iSixSigma

Hypothesis testing – Paired t tests

Six Sigma – iSixSigma Forums Old Forums General Hypothesis testing – Paired t tests

Viewing 3 posts - 1 through 3 (of 3 total)
  • Author
    Posts
  • #29228

    Cindy
    Participant

    if the differences between paired data is non-normal ,sample size is small and leptokurtic (alot of zero’s),  how valid is the p value in a paired t test? 
    Is this test robust enough to still trust the p value?
    thanks for any advice you can offer………
     

    0
    #74242

    Eileen
    Participant

    Cindy,
    Don’t use the paired t-test. The p-value won’t mean a thing!!
    If you have to use a test of statistical inference, your best bet is the Wilcoxan Sign-Rank Test (a nonparametric test of paired significance).
    This is the most valid test of statistical inference.
    Eileen
    Quality Disciplines

    0
    #74244

    KT
    Participant

    Be careful of nonparametric tests. The have much lower power than the parametric equivalents.
    The real question is how small is the sample size? and how non-normal are the data? Even relatively small samples and relatively non-normal data are covered by the Central Limit Theorem.
    You can test this by simulating your data in MINITAB or another package, getting multiple averages using your sample size, and looking at the distribution of those averages.
    On the other hand, if the sample size is, say, 2 and the distribution is incredibly skewed to one side, then the nonparametric test may be the only alternative.

    0
Viewing 3 posts - 1 through 3 (of 3 total)

The forum ‘General’ is closed to new topics and replies.