MSA paired test
Six Sigma – iSixSigma › Forums › Old Forums › General › MSA paired test
- This topic has 12 replies, 7 voices, and was last updated 12 years, 4 months ago by
Ken Feldman.
-
AuthorPosts
-
December 9, 2009 at 9:00 am #53016
Hi All,I must be wrong we doing the following statement, but I can’t see where…If someone could help me…I’ve done a MSA on a testing machine. This machine has 2 parallel equipments, which should give the same results. I considered them as 2 different operators. MSA result is very good (%Reproductibility = 6%, % Repeatability = 2%).
Then, I concluded that I can say Equipment 1 and Equipment 2 give the same acceptable result.But, if I do a paired t-test using the results of this MSA, I find p=0, than Equipment 1 seems to be different from Equipment 2.Where am I wrong??Thx for reading till the end!0December 9, 2009 at 9:15 am #187415
Ken FeldmanParticipant@DarthInclude @Darth in your post and this person will
be notified via email.You are mixing apples and oranges. Gage R&R tests the variability of your measurement system. Your hypothesis test examines whether the means of the two samples come from different populations. Try a two variance test and see what happens. There should be no assumption that the paired t test and Gage R&R will result in similar decisions regarding the equipment.
0December 9, 2009 at 9:21 am #187416That’s clear, thank you Darth
0December 9, 2009 at 10:29 am #187417sounds like you are evaluating repeatabilty/ reproducibility as well as bias between you measurement systems.
-what do the % values represent below? (% tolerance, % study variation,?)
– how did you select your range of samples?
– what was your total sample size?
0December 9, 2009 at 11:26 am #187418– It’s the % of study variation.
– The samples have been choosen in order to cover the whole range of possible production.
– 10 different samples, measured 6 times with each equipmentOther information I didn’t mention earlier:
I find a perfect correlation between the results of my 2 equipments (R²>99.99%)0December 10, 2009 at 10:02 am #187451Interesting…
Now you have a repeatable and reproducible measurement system comprised of two perfectly correlated measurement tools, and they are different!
Just out of curiosity, what is the tolerance on the parts and what is the range span of parts in production that you chose for the study? What was the % RR as compared to the tolerance (P/T ratio)?
0January 11, 2010 at 8:00 am #188203
ThothathiriMember@ThothathiriInclude @Thothathiri in your post and this person will
be notified via email.Angler,
Sorry for responding late.For you ideal statistical test is paird-t test, test
10 parts in equipment 1 and the same parts to be
used in testing in the equipment 2.You can analyze the result using Excel, the command
is ttest and mention 1 in the type of test. Result
what you get is p value, if its0January 11, 2010 at 8:37 am #188204if its
0January 11, 2010 at 9:35 am #188206
THOTHATHRIMember@THOTHATHRIInclude @THOTHATHRI in your post and this person will
be notified via email.its p value based on the alpha level of 5%p values decides statistical significance of test. P value 0.05 means fail to reject Ho.
0January 11, 2010 at 10:45 am #188208Much better – maybe that answer can be understood.
0January 11, 2010 at 6:20 pm #188222
Ken FeldmanParticipant@DarthInclude @Darth in your post and this person will
be notified via email.Stan, while the answer may be more understandable it is still not clear whether he is correct or not. The null for a test of means is that the two population means from which the samples are drawn are the same. If he believes the test is to determine whether the sample means of the two pieces of equipment are equal then he is wrong. Of course they are different since the means will likely be mathematically different. The selected alpha says you are willing to be wrong 5% of the time if you state that the samples came from different populations. The p value is the true risk or % of the time you will be wrong if you reject the null. If the % probability of being wrong is greater than your desired risk then you don’t reject (p is high). If it is less than your acceptable risk (p is low)then you can reject the null with comfort. In hypothesis testing you are testing the difference in parameter means not the sample means. But then, maybe that is what he said after all.
0January 12, 2010 at 5:10 am #188232
SeverinoParticipant@Jsev607Include @Jsev607 in your post and this person will
be notified via email.I like how he snuck an ad-hoc sample size in there as well…
0January 12, 2010 at 5:48 am #188234
Ken FeldmanParticipant@DarthInclude @Darth in your post and this person will
be notified via email.Yeah, wonder what the power is for a sample size of 10.
0 -
AuthorPosts
The forum ‘General’ is closed to new topics and replies.