Measuring public support for European integration using a Bayesian item response theory model
| Author | Michele Scotto di Vettimo |
| DOI | 10.1177/14651165221080400 |
| Published date | 01 June 2022 |
| Date | 01 June 2022 |
| Subject Matter | Articles |
Measuring public support
for European integration
using a Bayesian item
response theory model
Michele Scotto di Vettimo
Department of Politics, University of Exeter, Exeter, UK
Abstract
This study proposes the use of Bayesian item response theory models to measure aggre-
gate public support for European integration. This approach addresses the limitations of
other indicators and produces valid estimates of public attitudes over long time periods,
even when available indicators change over time or present interruptions. I compare
Bayesian item response theory models with alternative approaches used in the study
of support for European integration, and demonstrate that they produce more accurate
estimates of latent public opinion. The estimates are validated by showing their associ-
ation both to alternative public opinion measures and to the vote share of Eurosceptic
parties across Europe. I show that Bayesian models solve unaddressed issues like ensur-
ing cross-country comparability of the estimates and modelling responses with multiple
answer options.
Keywords
European integration, Euroscepticism, item response theory, public opinion, public support
Introduction
Against a background of growing politicisation of the European Union (EU), public atti-
tudes towards EU integration have become central to understanding European-level
policy-making (e.g. Bølstad, 2015; Toshkov, 2011; Wratil, 2019) as well as national-
level party strategies (e.g. Hutter and Grande, 2014). Yet, the precise measurement of
Corresponding author:
Michele Scotto di Vettimo, Department of Politics, University of Exeter, Amory Building, Rennes Drive,
EX4 4RJ, UK.
Email: m.scotto-di-vettimo@exeter.ac.uk
Article
European Union Politics
2022, Vol. 23(2) 171–191
© The Author(s) 2022
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/14651165221080400
journals.sagepub.com/home/eup
aggregate public opinion towards the EU has received relatively limited attention. A wide
range of measures have been employed, but few of these efforts have been justified with
reference to established conceptualisations of EU support (Hobolt and De Vries, 2016).
In particular, EU support possesses two properties calling for a careful empirical treat-
ment. Firstly, people might like certain aspects of the EU, but not others and present
ambivalent attitudes towards the EU integration rather than just showing support or oppo-
sition (De Vries, 2013). Secondly, EU support has a multilevel in nature. According to the
‘benchmark theory’, EU attitudes are the result of a comparison between citizens’EU and
national evaluations. National contextual factors are ‘part-and-parcel of the way in which
people evaluate the EU level, and vice versa’(De Vries, 2018: 28). Hence, survey indi-
cators tapping into EU support are rarely understood in the same way across countries,
raising serious concerns for the cross-national comparability of measures of EU
support based on survey indicators (Ariely and Davidov, 2011: 272).
Yet, existing measurement strategies are rarely capable of dealing with these two
issues satisfactorily. Scholars exploring aggregate-level EU support have mainly relied
on existing single-question indicators, selected mostly because they constitute the only
data source that allows for cross-national and longitudinal comparisons (Hobolt and
De Vries, 2016: 416). However, the cross-national ‘equivalence’of these indicators is
rarely assessed. Furthermore, they often have to be retrofitted to suit a particular analysis
(Anderson and Hecht, 2018: 621).
Dimension reduction techniques, like the Dyad Ratios (DR) algorithm (Stimson, 1991,
2018), have also been employed to estimate latent public EU support from a set of dif-
ferent indicators. In the EU context, the algorithm represents the most advanced approach
so far employed to estimate EU support starting from aggregate-level data (Anderson and
Hecht, 2018; Guinaudeau and Schnatterer, 2019). Nevertheless, DR estimates inherit all
the problems related to the lack of cross-national comparability of the raw indicators, and
the algorithm is poorly designed to deal with neutral answer options.
Thus, I propose the use of Bayesian item response theory (IRT) models for the estima-
tion of public EU support starting from aggregate-level data. The EU context represents a
promising area of extension for this technique. Existing surveys provide for identically
worded questions over relatively long, though irregular, time periods. Additionally,
current approaches to the measurement of aggregate-level EU support fail in appropri-
ately dealing with key issues like ambivalence and the comparability over time and
across countries. Bayesian IRT, instead, can produce estimates of aggregate EU
support with a sounder theoretical grounding (Caughey and Warshaw, 2015; McGann,
2014), and that are more tailored to the established conceptualisations of EU support.
This article contributes to the literature about the measurement of aggregate-level EU
support by applying, for the first time, a Bayesian IRT model. I explain why this approach
is superior to other available alternative techniques as it starts from an explicit model of
individual behaviour, offers a theory-based approach to deal with neutral responses and to
appropriately capture more ambivalent attitudes towards the EU, and produces measures
of public preferences that are comparable both over time and between countries (McGann
et al., 2019). I show that these advantages enable Bayesian IRT to produce measures of
EU support that are both more precise and more in line with the conceptual properties of
172 European Union Politics 23(2)
To continue reading
Request your trial