網路聲量與傳統民調之資料比對研究 -以2020台灣總統大選為例

2019 年臺灣民主進步黨黨內總統初選，⾸次在初選民調中加⼊⼿機加權，為市話調查提供了另⼀種可能性。事實上從業界經驗與結果判讀，近幾年臺灣幾次⼤型的選舉預測，市話仍然能在選前就準確的預估選戰結果，然⽽調查⽅法與時俱進，⼤數據探勘系統逐漸成熟，坊間民調公司幾乎都開始提供⼤數據服務，配合市話調查，提供客⼾更多元的參考訊息。早期民意調查多採市話調查的⽅式，趁著晚上民眾回家的時刻，撥打電話進⾏訪問，隨著外租族與年輕⼈使⽤⾏動電話變成習慣，不裝市話的年輕家庭愈來愈多，⾃然被排除在市話調查樣本之外。從統計原理來說，只要樣本加權後對應母體，這些被排除的調查對象，其實並不影響市話調查的結果。這形成⼆派學說的意⾒，傳統派主張根基於母體對應樣本之統計學理，市話調查仍具參考價值；另⼀派專家學者則認為，市話已經不⾜以涵蓋對應母體的⾜夠樣本。然⽽在⼆派意⾒相持下，並沒有實際的研究結果，⾜以⽀撐或證明任何⼀⽅的說法。市話加⼊⼿機加權後的雙重調查組合，和原本的市話調查有沒有差異性？是否更為準確？和⼤數據探勘的網路聲量⽐較，能為政治候選⼈提供什麼樣的資訊？從學術研究上來說，市話加⼊⼿機調查為近年來才開啟的新的民意調查⽅式，相關的研究很少，也沒有太多實際的結論得以對照與驗證其結果為何？本研究以臺灣2020 年的總統⼤選結果做為⽐較基礎，探究上述三種調查⽅式的差異與特性，在接近半年的重覆調查中，截取各次調查中相同的對象（即總統候選⼈蔡英⽂與韓國瑜）與相同的提問（⼆⼈⽀持度⽐對），做為⽐較基礎。研究範圍包含4 次傳統市話民調與1 次市話加⼊⼿機加權調查，同時對⽐了網路⼤數據資料庫在調查同時期的聲量，對應實際的選舉結果後，發現市話加⼊⼿機加權結果並沒有⽐傳統市話調查來得特別準確，無論是單獨看待蔡英⽂之得票結果或是韓國瑜最終的得票結果，市話加⼊⼿機的調都沒有傳統市調或⼤數據資料庫的聲量判讀來得有優勢，⼿機的使⽤族群加權是否過度矯正了樣本的對應性，還需更多的研究，當然本研究僅就單⼀次的數據資料，可能包括了其它變項與誤差影響研究結果，還需要更多實質的研究來證明，做為⼀個開端，本研究期能為政治民意調查提供⼀個不⼀樣的觀點。

關鍵字

大數據；總統選舉；網路聲量；輿情探勘；市話民調

並列摘要

In the rapidly changing information age. We must accept to the changes by the maturity of network communications and the explosion of information reception. Physical stores have become online shopping malls; Internet dial-up has become Wi-Fi, meetings Online; Line for chat, Netflix watching movies; Spotify listening to music, political communication and election investigations are no exception. Early public opinion surveys mostly adopted with home call. They dialed the phone to conduct interviews, when people go home in the evening. As young people use mobile phones to become a habit, more and more families do not use home call, and they are excluded from the sample in the room phone survey. In terms of statistical principles, as long as the sample is weighted to correspond to the population, these excluded survey subjects do not actually affect the results of the laboratory survey. In fact, several election predictions in Taiwan in recent years, home calls can still accurately predict the results of the election before the election. But eliminated without change. The big data system has improved, and almost all polling companies have begun to provide this service with multiple reference information. In the 2019 Taiwan Democratic Progressive Party’s inner-party presidential election, the mobile phone survey was added to the primary polls for the first time, which provided another possibility for home call surveys. Is there any difference between this dual survey combination and the original surveys? Is it more accurate? Compared with the network volume by big data, what kind of information can be provided for political candidates? This study uses the results of Taiwan’s 2020 presidential election as a comparison base to explore the differences and characteristics of the above three survey methods, hoping to provide a different perspective for political opinion polls research.