ArticlePDF Available

Abstract and Figures

The Council of the European Union is the European Union’s most powerful legislative body. Yet, we still have limited information about Council politics because of the lack of suitable data. This paper validates a new approach to studying Council politics entitled DICEU – Debates in the Council of the European Union. This approach is the first to leverage the public videos of Council deliberations as a data source. We demonstrate the face, convergent, and predictive validity of DICEU data. Governments’ ideal points scaled from these videos yield meaningful and well-known conflict dimensions. Moreover, governments’ positions during Council negotiations correlate highly with expert assessments and predict subsequent votes on legislative acts. We conclude that DICEU data provide a promising new approach to studying Council politics and multilevel governance.
Content may be subject to copyright.
Public deliberations
in the Council of
the European Union:
Introducing and
validating DICEU
Christopher Wratil
Minda de Gunzburg Center for European Studies, Harvard
University, Cambridge, MA, USA; Cologne Center for
Comparative Politics, University of Cologne,
Cologne, Germany
Sara B Hobolt
Department of Government, London School of Economics
and Political Science, London, UK
The Council of the European Union is the European Union’s most powerful legislative
body. Yet, we still have limited information about Council politics because of the lack of
suitable data. This paper validates a new approach to studying Council politics entitled
DICEU – Debates in the Council of the European Union. This approach is the first to
leverage the public videos of Council deliberations as a data source. We demonstrate
the face, convergent, and predictive validity of DICEU data. Governments’ ideal points
scaled from these videos yield meaningful and well-known conflict dimensions.
Moreover, governments’ positions during Council negotiations correlate highly with
expert assessments and predict subsequent votes on legislative acts. We conclude
that DICEU data provide a promising new approach to studying Council politics and
multilevel governance.
Corresponding author:
Christopher Wratil, Minda de Gunzburg Center for European Studies, Harvard University, Adolphus Busch
Hall, 27 Kirkland Street, Cambridge, MA 02138, USA.
European Union Politics
0(0) 1–21
!The Author(s) 2019
Article reuse guidelines:
DOI: 10.1177/1465116519839152
Council, dataset, governments, quantitative text analysis, videos
The Council of the European Union (EU) is the EU’s primary legislative body,
where national ministers discuss, negotiate, and vote on legislative proposals. Most
research on politics in the Council is based on either expert interview data or
voting records. These data sources have generated key insights into intergovern-
mental politics in the Council, exploring what factors drive governments’ position-
taking and voting (e.g. Bailer et al., 2015; Hagemann et al., 2017; Hagemann and
Høyland, 2008; Thomson, 2011) and the nature of alignments and latent conflict
dimensions in Council politics (e.g. Mattila, 2009; Plechanovova, 2011; Thomson,
2009; Thomson et al., 2004; Zimmer et al., 2005). They have also been used to test
different bargaining models and estimate the distribution of power among member
states and across EU institutions (e.g. Bailer, 2004; Costello and Thomson, 2013;
Golub, 2012; Rasmussen and Reh, 2013; Thomson, 2008, 2011). However, existing
data sources have key limitations when it comes to studying legislative politics in
the Council. Expert data may be influenced by interviewee bias and are hard to
replicate for other researchers (e.g. Bueno de Mesquita, 2004). Voting data, in
turn, tell us little about the preceding negotiation process and lack variation,
since 98% of all votes are cast in favor of the proposal (Hagemann et al., 2017).
In addition, none of the available datasets on Council politics captures the evolu-
tion of negotiations on legislative files over time.
In this article, we propose a novel approach to studying the Council, which uses
video recordings of public deliberations. Since the entry into force of the Lisbon
Treaty, the Council must deliberate in public when negotiating legislative files and
discussing strategic questions
and all debates are video-streamed online. We pre-
sent the first study that rigorously collects, codes, and validates these public delib-
erations as a data source to study the Council. We label our data gathering
approach DICEU – Debates in the Council of the European Union”. The approach
produces textual data of all speeches by national ministers and other representa-
tives in the Council as well as human codings of intergovernmental conflicts.
