We wish to summarize several dichotomous variables of individual level data into an index using Principle Component Analysis (PCA). The code we use is below. We have two questions.
Code:
tetrachoric TeamDept Company Shares Benefits FixedInc, stats(rho obs) posdef mat def pay_mat = r(Rho) local n = r(N) pcamat pay_mat, n( `n' )
Q2: We wish to ensure the same Principle Component emerges in every country. This would mean (in our opinion) that comparable item loadings and eigenvalues are extracted across countries. The Stata documentation on PCA mentions we could test eigenvalues and loadings using testparm. However the data must have a multivariate normal distribution. Our data is not. Is there another way to test eigenvalues and item loadings across countries?
0 Response to Principle Component Analysis: using weights and comparison of PCA between countries
Post a Comment