{"id":"W4381333182","doi":"10.1007/s11423-023-10250-z","title":"Developing and validating the assessment of the skills of preservice teachers in implementing pedagogical model of collaborative problem solving","year":2023,"lang":"en","type":"article","venue":"Educational Technology Research and Development","topic":"Innovative Teaching and Learning Methods","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Reliability (semiconductor); Mathematics education; Multinomial logistic regression; Psychology; Validity; Item response theory; Alternative assessment; Computer science; Psychometrics; Machine learning; Developmental psychology","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":[],"consensus_categories":[],"category_scores_codex":[0.007964098,0.00008692951,0.0001821833,0.0004703347,0.0003268822,0.00001034746,0.0003016449,0.00009233531,0.00002103707],"category_scores_gemma":[0.001240233,0.00005742481,0.000008078972,0.002528894,0.0004572587,0.00003507439,0.000522534,0.0005971473,5.766934e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007510111,"about_ca_system_score_gemma":0.001594696,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005584869,"about_ca_topic_score_gemma":0.00003130035,"domain_scores_codex":[0.9979531,0.0006701206,0.0004683944,0.0002379736,0.0003390225,0.0003314219],"domain_scores_gemma":[0.9978175,0.001009671,0.0002212011,0.0001797485,0.0007541275,0.00001779692],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001025726,0.0001567593,0.6652174,0.0001106792,0.00009218632,2.712883e-7,0.07346284,0.00009973129,0.005579528,0.2408708,0.0002998226,0.01409972],"study_design_scores_gemma":[0.0004131957,0.0000676563,0.8910255,0.000244913,0.00000330045,0.000003070605,0.0737256,0.0004404076,0.004489574,0.02862786,0.0008540685,0.0001048743],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9895687,0.0001024175,0.001264141,0.006818565,0.00003878849,0.0004220664,0.00000387895,0.00001366286,0.001767847],"genre_scores_gemma":[0.8803334,0.00001254101,0.118787,0.00001330054,0.000008585161,0.0001487743,0.000006596678,0.000007432504,0.0006824205],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2258081,"threshold_uncertainty_score":0.2828923,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2502626672855353,"score_gpt":0.5470780597973545,"score_spread":0.2968153925118192,"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."}}