ECPR

Install the app

Install this application on your home screen for quick and easy access when you’re on the go.

Just tap Share then “Add to Home Screen”

ECPR

Install the app

Install this application on your home screen for quick and easy access when you’re on the go.

Just tap Share then “Add to Home Screen”

Nowcasting and Forecasting Potentials of VAAs: The Case of Smartwielen.lu

Advertising
Agenda-Setting
Activism
Raphael Kies
University of Luxembourg
Agnieszka Walczak
Luxembourg Institute of Socio-Economic Research - LISER
Raphael Kies
University of Luxembourg
Patrick Dumont
Australian National University

Abstract

The smartwielen.lu site is a so-called “voting advice application” that is well established in the Luxemburgish political landscape. Its usage increased steadily: in 2009, 36.000 matchings were established, in 2013 around 70.000, and for the last legislative election (2018), the number of matchings amounted to more than 200.000. This corresponds to around 50.000 unique users, i.e. 20% of the electorate in Luxembourg. The objective of this paper is to test to what extent the data from smarwielen.lu could become a valid (and cheap) method to measure political preferences to “nowcast” and “forecast” the electoral outcomes. The major question of the paper is whether the available data reflects the preferences of voters, after having been weighted according to a number of criteria. More particularly, we will test two sets of data derived from smartwielen. The first one are the partisan preferences, i.e. the matching for the political parties, expressed by smartwielen. The second one, are the responses to the optional questionnaire all the users were invited to respond to. This optional questionnaire - that was responded by more than 20.000 unique users - contained questions about the propensity to vote for each party (from a scale to 1 to 10) as well as questions on the socio-demographic data of users (age, gender, education, and zip code). In order to control for auto-selection bias, the latter will be calibrated on the socio-demographical variables (gender, age and education) at the municipal level. This paper should make an important contribution in the fast growing research domain of opinion mining and sentiment analysis (OMSA) that is rarely concerned with the measurement of political preferences (Piryani et al. 2017). The ones that are, essentially look at the capacity of twitter to nowcast and forecast electoral behavior (see Conover et al. 2010; O’Connor et al. 2010; Ceron et al. 2015). These studies suggest that twitter can in certain cases be more effective that traditional opinion survey to forecast electoral behavior despite the fact that active social media political users are sociologically not representative of the general population. They suggest that the ‘‘wisdom of crowds’’ may compensate for this partly unrepresentative information and that political preferences or ideology self-placement in social media do not differ fundamentally from the general population (Best and Krueger 2012; Ceron et al. 2015). Very few studies have attempted to do the same with the VAA (with the exception of Fournier et al. 2015) and, among them, none did benefit from such a large proportion of respondents for to general questionnaire (i.e. the VAA questionnaire) and the optional questionnaire.