Election Forecasting Techniques - Part I

This is the first of two special issues devoted to current topics and innovative approaches in the field of election forecasting techniques. The articles included in these special issues were submitted to the journal after a call for papers was circulated in mid-2013, soliciting contributions that advance the current state of the literature and/or promote novel approaches to political opinion polling, with special emphasis on uses of forecasting techniques of election results. The articles hosted in the two issues cover topics ranging from exit polls, explanatory statistical models based on structural variables (economic trends, government approval ratings, etc.), prediction markets, social media-based election forecasting, the web as a means to collect data on voting preferences, and measures of forecast accuracy. In the first contribution appearing in this issue, titled “Evolving approaches to election forecasting” (the only invited article), Jocelyn Evans examines major approaches to electoral forecasting and discusses their distinctive traits and the constraints which render them variably useful in specific research contexts. He also addresses the growing use of forecasting tools, stressing the need to adapt techniques originally developed in order to achieve other goals and to not lose track of researchers’ major purpose when employing these techniques, which is to say a greater comprehension of how elections actually work. As regards prediction markets, an interesting article submitted to the journal is “Accuracy and bias in European prediction markets”, by Sveinung Arnesen and Oliver Strijbis. The paper describes how prediction markets work, specifically the Iowa Electronic Markets (IEM), and provides a meta-analysis of the scores from 62 prediction market vote share contracts for elections in Switzerland, Germany, and Norway. The aim of the paper is to uncover potential biases in forecasting by comparing them with the actual results. The authors show that there is an aggregate bias in the predictions: the actual outcomes tend to have more extreme values than predicted, so that European prediction markets would be biased. Specifically, they show that small-sized vote share contracts tend to be overpredicted, and large-sized vote share contracts tend to be underestimated. The major result reported in this contribution appears to invite researchers to a cautious use of the logarithmic market scoring rule (LMSR) as an automated market maker in vote share markets. In “Assessing correct voting: A study based on a simulation of municipal elections in Italy”, Giancarlo Gasperoni and Debora Mantovani offer an empirical application of “correct voting” to the Italian political system. The authors estimate correct voting using data collected through the development of an on-line simulation of an Italian election campaign implemented via a “dynamic process-tracing environment”. A typology of voting behaviour is then proposed which combines both correct voting models and the traditional approach distinguishing between political subculture belonging and opinion-based voting among Italian voters. Four multinomial logistic regression models are developed in which the dependent variable is the above-mentioned typology of voting behaviour; the authors use these models to test hypotheses on voting behaviour according to which voters are more likely to vote “correctly” if they express high levels of interest in politics and high degrees of political competence, and if they are “active” seekers of information during the simulated election campaign. Findings show that voters are more likely to vote correctly if they express higher levels of interest in politics, but the effect of political competence is statistically insignificant. Moreover, voters are not more likely to vote correctly if they are “generally active” seekers of flow items concerning candidates’ issues orientations, but they are more likely to vote correctly if they are “specific active” seekers of information concerning their “correct” candidates’ issue orientations. A further article deals with “Forecasting elections with high volatility”, by Antonio F. Alaminos. In the article, Alaminos proposes the use of a combination of aggregated electoral data from the 1994 German Bundestag elections and the 1998 German Allbus social survey to estimate four probabilistic models of forecasting the German 1998 general elections. The models are built following the logic of Markov chains which, according to the author, make it possible to account for the large electoral volatility observed in the German elections across the 1990s. The forecasts based on the four models perform better than those provided by other techniques, in terms of predicting the winning party and the position of the second and third parties. In addition, the author demostrates that, among the four proposed models, the two corrected models – which assume that there are restrictions to electoral mobility – behave better than the two other pure Markov chain models, which assume that all voters can change their electoral choice. This first issue of the double-issue set concludes with contributions drawn from a round table discussion dedicated to election forecasting, which took place on February 15, 2013, in Milan during a national conference on “The Value of Statistics for Businesses and Society: Opinion and Market Research” promoted by the Association for Applied Statistics (ASA), the Association for Market, Social, and Opinion Research (ASSIRM), the Italian Statistics Society (SIS), and the Catholic University of the Sacred Heart of Milan.