We validate the DICEU approach in three steps. First, we demonstrate its face
validity with a pilot dataset covering Council meetings in the Economic and
Financial Affairs Council (ECOFIN) configuration between 2010 and 2015.
Scaling governments’ ideal points with an item-response theory (IRT) model as
well as different quantitative text analysis (QTA) models yields meaningful latent
conflict dimensions that are in line with prior knowledge about politics in the
Council. High correlations between the ideal points scaled from human coding
and those from text highlight the potential of QTA models for future analyses of
Council politics. Moreover, we demonstrate DICEU’s potential for generating
further discoveries by tracing the dynamic development of conflict in the
2European Union Politics 0(0)
Council over time as well as measuring new quantities of interest (such as minis-
ters’ sentiment or issue attention). Second, we assess convergent validity with
expert data from the Council, using the “Decision-making in the European
Union” (DEU) dataset as a benchmark (Thomson et al., 2006, 2012). We find
that DEU and DICEU positions on the exact same issues correlate highly,
which suggests that the public deliberations allow us to capture governments’
negotiation positions. Finally, we investigate predictive validity of our position
measures on Council votes. We find that governments’ DICEU positions are a
significant predictor of their voting behavior in the Council, even when controlling
for factors that are known to influence voting.
In conclusion, we argue that the DICEU approach offers a rich data source on
Council deliberations that can be used to study many aspects of politics in the EU
and its member states. To facilitate further research, we make our entire DICEU
pilot dataset on ECOFIN (2010–2015), as well as our DICEU codebook and
related documentation, publicly available.
Data on Council politics
The vast majority of empirical research on politics in the Council relies either on
expert interviews (e.g. Kleine, 2013; Thomson et al., 2006, 2012; Wasserfallen
et al., 2019) or on official voting records (e.g. Hagemann, 2007; Hagemann and
Høyland, 2008; Mattila, 2004, 2009) as the main data source. In addition, some
work has analyzed official documents from the Council (e.g. H
age, 2014; Obholzer,
2014; Sullivan and Selck, 2007) and the policy statements that governments can
submit with their voting record (Hagemann et al., 2019). While these data have
produced important insights, they also have limitations.
The drawbacks of relying on expert data on the Council are well known: first,
expert interviews must be conducted in close temporal proximity to negotiations in
the Council to reduce the potential problem of post-diction bias (Bueno de
Mesquita, 2004). The more distant the memory of negotiations, the more likely
are experts to use cues, such as the negotiation outcome or their personal political
convictions, when reconstructing actor positions. This likely introduces bias.
Second, the social scientific standard of replicability is ill-defined in the case of
expert interviews. While analyses of voting data, in contrast, are easily replicable,
they only reflect the very last phase of Council politics, when legislation is passed,
and cannot tell us about the actual proceedings of negotiations. Moreover,
since a strong consensus norm in Council voting prevails (Heisenberg, 2005;
Novak, 2013), with less than 1.5% of votes cast in opposition to legislative pro-
posals in the 1999–2011 time period, variation in voting data is very limited
(Hagemann et al., 2017). These consensual votes thus give relatively limited infor-
mation about government positions during the negotiations. A more recent inno-
vation in the literature on Council politics is large-N data created on the basis of
official Council documents; for instance, draft minutes of Council meetings as well
as presidency notes on the state of negotiations (see H
age, 2014; Obholzer, 2014).
Wratil and Hobolt 3
One concern with this approach is that we know little about how the national
presidencies may differ in the degree to which they make Council documents public
as well as the amount of information they include in these documents (see also
Hillebrandt et al., 2014).
Finally, existing data sources also do not allow us to study the dynamics of the
negotiation process, as they are usually snapshots of one time point. Even the most
extensive expert project on the Council, the DEU data (Thomson et al., 2006,
2012), does not contain sufficient data to estimate how government positions on
key conflict dimensions (e.g. left–right, pro-anti regulation, pro-anti integration)
develop over time. As a result, we still lack dynamic time-series-cross-section data
in Council research. This prevents us from investigating research questions on the
dynamics of Council negotiations, and it also makes the causal identification of
effects more difficult, since many causal inference methods rely on tempo-
ral ordering.
