Issues
home Home navigate_next Issues navigate_next Backissues navigate_next Volume 31, No.1 navigate_next The Examination of Taiwan's Election and Democratization Study Panel Data

The Examination of Taiwan's Election and Democratization Study Panel Data

  •  Kuang-Hui Chen and Tsung-Wei Liu
  •  2006 / 11  

    Volume 13, No.2

     

    pp.75-116

  •  10.6612/tjes.2006.13.02.75-116

Abstract

TEDS conducted two waves of panel studies. These panel data can be used to describe the dynamics of Taiwanese voters and to develop related causal models. However, because of panel attrition and panel effect, there may be problems of internal and external validity. The examination of panel data shows that the panel attrition did not occur randomly. There are significant differences between those respondents who participated in the second interview and those who dropped out in terms of demographic characteristics, but no significant difference was found in terms of political attitudes. Both TEDS 2003 and 2004P were composed of panel samples and independent samples. Panel samples are those respondents who were interviewed in TEDS 2001 and independent samples are those respondents who were never interviewed before. To be interviewed by academic research staff is a special experience, so the respondents may be intrigued to access more political information and become more willing to participate in political activities afterwards. Therefore, the three TEDS surveys could be treated as a quasi-experiment. While the panel samples is treatment group, the independent samples is control group, and the interview is the treatment. This quasi-experiment demonstrates that panel effect did change the respondents' political attitudes and increase their political participation. To sum up the consequences of panel attrition and panel effect, TEDS panel data are biased. Researchers who analyze this data set should be attentive to the issue of biased sample and think about the methods to correct the bias before drawing conclusions or making inferences.