在資料包絡法的文獻中，有一種對數線性模式可用以估測生產曲面，此 種模式可以解釋邊際產量遞增的現象。本文由對偶的角度探討此種模式之一些特 性，最重要的一項發現是對於多產品的決策單位，其整體效率可表示為各單一產 品效率之加權幾何平均;另外一項發現則是在此模式下，產品轉換曲線為凸性函 數，並非一般所預期的凹性型態，如此，有些決策單位之效率將會高估。基於此 點，本文最後建議採用一種二階段之效率評估方式，以降低線性與對數線性所造 成之影響。

　　　　　In the literature of data envelopment analysis (DEA) the piecewise loglinear model for estimating efficient production surfaces which permits cases of both increasing and nonincreasing marginalproducts receives less discussion than it should have. This paper takes a dual point of view to discusssome properties that this model possesses. One major point is that the efficiency of a multioutput decision making unit (DMU) can be expressed as a weighted geometric mean of the efficiencies of eachoutput. Another point is that the product transformation curve of this model is convex, not of the usually expected concave form. A consequence is that the efficiencies of some DMUs will be overestimated. In conclusion, a two-stage method for measuring efficiency is suggested to alleviate the effects caused by linearity and loglinearity.