1
$\begingroup$

I have some people to make choices on two sets A(a1, a2, ..., an) and B(b1, b2,..., bm). That is, each person shall select a few elements from A and a few elements from B. Then I want to find out if A and B have correlations, particularly if any elements of A are correlated to any elements of B. How could I do it? Many thanks.

$\endgroup$
2
  • $\begingroup$ Does that mean that the values of a1, a2, ... an and of b1, b2, ... bm are all binomial? $\endgroup$ – russellpierce Jan 21 '13 at 17:45
  • $\begingroup$ no, some values of a1, a2, ... an and of b1, b2, ... bm may be an integer, indicating the weights. Thanks. $\endgroup$ – Ed Cloud Jan 21 '13 at 18:32
1
$\begingroup$

That sounds like a good application for Canonical Correlations Analysis. If you use R, you could check out the yacca package (Yet Another Canonical Correlations Analysis). The manual that comes with it explains things well.

CCA tries to find a linear combination of the X variables (Matrix 1) and a another linear combination of the Y variables (Matrix 2) with maximal correlation. You can examine the coefficients of these linear combinations to see which variables of X weight high with those of Y. You can do some significance testing as well (based on eigenvalues, as in PCA), but it is more of an exploratory tool in my view.

There is a green Sage book on CCA that explains the mechanics very well. "Canonical Correlations Analysis: Quantitative Applications in the Social Sciences"

CCA looks at correlations between the variables themselves. If you believe the correlations exist at the level of latent factors, you need to get into structural equation modelling. CCA would be a place to start, however.

$\endgroup$
1
  • $\begingroup$ Many hanks. I will have a look at CCA. $\endgroup$ – Ed Cloud Jan 21 '13 at 18:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.