Skip to contents

svycor extends the survey package by calculating correlations with syntax similar to the original package, which for reasons unknown lacks such a function.


  na.rm = FALSE,
  digits = getOption("jtools-digits", default = 2),
  sig.stats = FALSE,
  bootn = 1000,
  mean1 = TRUE,



A formula (e.g., ~var1+var2) specifying the terms to correlate.


The or object.


Logical. Should cases with missing values be dropped?


An integer specifying the number of digits past the decimal to report in the output. Default is 2. You can change the default number of digits for all jtools functions with options("jtools-digits" = digits) where digits is the desired number.


Logical. Perform non-parametric bootstrapping (using wtd.cor) to generate standard errors and associated t- and p-values. See details for some considerations when doing null hypothesis testing with complex survey correlations.


If sig.stats is TRUE, this defines the number of bootstraps to be run to generate the standard errors and p-values. For large values and large datasets, this can contribute considerably to processing time.


If sig.stats is TRUE, it is important to know whether the sampling weights should have a mean of 1. That is, should the standard errors be calculated as if the number of rows in your dataset is the total number of observations (TRUE) or as if the sum of the weights in your dataset is the total number of observations (FALSE)?


Additional arguments passed to svyvar().


If significance tests are not requested, there is one returned value:


The correlation matrix (without rounding)

If significance tests are requested, the following are also returned:


A matrix of p values


A matrix of t values


A matrix of standard errors


This function extends the survey package by calculating the correlations for user-specified variables in survey design and returning a correlation matrix.

Using the wtd.cor function, this function also returns standard errors and p-values for the correlation terms using a sample-weighted bootstrapping procedure. While correlations do not require distributional assumptions, hypothesis testing (i.e., \(r > 0\)) does. The appropriate way to calculate standard errors and use them to define a probability is not straightforward in this scenario since the weighting causes heteroskedasticity, thereby violating an assumption inherent in the commonly used methods for converting Pearson's correlations into t-values. The method provided here is defensible, but if reporting in scientific publications the method should be spelled out.


This function was designed in part on the procedure recommended by Thomas Lumley, the author of the survey package, on Stack Overflow. However, he has not reviewed or endorsed this implementation. All defects are attributed to the author.

See also

wtd.cor, svymean()

Other survey package extensions: svysd()

Other survey tools: pf_sv_test(), svysd(), weights_tests(), wgttest()


Jacob Long


if (requireNamespace("survey")) {
 # Create survey design object
 dstrat <- svydesign(id = ~1, strata = ~stype, weights = ~pw,
                     data = apistrat, fpc = ~fpc)

 # Print correlation matrix
 svycor(~api00 + api99 + dnum, design = dstrat)

 # Save the results, extract correlation matrix
 out <- svycor(~api00 + api99 + dnum, design = dstrat)

#>           api00     api99      dnum
#> api00 1.0000000 0.9759047 0.2543484
#> api99 0.9759047 1.0000000 0.2441910
#> dnum  0.2543484 0.2441910 1.0000000