Linear Mixed Model Interpretation

[Damian Ianno]

Hi All,
Have a dataset I am wishing to apply a linear mixed model to and have some questions hopefully the community can help with

Data:

• Dependent (Y) variable: 10 point scale (0-10)
• Independent (X) variables: Group (0 control/1 intervention)
• Repeated measures on same individual across 5 time points (y1-y5) since intervention.

Below is the raw data (reduced data-set from original)

* Example generated by -dataex-. To install: ssc install dataex
clear
input byte(id y1 y2 y3 y4 y5 group)
 1 6  .  . . 2 0
 2 6  8  . . . 0
 3 0  .  . . . 0
 4 0  2  2 1 1 0
 5 0  0  0 0 0 0
 6 2  .  . . 4 0
 7 3  .  . . 0 0
 8 .  .  . 4 4 0
 9 3  4  . 2 4 0
10 2  6  6 4 6 0
11 4  4  4 . 0 0
12 .  .  8 8 8 0
13 2  .  . 1 5 0
14 .  .  . 0 5 0
15 3  2  . 2 . 0
16 0  0  0 . 0 0
17 0  2  0 4 5 0
18 .  3  3 3 2 0
19 0  0  0 8 5 0
20 7  3  4 5 2 0
21 5 10  7 6 5 0
22 8  6  5 2 . 0
23 0  8  6 6 6 0
24 4  5  . 3 0 0
25 .  2  2 1 1 0
26 4  0  0 0 . 0
27 2  3  0 1 1 0
28 5  .  4 3 . 0
29 0  0  . . . 0
30 0  6  4 4 4 0
31 9  7  7 7 5 0
32 3  2  0 . 0 1
33 .  .  . . 2 1
34 5  4  . . . 1
35 0  0  0 0 2 1
36 2  1  0 1 1 1
37 .  .  . 4 3 1
38 .  .  4 4 5 1
39 .  2  2 . 0 1
41 .  5  5 4 3 1
42 .  .  . . . 1
43 4  2  2 3 3 1
44 0  0  0 2 0 1
45 3  2  3 0 2 1
46 .  0  0 0 . 1
47 2  0  2 . 1 1
48 0  3  2 2 3 1
49 0  3  3 . . 1
50 .  5  4 3 3 1
51 5  5  5 5 5 1
52 0  .  0 0 0 1
53 5  . 10 7 8 1
54 .  0  0 0 . 1
55 .  0  1 1 5 1
56 2  2  0 . 2 1
57 6  6  5 5 . 1
end

I then reshape the data to long format
reshape long y, i(id) j(time)

And apply the model
mixed y i.group##i.time || id:, covariance(unstructured) var

Then run a contrast command to assess treatment at each of the 5 time points
contrast group##time, effect

My question is on the interpretation of the STATA output:

1. Many papers produce box plots for two groups across multiple time points and state a ‘group time interaction effect’ as the final quoted result (see below, p=0.38). Does my study value indicated by red arrow above (p=0.7685) correspond to what the paper below mentions as their ‘group time interaction effect’?

2. In the study from which the figure above originates, the methods section of the paper states:

They have used time and group as fixed effects, study ID as a random effect and an interaction between time*group in the model. Based on the data above it seems I have an identical format to what this paper has used, would this be a correct interpretation based on the information available?

Many thanks and eagerly awaiting peoples reply.

Regards, Damian

[Damian Ianno]

.