rockchalk (version 1.6.2)

plotSlopes: Assists creation of predicted value lines for values of a moderator variable.

Description

This is a "simple slope" plotter for linear regression. The term "simple slopes" was coined by psychologists (Aiken and West, 1991; Cohen, et al 2002) to refer to analysis of interaction effects for particular values of a moderating variable, be it continuous or categorical. To use this function, the user should estimate a regression (with as many variables as desired, including interactions) and the resulting regression object is then supplied to this function, along with user requests for plots of certain variables.

Usage

plotSlopes(model = NULL, plotx = NULL, modx = NULL,
    modxVals = NULL, plotPoints = TRUE, plotLegend = TRUE,
    col, llwd, ...)

Arguments

model

Required. Fitted regression object. Must have a predict method

plotx

Required. String with name of IV to be plotted on x axis

modx

Required. String for moderator variable name. May be either numeric or factor.

modxVals

Optional. If modx is numeric, either a character string, "quantile", "std.dev.", or "table", or a vector of values for which plotted lines are sought. If modx is a factor, the default approach will create one line for each level, but the user can

plotPoints

Optional. TRUE or FALSE: Should the plot include the scatterplot points along with the lines.

plotLegend

Optional. TRUE or FALSE: Include a default legend. Set to FALSE if use wants to run a different legend command after the plot has been drawn.

col

Optional. A color vector. By default, the R's builtin colors will be used, which are "black", "red", and so forth. Instead, a vector of color names can be supplied, as in c("pink","black", "gray70"). A color-vector generating function like rain

llwd

An optional vector of line widths used while plotting the lines that represent the values of the factor. This applies only to the lines in the plot. The ... argument will also allow one to pass options that are parsed by plot, such as lwd. That de

...

further arguments that are passed to plot

Value

• The plot is drawn on the screen, and the return object includes the "newdata" object that was used to create the plot, along with the "modxVals" vector, the values of the moderator for which lines were drawn. It also includes the call that generated the plot.

Details

The variable plotx will be the horizontal plotting variable; it must be numeric. The variable modx is the moderator variable. It may be either a numeric or a factor variable. A line will be drawn to represent the predicted value for selected values of the moderator.

The parameter modxVals is optional. It is used to fine-tune the values of the moderator that are used to create the simple slope plot. Numeric and factor moderators are treated differently. If the moderator is a numeric variable, then some particular values must be chosen for plottings. If the user does not specify the parameter modxVals, then lines will be drawn for the quantile values of the moderator. If the moderator is a factor, then lines are drawn for each different value of the factor variable, unless the user specifies a subset of levels with the modxVals parameter.

For numeric moderators, the user may specify a vector of values for the numeric moderator variable, such as c(1,2,3). The user may also specify an algorithm, either "quantile" (which would be selected by default) or "std.dev." The alternative method at this time is "std.dev.", which causes 5 lines to be drawn. These lines are the "standard deviations about the mean of modx" lines, at which modx is set at mean - k* standard deviation, and k takes on values -2, -1, 0, 1, 2.

Here is a wrinkle. There can be many variables in a regression model, and we are plotting only for the plotx and modx variables. How should we calculate predicted values when the values of the other variables are required? For the other variables, the ones that are not explicitly inlcluded in the plot, we use the mean and mode, for numeric or factor variables (respectively). Those values can be reviewed in the newdata object that is created as a part of the output from this function

References

Aiken, L. S. and West, S.G. (1991). Multiple Regression: Testing and Interpreting Interactions. Newbury Park, Calif: Sage Publications.

Cohen, J., Cohen, P., West, S. G., and Aiken, L. S. (2002). Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences (Third.). Routledge Academic.

See Also

plotCurves and testSlopes

Examples

set.seed(12345)
x1 <- rnorm(100)
x2 <- rnorm(100)
x3 <- rnorm(100)
x4 <- rnorm(100)
xcat1 <- gl(2,50, labels=c("M","F"))
xcat2 <- cut(rnorm(100), breaks=c(-Inf, 0, 0.4, 0.9, 1, Inf), labels=c("R", "M", "D", "P", "G"))
dat <- data.frame(x1, x2, x3, x4, xcat1, xcat2)
rm(x1, x2, x3, x4, xcat1, xcat2)

##ordinary regression. 
dat$y <- with(dat, 0.03 + 0.1*x1 + 0.1*x2 + 0.4*x3 -0.1*x4 + 2*rnorm(100))
m1 <- lm(y ~ x1 + x2 +x3 + x4, data=dat)
## These will be parallel lines emf 
plotSlopes(m1, plotx="x1", modx="x2")
plotSlopes(m1, plotx="x1", modx="x2", modxVals=c(-0.5,0,0.5))
plotSlopes(m1, plotx="x1", modx="x2", modxVals="std.dev.", main="A plotSlopes result with \"std.dev.\"values of modx")


plotSlopes(m1, plotx="x1", modx="x2", modxVals="std.dev.", ylab="Call Y What You Want")
plotSlopes(m1, plotx="x1", modx="x2")
plotSlopes(m1, plotx="x4", modx="x1")


