Hi Community,
for a research paper I gleaned p-values from Granger non-causality tests and would like to, as a next step, transform these via a probit tranformation. Unfortunately quite a few of them ( ~10%) are reported as exactlxy "0" which makes a probit transformation impossible as this would result in an infinitely large number.
Is there anyone who has already come across this problem or knows some way how to elegantly circumvent it other then leaving out these numbers altogether? E.g., I thought about replacing all the zeros by a very small number that R or STATA can still handle; but I fear that there is too much subjectivity involved and could result in a highly distorted upper range.
Otherwise, one could replace all zero values with uniformly/randomly distributed numbers for the upper range; but yet again, where to draw the lines of that range?
I am grateful for everythink!
Thanks in advance!