1
$\begingroup$

I have a set of 30 patients. Each patient has a variable number of observations over a variable length of time(ranging from 4 to 10 data points per patient).

My outcome variable is whether they have gained or loss/stabilised their weight at any given point in time. I calculate this weight gain/loss by taking their weight on a given observation and compare it to the previous observation. Creating a dichotomous nominal Loss / Gained Weight variable.

I have a few continuous explanatory variables. some are normally distributed and others non-normal.

Can i treat my outcome variable as "independent observations" and use logistic regression modeling? Otherwise, is there any suggestion for which model to use?

$\endgroup$
0

1 Answer 1

2
$\begingroup$
  1. Yes, this is a repeated measures situation.

  2. It would probably be better to model the weight directly and not the change between points. But even if the change is what you are interested in, you should definitely not just turn it into a binary gained/lost variable.

  3. The distribution of your predictors is not relevant. The distribution of the model residuals is generally important.

Overall, this is not a lot of data to deal with the complexities you are after (30 patients, 4-10 time points, many predictors).

$\endgroup$

Your Answer

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

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