Skip to contents

Polyserial correlation

Source: R/het_cor.R
polyserial.Rd

Estimates the correlation between a continuous variable and an ordinal variable, assuming the ordinal variable arises from cutting an underlying normal variable into categories. Uses the two-step estimator: thresholds from the marginal proportions, then maximum likelihood for the correlation (Olsson, Drasgow and Dorans 1982).

Usage

polyserial(x, y, max_cor = 0.9999)

Arguments

x

A numeric vector.

y

A factor of the same length as x. Its level order is used as the ordinal scale; an unordered factor gives a warning.

max_cor

Largest absolute correlation searched, between 0 and 1.

Value

The estimated correlation, using cases complete for both variables, or NA if y has fewer than two observed levels or x is constant.

References

Olsson U, Drasgow F, Dorans NJ (1982). The polyserial correlation coefficient. Psychometrika, 47, 337-347.

See also

het_cor(), polychoric()

Examples

set.seed(1)
x <- rnorm(500)
latent <- 0.6 * x + sqrt(1 - 0.6^2) * rnorm(500)
y <- cut(latent, c(-Inf, -1, 0, 1, Inf), ordered_result = TRUE)
polyserial(x, y)
#> [1] 0.5373007