Longitudinal data analysis

[Aleks Leligdowicz]

I am analyzing a data set of before/after biomarker values in 2 groups. I have baseline continuous characteristic variables for samples in both groups which I'd like to include in some models and compare them to unadjusted models of only before/after change in values between groups. What is the most appropriate way to model the data?

I am considering:
1) long format data with before/after time points using xtgee --> xtgee biomarker time##i.group, i(DonorID) robust
*to test effect of baseline variable, followed the above model with --> xtgee biomarker time##i.group adjustmentVar , i(DonorID) robust
2) long format data with before/after time points using mixed effects model
3) wide format data linear regression predicting the "after" biomarker variable, adjusting for the "before" biomarker variable, using lrtest for comparison of models without vs with adjustment variable
4) wide format ttest of change in biomarker values, comparing condition 1 to condition 2 (not sure how to adjust for baseline characteristics if this format is used).

I have between 17-22 samples per group
For linear regression (option 3), the data would be log10 transformed for the biomarker values, but the adjustment variables are not normally distributed (even after transformation).
I think the xtgee is most appropriate but would greatly value help with this analysis.

[Phil Bromiley]

Welcome to Statalist. You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Being able to replicate your problem is often essential to helping you.

You only have 2 groups and roughly 40 observations? I don't think many of these panel estimators are really legitimate with such small samples. Is group membership uncorrelated with the other included variables?

With 40 observations, I would be strongly inclined to something simple like xtreg or regress with i.group as a control variable. However, I'm not from your area and norms vary.