To address some of these limitations, we present a new complementary
approach to studying Council politics. We argue that videos of public deliberations
of the Council provide a promising, and hitherto largely unexplored, data source
for understanding the Council. While the Council has historically been a largely
opaque institution of EU-level policy-making detached from the public eye, it has
gradually become more transparent (see e.g. Hillebrandt et al., 2014). In June 2006,
the European Council decided to make a large proportion of its legislative delib-
erations in the Council open to the public (European Council, 2006). From late
2006, webcasts of the open deliberations were made available on the Council’s
website in 20 different languages. Since the adoption of the Lisbon Treaty in
December 2009, the Council configurations have to meet in public when they
deliberate or vote on draft legislative acts, or discuss presidency work programs
and other questions of strategic relevance. All these meetings must be broadcast on
the Council’s video streaming website (
While Council deliberations are now video-recorded, this increased transparen-
cy has not turned the Council into a classical legislature where representatives
address a public audience in plenary debates. Instead, ministers’ speeches during
Council deliberations are much more technical and forthright than most parlia-
mentary speeches (see the Online appendix for examples). Ministers discuss the
details of policy-making and they openly advise colleagues on how to deal with
domestic voters. This suggests the videos show actual negotiations, or at least
part thereof.
Videos of public deliberations overcome key limitations of existing data sources.
They are not affected by post-diction bias, such as interviews may be, and they can
be analyzed at any point in time (if archived). They also provide the most direct
measurement of actor positions, free of any information filters applied by experts
or officials. Compared to Council documents, such as draft minutes of Council
meetings, they provide a full account of all public deliberations rather than short
summaries. Analyses using video data are also fully replicable, as researchers have
access to the exact same source material. Importantly, this is also a very rich data
4European Union Politics 0(0)
source, as the Council produces around 100 hours of video material per calendar
year, which allows researchers to observe the dynamics of policy-making in the
Council across a large number of policy issues.
The DICEU approach
To study the Council using videos of public deliberations, we have developed the
DICEU data gathering approach based on two components: textual data of min-
isters’/representatives’ speeches and human codings of these speeches. The data
extraction process firstly involves the transcription of the video footage of public
deliberations in English
(since meetings are not transcribed) and secondly, the
identification and coding of debates according to a comprehensive codebook,
which contains variables on the debate topic, characteristics of the speaker, the
language of the original/non-translated speech, and the general approval a speaker
expresses during a debate. Full details are in the Online appendix.
The most important information contained in DICEU data for substantive
purposes relates to the speaker’s general approval of the proposal(s) being nego-
tiated. We therefore introduce the general approval variable in some depth here. To
generate this variable, coders identify a major dimension of contestation within
each debate. For instance, this may be “Approval of the Commission’s proposal
on X”. Due to rather uniform practices used by Council presidencies to structure
and moderate debates, this dimension is almost without exception about
(dis)approval of either (a) a legislative proposal presented by the European
Commission; (b) the Council state of play on a proposal (e.g. a presidency com-
promise proposal, negotiations at lower bureaucratic levels); or (c) the state of
negotiations with the European Parliament (EP) (e.g. a presidency’s mandate for
negotiations with the EP). In fact, many presidencies close their entry statement by
asking whether delegations “approve” of a proposal or compromise proposal.
Coders use a five-point Likert item to assess the degree of approval of each
actor speaking in the debate:
1. The speaker expresses full approval.
2. The speaker expresses more approval than disapproval.
3. The speaker expresses a balance of approval and disapproval.
4. The speaker expresses more disapproval than approval.
5. The speaker expresses full disapproval.
6. Degree of approval cannot be assessed (applies to very short or irrelevant
The resulting dataset (excluding descriptive variables, such as the position of the
speaker) can be thought of as a DAmatrix, in which Ddenotes debates and A
denotes the actors/speakers. Each element of the matrix is a coding of the actor on
the general approval item. We assessed the inter-coder reliability of our human
codings using Krippendorff’s alpha. For the general approval variable that is
Wratil and Hobolt 5
measured on the same scale in each debate, alpha is 0.82, based on a random
sample of double-coded debates (see the Online appendix). In the following, we
assume that national delegations are unitary actors. Hence, we analytically pool all
speakers giving speeches for the same national delegation (e.g. the French perma-
nent representative and the French minister).
