2
$\begingroup$

I am measuring the association between advance pathology and functional outcome. However, I want to correct for another measure reflecting pathology in an earlier stage. Adding it as a covariate is problematic, because both measures of systemic pathology are highly collinear. Could I add the early pathology index as a weight to my regression model? Would that be a good approach? Appreciate your thoughts on this!

Update on what I am exactly aiming for:

I want to correct for the fact that some people had higher values in the early marker of systemic pathology and measure the association between the late marker and mobility, making sure that it is independent from the higher values of the early markers. But I cannot put it as a covariate because of collinearity. So, I am searching for another method to deal with this

Improve this question
$\endgroup$
3
  • 1
    $\begingroup$ It's unclear what you're proposing; a naive reading of your plain words makes no sense to me at all. How would making the covariate a weight be a suitable substitute for it being a covariate? One is talking about using it to model the conditional mean of the response while the other uses it to describe the relative precision of the response. Please clarify how the two models would be related. $\endgroup$
    – Glen_b
    Commented Apr 5, 2017 at 23:02
  • $\begingroup$ I want to correct for the fact that some people had higher values in the early marker of systemic pathology and measure the association between the late marker and mobility, making sure that it is independent from the higher values of the early markers. But I cannot put it as a covariate because of collinearity. So, I am searching for another method to deal with this. $\endgroup$
    – HIL
    Commented Apr 6, 2017 at 0:49
  • $\begingroup$ That's good information ... please put it in your question. $\endgroup$
    – Glen_b
    Commented Apr 6, 2017 at 1:19

1 Answer 1

Reset to default
0
$\begingroup$

One can use a weight estimation procedure, to determine an estimate of the weights to use best for weighted least squares regression, e.g., see in SPSS. I will expand on this answer a bit later on.

Improve this answer
$\endgroup$

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.