Stats question! Help! Please!

[TruthSeeker]

I've Googled unsuccessfully, and I'm hoping someone here will know the answer to this:

I'm wondering about planned comparisons in a mixed model. Details:

2 (between subjects: sick versus healthy) X 4 (within subjects: assessment time: T0, T1, T2, T3) design.

N=13 sick and 22 healthy.

They run a mixed model ANOVA and find a significant within subjects main effect (F(3,99)=20.4; p <.001). No other main effects or interactions.

Then, they use LSD test for planned comparions of T0-T3. The stat for these comparisons is a t with 34 degrees of freedom suggesting that they are treating the two assessment times as independent, but these are actually repeated measures.

So, my questions are:
Is this an appropriate way to do a planned comparison of repeated measures?

If not, what might I recommend instead?

Thanks!

 

[T'ai Chi]

What stat. program are they/you using?

If not, what might I recommend instead?

Depending on what is being measured, often when repeated measures is unwieldy, one can calculate a univariate measure from their repeated measures, such as T3-T0, mean(T0, T1, T2, T3), or median(T0, T1, T2, T3), for example.

 

[TruthSeeker]

What stat. program are they/you using?

It is taken from a manuscript I'm reviewing. They do not mention the program they are using. I would guess SPSS

Depending on what is being measured, often when repeated measures is unwieldy, one can calculate a univariate measure from their repeated measures, such as T3-T0, mean(T0, T1, T2, T3), or median(T0, T1, T2, T3), for example.

Their question is whether the outcome changes in a hypothesized direction (it should increase) from t0 to T3 and whether the outcome is different between ill and healthy.

Are you implying in your response that testing segments (e.g. t2 versus t0) is not appropriate?

Thanks so much!

 

[Jorghnassen]

Damn, I did that kind of stuff back in the day. Just look up anova with repeated measures. You just add like one line in your SAS code (if SAS was used), and probably need to specify some correlation structure (AR1 will probably do the trick, if the time points are equally spaced).

Here's an example, though you'd want to replace the TYPE=CS by TYPE=AR(1) in the PROC MIXED example...