Dear all,
I have a question as follows: I want calculate the partial correlation coefficient between X and Y. The results of simple linear regression showed that variable A has significant correlation with X, but not with Y; variable B has significant correlation with Y, but not with X; A has significant correlation with B.
So, can I choose A and B as control variables?
Yes, you can use A and/or B as control variables. But first, consider the following: 1) if one has observed correlation (between A and X, or between B and Y), that correlation alone does not establish a relationship between A and X (or between B and Y). Furthermore, correlation exists on a continuum from -1 to +1. What do you mean by "uncorrelated"? CORR = 0? CORR 0, but p>0.05? >0.10? 95%CI = (-0.01 to 0.25)? (etc). 2) When considering whether/how to adjust your model by adding A or B, you have to ask yourself whether A or B make sense as covariates, based on the literature relevant to your study question. If B is age, for example, and Y is colon cancer risk, then it makes sense to adjust for age. 3) you haven't mentioned whether A and B are correlated with each other. If they are moderately to strongly correlated within your sample, then adding both of them to your correlation model may introduce a level of collinearity that will obscure the underlying association.
Thank you very much for your good answers! In my question, "uncorrelated " means no significant correlation (P>0.05). Moreover, A has significant correlation with B, but no multicollinearity were found among them (variance inflaction factor
I also agree with Mary's opinion.
I want to add one mention. Physicians must have consider, first of all, a logical association between variables in their academic feild. If there are enough the logically association, you can use the variable.
For instance Basically age is used co-related factor in health science.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis