主題至屬性指派受試者潛在特徵：醫學資料映射力的計算與應用

Theme—Attribute Assignment Analysis (TAAA) Applied to Person Latent Traits: Computation and Applications of Mapping Power for Medical Data

臨床量表研究常以總分分類受試者，或以邏輯斯迴歸區分群體，但較少評估雙向度結構與外在效標間的一致性。本研究提出「主題至屬性指派分析」（Theme-Attribute Assignment Analysis, TAAA），依雙題組結構進行主題命名並映射至受試者屬性，並以Cohen's kappa衡量主題與效標之一致性，定義為映射力（MP）。本研究以三項實證資料（樣本數約N=4-200，含8-20題）進行示範，效標為既定分群或專家分類；另以四組題數比例資料探討方法限制。結果顯示，在題數均衡情境下，TAAA之MP為1.00、1.00與0.76，與正則化輯斯迴歸表現相近；當雙向度題數比例偏離1:1且低於1:2時，MP可能顯著下降並出現單群退化。TAAA為一結構導向之一致性評估方法，而非預測模型，適用於需由測量結構映射至屬性分類之研究情境。

In clinical scale research, participants are often classified by total scores or analyzed using logistic regression; however, the agreement between bidimensional item structures and external criteria is rarely quantified. This study proposes Theme—Attribute Assignment Analysis (TAAA), a structure-oriented framework that assigns theme labels derived from dual testlet structures to participant attributes. Cohen's kappa is used to quantify agreement with external criteria, defined here as Mapping Power (MP). Three empirical datasets (sample sizes approximately ranging from 40 to 200, with 8-20 items) were analyzed using predefined group labels or expert classifications as criteria, and four additional datasets were constructed to examine the effect of item-count imbalance. Under relatively balanced conditions, TAAA achieved MP values of 1.00, 1.00, and 0.76, which were broadly comparable to those obtained using regularized logistic regression. However, when the item-count ratio deviated markedly from 1:1 and dropped below 1:2, MP declined substantially and degenerate single-cluster solutions emerged. TAAA is therefore positioned as a structure-oriented agreement evaluation method rather than a predictive model and is particularly suitable for studies requiring the mapping of interpretable bidimensional structures to categorical attributes.