No in the null hypothesis, for the distribution, don't have to care, can be distribution free, except assume that the expected values on the measurements are the same. For details, see my post

https://news.ycombinator.com/item?id=10484602