1

Sometimes I have a research question where I have maybe 10 different outcomes and I am interested in whether any of these 10 outcomes are associated with several other variables, such as weight, height, age, sex, alcohol use, symptoms etc.

Now the only way I can answer the question "does any of these 10 outcomes correlate with any of these predictors?" is by running 10 regression models, one for each outcome.

Is there a more elegant way? I've thought of multivariate regression but I'm not sure if it's fitting here?

Paze
  • 1,751
  • 7
  • 21
  • 1
    That's pretty much what [multivariate regression](https://stats.stackexchange.com/q/254254/28500) does. The regressions for the individual outcomes don't differ, but inference on the models can take into account the correlations among outcome types. – EdM Nov 18 '20 at 20:57
  • Would it be correct that I would need to adjust for every p value obtained, for each outcome? The same way I would if I ran 10 models? – Paze Nov 18 '20 at 21:03
  • If the question is simply that "does any of these 10 outcomes correlate with any of these predictors?", then I would prefer to do 10 regression models. – TrungDung Nov 18 '20 at 21:10
  • TDT thank you for chiming in. I'm just not sure because the outcomes in this case are a questionnaire questions concerning symptoms and quality of life and they may be correlated (such as memory loss and cognitive decline). – Paze Nov 18 '20 at 21:18
  • 2
    Maybe describe your actual problem and the data you have rather than theoretical one? If you *just* want to learn about correlation between multiple variables, calculate correlation matrix. – Tim Nov 18 '20 at 21:20

0 Answers0