環境財需求函數之估計：封閉式條件評估模型之比較分析

Estimation of Demand Functions for Environmental Quality: Comparing Models for Contingent Valuation Experiments

封閉式條件評估資料之分析，對於民眾平均願付價格或願受價格的求得，反應函數之設定有兩種：第一種是由Hanemann提出的間接效用函數法，其方式是以間接效用函數之差做為反應函數，另外一種是由Cameron和James提出的支出函數法，其方法是以支出函數之差作為反應函數；即使McConnell(1990)曾比較兩種反應函數之對偶情形，而Park和Loomis(1992)也曾對McConnell提出的對偶情況進行檢定，然而文獻上尚未於一研究中，同時以間接效用函數和支出函數所設定之反應函數進行綜合的比較與分析。本研究針對此，在各類反應函數下，設定線型、線型對數、半對數型及雙對數型等函數型式，並以羅吉(logit)和波比(probit)模型進行估計，同時比較檢定不同函數型式及不同估計模型下之結果，並解釋由一封閉式條件評估法所收集來之資料，在不同方法及模型下是否有顯著差異。本文同時利用一組以條件評估法收集來之資料做為實證分析之用，此資料是以高雄市之自來水家庭用戶為對象，評估東港溪水質改善帶給自來水用戶之經濟效益。綜合而言，實證結果顯示，平均10種函數式與2種估計模型所求得的18個連續函數之每戶每兩個月對於水質改善的平均願付價格為616.93元，此值落於大部份函數型式下之信賴區間內，此顯示若無法判斷何種函數型式為最佳的情況下，此值具有相當之可信度。另外，直接假設效用差或支出差為線型推導所得的水質受補償需求函數符合經濟理論性質；而在估計方法上，以支出函數法下的羅吉模型最為簡易，而間接效用函數法下的波比模型則最為困難。所以在分析封閉式條件評估資料時，除了考慮反應函數理論之合理性外，亦需考慮其在實證估計上之簡便與可行性。

A comprehensive discussion and comparison of two competing models, Hanemann's utility difference model and Cameron's expenditure difference model, in the analysis of contingent policy referendum data are presented. The comparison of Hanemann's and Cameron's model, an extension of earlier work, is analyzed by logit and probit model under different specification of utility and expenditure difference functional forms. The welfare measure of mean willingness to pay and the corresponding confidence interval are estimated. The Hicksian demand function for each specification is also derived. A set of data with 540 households from a contingent policy referendum experiment, a study conducted for the improvements of running water quality in Tung-kang creek of Ping-tung shian in Taiwan, is employed for our purpose. In summary, choice of models and functional specifications should consider not only the theoretical plausibility but also the empirical applicability and convenience. For all functional forms specified in this research, linear, linear-log, semi-log, and double-log, interpretation of response function as expenditure difference results a better performance in model estimation, mean welfare measure calculation, confidence intervals computation, and Hicksian demand function derivation if the function exists for good to be valued.