{"id":"W4394687365","doi":"10.2196/50315","title":"Measuring the Reliability of a Gamified Stroop Task: Quantitative Experiment","year":2024,"lang":"en","type":"article","venue":"JMIR Serious Games","topic":"Cognitive Abilities and Testing","field":"Psychology","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria; University of Saskatchewan","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Stroop effect; Intraclass correlation; Task (project management); Reliability (semiconductor); Cognition; Consistency (knowledge bases); Psychology; Cognitive psychology; Computer science; Psychometrics; Developmental psychology; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"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":[0.0004463444,0.0001555923,0.0002127321,0.00005721933,0.00007778673,0.00004703894,0.0001713115,0.00005484177,0.000566021],"category_scores_gemma":[0.0002185088,0.0001071651,0.0001382245,0.000219644,0.0001872252,0.00006826208,0.0000844276,0.0002075712,0.00008609053],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005214137,"about_ca_system_score_gemma":0.00004923495,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003361586,"about_ca_topic_score_gemma":0.0000453863,"domain_scores_codex":[0.9986899,0.0001457427,0.00031877,0.0003757208,0.0001903987,0.0002794451],"domain_scores_gemma":[0.9984615,0.0009043289,0.00007025539,0.000403942,0.0001177007,0.00004226718],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.001166019,0.001453934,0.01620197,0.001304284,0.001147932,0.0002349753,0.3604901,0.0001535028,0.06965026,0.1905035,0.01233921,0.3453543],"study_design_scores_gemma":[0.002881357,0.003877489,0.5293644,0.001845454,0.0003121406,0.0001621149,0.2702491,0.003426132,0.03034509,0.02079917,0.1349211,0.001816448],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9753467,0.003482149,0.0001131609,0.0005269807,0.0006498661,0.0005637337,0.00002832526,0.0001479738,0.01914111],"genre_scores_gemma":[0.9982458,0.000009208451,0.0002042765,0.00004812552,0.00008402848,0.0003400205,0.000003108302,0.00002522821,0.001040146],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5131624,"threshold_uncertainty_score":0.6197533,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05922230846763064,"score_gpt":0.3578098590581152,"score_spread":0.2985875505904845,"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."}}