本文設計「一般化內生抽樣法上即在一種財貨或服務的各選擇之需求發
生地點進行系統抽樣，獲得各樣本點之選擇、需求量、個人性質與選擇性質等資
料 此內生抽樣方法係同時基於選擇和需求量此二個內生變數的分層抽樣，故為
現場抽樣法與擇基系統抽樣法之綜合與一般化，其樣本可用於下述五種需求函數
之估計:單一選擇需求函數模式、混合需求函數模式、需求函數體系、間斷選擇
模式與二段模式;因此可解決現場抽樣樣本僅可用以估計單一選擇需求函數，及
擇基系統抽樣樣本僅可用以估計間斷選擇模式之缺點。按著，本文分別導出五種
需求函數模式中，混合需求函數模式、間斷選擇模式與二段模式等三者應用一般
化內生樣本之估計方法，並證明混合需求函數的估計式具有一致性和漸近常態分
配性質。最後，經Monte Carlo模擬方法，證實估計方法之正確性，並且顯示在常
態分配母體之情況下，一般化內生樣本之混合需求函數模式估計值遠優於傳統的
在母體隨機抽樣樣本之設限迴歸估計值(censoredregression estimates)。

There are five kinds of demand function models that have been used in
applied economics studies: a single-choice demand function, a pooled demand
function, a system of demand functions, adiscrete-choice model, and a two-part model.
The on-site samplingand the choice-based systematic sampling have been applied,
respectively) to estimate single-choice demand functions and discretechoice models.
This paper extends and generalizes the above twosampling strategies to be a
'generalized endogenous sampling' strategy which draws observations on the basis of
two endogenous variables, namely, choices and quantity demanded. Then, the
paperdevelops three estimation methods for the pooled demand function, the
discrete-choice model, and the two-part model using thegeneralized endogenous
samples. The estimators are proved to beconsistent and asymptotically normal. A
numerical study confirmsthe desirable performance of the estimators.