0
$\begingroup$

I am dealing with several clinically used variables that are not linked to each others. Those variables are correlated to worsened prognosis (some are numeral and asymmetric, therefore correlations were considered using Mann-Whitney-U, others were nominal variables). All of these variables do have statistical significance and some even high sensitivities, specificities and positive predictive values. But as human lives are considered, there's always need for perfection.

Therefore, I would like to test several combinations of variables. How do I do that elegantly?

Improve this question
$\endgroup$
2
  • $\begingroup$ How does one consider correlations using Mann-Whitney? What kind of correlation do you intend? $\endgroup$
    – Glen_b
    Commented Dec 17, 2014 at 14:15
  • $\begingroup$ Is there some reason why you can't use all of the variables? For prediction/prognostication, throwing away predictor variables just throws away potentially useful information. $\endgroup$
    – EdM
    Commented Dec 18, 2014 at 19:28

1 Answer 1

Reset to default
0
$\begingroup$

Numeric variables can be analyzed with Principal Components Analysis to see if they measure the same underlying component. Variables that measure the same underlying component as determined by factor loadings in the PCA can be aggregated into a single variable. The identity of the components can be identified by determining what is similar about the variables loading onto the same component. The aggregated score representing a component can then be used in other analyses such as t-tests comparing whether gender influences the aggregated component score.

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.