I was comparing the similarity of distribution of two data sets using KS test. Two data sets are not pure normal. I was wondering how much KS test depend on the distribution of samples. If the samples are Poisson or Bernoulli distributed instead of Normal distribution, how much the test will be affected.
Comparing with other current tests like Shapiro-Wilk (SW) and Anderson-Darling (AD) I found that, KS test is less sensitive to normality compared to SW and AD test. Can I say that, KS test is robust in terms of distribution of samples?