客家族群投票抉擇的時空變異：由地理加權迴歸檢視歷屆總統選舉

Abstract

客家選民的投票抉擇往往被視作同質性的，然而來自歷史、語言的事實反映出不同地區的客家族群不僅不同源，甚至彼此間難以溝通。也因此諸多學者提出「北藍南綠」的說法，描述不同地區客家族群的投票傾向。近期又有學者提出「北部濱海客家偏綠」、「三分天下」的概念。本文認為客家族群的投票傾向確實具有空間異質性，而空間異質性的探索應以空間分析研究途徑。本文也正視時間上的異質性，即客家族群的投票抉擇不一定是全然穩定的，而是會隨時間變化的。在建構理論的過程中，需要以更長遠、更宏觀的角度，也因此本文採用定群追蹤分析（panel analysis），更能掌握客家族群投票抉擇的穩定與變遷。 本研究的分析單位為我國除去金門、連江的 358 個鄉鎮市區；選舉的層級為總統選舉；時間上自 1996 至 2020 年共有 7 個年度。為避免人為的主觀偏見干擾，本研究透過地理加權迴歸（geographically weighted regression），讓空間資料自行呈現每個鄉鎮市區中的客家人口比例對政黨得票率的影響，並以地圖繪製其估計結果。本研究的實證分析包含 7 屆總統選舉，也包含以固定效果模型（fixed effect model）下 7 屆選舉的定群追蹤分析。 研究結果表明其具有時空間的異質性。我們發現「北藍南綠」具有一定的解釋力，但此項說法過度化約了我國客家族群的投票傾向，並讓後續研究者的關注焦點限縮在南與北。而中部地區的客家族群在逐屆分析中呈現對藍營的強力支持，但固定效果之下這種支持被消除了，本文認為可能是產業別的原因，使得以農業為主的地區會在總統層級的選舉中青睞藍營候選人。「北部濱海客家偏綠」這項說法則並不正確，在各項分析中都沒有呈現這種現象。「三分天下」則包含了多種說法，其中提到了新竹、桃園、高雄的客家族群會逐漸不支持國民黨，但在本文的實證資料中，這種現象也不明顯；也提到屏東地區的客家族群傾向支持民進黨，這在定群追蹤分析中可以發現，本文認為「三分天下」尚需要更多的驗證才能判斷。

The voting choices of Hakka voters are often perceived as homogenous; however, historical and linguistic facts reveal that different Hakka people in various regions not only have different origins but also face difficulties in communicating with each other. As a result, many scholars have proposed the " North are blue, South are green" concept to describe the voting tendencies of different Hakka people in different regions. Recently, scholars have also proposed the concept of "North coastal Hakka are green" and the idea of "the three divides". This study acknowledges that the voting tendencies of Hakka people do exhibit spatial heterogeneity, and exploring this spatial heterogeneity requires a spatial analysis research approach. The study also recognizes the temporal heterogeneity, implying that the voting choices of Hakka people may not be entirely stable but can change over time. In the process of constructing theories, a longer-term and more macroscopic perspective is needed. Hence, this study adopts a panel analysis approach to better understand the stability and changes in the voting choices of Hakka people. The analytical units of this study are 358 townships and districts in Taiwan, excluding Kinmen and Lienchiang. The level of elections considered is the presidential elections, spanning seven years from 1996 to 2020. To avoid subjective biases, this study employs Geographically Weighted Regression (GWR), allowing spatial data to reveal the impact of the proportion of Hakka population in each township and district on the party's vote share. The estimated results are presented by maps. The empirical analysis of this study covers seven presidential elections and includes fixed-effect models for panel analysis across the seven election periods. The research findings demonstrate temporal and spatial heterogeneity. We found that the concept of "North are blue, South are green" has some explanatory power, butit oversimplifies the voting tendencies of the Hakka ethnic group in Taiwan and narrows the focus of subsequent researchers to just the north and south regions. In the central region, the Hakka ethnic group showed strong support for the Blue camp in each election, but this support diminished under the fixed-effect model. The study suggests that this might be due to the influence of different industries, with agriculturally dominant regions leaning towards the Blue camp in presidential elections. The notion of "North coastal Hakka are green" is not supported by the various analyses, as it did not manifest in any significant way. The concept of "Tthe three divides" includes several hypotheses, one of which suggests that the Hakka people in Hsinchu, Taoyuan, and Kaohsiung would gradually shift away from supporting the Kuomintang (KMT). However, such a trend was not evident in the empirical data analyzed in this study. Another hypothesis in the "the three divides" concept posits that the Hakka people in Pingtung tends to support the Democratic Progressive Party (DPP), which was found in the fixed effect panel analysis. The study concludes that the "the three divides" hypothesis requires further verification before drawing definitive conclusions.