Hello Stata Forum,

I hope you all are doing well!

I wanted to clarify my utilization of the mixed linear model and whether it would more appropriate to say do an ANOVA or control for other variables.

I have a single-arm study evaluating the efficacy of a weight loss drug, drug A, in reducing a calculated weight score at 3 months from pre-intervention (month4). There are 10 study subjects as shown below with weight scores recorded at months 1, 2, and 3 (all recorded) after administration.

My objective is to determine if there is a significant difference between pre-intervention (month1) and post-intervention at 3 months (month4). Rather than complete a simple paired-sample T-test, I wanted to trend the change over time.
subject month1 month2 month3 month4
1 1.91 1.50 1.66 1.45
2 1.50 1.20 1.10 0.90
3 1.64 2.03 1.50 1.44
4 2.03 2.00 1.60 1.50
5 1.49 1.65 1.55 1.67
6 1.65 1.50 1.45 1.33
7 1.38 1.32 1.10 1.40
8 1.55 1.50 1.34 1.20
9 1.39 1.55 1.67 1.20
10 1.24 1.10 0.90 0.60
I first converted to long format from my data collection:

reshape long month, i(subject) j(time)

then used the repeated measures mixed model with random effects for time


mixed month time || subject:, var reml
margins, at (time=(1(1)4))
marginsplot, x(time)


Computing standard errors:

Mixed-effects REML regression Number of obs = 40
Group variable: subject Number of groups = 10

Obs per group:
min = 4
avg = 4.0
max = 4

Wald chi2(1) = 22.83
Log restricted-likelihood = 1.9729634 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
month | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
time | -.1075 .0224969 -4.78 0.000 -.1515932 -.0634068
_cons | 1.6035 .0848196 18.90 0.000 1.437257 1.769743
------------------------------------------------------------------------------

------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
subject: Identity |
var(_cons) | .0542298 .0285948 .0192934 .1524288
-----------------------------+------------------------------------------------
var(Residual) | .0253056 .0066456 .0151245 .0423402
------------------------------------------------------------------------------
LR test vs. linear model: chibar2(01) = 21.80 Prob >= chibar2 = 0.0000


I see it is significant but I also get a fixed portion prediction - am I controlling for time incorrectly? Also, would I be best served performing an ANOVA?

Thank you all