หนัาหลัก
 
Original article
Title Forecasting the election results by applying Pavia's method
Author Nattapon Yamchim* Pattrawadee Makmee and Kanok Pantong
Page 64-70

Abstract

  This research aims to forecast the election results by applying Pavia’s method. In this paper, the information of opinion toward general election, which is reflected as one of behavioural science is applied, including using applied statistics to forecast the election results beforeannouncing the election results.The process of data collection about people opinion wasproceeded by survey related to election issues. In this survey, the sample was 3,600 electorates in the general election on24thMarch 2019 from 30 electoral zones in Bangkok and the questionnaire about opinion of the general election was used as the tool for data collection. The applied statistics methods in this survey are percentage, Pavia’s method analysis (Mean AbsolutePercent Error: MAPE). The poll revealed that five parties received major scores, 22.69% for Pheu Thai, 21.94% for Democrat, 20.39% for Palang Pracharath and 16.69% for Future Forward Party. In terms of Analysis by using Pavia’s Method, the poll showed different results, 23.96% for Palang Pracharath, 22.45% for Future Forward Party, 21.25% for Pheu Thai Party and 19.12% for Democrat Party. Whenthe poll results by using Pavia’s method was compared with actual election, The percent of accuracy indicated at 82.28% or 17.12% of error.

Keyword:Forecast, Election, Opinion Survey, Poll

Introduction

  The opinion survey, which is correctly proceeded contributes to well-accepted result of the survey. The public opinion survey or polling has been well accepted and received popularity in the United State of America. This type of survey has been proceeded in order to predict or forecast the election results or interesting public issues.

  The first forecasting simulation which was designed to forecast the result of US presidential election was used in Econometrics class, studying of the effect of economic activity on voting. The purpose of this study was to present the votingbehavioral and analytical pattern from the effect of economic activity on presidential voting (Fair, 1978)

  In case of US presidential election, forecasting with regression model was ever used to forecast the state’s voting results by using thepast record of national polls and all information of all states in US. This forecast revealed that leading presidential candidate in any state on September before the election seemed to be elected on November. Both information of pre-election polls and post-variable was together applied for data analysis. This method increased more punctuality and accuracy of US presidential election forecasting. (Holbrook and Desart, 1999)

  In 2010, Pavia had improved the accuracy of the forecasting of election results based on the results of the pullsfrom pullingboots. The improvement consisted of three parts of the forecasting by using raw data from the direct surveys;part 1was the process of using bias checking to measure improvement of the use of Nonresponse Bias. Part 2 was an approximationto operate after the use of bias checking of Nonresponse to help forecast.In addition,part 3 was an implementation after the process in the second part was completed, resulting in the integration of different variables. When forecasting, regardless of the method forecasting results,were not real results so that errors can occur. In 1948, the forecastingof the US presidential election resultsshowed an error between Dewey's and Truman's popularity. The result revealed that Dewey will win but the election results with a score of more than 5% were Truman was elected president. This mistake caused a crisis of faith to the polls and polling agencies. The polls have been greatly improved (Wapor, 2006; Jounes, 2008) and the 2016 election, the error in the US presidential election forecasting was once again wrong when the results of the polls showed that Clinton will win, however, in fact, Trump was the winner.

 

  In terms of Thailand, a general election poll was firstly proceeded in 1975 and the poll has been continuouslyperformed by many pollingagencies. When members of parliament election taken place on 3rdJuly 2011, many pre-election polls surveyed by several agencies, Ramkhamhaeng Poll, ABAC Poll and Suan Dusit Poll had error. For Suan Dusit Poll’s result, 162 seats of member of parliament werepredicted to be taken by Democrats, this forecasted figure had a little error from the actual result, 165 seats. However, when considering the poll in each electoral zone, the forecasted number of membersof parliament (MPs) from Suan Dusit Poll had significant error. 3 seats of MPs in Bangkok were forecasted to be 25 seats for Pheu Thai and 5 seats for Democrat. But, the actual election results showed dramatic figures, 27 seats for Democrat and only 9 seats for Pheu Thai. This figure gave significant error around 44.45%

As a result of this forecasting error, investigators determined to find out a forecasting method of election poll and improve forecasting results of polling by Pavia’s Method (2010). This forecasting method specifies how to improve accuracy of

election forecasting from exit poll results and also uses bias testing for improving personal bias when answering the question. However, personal information or background of theelectorate was not applied in this forecasting improvement. Also, Trangucci, Ali, Gelman and Rivers (2018) explains voting pattern in 2016, analysis of pre-election poll in 2012 and 2016.This research shows difference between voting result in each electorate group.Gender and education level playthesignificant role in thedifferent voting decision. Moreover, personal and background information of electorate wasapplied together with Pavia’s method adjustment in order to improve theaccuracy of election forecasting, when comparing withthe actual election result.


 
RDI Suan Dusit Rajabhat University.
295 Ratchasima Road, Dusit, Bangkok 10300 Tel 0-2244-5280-2 Fax 0-2668-7460

© Copyright 2013 www.research.dusit.ac.th