0
$\begingroup$

I'm struggling to find a goodness of fit test of the above. The non-parametric tests I have looked at (KS) seem to be unable to deal with estimated parameters - can someone help?

$\endgroup$
  • $\begingroup$ I don't find the wording clear here. Are you testing for normality but want your test to be robust in the face of non-normality? Or something else. $\endgroup$ – Nick Cox Mar 5 '14 at 19:35
  • $\begingroup$ No - I simply have a range of non-normal distributions (t, skew-t, skew exponential power) I have generated parameter estimates for my data under all of these, and now want to conduct goodness of fit tests. $\endgroup$ – user40124 Mar 5 '14 at 20:31
  • $\begingroup$ What do you mean by robustness then? $\endgroup$ – Nick Cox Mar 6 '14 at 1:16
1
$\begingroup$

Unfortunately, there is no simple answer to your problem. The KS test methodology is generally used, but the critical regions need to be determined via simulation if you are not trying to test the normality of the data. See this link for details, it explains how one does this for non-normal tests. This paper is also very helpful, as it addresses exactly how you calculate these bounds using a computer.

$\endgroup$
0
$\begingroup$

For dealing with non-normality, you could consider using the Anderson-Darling test if you have an idea what your hypothesized distribution function is. This test is implemented in R in the ADGofTest. I know for testing normality there are modifications made to the test statistic to account for estimated parameters. I haven't thoroughly researched other distributions, but I would assume it is possible to make similar adjustments. The other option would be to just bootstrap the test statistic to estimate its distribution.

After some cursory research, it also looks like there is a non-parametric version of this test. See this paper: http://www.cithep.caltech.edu/~fcp/statistics/hypothesisTest/PoissonConsistency/ScholzStephens1987.pdf

It appears this test is also be implemented in R, in the package kSamples: http://cran.r-project.org/web/packages/kSamples/kSamples.pdf

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.