The status-quo and challenges of CAI surveys in political science: Comparisons of waseda-CASI surveys and official election results

In this paper, we illustrate how CAI surveys in political science have developed in the context of election studies, in particular, how CASI (Computer Assisted Self-administered Interviews) surveys have been applied to election studies in Japan. We then argue that various advantages exist for a survey mode of CASI, ranging from a quick data output, randomization and customization of questions and response choices, recording response time, implementation of survey experiments, mitigating social desirability bias, and to high representativeness. At the same time, we reckon two disadvantages of CASI, which are a lower response rate and a higher cost than other survey modes. To examine if a lower response rate really undermines a representativeness of data collected by CASI, we constructed a database of 79 post-election surveys conducted in Japan. Our analyses demonstrated that a lower response rate does not necessarily increase a turnout bias, i.e. a gap between an estimated turnout in surveys and the actual turnout at an election. Our analyses also confirmed that a lower response rate neither increases a vote percentage bias, i.e. a gap between estimated percentages of votes garnered by political parties and the official election results. Among the several survey modes, CASI appears to be closest to the actual electoral records. The concerns over a higher cost of CAI remain but its advantages seem to be not only theoretically assured but also empirically grounded.