Implications of Errors in Survey Data: A Bayesian Model
针对二分调查数据中的误差问题，开发了一个贝叶斯模型，并用购买回忆数据展示误差会严重扭曲行为推断，导致点估计偏高、区间估计过窄，提醒营销等领域的研究者重视数据误差。
Data from surveys often include errors, and such errors can have a serious effect on inferences about behavior or perceptions. In this paper a model is developed for making inferences based on dichotomous survey data with possible errors. A likelihood analysis reveals an identification problem, which can be avoided when a Bayesian approach is taken. The model is illustrated with purchase recall data from two previous studies, and the analysis shows that errors can have a significant impact on inferences about behavior. Ignoring such errors leads to point estimates that are systematically too high in many cases and to interval estimates that are unrealistically narrow. The effective amount of information in the survey data is reduced dramatically by the presence of errors. These results have important implications for the use and value of survey data in marketing and in many other areas.