visreg: Visualization of regression functions for a single variable

Description

A function used to visualize regression models quickly and easily. Default plots contain a confidence band, prediction line, and partial residuals. Factors, transformations, conditioning, interactions, and a variety of other options are supported.

Usage

visreg(fit, xvar, by, breaks=4, type=c("conditional","effect"),
trans=as.numeric, scale=c("linear","response"), xtrans, alpha=.05,
nn=101, cond=list(), whitespace=0.2, partial=TRUE, jitter=FALSE,
strip.names=FALSE, ...)

Arguments

fit

This is the fitted model object you wish to visualize. lm, glm, gam, rlm, and coxph data classes are all supported.

xvar

This is the variable to be put on the x-axis of your plot. Both continuous variables and factors are supported.

An option to use if you wish to divide your plot by a third variable. Uses the lattice package. Both continuous variables and factors are supported.

breaks

If a continuous variable is used for the 'by' option, the 'breaks' argument determines how many quantiles will be divided to be used for panels.

type

The type of plot to be produced. The following options are supported:

If 'conditional' is selected, the plot returned shows the value of the variable on the x-axis and the change in response on the y-axis, holding all other variables constant

trans

Allows transformations to the response. A function should be passed (see examples).

scale

If response is selected and a glm is passed as the fit parameter, the response will automatically be transformed.

xtrans

Allows transformations to the x-axis. A function should be passed. Note that inverse transformations to explanatory variables are performed automatically by visreg.

alpha

Alpha level for the confidence band displayed in the plot.

Controls the smoothness of the line and confidence band. The higher the number, the smoother they will appear.

cond

Used to set variables to specific values. Can also be used to select the reference point for an effect plot. Named lists should be passed. All variables left unspecified will be filled in with the median/most common category (see examples).

whitespace

When xvar is a factor, whitespace determines the ammount of space in between factors on the x-axis.

partial

Determines if partial residuals are shown on the plot.

jitter

Adds a small amount of noise to xvar. Potentially useful if many observations have exactly the same value. Default is FALSE.

strip.names

When by=TRUE, adds the name of the by variable to the strip at the top of each panel. Default is FALSE.

...

Graphical parameters can be passed to the function to customize the plots. If by=TRUE, lattice parameters can be passed, such as layout (see examples below).

Value

In addition to providing plots, the visreg function also invisibly returns the data frames, estimates, confidence intervals, and residuals used in the creation of its plots (see last example).

References

Breheny, P. and Burchett, W. (2012), Visualizing regression models using visreg. http://web.as.uky.edu/statistics/users/pbreheny/publications/visreg.pdf

Examples

Run this code

## Linear models

## Basic
fit <- lm(Ozone ~ Solar.R + Wind + Temp,data=airquality)
visreg(fit)
visreg(fit,"Wind",type="effect")
visreg(fit,"Wind",type="conditional")
visreg(fit,c("Solar.R","Wind","Temp"))

## Factors
airquality$Heat <- cut(airquality$Temp,3,labels=c("Cool","Mild","Hot"))
fit.heat <- lm(Ozone ~ Solar.R + Wind + Heat,data=airquality)
visreg(fit.heat,"Heat",type="effect")
visreg(fit.heat,"Heat",type="conditional")
## Use of whitespace option
visreg(fit.heat,"Heat",whitespace=.1)
visreg(fit.heat,"Heat",whitespace=.5)

## Transformations
fit1 <- lm(Ozone ~ Solar.R + Wind + Temp + I(Wind^2),data=airquality)
fit2 <- lm(log(Ozone) ~ Solar.R + Wind + Temp,data=airquality)
fit3 <- lm(log(Ozone) ~ Solar.R + Wind + Temp + I(Wind^2),data=airquality)
visreg(fit1,"Wind")
visreg(fit2,"Wind",trans=exp,ylab="Ozone")
visreg(fit3,"Wind",trans=exp,ylab="Ozone")

## Conditioning
visreg(fit,"Wind",cond=list(Temp=50))
visreg(fit,"Wind")
visreg(fit,"Wind",cond=list(Temp=100))

## Interactions
fit.in1 <- lm(Ozone~ Solar.R + Wind*Heat,data=airquality)
visreg(fit.in1,"Wind",by="Heat")
visreg(fit.in1,"Heat",by="Wind")
visreg(fit.in1,"Wind",by="Heat",type="effect")
visreg(fit.in1,"Wind",cond=list(Heat="Cool"),type="effect")
visreg(fit.in1,"Wind",cond=list(Heat="Hot"),type="effect")


## Nonlinear models

data("birthwt",package="MASS")
birthwt$race <- factor(birthwt$race,labels=c("White","Black","Other"))
birthwt$smoke <- factor(birthwt$smoke,labels=c("Nonsmoker","Smoker"))

fit <- glm(low~age+race+smoke+lwt,data=birthwt,family="binomial")
visreg(fit,"lwt",xlab="Mother's Weight",ylab="Log odds (low birthweight)")
visreg(fit,"lwt",scale="response",partial=FALSE,xlab="Mother's Weight",ylab="P(low birthweight)")

require(survival)
data(ovarian)
ovarian$rx <- factor(ovarian$rx)

fit <- coxph(Surv(futime,fustat)~age+rx,data=ovarian)

visreg(fit,"age",ylab="log(Hazard ratio)")

require(MASS)
fit <- rlm(Ozone ~ Solar.R + Wind*Heat,data=airquality)
visreg(fit,"Wind",cond=list(Heat="Mild"))

v <- visreg(fit,"Wind",cond=list(Heat="Mild"))