Performance Evaluation: Three Methods of Election Prediction on Election Day
選舉預測 ； 績效評估 ； 政治版圖 ； 典型鄉鎮 ； 典型村里 ； 出口民調 ； election prediction ； performance evaluation ； political map ； typical township ； typical village ； exit poll
|Volume or Term/Year and Month of Publication||
|Academic Degree Category||
選舉預測是選舉研究者極感興趣的議題之一，過去台灣對於選舉預測的研究多集中在以民意調查結果來進行預測工作，然受到選罷法規定選前十天禁止發佈民調結果，在民意易受到短期競選事件影響下，採用民調結果進行選舉預測工作有其侷限性存在，因而衍生出了投票日當天的選舉預測方法，從實務經驗中證實了該方法有其預測的準確性，然而在學術研究領域中卻是付之闕如。 是以，本文針對政治版圖估票模式、典型鄉鎮或典型村里以及出口民調三種投票日當天選舉預測方法進行探討。引介其概念界定、源流與執行流程，並對於實際進行選舉預測的經驗進行評述，針對其預測準確性、選舉適用性與樣本代表性進行績效評估，亦對於影響預測表現準確性的因素提出解釋，並嘗試提供解決之道。 本文發現，政治版圖估票模式與出口民調不論在準確性、適用性與代表性等績效評估的面向中皆有相當良好的表現，乃是因其具備了完善的抽樣設計與統計方法保護所致，兩種方法各有其學術研究上之意涵存在，是為日後在進行選舉預測時可茲採用的方法；而典型鄉鎮或典型村里雖在選舉預測時亦有其準確性，然卻難以由人為控制其預測表現，且本文亦證實預測準確之因並非由於人口結構一致所致，僅只能將該鄉鎮或村里視為是具有政黨支持程度的代表性。 此外就方法論上，本文建議政治版圖估票模式可採用PPS抽樣與實際得票率的估計的方式取代現有以等距抽樣與用平均得票率估計的方法，預測時會較為準確與合乎邏輯；出口民調則是各投開票所成功樣本數應達五十個樣本數以上以降低抽樣誤差，並建議參照國外立法例，將出口民調的執行納入法規範之下；再者，本文亦嘗試結合政治版圖估票模式與出口民調，建議執行出口民調時可採用政治版圖估票模式的抽樣設計，藉以降低人力與物力成本之耗費。
This paper is intended as an investigation of election prediction, one of the most interesting topics by scholars. So far, related studies in Taiwan have been concentrated on the method of public opinion survey. However, under the law of election and recall, which prohibits publishing the result of survey during the ten days before Election Day, the method is limited substantially because public opinion are more easily manipulated in the last days of the campaign. To improve the accuracy of predictions, another method, prediction based on the Election Day, has proved its capability by empirical experience but lacks further research. Given the circumstances, the purpose of this paper is to explore the three methods available to the analyzer – political map, typical township or typical village, and exit polls- based on Election Day. Thus, according to the conception, inception and procedure used in these three main methods, we examine the performance of each prediction in terms of its accuracy, applicability and sample representativeness. Furthermore, we attempt to explain the reasons that cause different predictability and provide various solutions. The main finding of this paper is that the political map and exit polls, which incorporated sampling design and statistical methodology, proved to be effective in many aspects. Thus, the methods are capable of application in the future and worthy of further studies. In spite of the good performance of typical township and typical village methods in the prediction of elections, it is difficult to manipulate and design the whole procedure. Also, the outcome of typical village does not result from the consistency of population structure but from the representativeness of party support. For methodology, we suggest that it would be more precise and logical in the prediction with political maps when PPS sampling and estimation by the actual rate of ballot are adopted rather than systematic sampling and estimation by the average rate of ballot. Furthermore, we suggest that the exit poll ought to reach more than fifty samples for each poll to lower the sampling error. The execution of exit poll should under the regulation by law, which could draw on foreign legislation. A combination of the two methods would mean less manpower and resources devoted as the sampling design of political map could be introduced into the use of exit poll.
社會科學 > 政治學