Prediction Contrasts, Rankings, and Groups

Prediction grids can be summarised by observed design variables without expanding the approved model-family scope.

grid <- create_prediction_grid(
  fit,
  at = list(condition = c("control", "treatment"))
)

pred <- predict_model(
  fit,
  newdata = grid,
  type = "expected",
  include_group_effects = FALSE
)

prediction_pairwise_contrasts(pred)
prediction_interval_width(pred)
prediction_rank_probabilities(pred)

The ranking function is deliberately descriptive. A probability of rank one is not converted into an automatic selection.

When the prediction data contain multiple rows per substantive group:

grouped <- group_prediction_summary(pred, by = "condition")
grouped
plot_group_predictions(grouped, "condition")

This makes aggregation explicit and reproducible rather than hiding it inside plotting code.