11-08-2012 12:25 AM
I am interested in estimating Spearman correlations with survey data. As I understand, the PROC CORR Spearman option is not valid when a WEIGHT statement is used. I have not seen an option for Spearman correlations in the SURVEY procedures. Any suggestions on how to calculate Spearman correlations and use a survey weight? Ideally, I would like to use the bootstrap weights for estimating the p-value for the Spearman coefficient
11-16-2012 10:44 AM
Spearman correlations are equivalent to Pearson correlations between the ranks of two variables and Pearson correlations are functions of totals. Therefore, you can estimate Spearman correlations using PROC SURVEYMEANS and a little programming. The method is described in the book "Model Assisted Survey Sampling" by Sardal, Swensson and Wretman. The essential formula that you will need for correlations is found in exercise 13.1.
There are also two Web Examples available at support.sas.com that demonstrate this method. The Web Examples demonstrate how to estimate the variance of a variable and the standard deviation of a variable; correlations are just an extension of the same idea.