0

p-value computed by the Kolmogorov-Smirnov test is known to be biased if the parameters of the theoretical model are not fixed but, instead, they are estimated from the observed data. In particular, the computed p-value is overestimated, hence we run the risk of accepting the fitting hypothesis also when the data does not really come from the theoretical distribution.

I do not understand what (the computed p-value is overestimated) mean, does it mean that the computed p-value is larger than it should be than one that we would have its table

Statistical scientist
  • 29
  • 2
  • 1
    When you calculate the K-S statistic on a sample where the cdf is based on parameter values estimated from the same sample, the cdf is a better fit to the data than the cdf with the population parameter values would be. Consequently, the p-values you get will be on average larger than they should be; equivalently, the significance level (alpha) is lower than you would otherwise think it is. – Glen_b Nov 10 '21 at 02:05

0 Answers0