Skip to contents

Spearman Rank Correlation

Source: R/spearmanCor.R
spearmanCor.Rd

Calculate Spearman correlation coefficient and its confidence interval. In addition to the base R function [cor](x, method="spearman"), frequency tables are also accepted as arguments (i.e. actually weights are used).

Usage

spearmanCor(
  x,
  y = NULL,
  conf.level = NA,
  sides = c("two.sided", "left", "right"),
  na.rm = FALSE
)

Arguments

x

either a contingency table, a two-column object (matrix, data.frame or list), or a vector of observations (together with y)

y

optional second vector. If x is not a vector, y must be NULL.

conf.level

confidence level of the interval. If set to NA (the default), only the point estimate is returned.

sides

character string specifying the sidedness of the confidence interval (one of "two.sided" (default), "left" or "right"). See ConfidenceIntervals().

na.rm

logical; whether to remove incomplete pairs. Applies to the vector interface; a frequency table must not contain missing counts.

Value

if conf.level = NA, a numeric scalar. Otherwise a named numeric vector with elements:

est

point estimate of Spearman's rank correlation

lci

lower confidence interval bound

uci

upper confidence interval bound

Details

The function calculates Spearman's rho statistic by means of cor(..., method="spearman") when two variables x and y are supplied. If a frequency table is provided an implementation based on SAS documentation is used. Both routes use midranks for ties and agree exactly; see the examples.
The confidence intervals are calculated via z-Transformation.

The number of observations entering the z-transformation is length(x) in the vector interface and sum(x), the table total, in the table interface. At least four observations are needed for an interval.

sides names the side of the interval that carries the finite bound, so "left" returns [lci, 1] and "right" returns [-1, uci]. Since rho is bounded, the open side is reported at the range boundary rather than as infinite.

References

Conover W. J. (1999) Practical Nonparametric Statistics (3rd edition). Wiley

See also

Association()

Other assoc.continuous: corPart(), corPolychor(), findCorrX(), hoeffdingD(), keepSig(), pearsonCor()

Examples


# Example from SAS documentation (PROC FREQ)
pain <- as.table(matrix(c(26,  6, 26, 7, 23, 
                           9, 18, 14, 9, 23), 
                           ncol=5, 
        dimnames=list(adverse=c("no", "yes"), dose=1:5)))

spearmanCor(pain)
#> [1] 0.3770609

spearmanCor(pain, conf.level=0.95)
#>       est       lci       uci 
#> 0.3770609 0.2361593 0.5024329 
  
# must be the same as
with(lapply(
       bedrock::untable(pain, 
                        colnames = c("adverse","dose")), 
       ordered), 
     spearmanCor(adverse, dose, conf.level=0.95))
#>       est       lci       uci 
#> 0.3770609 0.2361593 0.5024329