## now some numeric interactions worth plotting
dat$y2 <- with(dat, 0.03 + 0.1*x1 + 0.1*x2 + 0.25*x1*x2 + 0.4*x3 -0.1*x4 + 1*rnorm(100))

m2 <- lm(y2 ~ x1*x2 + x3 + x4, data=dat)
summary(m2)
plotSlopes(m2, plotx="x1", modx="x2")
plotSlopes(m2, plotx="x1", modx="x2", modxVals=c( -2, -1, 0, 1))
plotSlopes(m2, plotx="x2", modx="x1", modxVals="std.dev.")
plotSlopes(m2, plotx="x2", modx="x1", modxVals="std.dev.", xlab="Any label You Want")

## Catch output, send to testSlopes

m2ps1 <- plotSlopes(m2, plotx="x1", modx="x2")
testSlopes(m2ps1)

### Examples with categorical Moderator variable

stde <- 8
dat$y3 <- with(dat, 3 + 0.5*x1 + 1.2 * (as.numeric(xcat1)-1) +
-0.8* (as.numeric(xcat1)-1) * x1 +  stde * rnorm(100))

m3 <- lm (y3 ~ x1 + xcat1, data=dat)
plotSlopes(m3, modx = "xcat1", plotx = "x1")

m4 <- lm (y ~ x1 * xcat1, data=dat)
summary(m4)
plotSlopes(m4, modx = "xcat1", plotx = "x1")

dat$xcat2n <- with(dat, contrasts(xcat2)[xcat2, ])
dat$y4 <- with(dat, 3 + 0.5*x1 + xcat2n %*% c(0.1, -0.2, 0.3, 0.05)  + stde * rnorm(100))
m5 <- lm(y4 ~ x1 + xcat2, data=dat)
plotSlopes(m5, plotx="x1", modx="xcat2")
m6 <- lm(y4 ~ x1 * xcat2, data=dat)
plotSlopes(m6, plotx="x1", modx="xcat2")

## Make data with a more pronounced interaction
dat$y5 <- with(dat, 3 + 0.5*x1 + xcat2n %*% c(0.1, -0.2, 0.3, 0.05)  + (xcat2n %*% c(0.-1, 0.2, -0.3, 0.25)  )*x1 + stde * rnorm(100))
m7 <- lm(y4 ~ x1 * xcat2, data=dat)
plotSlopes(m7, plotx="x1", modx="xcat2")
##only plot first and third levels
m7ps <- plotSlopes(m7, plotx="x1", modx="xcat2", modxVals=levels(dat$xcat2)[c(1,3)]) 
##see what testSlopes says about this one
##testSlopes(m7ps)

## Now examples with real data
library(car)
m3 <- lm(statusquo ~ income * sex, data = Chile)
summary(m3)
plotSlopes(m3, modx = "sex", plotx = "income")


m4 <- lm(statusquo ~ region * income, data= Chile)
summary(m4)
plotSlopes(m4, modx = "region", plotx = "income")

plotSlopes(m4, modx = "region", plotx = "income", plotPoints=FALSE)


m5 <- lm(statusquo ~ region * income + sex + age, data= Chile)
summary(m5)
plotSlopes(m5, modx = "region", plotx = "income")

m6 <- lm(statusquo ~ income * age + education + sex + age, data=Chile)
summary(m6)
plotSlopes(m6, modx = "income", plotx = "age")

plotSlopes(m6, modx = "income", plotx = "age", plotPoints=FALSE)


## Should cause error because education is not numeric
## m7 <- lm(statusquo ~ income * age + education + sex + age, data=Chile)
## summary(m7)
## plotSlopes(m7, modx = "income", plotx = "education")

## Should cause error because "as.numeric(education") not same as
## plotx="education"
## m8 <- lm(statusquo ~ income * age + as.numeric(education) + sex + age, data=Chile)
## summary(m8)
## plotSlopes(m8, modx = "income", plotx = "education")

## Still fails. 
## plotSlopes(m8, modx = "income", plotx = "as.numeric(education)")

## Must recode variable first so that variable name is coherent
Chile$educationn <- as.numeric(Chile$education)
m9 <- lm(statusquo ~ income * age + educationn + sex + age, data=Chile)
summary(m9)
plotSlopes(m9, modx = "income", plotx = "educationn")