statistics.com

[Vlob]

Hello!

I am trying to generate a 9-dimensional Latin Hypercube sample (1000 draws or so) where each dimension variable is normally distributed with zero average and standard deviation equal to 1. The variables are required to be correlated in a particular way, the desired correation matrix is known.

Using the Iman-Conover technique I can get a sample with the correlation matrix which is pretty close to the required one. However I found out that the obtained matrix was always biased a little. The difference from the required matrix is a few percents, and it is stable as I repeat the procedure or try bigger sample.

I suspect my problem is that the Iman-Conover method reproduces the Rank correlations while my goal is to reproduce Pearson correlations.

Could anybody help? How can I generate a Latin Hypercube sample with a required Pearson correlation matrix?

[alethephant]

See:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.30.5269&rep=rep1&type=pdf

http://www3.stat.sinica.edu.tw/statistica/password.asp?vol=8&num=3&art=19

http://epp.eurostat.ec.europa.eu/portal/page/portal/research_methodology/documents/83.pdf

Note the comparison or rank and Pearson correlation in the last article.

[Vlob]

Thanks a lot!
This helps.