Pair t test: defininng hypothesis
Six Sigma – iSixSigma › Forums › Old Forums › General › Pair t test: defininng hypothesis
 This topic has 6 replies, 4 voices, and was last updated 12 years, 1 month ago by Klerx.

AuthorPosts

October 27, 2009 at 3:42 am #52841
I am performing pair t test for my life from FEA and actual testing. my aim is to statistically check that my FEA predicts life within 10% of test life, so I am using pairt test.
The null hypothesis is
Ho= difference in mean of test life and mean of fea life is 10%.
Ha= difference in mean of test life and mean of fea life is <10%.
pairt test is giving pvalue of 0.029 which states that I have to reject Ho and accept Ha.
Now question is, Is my null hypothesis is correct? or what will be my null hypothesis based on problem statement.
Please advice.
0October 27, 2009 at 5:19 am #186395Since you are performing a paired t – test For Your Life , the p – value indicates there’s a 3% chance that you will outlive the mean life by 10%( whatever that is) but still think you won’t !Anyways , the paired ttest appears inappropriate for the case in question , a 2 sample ttest appears more appropriate – of course test for normality , equal variances etc.BBUSA
0October 27, 2009 at 5:25 am #186396Forgot to mention your hypothesis statement is incomplete ( null hyp is incorrect ?)
0October 27, 2009 at 5:35 am #186397Than what can be my null hypothesis statement? Do I alway say that Ho= difference in mean life of test and mean life of fea is ‘0’ (zero)
0October 27, 2009 at 5:51 am #186398Your hyp statement could be ( For verifying if the test life is 10% lesser than the mean life )H0 : mu1 – mu2 < 10% ( whatever the 10% means)
Ha : mu1 – mu2 >= 10%Establish your hyp. with a 2 sample t – testBBUSA0October 27, 2009 at 3:24 pm #186407SAM:
Here is the general guideline for a paired ttest versus a 2sample test. If the measurments are made on the same experimental units, and those units are expected to differ from one another in a material way, then the paired ttest is appropriate, as it removes a source of variation (that due to the units themselves) and allows the test to clearly focus on the treatment of interest. It sounds as if you are comparing predicted life against actual life, but what units are being measured?
I can’t tell from your original description whether this should be a paired ttest or a 2sample test. I’m also not certain if you need to convert to a percentage metric, or not. Is it necessary to convert each measurement to a percentage? Just so you are aware, this is not a requirement of running the paired or 2sample tests, but it might be appropriate if the data covers a large range.0October 30, 2009 at 1:34 pm #186489SAM:
You can NOT, I repeat, you can not use a ttest, not 2sample nor paired.
Your conclusions are simply WRONG.
Life is Weibull distributed and the ttest uses the Normal distribution. So you have to calculate confidence intervals in order to be able to do hypothesis testing.0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.