條件評估法中受訪者內在相關之應用

Application of Intra-Respondent Correlation in Contingent Valuation

條件評估（contingent valuation）法常用封閉式（closed-ended）的問卷設計來蒐集願付代價（willingness-to-pay, WTP）或願受補償（willingness-to-accept, WTA）的資料，並以LDV（limited dependent variable）模型配合最大概似（maximum likelihood, ML）法進行迴歸分析。大部份的CV問卷調查中，受訪者會被告知不只一種環境改善計畫並被詢及其對每一個別計畫的WTP，從計量模型的架構來看，同一受訪者對於不同環境改善計畫的WTP函數很可能是相關而非獨立的，而此一受訪者內在相關（intra-respondent correlation）的起因，很可能是源於模型中被忽略的自變數。ML法對於模型的設定（specification）相當敏感，當ML法被用於估測LDV模型時，就算是被忽略的自變數與包含在模型內的自變數不相關時，其ML參數估測亦不具一致性。本研究討論當受訪者內在相關出現在LDV模型中時，對數概似函數（log-likelihood function）的推導，並介紹一檢定受訪者內在相關存在與否的拉氏乘數（Lagrange multiplier, LM）檢定法。實證結果顯示，本文所介紹之受訪者內在相關模型，其ML估測不但具備一致性且其估測效率亦提高。 In contingent valuation (CV) surveys, very often, a respondent is asked several willingness-to-pay questions regarding different environmental programs. The multiple responses from the same respondent may in fact be correlated, and by appropriately modeling, the intra-respondent correlation can be used to correct for the possible inconsistency as well as to improve the efficiency for parameter estimates. In this study, a limited dependent variable with model with intra-respondent correlation is modeled using a random-effect model as suggested by some panel data literatures. A Lagrange Multiplier test for testing the existence of the intra-respondent correlation is then being discussed. Finally, an empirical application is given to support the conclusions.

In contingent valuation (CV) surveys, very often, a respondent is asked several willingness-to-pay questions regarding different environmental programs. The multiple responses from the same respondent may in fact be correlated, and by appropriately modeling, the intra-respondent correlation can be used to correct for the possible inconsistency as well as to improve the efficiency for parameter estimates. In this study, a limited dependent variable with model with intra-respondent correlation is modeled using a random-effect model as suggested by some panel data literatures. A Lagrange Multiplier test for testing the existence of the intra-respondent correlation is then being discussed. Finally, an empirical application is given to support the conclusions.