DOI: 10.6129/CJP.20140411
中華心理學刊 民103，56卷，4期，453-466
Chinese Journal of Psychology 2014, Vol.56, No.4, 453-466

李家慈（國立台灣大學心理學系暨研究所）；賴文崧（國立台灣大學心理學系暨研究所；國立台灣大學腦與心智科學研究所；國立台灣大學神經生物與認知科學研究中心）；徐永豐（國立台灣大學心理學系暨研究所；國立台灣大學腦與心智科學研究所；國立台灣大學神經生物與認知科學研究中心）

摘要

本研究使用動態酬賞作業與三向度性格量表以大學生為受試探討酬賞預測誤差的男女差異及性格相關。我們用一增強學習模型作適配，並藉貝氏因子作參數組間比較。配對法則分析結果顯示男性酬賞敏感度比女性高，表示男性選擇優勢牌的傾向比女性佳。增強學習模型參數估計發現女性的學習速率較男性略高，亦即當預期與實際經驗不符時，女性傾向給予預測誤差較大權重，較快改變信念。而女性的選擇固著性較男性小，代表女性在作業中探索的程度高於男性。另外，相較於男性組，女性受試者的學習速率與作業總分呈負相關，而男女受試的選擇固著性皆與作業總分呈正相關。這些結果顯示男性相對於女性，平均而言有較接近「最佳化」的學習速率與較大的選擇固著性，為此作業男性較佳的選擇優勢牌傾向提供了一可能解釋。男女性在三向度性格量表各向度分數間無甚差異，然而，相關分析發現量表部分次向度的分數與增強學習模型參數間存有性別差異：女性受試學習速率與「新奇追求」分向度「揮豁性」呈負相關；類似負相關亦存在於女性受試選擇固著性與「酬賞依賴」分向度「親近性」。男性組則沒有發現任何顯著相關。這些結果顯示女性相對於男性，部分性格特徵與學習表現有關，然確切關聯有賴後續研究。

關鍵詞：三向度性格量表、貝氏因子、增強學習模型、動態酬賞作業、酬賞預測誤差

Using a Reinforcement Learning Model and Bayes Factors to Study Gender Differences and Personality Correlations in Reward Prediction Errors

Chia-Tzu Li（Department of Psychology, National Taiwan University）；Wen-Sung Lai（Department of Psychology, National Taiwan University；Graduate Institute of Brain and Mind Sciences, National Taiwan University；Neurobiology and Cognitive Science Center, National Taiwan University）；Yung-Fong Hsu（Department of Psychology, National Taiwan University；Graduate Institute of Brain and Mind Sciences, National Taiwan University；Neurobiology and Cognitive Science Center, National Taiwan University）

Abstract

We used a dynamic reward task to study gender differences in reward prediction errors in college students.
Experimental data were fit by a reinforcement learning model using the Bayesian estimation approach. We also
correlated subjects’ performance with their personality traits assessed by the Tridimensional Personality Questionnaire (TPQ). Matching law analysis revealed that males are more sensitive than females, meaning that males are better than females in choosing the advantageous card. Model fitting together with Bayes factors revealed that females update their values slightly faster than males, indicating that females have the tendency to give the trial-by-trial prediction error more weight. Further, females showed less perseveration than males, indicating that females tend to explore the uncertainty of future events more often. We also found that females’ learning rate is negatively correlated with their task performance, whereas both males and females’ degree of perseveration is positively correlated with their task scores. These results indicate that on average, males have a more ‘optimal’ learning rate and a larger degree of perseveration than females, providing a plausible explanation of why males performed slightly better than females in the task. Regarding the relation of task performance and personality traits, correlational analysis revealed that females with higher scores in “extravagance” (which is in the novelty-seeking dimension of TPQ) have lower value-updating rates. Moreover, females with higher scores in “attachment” (which is in the reward-dependence dimension of TPQ) have lower degrees of perseveration. No such findings were found in males. Further work is needed to verify the findings. 

Keywords: Tridimensional Personality Questionnaire, Bayes factor, reinforcement learning model, dynamic reward task, reward prediction error