Repeated measures analysis - anova or mixed

[Megan Moreton]

Hi,

My data has 3 treatment groups each with results at 3 time points (baseline & 2 equally spaced follow-ups). I want to assess the effects of treatment group and time on the outcome. I use dataex to provide an example of my data structure below.

I have tried a few ways of doing this & would appreciate any advice on which is better/any preferences. Initially, to get a sense of the data, I tried simple t-tests comparing the results at each timepoint between treatment groups (2 at a time), and using paired tests to compare each visit to baseline by treatment group.

I would now like to use either a repeated measures anova or a mixed model to evaluate the effects. I use the following commands:

// ANOVA
anova result group / id|group interval group#interval, repeated(interval)
contrast interval@group, effect mcompare(bonferroni)
anova result group##interval
contrast group@interval, effect mcompare(bonferroni)

// MIXED
mixed result group##interval || id:, var reml
contrast group##interval
contrast interval@group, effect mcompare(bonferroni)
contrast group@interval, effect mcompare(bonferroni)

Do these codes seem appropriate? & is there a preference (statistical or other) over which to use?

Thank you for any help, Megan.

* Example generated by -dataex-. To install: ssc install dataex
clear
input long(group id) float interval double result
3  1 1   3.4
3  1 2 1.425
3  1 3  2.55
2  2 1   4.3
2  2 2 2.325
2  2 3 1.875
3  3 1  9.45
3  3 2   7.4
3  3 3 6.975
3  4 1 5.925
3  4 2  4.15
3  4 3  4.25
2  5 1  4.45
2  5 2 2.525
2  5 3 7.575
2  6 1 4.175
2  6 2   6.4
2  6 3 6.825
1  7 1 6.125
1  7 2 7.075
1  7 3   5.1
3  8 1 4.975
3  8 2 2.675
3  8 3 2.625
1  9 1   3.7
1  9 2   7.1
1  9 3 7.875
2 10 1 6.675
2 10 2  6.25
2 10 3   5.2
1 11 1   3.2
1 11 2 4.125
1 11 3 4.075
2 12 1 5.025
2 12 2  3.05
2 12 3 6.325
1 13 1   1.9
1 13 2  4.25
1 13 3 5.475
1 14 1   5.2
1 14 2 4.025
1 14 3 7.625
3 15 1 4.775
3 15 2   4.9
3 15 3  4.45
1 16 1 2.325
1 16 2     5
1 16 3 6.175
1 17 1  5.05
end
label values group group
label values id id

[Marcos Almeida]

Welcome to the Stata Forum / Statalist,

Thanks for having used - dataex - as well as for having clearly stated your query.

In such a study design, I tend to use - mixed - instead of ANOVA.

That said, if you type - margins interval#group - followed by - marginsplot - you may probably understand what is happening.

P.S: please keep in mind that we have to use factor notation with - mixed - before apply the above-mentioned - margins - command.