Constructs plots of diagnostic measures for linear models.

Description

Used to construct plots of diagnostic measures for linear models. Also used to identify “extreme” values of diagnostic measures for a linear model.

Usage

  diagPlot(mdl)

Arguments

Details

This function produces a graphic that consists of at most six separate plots –

1. Studentized residuals versus leverages,

2. COVRATIO-1 versus fitted values,

3. Cook's Distance versus fitted values,

4. DFFits versus fitted values,

5. DFBetas for slope versus DFBetas for intercept, and

6. fitted line plot.

Each separate graph may have various individuals marked with their observation number. Observation numbers in red are the most extreme value that exceeds the cutoff value for the diagnostic measure plotted on that particular graph. Observation numbers in blue are observations that exceeded a cutoff value for at least one of the diagnostic measures NOT plotted on that particular graph. Thus, observations marked in red are “unusual” observations for the diagnostic measure shown on the plot whereas observations marked in blue are “unusual” observations for some other diagnostic measure but not for the diagnostic measure shown on the plot. The fitted line plot has all “unusual” observations marked with separate colors and the fitted line with that observation removed shown in the same color.

If more than one observation has the same extreme value for one of the diagnostics then only the first individual with the value is returned.

If the linear model object is other than a simple linear regression then only the first four plots are constructed.

Diagnostic statistic values are computed with the rstudent and influence.measaures functions.

Value

In addition to the graphic described in the details, a vector containing the row numbers of observations that were flagged as unusual by one of the diagnostic statistics. This vector can be assigned to an object and used to modify plots or easily remove the individuals from the data frame.

Note

This function is meant to allow newbie students the ability to easily construct plots for diagnosing “problem individuals” for one-way ANOVA, two-way ANOVA, simple linear regression, and indicator variable regressions. The plots can generally be constructed simply by submitting a saved linear model to this function. This function thus allows newbie students to interact with and visualize moderately complex linear models in a fairly easy and efficient manner.

See Also

fitPlot and residPlot from FSA; highlight; and influence.measures, outlierTest, and influence.plot in car.

Examples

require(FSA)      # for fitPlot
require(FSAdata)  # for Mirex data

data(Mirex)
Mirex$year <- factor(Mirex$year)

## Indicator variable regression with two factors -- As a general example
lm1 <- lm(mirex~weight*year*species,data=Mirex)
diagPlot(lm1)
## Simple linear regression
lm4 <- lm(mirex~weight,data=Mirex)
# Produces plot and saves flagged observations
pts <- diagPlot(lm4)
# Constructs a fitted line plot
fitPlot(lm4,cex.main=0.8)
# Highlights flagged observations on fitted-line plot
highlight(Mirex$mirex~Mirex$weight,pts=pts)


## Example showing outlier detection
x <- c(runif(100))
y <- c(7,runif(99))
lma <- lm(y~x)
diagPlot(lma)