Analysis of a dataset with respect to group and time course

[Diletta Malloggi]

Hi everyone,
I have a dataset where an index of functional recovery has been calculated in 6 different groups of animals (12 animals per group, each group received a different treatment). The index has been recorded for 34 weeks. So, I have 34 records (time points) for each animal.
I would like to assess if there are differences among the 6 groups, but also if there are differences over time.
I was wondering which test would be more appropriate to do this.
Is it correct to use repeated-measure two-way ANOVA?
Are there other models I could use?
Any suggestion is really appreciated.
Thank you for any help.

[Felix Bittmann]

Welcome to the forum! One solution to this kind of problem is as follows:

mixed depvar i.groupvar c.timevar || idvar:
margins group

*now testing for different trajectories
mixed depvar i.groupvar##c.timevar || idvar:


*testing for nonlinear effects
mixed depvar i.groupvar##c.timevar##c.timevar || idvar:
margins groupvar, at(timevar=(1(1)34))
marginsplot

[Diletta Malloggi]

Thank you very much!

What if the relationship between the index of functional recovery and the time is not linear?

[Felix Bittmann]

You can start with a purely descriptive approach to check this, like

scatter depvar time, by(groupvar)

[Diletta Malloggi]

Yes, sure.
The relationship between the index of functional recovery and time is not linear, so I asked if the mixed model was correct.

Many thanks!

[Felix Bittmann]

Well it depends on the kind of form. By introducing higher order terms (e.g. c.time##c.time) you can model squared, cubed,... forms. If this is not enough you could partition the time points into a few groups and then treat them as categorical (like i.timevar). Even more complex, you can estimate generalized ml models with meglm.

[Rich Goldstein]

if you believe the relationship is non-linear but are unsure of the form, try restricted cubic splines; see

h mkspline
or
h makespline