{"id":"W128966544","doi":"10.1007/978-3-319-07794-9_6","title":"What Counts as Evidence: A Review of Validity Studies in Educational and Psychological Measurement","year":2014,"lang":"en","type":"review","venue":"Social indicators research series","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":25,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Psychology; External validity; Test (biology); Standards for Educational and Psychological Testing; Test validity; Criterion validity; Validity; Incremental validity; Applied psychology; Construct validity; Psychometrics; Social psychology; Clinical psychology; Higher education; Education theory; Political science","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07015577,0.00194537,0.01169804,0.0243395,0.001813278,0.009680318,0.00398367,0.004548117,0.004178306],"category_scores_gemma":[0.2527719,0.001974546,0.00554913,0.02600121,0.006489686,0.008506058,0.004853215,0.004537499,0.0005229638],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00925348,"about_ca_system_score_gemma":0.03555022,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01008267,"about_ca_topic_score_gemma":0.02769852,"domain_scores_codex":[0.9408059,0.02326188,0.02105478,0.003243015,0.01106609,0.0005683695],"domain_scores_gemma":[0.6232536,0.3361834,0.02231556,0.003093189,0.01373376,0.001420492],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000201481,0.00006123565,0.002150939,0.5994952,0.004343507,0.00008629823,0.0007434131,0.0002367885,0.0002326346,0.004707403,0.005398992,0.3823422],"study_design_scores_gemma":[0.0001471064,0.0001206903,0.003574479,0.9022233,0.01623019,0.0003129599,0.0008685706,0.0002296876,0.0002931117,0.009305791,0.06660157,0.00009249435],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0001783264,0.997225,0.0003703511,0.001622913,0.0001972969,0.00008166846,0.00009352733,0.000004534046,0.0002263818],"genre_scores_gemma":[0.004583326,0.9903715,0.002798315,0.001624962,0.0002751832,0.0002114528,0.00007871748,0.000009469229,0.00004711144],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9298443,"threshold_uncertainty_score":0.3710236,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.96993829252238,"score_gpt":0.7425690907147161,"score_spread":0.2273692018076638,"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."}}