{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06287167,0.000210394,0.001840487,0.001239664,0.000257551,0.0001738937,0.001004596,0.0002547002,0.0005483163],"category_scores_gemma":[0.3538786,0.0001318928,0.0002356087,0.004848207,0.001447635,0.0003069988,0.0004420045,0.0007279498,0.0001237283],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002275585,"about_ca_system_score_gemma":0.0007331192,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008826975,"about_ca_topic_score_gemma":0.000004199469,"domain_scores_codex":[0.9869681,0.006746337,0.001253849,0.000649069,0.004013357,0.0003692935],"domain_scores_gemma":[0.9451019,0.05106882,0.001194824,0.0006864273,0.001777808,0.0001701964],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000005678979,0.00004924354,0.0001748672,0.0195844,0.00003991917,0.000002081663,0.0002561504,1.298645e-9,4.825073e-8,0.00162452,0.04306836,0.9351947],"study_design_scores_gemma":[0.00003288391,0.0001350744,0.0004673572,0.06414809,0.00003071234,0.00000922529,0.0007935741,1.972921e-8,1.752492e-7,0.02909406,0.9051627,0.0001261058],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0000736181,0.9953905,0.000001890799,0.001115897,0.0008428867,0.0005995442,0.00000727326,0.000007872752,0.00196047],"genre_scores_gemma":[0.00009423034,0.9986724,0.0002164505,0.00005836141,0.0002900197,0.0001697238,0.000002587687,0.00001017607,0.0004860355],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9350686,"threshold_uncertainty_score":0.9649708,"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."}}