I am doing an event study. I use two parametric t-tests and two non-parametric tests (sign and rank). The values of the two parametric t-tests are around 25 while the values of the two non-parametric tests are around 8. Thus, there seems to be systematical difference between the values of the categories of test.
The data does not seem to be normally distributed.
My question is now. Why do I see this difference?
Two explanations pop to my mind: 1. The observations does approximately follow normal distribution and as a result the parametric tests are more powerful than the non-parametric test. 2. The parametric tests simply overestimate the results because they are not normally distributed hence the non-parametric distributions are the most reliable.
What do you think?
Advertise on Talk Stats