{"id":"W4396978444","doi":"10.21449/ijate.1376160","title":"The difference between estimated and perceived item difficulty: An empirical study","year":2024,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Psychology; Statistics; Empirical research; Econometrics; Mathematics education; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.001224334,0.00009533785,0.0001336232,0.0001908442,0.0001962712,0.001256396,0.0004996337,0.00004666487,0.0000359272],"category_scores_gemma":[0.0002454869,0.0000663117,0.00004395479,0.0002017955,0.00009665512,0.0007077001,0.00005029445,0.0002836499,0.000002639126],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004495859,"about_ca_system_score_gemma":0.0008251739,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001869477,"about_ca_topic_score_gemma":0.0003801784,"domain_scores_codex":[0.9980845,0.0003184718,0.0004253422,0.0001538357,0.0008760162,0.0001418045],"domain_scores_gemma":[0.9984225,0.0009151304,0.0001533783,0.00007477999,0.0003472352,0.0000870018],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001230981,0.0004287825,0.8492441,0.000001919192,0.00008095254,0.000007748953,0.014092,0.000002351647,0.0001133291,0.001107323,0.0002475331,0.1346617],"study_design_scores_gemma":[0.0002189144,0.0001335453,0.9597335,0.00009437176,0.00003134816,0.000003833351,0.03638928,0.000207659,0.000002540657,0.001026582,0.002084604,0.00007384711],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9901327,0.0001627158,0.000151456,0.005618794,0.002500994,0.0002258512,0.000002639989,0.00001823925,0.001186571],"genre_scores_gemma":[0.9977667,0.0003878434,0.0004843641,0.00004812746,0.0009648814,0.00001499817,0.000009163883,0.000007304943,0.0003165973],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1345878,"threshold_uncertainty_score":0.9997804,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08975046533286718,"score_gpt":0.5107784354602035,"score_spread":0.4210279701273363,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}