plotVIF: Variance Inflation Factor Plot

In abnormally-distributed/Bayezilla: Bayesian Models With JAGS and Several Utilities for Data Exploration and Model Evaluation.

Description

Calculate the VIF for a least squares or generalized linear model. This is used to diagnose the ill effects of multicollinearity and collinearity in a regression model. This can be helpful in deciding to keep or drop a variable, or more preferably in some cases, use a regularized model. If the VIFs are all rather low, then using glmBayes is safe. If some are higher, but the matrix is still full rank, you might wish to use apcGlm. Otherwise, if the model is not full rank, not positive definite, or has a very high conditioning number, you may wish to use a ridge regression estimatior such as ridge. To obtain information about rank, positive definiteness, and the condition number, use the vitals function.

To use this simply input the formula, data, and family exactly as you would do with the glm() function. A horizontal dash is marked at 5, indicating a common point where many argue the variance inflation is problematic. Some have lower conservative (2) thresholds and some have higher liberal (10) thresholds, but 5 is one of the more common figures, i.e., Sheather (2009).

What this means is that if a variable is inflating the variance of the estimation by a factor of 5, the standard error of the corresponding coefficient is 2.236068 higher than it would be if it were not correlated with other variables.
An example of output:

Usage

plotVIF(formula, data, family = "gaussian")

Arguments

formula	the formula
data	the data
family	the glm family. Defaults to "gaussian"

References

Sheather, Simon (2009). A modern approach to regression with R. New York, NY: Springer. ISBN 978-0-387-09607-0.

Examples

plotVIF()

abnormally-distributed/Bayezilla documentation built on Oct. 31, 2019, 1:57 a.m.