因果推論:分析方法之探討及其對混合選制中「污染效果」之應用研究

Abstract

因果關係在社會科學研究中舉足輕重，但因果推論方法卻常有爭論。方法學當務之急在建立一個共通的思維架構，能貫穿隨機分派實驗、準實驗、自然實驗、以及觀察研究，不但邏輯一貫，而且更能落實到具體可行的分析方法。近年日益受到重視的「反事實的因果模型」（counterfactual model of causality）是個可以由評估型研究逐步擴展為因果機制的途徑，因為這個模型一方面能刺激反事實的逆向思考，但另一方面又能將這觀察不到的、假想的反事實操作化為控制（比較）組，對澄清觀念甚有助益。本三年期計畫的目標有二：一、在方法方面：從方法學的角度檢視文獻中溯因型及評估型兩類研究的方法，進而以「反事實的因果模型」（counterfactual model of causality）為基礎予以整合，發展最適合政治學之觀察研究的因果推論方法。二、在應用研究方面：將此一因果推論方法應用至「單一選區兩票制」下所謂「污染效果」（contamination effect）的議題，並以日本及我國的集體（選區）與個體（選民）資料，分析混合選制下政黨在選區提名候選人，是否有提升其政黨票的效果。

Establishing causal relationship among variables is one of the central foci of social science research. However, causal inference is also full of debates and controversies. It is critical for empirical researchers to find a unified framework which is able to integrate various research designs such as randomized experiments, quasi-experiments, natural experiments, as well as observational studies. The counterfactual model of causality (CMC) that attracts many students’ attention in the recent decade seems to be a promising approach. CMC not only stimulates alternative thinking but also operationalizes the unobservable “counterfactuals” into observable control/comparison groups. The purpose of this three-year research project is twofold. 1. To develop a causal inference model which is built upon the counterfactual model of causality and yet also takes account the underlying causal mechanism. Hopefully, this integrated approach will be suitable for most observational studies found in political science. 2. To apply the integrated approach of causal inference to a heatedly debated issue in the field of comparative electoral systems, i.e., the existence of the so-called contamination effect between the single-member district (SMD) ballot and the proportional representation (PR) ballot in mixed-member electoral systems. District-level and survey data of elections in Japan since 1996 and in Taiwan since 2008 under the MMM electoral system will be analyzed. Clarifying this issue with our rigorous approach to causal analysis will contribute substantially to the growth of the literature.