Repeated Measures Data: Treatment and Time Interaction in xtmixed command

[Sule Yaylaci]

Dear Statalist users,
I am using Stata 14 SE. The dataset I am working on is repeated measures survey data, gathered before and after an experiment (one year after). ID is the code for each individual, and wave refers to Wave 1 and Wave 3. (There was also a wave 2 but it is not part of this dataset). The dependent variable is 'factor1_w'. The treatment variable is 'Group'. Group that received the treatment is Group==1.
I am trying to understand if the treatment had caused a significant change on the variable titled 'factor1_w'.


I ran three xtmixed commands: one where I interact Group and the time variable (wave), and another where I limit the analysis to the Control and then the Treatment group and add 'wave' as a covariate.
I expect the results to be the same, but they are not.
While the interaction of time and treatment shows no significant effect, the analysis where I look at only the Treatment Group shows that 'wave' has a statistically significant negative effect, and the analysis on Control Group only shows that 'wave' does not have any significant effect.

If time (variable 'wave') has a significant effect on the dependent variable in the Treatment Group, should not it also be present in the effect of the interaction term?

I placed the sample data below, using dataex .
The two commands and the output is also below.

Any help is appreciated.
Regards,
Sule

clear
input double ID byte wave double(Group Male NRGAgeN NRGEducN) float factor1_w
11001 1 0 0 2 1  .7437461
11001 3 0 0 2 1  .5075265
11002 1 0 1 3 2  .9770322
11002 3 0 1 3 2         .
11003 1 0 0 3 3  1.579619
11003 3 0 0 3 3         .
11004 1 0 0 2 3 1.9388603
11004 3 0 0 2 3 1.1451527
11005 1 0 0 3 4 1.5843916
11005 3 0 0 3 4  .9314132
end

xtmixed factor1_w wave##Group || ID:, var reml

xtmixed factor1_w i.wave if Group==1 || ID:, var reml
xtmixed factor1_w i.wave if Group==0 || ID:, var reml

[Joseph Coveney]

While the interaction of time and treatment shows no significant effect, the analysis where I look at only the Treatment Group shows that 'wave' has a statistically significant negative effect, and the analysis on Control Group only shows that 'wave' does not have any significant effect.

If time (variable 'wave') has a significant effect on the dependent variable in the Treatment Group, should not it also be present in the effect of the interaction term?

Nope.

regress trunk c.weight##i.foreign
regress trunk c.weight if foreign
regress trunk c.weight if !foreign

[Sule Yaylaci]

Thanks for the link, Joseph. Much appreciated.