simple_slopes: Simple slopes of an interaction.

In reghelper: Helper Functions for Regression Analysis

Simple slopes of an interaction.

Description

simple_slopes calculates all the simple effects of an interaction in a fitted model (linear, generalized linear, hierarchical linear, or ANOVA).

Usage

simple_slopes(model, ...)

## S3 method for class 'lm'
simple_slopes(model, levels = NULL, confint = FALSE, ci.width = 0.95, ...)

## S3 method for class 'glm'
simple_slopes(model, levels = NULL, confint = FALSE, ci.width = 0.95, ...)

## S3 method for class 'lme'
simple_slopes(model, levels = NULL, confint = FALSE, ci.width = 0.95, ...)

## S3 method for class 'merMod'
simple_slopes(
  model,
  levels = NULL,
  confint = FALSE,
  ci.width = 0.95,
  confint.method = c("Wald", "profile", "boot"),
  ...
)

Arguments

model	A fitted linear model of type 'lm', 'glm', 'aov', 'lme' (nlme), or 'merMod' (lme4), with at least one interaction term.
...	Additional parameters to be passed on to the 'confint' method, if 'confint' is TRUE.
levels	A list with element names corresponding to some or all of the variables in the model. Each list element should be a vector with the names of factor levels (for categorical variables) or numeric values (for continuous variables) at which to test that variable. Note: If you do not include 'sstest' as one of these levels, the function will not test the simple effects for that variable.
confint	Whether or not to include confidence intervals for each estimate.
ci.width	If 'confint' is TRUE, this represents the width of the confidence intervals to calculate, as a proportion from 0 to 1.
confint.method	For 'merMod' models only, specifies what method to use for computing the confidence intervals.

Details

If the model includes interactions at different levels (e.g., three two-way interactions and one three-way interaction), the function will test the simple effects of the highest-order interaction. If there are multiple interactions in the highest order, it will test the first one in the model. If you wish to test simple effects for a different interaction, simply switch the order in the formula.

By default, this function will provide slopes at -1 SD, the mean, and +1 SD for continuous variables, and at each level of categorical variables. This can be overridden with the levels parameter.

If a categorical variable with more than two levels is being tested, you may see multiple rows for that test. One row will be shown for each contrast for that variable; the name of the contrast is identified in parentheses after the 'sstest' label.

Value

A data frame with a row for each simple effect. The first few columns identify the level at which each variable in your model was set for that test. A 'sstest' value in a particular column indicates that the simple slope for this variable was being tested. After columns for each variable, the data frame has columns for the slope of the test variable, the standard error, t-value, p-value, and degrees of freedom for the model. For 'merMod' models, the degrees of freedom and p-values will not appear, as these are not calculated by the lme4 package.

Methods (by class)

• simple_slopes(lm): Simple slopes for linear models.

• simple_slopes(glm): Simple slopes for generalized linear models.

• simple_slopes(lme): Simple slopes for hierarchical linear models (nlme).

• simple_slopes(merMod): Simple slopes for hierarchical linear models (lme4).

Examples

# linear model
mtcars$am <- factor(mtcars$am)  # make 'am' categorical
model <- lm(mpg ~ wt * am, data=mtcars)
summary(model)  # significant interaction
simple_slopes(model)
simple_slopes(model,
    levels=list(wt=c(2, 3, 4, 'sstest'), am=c(0, 1, 'sstest')))  # test at specific levels

# generalized linear model
model <- glm(vs ~ gear * wt, data=mtcars, family='binomial')
summary(model)  # marginal interaction
simple_slopes(model)
simple_slopes(model,
    levels=list(gear=c(2, 3, 4, 'sstest'), wt=c(2, 3, 'sstest')))  # test at specific levels

# hierarchical linear model (nlme)
if (require(nlme, quietly=TRUE)) {
    model <- lme(Sepal.Width ~ Sepal.Length * Petal.Length, random=~1|Species, data=iris)
    summary(model)  # significant interaction
    simple_slopes(model)
    simple_slopes(model,
        levels=list(Sepal.Length=c(4, 5, 6, 'sstest'),
        Petal.Length=c(2, 3, 'sstest')))  # test at specific levels
}

# hierarchical linear model (lme4)
if (require(lme4, quietly=TRUE)) {
    model <- lmer(Sepal.Width ~ Sepal.Length * Petal.Length + (1|Species), data=iris)
    summary(model)
    simple_slopes(model)
    simple_slopes(model,
        levels=list(Sepal.Length=c(4, 5, 6, 'sstest'),
        Petal.Length=c(2, 3, 'sstest')))  # test at specific levels
}

reghelper documentation built on Sept. 3, 2023, 1:07 a.m.