美国政治研究中的抽样调查方法

Survey Research Methods in American Political Studies

抽样调查是研究美国政治的重要观测工具。自19世纪末至今,美国政治研究中的抽样调查从起源、发展,到21世纪以来遇到挑战。面对概率抽样调查覆盖误差增大、应答率下降和成本上升,非概率抽样调查的兴起,以及来自大数据的竞争等问题,抽样调查研究者们不断创新,正在探索响应式调查设计,非概率样本的统计推断,以及与大数据结合应用等方法。本文采用总调查误差的框架,从测量误差、覆盖误差、无应答误差和调整误差四个方面分析了2016年美国总统大选前民调失灵的原因。概率抽样调查、非概率抽样调查和大数据各有自己的主要应用场域,未来的发展中三种数据采集手段会相互校验、融合使用,而高质量的概率抽样调查是衡量非概率抽样调查或大数据质量的参照基准。

Survey research is an important observational instrument for American political studies.It has quickly developed since it originated in the late 19th century,but it confronts serious challenges in the 21st century.To deal with the problems such as growing coverage errors,decreasing response rates and increasing costs,non-probability sampling survey’s rise,and big data’s competition,survey researchers are innovating,proposing responsive survey design,inventing statistical inferential techniques for non-probability sampling surveys,and exploring possible joint usage with big data.Using the total survey error frame,this article analyzes the reasons of the polling’s predication failures in the 2016 presidential election from the aspects of measurement error,coverage error,non-response error,and adjustment error.It concludes that probability sampling surveys,non-probability sampling survey,and big data have their own domains of application.