a non-asymptotic procedure were used (e.g., -permute-), it would take a huge number of repetitions to get such a precise approximation of the p-values. So, I wonder if there is some way to approach the problem at hand that doesn't require unrealistic assumptions about how precise a p-value is.