Predict ordered comparisons from an experimental lapse BTL fit

Calculate (1-epsilon) * plogis(theta1-theta2+beta) + epsilon/2. Positive beta favors the first presented item. Swapping the items generally does not give complementary probabilities unless beta is zero. These are plug-in probabilities, without integration over estimation uncertainty. Valid epsilon-zero boundary fits also support prediction; their unavailable joint uncertainty does not invalidate the point estimates.

Usage

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

Arguments

object

A validated experimental fit from fit_bt_model() with engine = "lapse".

newdata

A data frame containing object1 and object2 IDs known to the fit. NULL uses the original comparisons in their original order. Repeated pairs and self-predictions are allowed; a self-prediction includes positional bias and need not equal 0.5. Empty input returns numeric(0).

...

Reserved; additional arguments are rejected.

Value

Numeric first-item win probabilities in input row order.

See also

fit_bt_model(), integrated CJ workflow

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_firth(), scale_separation_reliability(), summarize_bt_fit()