1996 年 11 巻 p. 3-22,143
This study reanalyzes the linkage between economic conditions and votes for an incumbent party using aggregate time-series data on U. S. federal and Japanese national elections. The uniqueness of the study is the use of a Seemingly Unrelated Regressions (SUR) technique that estimates vote functions for presidential, Senate, and House elections (House of Representatives and House of Councilors elections) jointly as system equations. The technique exploits all the information in aggregate election data and is capable of estimating vote functions more efficiently than existing techniques that omit part of the information in estimation. The SUR estimation uncovers the presence of the economic effect in Japanese elections that previous studies failed to find because of their inability to use all the information in aggregate data or because of Kramer's (1983) problem inherent in cross-sectional survey analyses.
Further, the study conducts formal cross-equation tests which show that, in the United States, the economic effect is greater in presidential than in legislative elections, whereas, in Japan, the effect is in similar magnitude in elections for both chambers of the Diet. Finally, the analysis detects a fundamental difference between the 1992 U. S. presidential and the 1993 Japanese parliamentary elections in which the incumbent conservative governments were defeated in the beginning of the new post-cold war era: the latter deviated significantly from the voting pattern in previous elections, while the former did not.
The study argues that retrospective voting shown to shape the standard electoral behavior in both states is rational and may not induce socially suboptimal political business cycles in the economy where large portion of economic growth is permanent. The electoral foundation of democracy is found to exist in Japan and the United States given the evidence that voters evaluate an incumbent party's economic performance correctly and critically in elections.