离散选择模型的基本原理及其发展演进评介

Research on The Theoretical Basis and Evolution of Discrete Choice Models

离散选择模型的研究真正兴起于19世纪50年代末,属于微观计量经济学的范畴。该模型能够对个体和家庭行为进行经验性的统计分析,因而在经济学和其他社会科学中得到广泛的应用。本文从离散选择模型的基本性质及效用最大化的理论背景出发,指出logit模型虽然使用的是最早并且最为广泛的离散选择模型,但是其存在着三大局限性:不能表示随机口味的变化、暗含成比例的替代形式和不能处理不可观测因素在不同期间相关的情形。GEV(含嵌套logit)、probit和混合logit模型等其他的离散选择模型,很大程度上都是为了避免这些限制而产生并发展起来的。

The boom of discrete choice models , which belong to the area of microeconometrics, were really beginning from the end of 1950s. The models can be used to analyse the behaviors of private and family experientially, and they are widely applied to economics and other social sciences. This paper begins with the theoretical basis and utility maximization, then demonstrates that there are three limitations of logit： it can not represent random taste variation; it implies proportional substitution across alternatives and cannot handle situations where unobserved factors are correlated over time, though logit is by far the easiest and most widely used discrete choice model. Other models, such as GEV （including nested logit）, probit and mixed logit, have arisen largely to avoid these limitations.