Compare Models With and Without Interaction Term — interaction

This function fits two models—one with and one without an interaction term between an exposure and a potential effect modifier— and compares them using either a likelihood ratio test (LRT) or Wald test. It is useful for assessing whether there is statistical evidence of interaction (effect modification).

Usage

interaction_models(
  data,
  outcome,
  exposure,
  covariates = NULL,
  effect_modifier,
  approach = "logit",
  test = c("LRT", "Wald"),
  verbose = TRUE
)

Arguments

data: A data frame containing all required variables.
outcome: The name of the outcome variable
exposure: The name of the main exposure variable.
covariates: character vector of additional covariates to adjust for
effect_modifier: The name of the variable to test for interaction
approach: The regression modeling approach to use. One of: "logit", "log-binomial", "poisson", "robpoisson", "negbin", or "linear".
test: Type of statistical test for model comparison. Either: "LRT" (likelihood ratio test, default) or "Wald".
verbose: Logical; if TRUE, prints a basic interpretation of whether interaction is likely present (default = FALSE).

Value

A list with the following elements:

model_no_interaction: The model without the interaction term.
model_with_interaction: The model with the interaction term.
p_value: The p-value for interaction (based on selected test).
interpretation: A brief text interpretation if verbose = TRUE.

Examples

data <- data_PimaIndiansDiabetes