{"id":"W3045815317","doi":"10.1103/physrevphyseducres.16.020107","title":"Development and validation of the moon phases concept inventory for middle school","year":2020,"lang":"en","type":"article","venue":"Physical Review Physics Education Research","topic":"Science Education and Pedagogy","field":"Social Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université du Québec à Montréal","funders":"Social Sciences and Humanities Research Council of Canada; Fonds de Recherche du Québec-Société et Culture; Université du Québec à Montréal","keywords":"Cronbach's alpha; Readability; Test (biology); Psychology; Reading (process); Reliability (semiconductor); Construct validity; Face validity; Perspective (graphical); Statistics; Psychometrics; Clinical psychology; Computer science; Mathematics; Artificial intelligence; Physics","routes":{"ca_aff":true,"ca_fund":true,"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":[0.001010517,0.00006112512,0.000137762,0.00001921122,0.0005257821,0.00004523342,0.0003152399,0.00001589251,0.0001104362],"category_scores_gemma":[0.002682407,0.00004733799,0.00005770417,0.0007549403,0.0003965329,0.0001778056,0.00005014801,0.0001515575,0.00002806168],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007846163,"about_ca_system_score_gemma":0.003271724,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001179184,"about_ca_topic_score_gemma":0.000009705089,"domain_scores_codex":[0.9983927,0.000418036,0.0001744187,0.0002038622,0.0006196032,0.0001913632],"domain_scores_gemma":[0.9985791,0.0003824039,0.0001012137,0.000147959,0.0005783153,0.000211006],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001570765,0.001266009,0.001995078,0.001710734,0.00002676924,1.972459e-8,0.1679826,0.000007013753,0.004286954,0.3961025,0.07789028,0.3487163],"study_design_scores_gemma":[0.000242016,0.00007611308,0.000918266,0.001004478,0.00002712588,8.645367e-8,0.03745775,0.0001517491,0.03981939,0.03217816,0.8878919,0.0002329374],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.923543,0.006463192,0.0003125428,0.05677993,0.0005160766,0.002962721,0.000009079924,0.000028472,0.009384928],"genre_scores_gemma":[0.9953716,0.0009816864,0.0002850972,0.001610245,0.0009626857,0.0003005424,0.00001334336,0.000006776598,0.0004680717],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8100016,"threshold_uncertainty_score":0.5803897,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.451895880993847,"score_gpt":0.5574689975373768,"score_spread":0.1055731165435297,"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."}}