polyCov: Polychoric Covariances

In SachaEpskamp/polyCov: Estimate polychoric correlations and covariances.

Description

This function can be used to estimate a polychoric (or tetrachoric) correlation matrix assuming standardized latent items. If thresholds are given, it estimates polychoric covariances, means and variances.

Usage

polyCov(x, y = NULL, thresholds = NULL, ...)

Arguments

x	A matrix or dataframe containing ordinal data, or a vector containing ordinal data.
y	A vector containing ordinal data. Only if x is also a vector.
thresholds	An optional list containing thresholds for each item.
...	Arguments sent to optimize

Details

If 'thresholds' is not assigned, then the latent means and variances of each item are fixed to 0 and 1 respectively resulting in a correlation matrix. If ‘thresholds’ is assigned then the means and variances are estimated for each item and a covariance matrix is returned. For dichotomous items the variance is fixed to 1.

The polychoric correlations are estimated in a two-step procedure (Luo, 2011). first the thresholds are obtained from the marginal distribution of each variable and then the covariances are optimized to the frequency table.

Value

covmat	The estimated covariance matrix.
thresholds	A list containing the estimated thresholds.
means	A vector of estimated means.
vars	A vector of estimated variances.

Author(s)

Sacha Epskamp

References

Luo, H. (2011) Some Aspects on Confirmatory Factor Analysis of Ordinal Variables and Generating Non-normal Data. Retrieved from http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-149423

Examples

# Simulate data:
dat <- rnorm(1000) + matrix(rnorm(1000*5),,5)

# Real correlation matrix:
cor(dat)

# Transform data in ordinal scale:
datOrd <- ifelse(dat < -1,1,ifelse(dat>1,3,2))

# Polychoric correlations:
polyCov(datOrd)

SachaEpskamp/polyCov documentation built on May 9, 2019, 12:09 p.m.