topic Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example in Discussions

Analyzing Repeated Measures Data in JMP® Software - masterthesis example

ThomasVH — Thu, 24 Nov 2016 12:58:51 GMT

Hi,

I was wondering if someone could verify if I took the right steps in setting up a mixed model for repeated measures for the following research:

In our research we had 2 groups of soccer players (amateur vs. professionals).

Each player of both groups performed 3 physical tests (a short vs. a long laboratory test vs. a field test), each test was performed on different days.Our outcome measure was VO2max (physical capacity).

Purpose: I mainly want to compare the VO2max between the amateur vs. professionals for each physical test. (2x3 table)

So I thought I should use a 2-way repeated measure ANOVA. In JMP pro 12-> a mixed model repeated measures.

But i'm not sure about which factor I should put in the fixed effect and/or random effect box when using Analyse -> fit model for my purpose.

Can someone please verify if I did the right thing in the two last pictures below? (The first picture is just a small selection from my data.)

Kind regards,

Thomas

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

Mark_Bailey — Sat, 26 Nov 2016 19:04:01 GMT

I don't think that this study requires a repeated measures analysis. It appears to simply be a two-way ANOVA:

V(O2max) = Intercept + Level + Physical Test + Level * Physical Test. The subject is your residual, so you do not need to enter an explicit term to estimate it. The factor levels are crossed, not nested in this case.

You should be able to use the Standard Least Squares personality in this case.

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

ThomasVH — Mon, 28 Nov 2016 15:58:50 GMT

Hi Mark,

Thank you for your answer.

The reason why I thought to use a mixed model instead of a 2-way ANOVA is the fact the VO2max values are based on repeated measures. For instance, subject 1 did three tests a short one (64,6 VO2max), the week after a long one (54,7 VO2max), again one week later a field test (52,86 VO2max). The same is for the other subjects.

Does this justify the use of the mixed model?

If it justifies the mixed model. Should I leave the random effect box empty like you suggest? Or fill in the subject factor in the random effect box?

Thanks in advance.

Kind regards,

Thomas

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

Mark_Bailey — Mon, 28 Nov 2016 16:33:41 GMT

I see your point. Subject is a blocking structure and its effect is random. I think that it makes sense to include it as a random effect after all.

I still don;t see this study as an example of 'repeated measures' but my definition might be too strict. It is just a two-way ANOVA with blocking.

Consider including the interaction term as I think it is estimabile and, if so, could be informative.

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

ThomasVH — Mon, 28 Nov 2016 18:10:28 GMT

I'm a little confused now. Perhaps I should clarify the whole study better.

We tested lots of players. Some of these players were professional and others were amateurs. But we tested them in group. For instance week 1 on monday each amateur did 1 test, the same week on tuesday each professional did 1 test. In week 2 on monday each amateur did another test, the same week on tuesday each professional did another test. Finally, in week 3 each amateur did a third test on monday and each professional did his third test on tuesday.

Now we want to test if the VO2max differs between the level (prof vs. amateur) with taking the different type of physical test (short vs. long vs. field) into account.

Taking your previous answers into account I think my analyse + data should look like this. I have also add the future 2x3 table I want to solve (I tried it once).

Do I still have to use just a 2-Way ANOVA?

Kind regards,

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

Mark_Bailey — Tue, 29 Nov 2016 11:21:09 GMT

I apologize for any confusion that I might have caused you.

Your design structure supports the two-way ANOVA for the fixed effects. Subject is a random effect, but I am not sure that you can estimate it separately from the residual random effect. I entered your example above and repeated your mixed model analysis. Note that subject is nested within athlete and physical test. This nesting should be indicated for this term in the Fit Model dialog before you run your model.

(I can't include an image at this point because the dialog to insert a picture completely changed. I can't figure out how to use it.)

My estimate for the subject variance is zeroed.

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

ThomasVH — Tue, 29 Nov 2016 12:25:20 GMT

Ok, I think I'm almost there.

First,

If I want to solve this table:

I should do my analysis as a 2-Way ANOVA without repeated measure. Because the repeated factor (physical test: short vs long vs field) are not compared with each other.

But if I want to solve it this way:

I should do my analysis as a 2-Way ANOVA repeated measure (thus a mixed model). Because the repeated factor (physical test: short vs long vs field) are in this way compared with each other?

Second, so I shouldn't include subject as a random effect?

Third, the fact subject is nested within physical test and level so my Fit Model dialog should look like this?

Kind regards,

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

Mark_Bailey — Tue, 29 Nov 2016 18:02:48 GMT

I conferred with my colleagues who are wiser, smarter, and much better looking than me. They suggest that you should not estimate the interaction term as I suggested. It will confound subject with the residual as I discovered.

So go back to the beginning and your original mixed model seems like the way to go after all.Specifically, start a mixed model including the two main fixed effects and the random subject effect. Indicate nesting of subject within the other two factors. That specification should do it.

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

ThomasVH — Tue, 29 Nov 2016 19:07:46 GMT

So I do the analysis this way:

Thank you and your helpfull colleagues,

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

Mark_Bailey — Tue, 29 Nov 2016 22:14:44 GMT

No. Remove the interaction fixed effect. The rest is correct!

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

Mark_Bailey — Wed, 07 Dec 2016 18:48:05 GMT

I discovered that the nesting will impact your estimation. Instead of nesting, simply label your subjects uniquely as 1-8 instead of 1-4 within each kind of athlete. I attached a data table with the correct data levels and a Model table script for this analysis.

I apologize for the delay in this answer.

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

ThomasVH — Wed, 07 Dec 2016 20:29:17 GMT

Thank you for mentioning it. I tried it and indeed, it has also an impact on the p-values.

Greetings

Re: Analyzing Repeated Measures Data in JMP® Software - masterthesis example

ThomasVH — Sat, 10 Dec 2016 22:29:00 GMT

Everything went well. I'm only a little worried about my SEM values.

In most of the parameters for instance the VO2max (see picture beneath) the SEM values are the same for each group. Is this normal?