Skip to contents

Conditional treatment effects

Source: R/conditional_effects.R
conditional_effects.Rd

Compute conditional (individual-level) treatment effects at specific covariate values from a fitted ML-UMR model. Unlike marginal_effects(), which averages over a population's covariate distribution, conditional effects evaluate the treatment effect at a particular covariate profile.

Usage

conditional_effects(
  object,
  newdata = NULL,
  effect = "all",
  summary = TRUE,
  probs = c(0.025, 0.5, 0.975)
)

Arguments

object

An mlumr_fit object

newdata

Data frame of covariate values at which to compute effects. Each row defines one covariate profile. Column names must match the covariates used in fitting. If NULL (default), uses the covariate means from the IPD as a single reference profile.

effect

Which effect measure. For binomial: "all", "link_effect", "rd", or "rr". For normal: "all" or "md". For Poisson: "all" or "rr". The legacy value "lor" is accepted as an alias for "link_effect" when the fitted link is logit.

summary

Return summary statistics (TRUE) or full posterior draws (FALSE)

probs

Quantiles for summary (default c(0.025, 0.5, 0.975))

Value

A data frame. If summary = TRUE, contains columns profile, effect, mean, sd, and quantile columns. If summary = FALSE, returns a single combined data frame of full posterior draws with a profile column indicating which covariate profile each draw belongs to.

Details

For SPFA models, the conditional link-scale treatment effect is constant across all covariate values because the shared beta cancels in the treatment contrast on the fitted link scale. However, risk difference (RD) and risk ratio (RR) still vary with covariates because they depend on absolute probability levels. For relaxed models, all conditional effects vary with covariate values because the index and comparator treatments have different regression coefficients.

The conditional link-scale effect is computed directly as eta_index - eta_comparator. This avoids numerical distortion from transforming extreme response-scale probabilities back through the link function.

Conditional vs marginal on non-identity links. Conditional effects are evaluated at a single covariate profile, so there is no averaging over a population and no Jensen's-inequality gap between the conditional and marginal response. Compare with marginal_effects() and predict.mlumr_fit(), which average over either the IPD individuals (index population) or the AgD integration points (comparator population) and therefore return E[g^{-1}(eta)], not g^{-1}(E[eta]).

See also

marginal_effects() for population-averaged treatment effects; conditional_predict() for absolute predictions at specific profiles; predict.mlumr_fit() for population-level predictions.

Examples

if (FALSE) { # \dontrun{
# Conditional effects at IPD covariate means (default)
conditional_effects(fit)

# At specific covariate values
conditional_effects(fit, newdata = data.frame(age = 60, sex = 1))

# Multiple profiles
profiles <- data.frame(age = c(50, 60, 70), sex = c(0, 0, 1))
conditional_effects(fit, newdata = profiles)
} # }