4
$\begingroup$

I am learning about mixed models and I have a question regarding the outcomes that can be considered. If I have hierarchical data, do the outcomes that I can consider need to belong to the lower level? I wonder if there is a need to aggregate the data when considering upper-level outcomes.

If I have measurements from several people belonging to different cities, does the outcome need to be patient-related? As life expectancy, for example.

And if I have data about certain measurements on different cells in various patients, can I apply mixed models to assess the relationship between the measurements and patient outcome?

Thanks!

Improve this question
$\endgroup$
3
  • 1
    $\begingroup$ Yes. You can use a mixed model if your outcome/dependent variable is on the lower level of your data hierarchy. You can't have a model predicting higher level variable from lower level variable (there was some attempt in the SEM world to create such a model, but I think the conclusion was that it was no better than averaging the lower-level variables within each cluster and using the averages as higher-level predictors). $\endgroup$
    – Sointu
    Commented Oct 17 at 11:05
  • $\begingroup$ "can I apply mixed models to assess the relationship between the measurements and patient outcome" - probably. Sounds like this would be a model with level 1 variable predicting level 1 variable. $\endgroup$
    – Sointu
    Commented Oct 17 at 11:06
  • $\begingroup$ The measurements belong to different cells, so the cells would be the lower level, right? @Sointu $\endgroup$
    – niqp
    Commented Oct 17 at 11:09

1 Answer 1

Reset to default
1
$\begingroup$

In general, everything you describe is possible, however, it is most easily accommodated in multilevel structural equation modeling (SEM). See this earlier thread for some more information.

The best program for having outcomes at both levels is Mplus, which allows you to specify variables that exist purely at the between group level and allows variables that are measured within groups to either be predicted by or predict variables at both levels. It employs latent mean centering of these latter variables, which accounts for sampling error in constructing group means for within group variables.

You can have up to three levels in your data hierarchy, which can be purely nested or crossed. For the distinction between these two should you not be certain, see this excellent response by @Robert Long.

I can see that there is some confusion about your data. If you edit your original question with more details about the data (the key variables of interest and what level they are measured at), I'm sure we can help you further.

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Thank you! I do not have specific data on this, I was just thinking about different scenarios of hierarchically structured data. $\endgroup$
    – niqp
    Commented Oct 18 at 13:42

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.