Hypothesis testing
Six Sigma – iSixSigma › Forums › Old Forums › General › Hypothesis testing
 This topic has 6 replies, 2 voices, and was last updated 13 years, 5 months ago by Sorour.

AuthorPosts

July 4, 2008 at 12:58 am #50467
Hi all.
As I understand it, equal variance at a factor level is a requirement to run a Kruskal Wallis test.
What to do if you don’t have equal variance? Am I trying to compare apples with pears?
Cheers,
Paul0July 4, 2008 at 1:43 am #173536
Michael MeadParticipant@MichaelMead Include @MichaelMead in your post and this person will
be notified via email.The KruskalWallis H test is a nonparametric test for deciding if samples come from the same population. While the variances may not be equal between the two or more samples, they should not be significantly different, which would indicate that the samples are from different populations.
This may also be accomplished using a t test or the MannWhitney U test0July 4, 2008 at 2:24 am #173537Thanks Michael.
Can I ask what is your definition of “different” when talking about variance?
As for the suggested alternative tests, a bit of background might help.
I am looking at 4 month’s worth of data where the output metric is report sign off time. I want to investigate if there is a difference in this Y between months.
Some month’s data is normally distributed and some is not.
Could you therefore suggest an appropriate test to run (or steps to determine an appropriate test)?
Thanks,
Paul0July 4, 2008 at 3:13 am #173539
Michael MeadParticipant@MichaelMead Include @MichaelMead in your post and this person will
be notified via email.Hello Paul,
You can do an ANOVA (analysis of variance) using Excel or any statistics package. Set up your data in columns. The ANOVA procedure is in the Data Analysis toolpak.
You can also compare two means using the z test or t test. That would be the simple way. A pvalue less than .05 is significant.
Good luck.0July 4, 2008 at 3:32 am #173540Thanks again Michael. Getting a little confused, however.
As some of my data is not normally distributed at a factor level (month), does that not invalidate a comparison of means?
Hmmmmmmm…
Paul0July 4, 2008 at 6:04 am #173541
Michael MeadParticipant@MichaelMead Include @MichaelMead in your post and this person will
be notified via email.Paul,
Don’t worry about the normality of your data. You are testing the means. Do you have any reason to believe one month’s distribution is a different shape than the next month? I don’t think so. Thus, the shape and spread of the distributions is irrelevant. And, after all, you are just checking at this point. If something useful appears you should do further investigation.
I can’t say that you should totally ignore the normailty assumption What I am trying to say is; you will probably get the information you need without that bit of worry. In economics, we would never base a theory just on a regression analysis. There must be some logic or reasoning behind it. Conversely, if there is no theory to support the idea that the distributions are of different shapes, relax and test the hypothesis.
By the way, where are you? We are having this discussion in the middle of the night in the US. For me it is easy, I am in China.
Good luck.0July 7, 2008 at 11:29 pm #173602Hi Michael.
Not much of a time difference between us, as I am in Oz.
Thanks again. I will plough on with the tests, happy I’m on the right road.
Cheers,
Paul0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.