Package index • mixedsubjectsirt

Fitting mixed-subjects models

Estimators for mixed-subjects IRT calibration. fit_mixed_subjects_mml() is the recommended marginal-likelihood estimator; fit_2pl() and fit_1pl() fit human-only baselines using a mirt backend

fit_mixed_subjects(): Fit a mixed-subjects 2PL calibration
fit_mixed_subjects_1pl(): Fit a mixed-subjects 1PL calibration (frozen expected-count)
fit_mixed_subjects_from_quadrature(): Fit from precomputed quadrature summaries
fit_mixed_subjects_iterative(): Fit a mixed-subjects 2PL calibration with iterative EM
fit_mixed_subjects_mml(): Fit a mixed-subjects 2PL calibration via marginal maximum likelihood
fit_mixed_subjects_mml_1pl(): Fit a mixed-subjects 1PL calibration via marginal maximum likelihood
fit_mixed_subjects_split(): Fit a split-sample mixed-subjects 2PL calibration
fit_2pl(): Fit a unidimensional 2PL IRT model
fit_1pl(): Fit a 1PL (one-parameter logistic) model

Choosing lambda

Power-tuning the mixed-subjects correction. tune_lambda_ability_risk() is the recommended practical criterion (downstream ability-score risk); tune_lambda_ppi_score() is a theoretical PPI++ diagnostic; diagnose_lambda_grid() is a sensitivity check

tune_lambda_ability_risk(): Tune lambda by downstream ability-score risk
tune_lambda_ability_risk_1pl(): Tune lambda by downstream ability-score risk for a 1PL model
tune_lambda_ability_risk_crossfit(): Cross-fit ability-score-risk lambda tuning
tune_lambda_ability_risk_item(): Per-item ability-risk lambda tuning via coordinate descent
tune_lambda_ppi_score(): Plug-in PPI++ optimal tuning parameter
tune_lambda_ppi_score_1pl(): Plug-in PPI++ optimal tuning parameter for a 1PL model
tune_lambda_ppi_score_item(): Per-item PPI++ optimal tuning parameters
diagnose_lambda_grid(): Diagnose lambda values over a grid

Covariance and uncertainty

Sandwich covariance for fitted models, used through the vcov() S3 method

vcov_mixed_subjects(): Sandwich covariance for a mixed-subjects fit
vcov_mixed_subjects_1pl(): Sandwich covariance for a 1PL mixed-subjects fit
vcov_mixed_subjects_mml(): Marginal-MML sandwich covariance for a mixed-subjects fit

Ability scoring and risk

Estimating abilities and propagating item-parameter uncertainty into scores

score_theta(): Estimate ability scores from a 2PL calibration
ability_gradient(): Gradient of ML ability scores with respect to item parameters
ability_gradient_1pl(): Gradient of ML ability scores w.r.t. 1PL item parameters
ability_risk(): Propagated ability risk from item-parameter uncertainty
ability_risk_1pl(): Propagated ability risk for a 1PL fit

Scale linking

Using established linking procedures to place LLM item parameters onto the human scale

link_item_parameters(): Link item parameters onto a target scale

Simulation and low-level building blocks

Data simulation, Gauss-Hermite quadrature, posterior weights, and expected-count summaries

simulate_2pl(): Simulate 2PL item responses
make_quadrature(): Create a standard-normal Gauss-Hermite quadrature grid
posterior_weights_2pl(): Compute posterior quadrature weights for a 2PL model
summarize_expected_counts(): Summarize response data as expected quadrature counts
mixed_subjects_quadrature(): Convert responses to quadrature form
mixed_subjects_loss(): Mixed-subjects objective function