本文評估民國88年我國公私立地區醫院之技術效率。目前國內外文獻所採用之兩階段非參數資料包絡分析法(nonparametric data envelopment analysis, DEA),其作法為:第一階段利用非參數方法計算效率值;第二階段,採用參數假設之截斷陶比迴歸模式(censored Tobit regression model),探討影響醫院效率表現的外部環境因素。就方法論而言,第一階段採用非參數估算效率值,卻於第二階段時改以參數假設之Tobit 迴歸進行分析,邏輯並不一致。再,若參數假設錯誤時,censored Tobit迴歸結果將有偏誤(biased)並且不一致(inconsistent)。本文修正傳統作法,於第二階段時採用假設誤差項為半參數(semiparametric)的對稱截斷最小平方法(STLS; Powell, 1986)與截斷最小絕對離差法(CLAD; Powell, 1984)兩種模式進行分析,並與censored Tobit迴歸之結果,作一比較。 第一階段估算結果發現:地區醫院產業確實存在技術無效率,且公立醫院效率表現低於私立醫院。從第二階段分析結果發現:權屬別、醫院規模(病床數及其平方項)、平均住院日、病床使用率,皆為影響效率的重要因素。而STLS、CLAD與Tobit分析結果之比較顯示:使用半參數假設之STLS與CLAD,其結果相對較採行Tobit迴歸結果穩定。
This study investigates the technical efficiency of local hospitals in Taiwan. The typical two-stage DEA method in the current literature computes non-parametric efficiency scores at the first stage. At the second stage, censored Tobit regressions are used to identify external factors that are thought to influence the estimated efficiency sores. Methodologically, it is inconsistent to use a non-parametric estimation at the first stage, while switching to a parametric method in the later stage for further discussions. Furthermore, it is well known that censored Tobit regression results are biased and inconsistent if the parametric assumptions are problematic. In this study, we suggest the symmetrically-trimmed least squares (STLS; Powell, 1986) and censored least absolute deviation (CLAD; Powell, 1984) for the second stage discussion, and compare their results with those of the censored Tobit regressions. The first stage estimation results suggest that technical inefficiency is pervasive among local hospitals. Moreover, public local hospitals performed worse than their private counterparts. The second stage analysis suggests ownership, size, average stay, and bed flows are the factors that influence efficiency. Finally, the estimation results of the semiparametric STLS and CLAD are more stable than those of the censored Tobit regressions.