topic Re: TOST Equivalence Test error in Discussions

TOST Equivalence Test error

tkutch — Sat, 10 Jun 2023 20:37:41 GMT

I followed the Equivalence Test (TOST) example but the output is different from the website.

https://www.jmp.com/support/help/en/15.1/index.shtml#page/jmp/example-of-an-equivalence-test.shtml#ww854229

Website:

>= 6 p=0.0124

<= -6 p<.0001

My output:

<= 6 p=0.0124

>= -6 p<.0001

Essentially, the null hypotheses have been reversed so that rejecting corresponds to non-equivalence. The hypothesis on website appear accurate as rejecting the null indicates evidence that data is within the equivalence region. Why is my software reversing the hypothesis?

I am using JMP 13.2.0

Re: TOST Equivalence Test error

Mark_Bailey — Thu, 16 Jul 2020 17:00:59 GMT

JMP uses alphabetic or numeric order by default except for a few special cases. Gender is not one of them. So the default order is F is first and M is second. That leads to the results shown in the on-line documentation.

This order is important. JMP will subtract the response of the second level from the response of the first level, so the mean height of sex = F from the mean height of sex = M.

If the order of the sex levels is reversed, then so is the difference (negated). I was able to reproduce your result by adding the Value Order column property and making M first and F second. Is that what you have for the order?

Re: TOST Equivalence Test error

tkutch — Sat, 18 Jul 2020 13:46:54 GMT

@Mark_Bailey I have F first and M second in the Value Order column property. I tried switching the order (M first, F second) and yes the mean difference flipped, but the null hypothesis are still <=6 and >= -6.

Re: TOST Equivalence Test error

Mark_Bailey — Sun, 19 Jul 2020 11:18:04 GMT

I see. I cannot remember if the labelling of the null hypotheses was an issue in JMP 13. You should report this behavior to JMP Technical Support (support@jmp.com). Please report here what you learn from them.