Correlation matrix and regression

[Lou France]

Hi!

This might be a very light question but when I read research papers, before giving the results of regressions, authors often show the correlation matrix.
I do understand what the correlation coefficient and its significance indicates but I don't understand the value of adding it in a research paper. What do you think?

[Paul Mueller]

Could you give an example?

[Lou France]

http://scholar.google.co.uk/scholar_...hVMVxoKHQb0DPo

[Rich Goldstein]

I have not looked at the paper you cite, but in my opinion there is no value; you might want to look up (i.e., google) the "society for the suppression of the correlation coefficient"

[Carlo Lazzaro]

Lou:
in similar fashion you can read: https://www.stat.berkeley.edu/~brill/Papers/jwtint4.pdf.
However, as much as I prefer -regress- to any correlation procedure, the correlation matrix (-estat vce, corr) makes sense after regression to sniff out (along with -estat vif-) quasi-extreme multicorrelation issues.

[Dick Campbell]

This was common practice in the social science some years back. One reason was, and still is, that with the correlation matrix, as well as means and sd's in hand, you can use the information to run additional regressions, experiment with alternative specifications etc. You can also use the matrix to run SEM models which allow one to extend published results in some cases. Also, back in the days before regression diagnostic packages, the correlation matrix gave some minimal information on collinearity etc. Of course for models with large numbers of variables printing the correlation matrix was not feasible and for many kinds of analyses such as logistic regression the correlation matrix is not sufficient. So, the practice has largely died out, at least in areas that I read regularly.