採用無母數因子分解法來解構美國的種族健康差距

Racial Health Disparities in the United States: A Nonparametric Decomposition Approach

過往研究常使用Blinder-Oaxaca（B-O）這個分解法來檢視美國的種族健康差異。然而，B-O方法假設每個可觀察變量均具備可比較特徵，這假設常會導致這些屬性差距的因子估計偏誤。為克服B-O方法的局限性，本文採用從配對模型（matching）所衍伸的無母數參數估計法。透過這方法，我們採用無母數參數估計法將白人和黑人間健康差異區分為可解釋和未解釋部分。我們的結果顯示，非母數參數估計與母數參數估計所得結果一致，也得到更為準確的估計結果。

Prior studies have used the Blinder-Oaxaca (B-O) decomposition method to examine racial health disparities in the United States. However, the B-O method assumes that the characteristics of each observed variable are comparable resulting in a misestimation of the components of the gap attributes. To overcome the limitations of the B-O model, this paper employs an alternative nonparametric decomposition approach using a matching technique. By adopting this method, the differences in racial health disparities between two groups–Whites and Blacks–are nonparametrically estimated and decomposed into explained and unexplained portions of the gap. Our results show that the nonparametric estimations are consistent with the parametric estimations, and the nonparametric analysis results from the matching are substantially more accurate.