Dimension Atherefore has 28 col-
umns for the national delegations plus one column for the European Commission
that usually participates in debates (e.g. introduces a new legislative proposal) and
one column for “other” actors (e.g. in ECOFIN usually the European Central
Bank (ECB) that contributes to some debates). The resulting text corpus contain-
ing the speech transcriptions is likewise a DAmatrix, in which each element is all
text produced by an actor in one debate, i.e. we simply pool all speech interven-
tions if an actor speaks more than once during a debate.
To validate the DICEU approach, we have applied it to all public deliberations
held in the ECOFIN Council configuration between 2010 and 2015. From mid-
2011 onwards, all videos of public deliberations are available on the Council’s
video streaming website. In addition, we were able to obtain the videos of several
meetings held in 2010 and the first half of 2011 from the Council Secretariat. We
focus on the ECOFIN configuration in this time frame for two reasons: first,
economic and financial affairs is a policy area in which we have much prior knowl-
edge of governments’ preferences and available criteria to cross-validate our data
with. For instance, we expect to see differences in the positions of net contributors
and net beneficiaries of the EU budget (e.g. Bailer et al., 2015). Likewise, we
anticipate a divide between creditors and debtors during the Eurozone crisis.
Second, during this period, ECOFIN dealt with a high number of very salient
and important legislative files and was arguably the main locus of EU secondary
law-making activity. Prominent legislative packages include the banking union, the
“two-pack” and “six-pack”, and several issues on taxation (such as the financial
transaction tax) as well as the always salient annual budget negotiations. In addi-
tion to our ECOFIN data, we also collected further data on a selection of files
negotiated between 2006 and 2008 for further validation purposes with expert
data, detailed below.
In total, our DICEU dataset from ECOFIN comprises of 1195 debate partic-
ipations by national delegations, the Commission and other actors (e.g. the ECB)
from 89 debates.
Debates lasted 42 minutes on average, but vary considerably in
length with a standard deviation of almost 37 minutes. The median number of
words spoken by a representative during a debate is 242 and the mean is 377.
Across all debates, the British delegation spoke more than any other national
delegation (about 29,000 words), but less than a third of the 91,000 words
spoken by Commission representatives. The Online appendix contains further
details on the data, including various descriptive statistics as well as a detailed
analysis of the missing data patterns in DICEU data, which shows that the under-
lying missing data processes are likely “ignorable”. In the subsequent sections, we
evaluate the validity of the key measures included in DICEU data.
6European Union Politics 0(0)
Face validity: Ideal points from human coding and text
We first assess face validity, i.e. the extent to which DICEU data appears to cap-
ture legislative politics in the Council accurately, by scaling governments’ ideal
from our ECOFIN debates between 2010 and 2015. To measure govern-
ments’ ideal points, we use our human codings of approval as well as our debate
transcriptions. We assume that the conflict space in the ECOFIN Council is uni-
dimensional, but this assumption could be relaxed in future work. To scale govern-
ments’ ideal points based on the general approval variable, we use a Bayesian
mixed factor analysis model that is equivalent to a standard two-parameter IRT
model with a probit link, when the items are ordinal as in our case (Quinn, 2004).
The model assumes that the observed responses xij , where idenotes governments/
actors and jdenotes items, are determined by a matrix of latent variables x
ij as well
as cut-points c(Quinn, 2004: 339). In our case, this simplifies to
xij ¼c(1)
ij 2cjðc1Þ;cjc
½ (2)
where cindexes the categories in our general approval items and takes values from
1 (full approval) to 5 (full disapproval). Equation (2) imposes a strict ordinal
relationship between the latent responses and the categories of our items. The
factor analytic model for the ideal points of the governments derives as
where X
iis the vector of government i’s latent responses, Kis a matrix of factor
loadings of the items on the estimated factors, hiis the vector of governments’
factor scores, and eiis the error term. Setting the first element of hi1to 1 turns the
first element of Kj1into the item’s difficulty parameter, Kj2into the item’s discrim-
ination parameter, and hi2into the government’s factor score on the first dimen-
sion, i.e. her ideal point (Armstrong et al. 2014: 296). The model is fitted via a
Markov Chain Monte Carlo (MCMC) algorithm (further details are in the
Online appendix).
