Skip to contents

Partial correlation controlling for covariates

Source: R/ops-timefreq.R
partialCorrelation.Rd

The correlation between x and y after linearly removing one or more covariates z from both – the standard tool for asking whether an association survives adjustment for a confound. With method = "spearman" the variables are rank-transformed first (a rank / robust partial correlation). The coefficient reproduces base R's residual method and the ppcor package to machine precision.

Usage

partialCorrelation(x, y = NULL, z = NULL, method = c("pearson", "spearman"))

Arguments

x

Either the first numeric variable (with y and z given), or a matrix/data frame whose column 1 is x, column 2 is y and the remaining columns are the covariates. Incomplete rows are dropped.

y

The second numeric variable when x is a single vector.

z

The covariate(s) to control for: a numeric vector or matrix.

method

"pearson" (default) or "spearman" (rank-based).

Value

A list with statistic (the partial correlation), p_value (two-sided t-test), df (n - 2 - k), n and k (number of covariates).

See also

correlationTest, kendallTau

Examples

x <- rnorm(50); z <- rnorm(50); y <- 0.5 * z + rnorm(50)
partialCorrelation(x, y, z)
#> $statistic
#> [1] -0.05034463
#> 
#> $p_value
#> [1] 0.7311977
#> 
#> $df
#> [1] 47
#> 
#> $n
#> [1] 50
#> 
#> $k
#> [1] 1
#>