How to perform a Two One Sided T (TOST) Test in Minitab

[statman78]

How do you perform a Two One Sided T Test in Minitab?

[Stijloor]

Originally Posted by statman78

How do you perform a Two One Sided T Test in Minitab?

Hello statman78,

Would this help? Two One Sided T Test in Minitab

Stijloor.

[cuadra]

Dear Statman78

Mr. Stijooor is trying to protect you. If you put numbers into Minitab (assuming you have the right format) you will always get an answer back. It is up to the user to make sure;

a) The right test was used (i.e. two sample t-test versus pair t-test)
b) The necessary statistical assumptions were verified (normality, etc)
c) You have the right sample size (enough power, ability to detect the desire difference, etc)
d) Etc, etc

[Tim Folkerts]

I'm not familiar with the term "two one sided T Test".

I have heard of one-tailed and two-tailed T Tests (AKA one-sided and two-sided).
I have heard of one sample and two sample T Tests.
I have heard of paired t-tests.


What are you trying to accomplish? If I knew the context, I might have a better handle how to address the question. Minitab has capabilities to do just about any statistics test around, so I expect there is a way to do what you need.


Tim F

[Dean Frederickson]

Originally Posted by Tim Folkerts

I'm not familiar with the term "two one sided T Test".

I have heard of one-tailed and two-tailed T Tests (AKA one-sided and two-sided).
I have heard of one sample and two sample T Tests.
I have heard of paired t-tests.


What are you trying to accomplish? If I knew the context, I might have a better handle how to address the question. Minitab has capabilities to do just about any statistics test around, so I expect there is a way to do what you need.


Tim F

Tim I don't know but this might help. http://www.statisticallysignificantc...FRqWGgod0A32Zg

[statman78]

Cuadra and to those who have responded to this thread. Thank you for some guidance. I think an explanation of what I am trying to do is in order to get some help.

Why do I want to do a TOST test?

It is a customer requirement for me. The TOST test should be performed if you are qualifying one part number from one machine to another machine with the same tonnage and process. In the past we would compare capability results from one machine to another. However, good capability results did not always yield equivalence.

Requirements:

(1) Measurement system meets requirements (2) Variances are equivalent (3) Data are normally distributed (4) Samples are
representative (5) Practical significance equates to the point of detection by customers, or performance degradation

This test is to be performed specifically for demonstrating practical equivalency between two data sets with 95% confidence.
A practical significance must defined before entering the data. Practical significance is defined as the positive half of the acceptable shift in mean. This value is how much they can differ with no practical on performance, or detectability.

For example: Nominal = 1.0 Tolerance = +/-0.05. The acceptable shift would usually be 20% of the tolerance. In this case the practical significance would be = 0.02(0.05) =0.01 (Not 1.0 + 0.01)


If the Confidence Interval is completely contained within the Acceptance Interval, then the means are practically equivalent.

There are 3 conclusions to be made after the test as per the customer requirement.

The means are practically equivalent significant at alpha = 0.05
The means are not practically equivalent significant at alpha = 0.05
Test is inconclusive due to non normal data or unequal variances.

If the last two conditions exist then I must go and perform a Mann Whitney Test for non normal data.

The customer provided a locked excel template but it is quite cumbersome to use as you can only use 2 dimension per template (We usually have 5 to 64 dimensions that have to be tested in such a way). If the test fails I constantly have to go back in to Minitab and do a Mann Whitney test. I would post the template here but its locked and I can't distribute it per confidentiality agreements. Ideally I would like to be able to run a macro so that if the test is inconclusive or not equivalent Minitab would automatically do Mann Whitney.

I hope this makes things clearer.

[Marc]

Any current thoughts on this?

[Allattar]

Still no clearer as to what they are doing.

If I read this right, you have two machines measuring parts.
You run a test where you put the same items in each machine and read up to 64 different dimensions off of each part.

If this is the case do you know what the part number is for each item tested, or is that unknown. (The difference being a paired test or two sample t.)

Have I got that part right?

As to excel you check for normality, run a t-test of some kind, then compare the means to the nominal.

I have all kinds of questions being flagged up about what t-test your using, and what you are checking the normality of, but these can wait till I understand just how the measurements are taken.