2005年台北縣選民之一致與分裂投票：縣長與鄉鎮市長選舉總體資料之分析

Abstract

本文使用總體層次的選舉結果資料，藉由區位推論方法，嘗試探討台北縣選民於2005年三合一選舉中，在縣長與鄉鎮市長這兩種不同層級行政首長選舉的一致與分裂投票情形。本文以「Goodman區位迴歸模型」及「Multinomial-Dirichlet階層模型」進行總體資料的分析。研究結果顯示，以Goodman區位迴歸模型僅能看出大致趨勢:不論泛藍或泛綠陣營，在兩項選舉中的一致投票率仍遠高於分裂投票率。而Multinomial- Dirichlet階層模型則更進一步提供每種投票模式的估計係數。大體而言，泛藍陣營的縣長候選人支持者，有將近八成同樣支持泛藍陣營鄉鎮市長候選人，而投給羅艾嘉的選民有將近64%在鄉鎮市長選舉也支持泛綠候選人。若以聯合機率分布來看，藍藍型一致投票則有約五成三的比例，綠綠型一致投票則有超過33%的比例，亦即藍藍型一致投票的比例要高於綠綠型一致投票。在分裂投票部分則是以綠藍型分裂投票率高於藍綠型分裂投票率。最後，本文以區位推論得出之結果，與TEDS2005M民調資料分析之結果進行比較，發現兩者結果大致相符，此亦顯示本文以總體資料進行個體層次行為推論之分析，有高的可信賴程度。

This article examines voting choices in the local elections of Taipei county and township magistrates in December 2005, by using the ecological inference method with aggregate-level data. Empirical evidence indicates that the result of Goodman`s ecological regression model provides an approximate pattern that most voters cast straight-ticket votes in the simultaneous elections. In addition, analysis from the Multinomial-Dirichlet hierarchical model presents nearly 80% of voters who vote for the KMT county magistrate candidate support the KMT/PFP township magistrate candidate, and nearly 64% of voters who vote for the DPP county magistrate candidate support the DPP/TSU township magistrate candidate. Moreover, with regard to the joint probability of the two types above, pan-blues seem to enjoy a higher straight-ticket rate (53%) than pan-greens (over 33%). As to the rate of split tickets between the county and township magistrate elections, we find that a green-blue pattern is noticeably higher than a blue-green pattern. Finally, in comparing the resu1ts of the ecological inference model with aggregate-level data with the resu1ts of cross-tabulation with individual-level data from surveys, we find that there is a similar pattern between these two different approaches.