Predicting the outcome of European Parliament elections is an exceptionally difficult task, not only because the vote covers multiple countries, but also because voters may vote differently to the way they would in national elections. Nicolò Fraccaroli and Nils Hernborg present a model that can compensate for deficiencies in European election polling, noting that current election projections could be underestimating support for Eurosceptic and Green parties. 

The European elections are approaching and a number of estimates of the future composition of the European Parliament are being released (even by the European Parliament itself). They all tell a similar story: votes for mainstream political groups such as the European People’s Party (EPP) and the Social Democrats (S&D) will drop substantially, whereas Eurosceptic forces are expected to gain new seats in the next parliament. Many of these forecasts are based on Poll of Polls, a website that regularly collects electoral polls from different pollsters in EU countries and makes the data freely available online. Several EU countries, however, lack polls asking respondents how they would vote in European elections. For this reason, a majority of the early forecasts utilise polls for national elections and assume that these are equivalent to polls for European elections.

In this article we highlight two problems related to this approach that make these predictions inaccurate, leading to potentially distorted scenarios of the next European Parliament. We propose a new methodology that aims at correcting some of these limitations. We will see that, while some inaccuracy is inherent also to our method, the outcome it displays is different from those of contemporary polls but closer to more recent polls for a number of cases. In particular, our approach predicts a weaker pro-EU majority vis-à-vis a more Eurosceptic, but also “greener”, parliament than traditional polls suggest.

The problems of using national polls for European elections

The main problem related to the polls that are currently being used is that they are based on national elections. This means that predictions for EU elections are based on the assumption that voters’ preferences for national elections can also be applied to European elections. However, we know from the literature that this is not the case. European elections are in fact considered second-order elections, meaning that citizens assign them less weight than domestic ones when voting. As a result, European elections not only feature lower turnouts, but also different voting patterns.

In particular, in European elections smaller and new parties tend to perform better, at the expenses of large and governing parties. Moreover, by using polling data on the latest European elections, Marco Giuliani has provided further evidence in support of this systematic error in pollsters’ predictions: the 2014 polls (including polls “tailored” for European elections and not only for national ones) tended to predict a lower electoral outcome for small and a larger one for big and governing parties than the actual result. In other words, polls predicted exactly the opposite of what the second-order theory suggested, and got it (partially) wrong.

Taking second order elections seriously: A new method

Our approach takes the insights of the second-order elections theory seriously. In simple terms, we weight current polls based on their predicting error in previous elections, at a comparative point in time to the current one. To do so, we have collected a large dataset of national polls in all EU countries from the autumn of 2014. Our error term is computed as the difference between the polls’ seat prediction per political group and the actual result gained in 2014. For example, if the national polls suggest an EP group would gain 105 seats, but then actually gained 100 seats, then the error would be +5%.

Having computed the average error by political group, we then assign it to each party’s polls according to its political group membership. Figure 1, which displays the errors by political group, is clearly in line with the second-order elections theory, as it displays overestimation errors for large governing parties such as many of those in the EPP, S&D, ECR and ALDE groups, and underestimation errors for small parties such as the Greens and Eurosceptic parties (EFDD and ENF). For example, EFDD was largely underestimated as its main parties, UKIP and the Five Star Movement were polled at 12% and 18% but then gained 26.6% and 21.2% of the votes respectively.

Figure 1: Error margin as a share of total seats per political group for the 2014 European elections

Source: Compiled by the authors.

We then plotted the results against the seats of the future European Parliament. In doing so, we took into account a number of details, such as country-specific electoral thresholds and the change in number of seats due to Brexit (see here for the new allocation). For simplicity, we assume one constituency per country and proportional elections in each Member State.

A more Eurosceptic and Green parliament?

Figure 2 plots the expected composition of the next European Parliament using polling data, whereas Figure 3 plots the same data adjusted for the errors we estimated. Similar to what most polls are predicting today (Figure 2), our estimates indicate that the EPP and S&D will no longer hold a majority in parliament (Figure 3). However, in our projections the support of ALDE will not be enough to reach 50% + 1 of seats, but will necessarily need the support of Macron’s En Marche (which has formed a partnership with ALDE, but may not act as one political group) or the Greens, which will hence play a more relevant role in the next European Parliament. In addition, adjusting for the errors would make the ENF group (the group of Salvini and Le Pen) the third largest political group in the parliament, rather than ALDE, in contrast to what polls are predicting. Still, even if Eurosceptic groups (ENF, EFDD and ECR) were to coalesce, they would still be unable to reach a majority.

Figure 2: Projection of EP Elections based on polls (July – October 2018)

Source: Compiled by the authors.

Figure 3: Projection of EP Elections after correcting for 2014 errors

Source: Compiled by the authors.

Figure 4 shows our estimate compared to the ones released by other institutes at different points in time. Interestingly, while our estimates (the blue diamond) were adjusting polls from Autumn 2018, they are closer to polls created in March 2019 (red and yellow squares) than to polls published in the same period (other colours) in a number of cases (S&D, En Marche, Greens, GUE, ECR, EFDD, Non-attached; EPP current estimates are being reviewed downward). Importantly, this might be due to the fact that recently national election polls started being replaced by EU elections polls, internalising therefore the second order elections factor.

Figure 4: Comparing estimates across different research institutes

Source: Compiled by the authors.

Overall, national polls are not bad predictors of EU elections, as recently shown by Abel Bojar on this blog. Nevertheless, as we have shown, they might commit some structural errors when it comes to EU elections. While our methodology is too simple to be ascribed as a forecasting technique, it sheds new light on the relationship between voters’ behaviour in EU elections and polls, serving as a helpful filter to read poll predictions in view of the next elections.

Please read our comments policy before commenting.

Note: This article gives the views of the authors, not the position of EUROPP – European Politics and Policy or the London School of Economics. Featured image, Ska Keller, credit: © European Union 2014 – European Parliament (Attribution-NonCommercial-NoDerivs Creative Commons license)


About the authors

Nicolò Fraccaroli – University of Rome, Tor Vergata
Nicolò Fraccaroli is a PhD candidate in Economics at the University of Rome Tor Vergata. He worked for the DG International & European Relations of the European Central Bank (2017-2019) and graduated from the LSE’s MSc in Political Economy of Europe (2015). He is co-author with Robert Skidelsky of the book Austerity vs Stimulus. The Political Future of Economic Recovery (Palgrave Macmillan, 2017).

Nils Hernborg – Sciences Po
Nils Hernborg is a Master’s student in International Economic Policy at Sciences Po, Paris. He worked for the DG International & European Relations of the European Central Bank (2018-2019) and prior to that he worked as data analyst in the fintech industry. He graduated from King’s College London with a BSc Political Economy (2015).

Print Friendly, PDF & Email