We also analyze the transcriptions of the debates without any form of human
coding using two closely related unsupervised text scaling models as well as sen-
timent analysis. First, we employ the standard Wordfish Poisson text scaling model
(Slapin and Proksch, 2008), which has already been used for applications in EU
politics (e.g. Franchino and Mariotto, 2012). We simply pool all interventions by a
government across all debates in one document. Second, we use the Wordshoal
model recently proposed by Lauderdale and Herzog (2016), which is a two-stage
procedure relying on Wordfish estimates for each debate that are subsequently
used as data in a Bayesian factor analysis. The Wordfish estimates of a government
Wratil and Hobolt 7
in the debates it participated in, wij, where jdenotes the debate, are used as data in
the second stage and assumed to be a linear function of the government’s ideal
point, hi(see Lauderdale and Herzog, 2016: 378)
where ajis a debate fixed effect and bja debate loading (or debate marginal effect)
that indicates how strongly a debate is associated with the common latent dimen-
sion. The Wordshoal model is fitted using an MCMC algorithm for the second
stage. Third, we also use the Lexicoder sentiment dictionary (Young and Soroka,
2012) to calculate a standard measure of “tone” for each government. Tone simply
is the difference between positive (þnegated negative) and negative (þnegated
positive) words divided by all words.
In total, we estimate four different types of ideal points for each government: IRT
approval is an estimate from human coding of a government’s approval of the state
of negotiations (general approval variable), Wordfish is the Poisson text scaling
estimate, Wordshoal is the two-stage estimate with Wordfish for each debate and
a Bayesian factor analysis on top, and tone is the Lexicoder sentiment. Table 1
displays correlations between the four types of ideal points for our 28 governments.
The correlations are strikingly high between 0.73 and 0.95. It is particularly
noteworthy that we find a high correlation of 0.81 between Wordfish and IRT
approval. This suggests that models based on Poisson text scaling primarily cap-
ture to what extent governments approve of the state of play of negotiations in the
Council. Hence, if researchers are only interested in a general conflict dimension
about the approval of EU legislation, they may reliably use Wordfish
(or Wordshoal) in future projects.
Next, we investigate more closely the actor alignments on the main conflict
dimension in public ECOFIN deliberations by plotting governments’ estimated
ideal points from the IRT approval model and the Wordfish model in Figure 1
(a) and (b). Figure 1(c) shows a strong correlation between the two types of ideal
points. The figure also reveals a clear north–west versus south–east cleavage with
the United Kingdom, Sweden, the Netherlands, or Denmark among the least
approving of ECOFIN politics and countries such as Greece, Cyprus, Estonia,
or Croatia among the most approving. Figure 1(d) confirms the findings from the
approval IRT by demonstrating that the member states in the north–west are also
Table 1. Correlations between four types of ideal points.
IRT approval Wordfish Wordshoal Tone
IRT approval 1 0.81 0.77 –0.78
Wordfish 0.81 1 0.95 –0.81
Wordshoal 0.77 0.95 1 –0.73
Tone –0.78 –0.81 –0.73 1
8European Union Politics 0(0)
Figure 1. Comparison of ideal points from different scaling methods: (a) Ideal points from IRT
approval, (b) Ideal points from Wordfish, (c) IRT approval and Wordfish, and (d) IRTapproval and
tone. Note: 95% credible/confidence intervals as vertical lines in panels (a) and (b); linear regres-
sion line in panels (c) and (d). AT: Austria; BE: Belgium; BG: Bulgaria; HR: Croatia; CY: Cyprus;
CZ: The Czech Republic; DK: Denmark; EE: Estonia; FI: Finland; FR: France; DE: Germany; EL:
Greece; HU: Hungary; IE: Ireland; IT: Italy; LV: Latvia; LT: Lithuania; LU: Luxembourg; MT: Malta;
NL: The Netherlands; PL: Poland; PT: Portugal; RO: Romania; SI: Slovenia; SK: Slovakia; ES: Spain;
SE: Sweden; UK: The United Kingdom. Since Croatia joined the EU in 2013, the measurement of
its position is less reliable; the country is also placed less extreme in the Wordshoal model
(not shown).
