In a statistical hypotheses testing a significance level $\alpha$ has to be set. The most often, $\alpha$ is set to be 5 %, sometimes 1 % and 10 % values are used.
Value of $\alpha$ tells us what is a maximal tolerated probability of error of first kind (rejecting true hypothesis). There is also error of second kind (not rejecting false hypothesis). The convention is to anchor maximal probability of the error of first kind. But there is no reason to prefer error of first kind to error of second kind.
My questions are these:
- Who first devised to use values 1 %, 5 % and 10 % for statistical signifance and why?
- Why error of first kind is preferred?