How to Measure Political Concepts Using Public Opinion Surveys : Learning from Cognitive Science

自从采用电脑辅助民意调查以来，问卷采用“随机抽样”的方法，已成为一种标准做法。随机化“消除”了总体统计中的排序效应。然而，由于“锚定效应”，当在个人层面上分析数据时，它会带来意想不到的、不可靠的性质。通过网络调查，我测试了几种解决问题的方法。我的结论是：1）通过改变问题格式来消除锚定效应是不容易的，以及2）“随机化”的使用应该是选择性的，而不是自动的。

A use of "randomization" of question items in a questionnaire has become a standard practice since the introduction of computer-assisted public opinion surveys. Randomi-zation "eliminates" the ordering effect in aggregate statistics. Because of an "anchoring effect," however, it introduces unexpected, unreliable nature when the data are analyzed at an individual level. Using web-based surveys, I have tested several ideas to remedy the problems. My conclusions are : 1) it is not easy to eliminate anchoring effect by way of altering question formats, and 2) a use of "randomization" should be selective, and not automatic.