The AI-GENIE paper is now PUBLISHED at Behavior Research Methods (FREE link in the comments), by the @Psychonomic_Soc ! In this paper we are not only setting Generative Psychometrics in stone, but we are showing a full pipeline from Item Generation using LLMs to FULL in-silico Structural Validity of the Items, after removing redundant and unstable items! Finally, we show that the in-silico dimensionality matches EXACTLY the dimensionality obtained in nationally representative samples for the best model! The appendix provides the benchmark for newer models, including the open-source GPT models from OpenAI! Big congratulations to @larafromUVA , for this is the first paper out of her PhD work.
新しい効果量指標を提案する論文が Educational and Psychological Measurement 誌に掲載されました!Hedges' g は正規分布を仮定しますが、実データはしばしば裾が重い分布です。本研究は尖度補正を加えた推定量を提案。裾が重い場合に g の残差バイアスを36–92%削減しました https://t.co/Bx4BxMcKcA