rockchalk (version 1.6.3)

testSlopes: Hypothesis tests for Simple Slopes Objects

Description

Conducts t-test of the hypothesis that the "simple slope" line for one predictor is statistically significantly different from zero for each value of a moderator variable. The user must first run plotSlopes(), which generates an object that can be investigated. This function scans that object to conduct the hypothesis tests.

Usage

testSlopes(pso, plot = TRUE)

Arguments

pso

(short for plotSlopesObject) Output from the plotSlopes function

plot

TRUE or FALSE, FALSE prevents creation of plots

Value

• A list including the hypothesis test table. For numeric modx variables, also the Johnson-Neyman (J-N) interval boundaries.

Details

Consider a regression with interactions

y <- b0 + b1*x1 + b2*x2 + b3*(x1*x2) + b4*x3 + ... + error

If plotSlopes has been run with the argument plotx="x1" and the argument modx="x2", then there will be several plotted lines, one for each of the chosen values of x2. The slope of each of these lines depends on x1's effect, b1, as well as the interactive part, b3*x2. We wonder if the combined effect of x1 is statistically significantly different from 0 for each of those values of x2.

This function performs a test of the null hypothesis of the slope of each fitted line in a plotSlopes object is statistically significant from zero. A simple t-test for each line is offered. No correction for the conduct of multiple hypothesis tests (no Bonferroni correction).

In addition, the so-called Johnson-Neyman (Johnson-Neyman, 1936; Preacher, Curran, and Bauer, 2006) interval is calculated and a couple of efforts are made to render it graphically. Where the t-test considers the question, is the slope of the line (b1 + b3*x2) different from 0, the J-N approach asks "for which values of x2 will that plotted line be statistically significant?"

References

Preacher, Kristopher J, Curran, Patrick J.,and Bauer, Daniel J. (2006). Computational Tools for Probing Interactions in Multiple Linear Regression, Multilevel Modeling, and Latent Curve Analysis. Journal of Educational and Behavioral Statistics. 31,4, 437-448.

Johnson, P.O. and Neyman, J. (1936). "Tests of certain linear hypotheses and their applications to some educational problems. Statistical Research Memoirs, 1, 57-93.

See Also

plotSlopes

Examples

library(car)
m3 <- lm(statusquo ~ income * sex, data = Chile)
m3ps <- plotSlopes(m3, modx = "sex", plotx = "income")
testSlopes(m3ps)

m4 <- lm(statusquo ~ region * income, data= Chile)
m4ps <- plotSlopes(m4, modx = "region", plotx = "income", plotPoints = FALSE)
##' testSlopes(m4ps)


m5 <- lm(statusquo ~ region * income + sex + age, data= Chile)
m5ps <- plotSlopes(m5, modx = "region", plotx = "income")
testSlopes(m5ps)

m6 <- lm(statusquo ~ income * age + education + sex + age, data=Chile)
m6ps <- plotSlopes(m6, modx = "income", plotx = "age")
testSlopes(m6ps)

## Finally, if model has no relevant interactions, testSlopes does nothing.
m9 <- lm(statusquo ~ age + income * education + sex + age, data=Chile)
m9ps <- plotSlopes(m9, modx = "education", plotx = "age", plotPoints = FALSE)
testSlopes(m9ps)