how to detect if two outcome variables are collinear

[anna campo]

Hello everybody,
i hope the question is not too stupid. How do I recognize if two variables are describing somehow the same thing? I used them both as outcomes for regression models and they show similar results. One is a clinical measure and the second a need for support that could be justified by the clinical measure or not (heart dysfunction and need for respiratory support, to simplify). Patients who need respiratory support have worse hearts in the dataset, as expected. But some patients in the dataset could require resp support for reasons other than cardiological. I would like to know if, describing respiratory support in the cohort and relative risk factors, I am only describing the same patients with heart dysfunction. Thank you very much!
anna

[Carlo Lazzaro]

Anna:
welcome to this forum.
Collinearity affects independent variables, whereas you are talking about outcomes (that I read as dependent variables).
Probably, the simplest answer to your question boils down to the number of patients requiring respiratory support that with a negative anamnesis for cardiovascular diseases.
That said, you can get more helpful replies acting on the FAQ in posting your queries. Thanks.

[anna campo]

and if the (dependent) cardiovascular variable is a continuous one, with normal distribution, is a t-test enough to declare the two variables are describing the same thing? I do not think so, it can tell me they are correlated but I want to understand if describing both outcomes is redundant or ok. Thank you a lot

[Carlo Lazzaro]

Anna:
the t-test tells you whether or noy the difference of the means of the two continuous variables reaches statistical significance.
The t-test can be one or two-tailed and works (to a given extent) even though the two distributions from which the two samples were drawn are not normal.
Other tests (say, -pwcorr-) are useful to investigate the collinearity of two variables.
That said, if you're actually interested in respiratory support due to a cardiovascular disease, you can be better off with ruling out from the analysis those patients who need ventilation due to, say, respiratory tract infection or disease.

[Bruce Weaver]

Hello anna campo. I wonder if your use of the word "collinear" in the subject might be leading people down the wrong track. You wrote:

I would like to know if, describing respiratory support in the cohort and relative risk factors, I am only describing the same patients with heart dysfunction. Thank you very much!

I wonder if you really mean that you want to estimate a model with need for respiratory support as the (dichotomous) DV, and with the (quantitative?) measure of heart dysfunction as one of the explanatory variables. And specifically, I wonder if you are interested in whether that heart dysfunction variables interacts with other risk factors. I.e., you want to know if the usefulness of those risk factors in predicting need for respiratory support depends on the heart dysfunction score. Is that more or less what you are asking?