Calculating Risk for Repeated Measures with Varying & Overlapping Predictors

[Malaika Schwartz]

I have data on several measures for 150 survey respondents: Infection (binary outcome of interest), Hormonal BC (binary predictor #1 of interest), Non-Hormonal BC (binary predictor #2 of interest) and Age, which have been collected for the years 2012-2016. I have been asked to determine the risk of infection due to hormonal BC use compared to non-hormonal BC use, adjusting for age.

My problem is, not only does the outcome of interest vary over time, the predictors do too, and there can be overlap between predictors. Respondents could have an infection in 2012 while on hormonal BC, have an infection in 2013 while on non-hormonal BC, and not have an infection in 2014 while on hormonal BC and non-hormonal BC. (See table below). This means I not only have to account for repeated measures, but also for interactions between BC use, and correlation between BC use and infection in the same year.

How do I go about setting up this analysis? This is way over my head and I don't think -xtgee using StudyID for the panel variable is quite right.

StudyID	Year	Infection	Age	Hormonal	NonHorm
1	2012	0	24	0	0
1	2013	1	25	0	0
1	2014	0	26	0	0
1	2015	1	27	1	0
1	2016	1	28	0	0
2	2012	1	26	1	1
2	2013	1	27	1	1
2	2014	1	28	1	1
2	2015	1	29	0	0
2	2016	1	30	0	0
3	2012	1	41	0	0
3	2013	1	42	0	0
3	2014	1	43	0	0
3	2015	1	44	0	0
3	2016	1	45	0	0
4	2012	0	23	0	1
4	2013	0	24	0	1
4	2014	0	25	0	1
4	2015	0	26	0	1
4	2016	0	27	0	1
5	2012	0	27	1	0
5	2013	0	28	1	0
5	2014	0	29	1	0
5	2015	0	30	0	0
5	2016	1	31	1	1

[Phil Bromiley]

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.

This has the feel of a course assignment - we don't help with homework. On the off chance is it not, let me note that you have panel data with a binary outcome so xtlogit or xtprobit look likely.

[Malaika Schwartz]

This has the feel of a course assignment - we don't help with homework. On the off chance is it not, let me note that you have panel data with a binary outcome so xtlogit or xtprobit look likely.

Thank you Phil. Unfortunately, this is not a course assignment, this was a work analysis that was assigned to someone who hasn't taken a stats course in many, many years and so has only done basic descriptive analyses and a couple regressions, all on cross-sectional data. If I had any training on how to deal with panel data, I've completely forgotten it. I'll look into xtlogit.