hypothesis tests on nonnormal data
Six Sigma – iSixSigma › Forums › Old Forums › General › hypothesis tests on nonnormal data
 This topic has 7 replies, 8 voices, and was last updated 19 years, 6 months ago by DrSeuss.

AuthorPosts

July 11, 2002 at 9:48 am #29845
I have two sets of time study datas taken by two different groups. They were done on the same workers, same condition (1st shifters), and same work. I would like to see if there is a significant difference between the two results. I was thinking of doing a paired ttest instead of 2 sample ttest. I first check their variances and found that they have equal variances. Later I found that both datas are not normal. My question is, should I proceed doing paired t? what type of hypo test is appropriate to nonnormal data like mine? Thanks.
0July 11, 2002 at 11:14 am #77136
Ovidiu ContrasParticipant@OvidiuContras Include @OvidiuContras in your post and this person will
be notified via email.The Mood’s median test will tell you if there’s a statistically significant difference (pvalue less than 0.05) between the medians of your two populations (distributions not normal) . If you have Minitab ,go to StatNonparametrics…
Good luck0July 11, 2002 at 11:45 am #77138Hi,
How many data points you have, if you have more than 30 points you may go for t – test, i am not sure whether this is correct or not.
The other way you can use not parametric tests like Anderson Darling, Man whiteny etc. which doesn’t depend on the kind of distribution.
thanks
A.Sridhar
0July 11, 2002 at 3:44 pm #77157Rko:
For nonnormal data you should use:
Levene’s test for equal variances, and:
Mann Whitney test to compare the medians (since your data is not normal, you can not compare means)
I Have never used the Moods test, but you can try both and compare the results.
In Minitab go to StatsNonparametricMann Whitney
Hope this helps,
FGM0July 15, 2002 at 4:50 am #77238Hi Rko!
I assume that you tried normalizing your data , inspite of which the data is non normal. Hypothesis tests such as t and ANOVA assume normality of data and hence are not appropriate when you have non normal data.
Mood’s Median test is what you could use to test the median value of your data before and after. This test does not assume normality of data and can be used to compare your sets of data. However your sample size before and after improvement should be same, to do this test.
Let me know, if this worked!
thanks
arun0July 15, 2002 at 6:03 am #77239Hi Rko,
I will also use Mann Whitely 2 Sample t test.
You may want to read up on nonnormal data here:
https://www.isixsigma.com/library/content/c020121a.asp
Hope it helps.
Best Regards,
C.T.
0July 16, 2002 at 5:25 pm #77283RKO,
All of the responses are good to the question. But, I wouldn’t completely throw out the ttest as a methodology. The ttest has been shown, in many nonnormal scenarios, to give good results. It’s always preferred to use as much of the data as possible to base your hypothesis tests on. Devolving to nonparametric means sacrifices hard won data. My recommendation, in situtations like this, is to run the appropriate parametric tests and use the nonparametric as a sanity check for the results of the hypothesis test.
Regards,
Erik0July 17, 2002 at 6:17 pm #77326
DrSeussParticipant@DrSeuss Include @DrSeuss in your post and this person will
be notified via email.Point of clairification – Kruskal Wallis vs Mood’s median
If your data has outliers, the Mood’s median will handle the data better than Kruskal Wallis. Otherwise ether test will work for testing median differences.0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.