{"id":"W2808430903","doi":"10.1007/s40037-018-0435-8","title":"Examining the effects of gaming and guessing on script concordance test scores","year":2018,"lang":"en","type":"article","venue":"Perspectives on Medical Education","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; McGill University","funders":"","keywords":"Concordance; Likert scale; Test (biology); Statistics; Psychology; Scale (ratio); Social psychology; Mathematics; Medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05332988,0.0006263404,0.0004221804,0.00122402,0.0005626013,0.001870164,0.001212512,0.0009964663,0.002440058],"category_scores_gemma":[0.2732798,0.0004301987,0.001154465,0.0007147027,0.001941929,0.001304886,0.00187079,0.0009174598,0.0004331632],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000959319,"about_ca_system_score_gemma":0.0007443066,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001539694,"about_ca_topic_score_gemma":0.002303845,"domain_scores_codex":[0.9393947,0.04328737,0.004160434,0.004956233,0.007364334,0.0008368684],"domain_scores_gemma":[0.4160828,0.5009346,0.04954749,0.02140422,0.009514351,0.002516558],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001088744,0.0001745372,0.9824117,0.00005127975,0.0003375853,0.00006707929,0.001142696,0.000911436,0.0009704813,0.0001655801,0.0001720001,0.01250682],"study_design_scores_gemma":[0.00003119685,0.001574511,0.9887516,0.00003900492,0.0001239487,0.0002156532,0.0003993245,0.005269623,0.002949972,0.0002846636,0.0003265708,0.00003390972],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9963272,0.00004301765,0.002410775,0.00005149535,0.000009321522,0.0001013787,0.0001191749,0.00002602122,0.0009117321],"genre_scores_gemma":[0.9972325,0.00001464972,0.002287442,0.00002971318,0.000006437937,0.0001048101,0.0001660692,0.0000112298,0.0001471171],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05332988,"threshold_uncertainty_score":0.2820387,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07302416389855387,"score_gpt":0.392594142714597,"score_spread":0.3195699788160431,"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."}}