INVESTIGATING BIAS BY PROXY RESPONSE IN OPEN-ENDED SURVEY DATA

Abstract

In social surveys, a response by one other than the target sample is called a “proxy response,” and poses several widely recognized problems in analysis. However, current attempts to systematically analyze the proxy-response bias are insufficient. This study aims to investigate the bias focus on open-ended questions. We use a group-based choice model of response behavior and topic model to analyze the open-ended response in 2012 Kumamoto Person Trip survey data. The results indicated that the number of characters and words in an open-ended response decreases by proxy response. We also found that the topics by self-response tend to be on their experience, while those by proxy response tend to be general one.