Predict pairwise win probabilities from a Firth Bradley-Terry fit

Calculate plogis(theta1 - theta2), the probability that the first item wins. Predictions use the fitted item strengths, without a positional, lapse, or tie parameter. They do not integrate over estimation uncertainty.

Usage

# S3 method for class 'pairwiseLLM_bt_firth'
predict(object, newdata = NULL, ...)

Arguments

object

A Firth fit returned by fit_bt_model() with engine = "brglm2".

newdata

A data frame containing object1 and object2 item IDs. NULL uses the original comparisons in their original order. IDs must be nonmissing and present in the fitted model. Repeated pairs are allowed; comparing an item with itself returns 0.5.

...

Reserved; additional arguments are rejected.

Value

A numeric vector of first-item win probabilities in input row order. An empty data frame returns numeric(0).

See also

fit_bt_model(), summarize_bt_fit()

Other frequentist models: bootstrap_bt_model(), build_bt_data(), build_elo_data(), fit_bt_model(), fit_elo_model(), predict.pairwiseLLM_bt_alpha(), predict.pairwiseLLM_bt_lapse(), scale_separation_reliability(), summarize_bt_fit()

Examples

if (requireNamespace("brglm2", quietly = TRUE)) {
  comparisons <- data.frame(object1 = c("a", "a", "b"),
                            object2 = c("b", "c", "c"), result = c(1, 1, 1))
  fit <- fit_bt_model(comparisons, engine = "brglm2")
  predict(fit)
  predict(fit, data.frame(object1 = "c", object2 = "a"))
}
#> [1] 0.1120953