The estimation of population correlations is not common for survey data, but point estimates can be used for exploratory purposes. Two methods for estimating correlations for survey data are presented here.

Method 1: SURVEYREG Procedure

This is the easier of the two methods to implement, but it provides only point estimates.

Since the Pearson correlation coefficient is the same as a standardized regression coefficient, you can use the STB option in the MODEL statement in PROC SURVEYREG to request the standardized regression coefficient. The standardized estimate for X in the following example provides the population correlation coefficient between X and Y, but you should not use the standard error that is computed here:

Method 2: Replication Methods

By using replication methods, you can compute both the point estimate of the population correlation and its standard error.

Refer to Poisson Regressions for Complex Surveys, which details a method using the GENMOD procedure with replication methods to compute Poisson regressions for survey data. You can use the same technique to compute correlation coefficients and their standard errors by replacing PROC GENMOD with the CORR procedure.

The computation of p-values requires distributional assumptions that are not trivial for survey data, and it is not shown here.

Operating System and Release Information