Decompose 2-way (x1*x2) interaction

This function decomposes a 2-way x1*x2 interaction into simple effects, and provides means, pairwise comparisons, and a plot.

Usage

decompose_interaction(
  model,
  df = NULL,
  x1,
  x2,
  at_x1 = NULL,
  at_x2 = NULL,
  ci = FALSE,
  ...
)

Arguments

model

A model object compatible with emmeans (e.g. lm)

df

data.frame or NULL if model has model$model

x1, x2

(character) variable name

at_x1, at_x2

(numeric or character vector) x1 and x2 values at which to calculate and compare means (if NULL, all unique values)

ci

(logical) Whether to plot confidence intervals (default FALSE)

...

additional arguments passed to the emmeans functions (see Details)

Value

List with elements:

slopes

Simple slopes and pairwise comparisons (from calling emmeans::emtrends())

means

(adjusted) means for every combination of x1*x2 (may need to specify at_x1 and/or at_x2 values)

contrasts

Comparisons of (adjusted) x1 means at each value of x2 (may need to specify at_x1 and/or at_x2 values)

plot

Plot (ggplot2 line graph using data from emmeans::emmip())

Details

This function uses the facilities provided by the library emmeans, and follows the procedure outlined by UCLA OARC Statistical and Data Analytics. It might be helpful to execute the function with x1 and x2 inverted, and see which set of results and plot make most sense for your scenario.

Note that if x1 or x2 has many different distinct values and at_x1 or at_x2 is not specified, then not all elements of the returned list will be useful ($means and $contrasts in particular may be difficult to use).

If model is of class lavaan, an additional argument lavaan.DV must be passed to specify the name of the outcome variable. Transfer of model information to emmeans is handled by semTools::lavaan2emmeans().

See also

emmeans::emmeans(), emmeans::emtrends(), emmeans::emmip()

Examples

if (FALSE) {
fit <- lm(SelfEsteem ~ Gender*AttitudeTowardsSchool, df)
fit |> decompose_interaction(x1="Gender", x2="AttitudeTowardsSchool")
}