1
$\begingroup$

I want to do logistic regression to test whether certain factors predict a loss of movement following an injury.

I have measured movement from 150 muscle groups that have wounded from about 40 people. I want to see if stitching has an effect on the range of movement that remains.

My independent variable is received stitching: yes/no
I also have a continuous independent variable (time, in hours until surgery) My dependent variable is movement lost: (yes/no) ...

Not everyone has sustained damage to all muscles, but some people have had a number of muscles "at-risk" of losing movement. When considered at a personal level, it means that these observations are, therefore, not independent. However, I intend to make no interpretations of this data at the person-level... only muscle-group level ... In which case, one person will only have one measurement taken of their left leg for example.

Does that mean that I can defend a decision to do a logistic regression if I only make interpretations at the muscular-group level?

If this isn't possible, does anyone have any ideas as to what I can do?

I've done some reading but I can't seem to get any answers ... The answer is simply "no", but I can't think of any alternative. I would appreciate any help and opinions.

Let's assume that my residuals are looking appropriately independent.

$\endgroup$
3
  • 2
    $\begingroup$ Why not use "person" as a random effect in your logistic regression model? $\endgroup$ Aug 5, 2021 at 15:20
  • $\begingroup$ I'd thought about that, does that mean that it ameliorates the "problem" of non-independence of my observations? $\endgroup$ Aug 6, 2021 at 11:31
  • $\begingroup$ That model does allow that observations are dependent, yes. $\endgroup$ Aug 6, 2021 at 14:06

1 Answer 1

1
$\begingroup$

As @BigBendRegion suggest, i think you should use a mixed model with random intercept for id. Further, do you expect that the surgery will take longer for people who get stitched, simply because stitching takes time. So maybe consider an interaction between your two independent variables.

Also, what about the age of people? It might affect the result - if you have access to that information?

$\endgroup$
2
  • $\begingroup$ I think the variable regarding surgery describes how much time has passed until the surgery happened, not how long the surgery took. Therefore, stitching and time until surgery should not have the correlation you mentioned. $\endgroup$
    – yuki
    Aug 6, 2021 at 8:49
  • 2
    $\begingroup$ - Kirsten, fantastic! - yes I was considering the interactions between variables, thanks for reminding me. - Yuki - You are right, the length of time to surgery is the factor... There are lots of other interesting observations I can look at though! Thanks for your input both :) $\endgroup$ Aug 6, 2021 at 11:37

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.