1
$\begingroup$

I have a question on modelling a big data set (about 5000 subjects). I want to model a random intercept+slope, many of the subjects have only two observations (baseline and one follow-up), but some have more observations. I want to include baseline as a covariate and I am wondering whether I have to remove the observations in t=0 or if I can use them "twice" by leaving them in the data set but also using them for estimating the coefficient of baseline... hopefully my problem became clear!

The second thing concerns the scaling of the variables- I am using R and just implementing it in the formula by scale(Time), scale(Baseline) etc. However, I am a bit confused about interpreting the results, would it be better to somehow do scaling by hand in advance?

Thanks a lot!

Improve this question
$\endgroup$

1 Answer 1

Reset to default
3
$\begingroup$

If you are going to include the value of your outcome at baseline as a covariate, then indeed, you should not also leave this baseline value in the response variable.

Nonetheless, if you are going to fit a mixed model, you are not required to include the baseline value as a covariate. Especially when you have more than two measurements, doing so implies that the baseline value is equally correlated with all subsequent measurements, which is often not the case.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Thanks a lot for your reply! Actually the baseline covariate was chosen because we are interested in showing that this value has a large impact on the outcome. However, if we remove the baseline value from the response as it would be technically correct, the model, incorporating a random slope, can't be fit anymore as there are numerous subjects with only one observation. So you would recommend to remove the baseline covariate instead? But how can we then interpret the effect of this baseline value on the response? $\endgroup$
    – Kathrin
    Jul 6, 2020 at 12:13
  • $\begingroup$ @Dimitris Rizopoulos I have the same question can you perhaps explain it more? is it correct to adjust for baseline if you also include baseline (t = 0 ) in your response in linear mixed model? $\endgroup$
    – user358238
    Sep 29 at 12:26
  • $\begingroup$ @Frank Harrell your comment is valuable. $\endgroup$
    – user358238
    Sep 29 at 12:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.