How to Reconstruct a Trend when Survey Questions Have Changed Over Time.

Abstract

Many trend studies draw on survey data and compare responses to questions on the same topic that has been asked over time. A problem with such studies is that the questions often do not remain identical, due to changes in phrasing and response formats. We present ways to deal with this problem using trend data on life satisfaction in Japan as an illustrative case. Life satisfaction has been measured in the Life in Nation survey in Japan since 1958 and the question used has been changed several times. We looked at three methods published by scholars who tried to reconstruct a main trend in life satisfaction from these broken time-series, coming to different conclusions. In this paper we discuss their methods and present two new techniques for dealing with changes in survey questions on the same topic.