Wratil and Hobolt 9
speaking in a more negative tone (in line with their disapproval of Council politics)
than those from the south–east. These actor alignments replicate prominent find-
ings in the literature on north–south, north–south–east, and east–west divides
between governments in the EU (Mattila, 2009; Plechanovova, 2011; Thomson,
2011; Veen, 2011).
They also provide new insights into Council politics, as they suggest that in the
context of ECOFIN negotiations, Western and Northern member states were the
least approving of and most negative about legislative politics. In the Online
appendix, we analyze the substantive meaning of the north–west versus south–
east conflict dimension present in our sample of ECOFIN debates. This reveals
that, broadly speaking, the rich member states in the north–west preferred smaller
EU budgets and emphasized economic competitiveness, while the poorer states in
the south–east preferred larger budgets and emphasized economic regulation.
An additional benefit of the DICEU approach is that it enables us to investigate
the dynamics of legislative politics in the Council over time. As an example, we can
track the relevance of the latent conflict dimension as scaled from the Wordshoal
model over time. This allows us to asses when the actor alignments in public
deliberations in ECOFIN were particularly aligned with the main latent conflict
dimension. Or less technically, when was the north–west versus south–east divide
most clearly visible? To answer this question, Figure 2 plots the absolute
Wordshoal debate loadings, bj, over time with a simple locally estimated scatter-
plot smoothing (LOESS) regression line. This illustrates that the conflict between
member states in the north–west versus south–east became more prevalent in
ECOFIN debates between mid-2012 and mid-2013 and remained important.
the Online appendix, we also plot the debate-specific ideal points of all national
governments from our Wordshoal results, wij bj, to further illustrate DICEU
data’s potential to investigate dynamics.
In sum, our analyses demonstrate a high level of face validity of DICEU data.
Moreover, DICEU data provide us with new research opportunities, for instance,
to investigate the Council deliberations in a temporal dimension or measure actors’
attention to different political issues, which we illustrate in the Online appendix
using dictionary approaches as an example.
Convergent validity: Comparison with expert data
Next, we assess convergent validity, i.e. the convergence of information retrieved
from Council deliberations with existing measures expected to capture similar
constructs. Given the limited data available on government positions in the
Council, DEU data represents the closest we get to a “gold standard” for measure-
ments of actors’ positions in the Council to be used for cross-validation. Following
the DEU project’s assumption that experts report the most important issues relat-
ed to a proposal as well as actors’ initial negotiation positions on these issues (that
is, immediately after the introduction of a proposal), we expect that DICEU
10 European Union Politics 0(0)
positions should converge to information from DEU under the following
1. Actors discuss the most important issues regarding a legislative proposal in
public deliberations in the Council.
2. Actors reveal their negotiation positions in public deliberations.
3. Actors do not change their negotiation positions between the introduction of a
proposal and its discussion in public deliberations.
We can check if (1) holds true by ascertaining how many issues from the DEU data
are discussed in public deliberations. Moreover, assumption (3) should hold for
Council deliberations that coincide with the introduction of a proposal or shortly
thereafter. In case assumption (2) holds, we would thus expect to find high con-
vergence between DEU and DICEU positions if the Council deliberations are held
in close proximity to the introduction of the proposal. To ascertain this conver-
gence, we identified a total of 21 legislative proposals covered in the DEU data that
were placed as “B items” for first discussion on agendas of Council meetings after
spring 2006, and could therefore, in principle, have been discussed in public. We
were able to gather a dataset of re-digitized video footage from the Council’s
Figure 2. Absolute values of debate loadings from Wordshoal over time.
Note: Solid line is LOESS regression line with span ¼0.75; dots are bj.
Wratil and Hobolt 11
archives covering public deliberations on 17 of these 21 proposals that were dis-
cussed in different Council configurations (not only ECOFIN).
This dataset provides us with a unique opportunity to cross-validate video-
based with expert-based data. According to DEU experts, actors were divided
on a total of 50 issues regarding these 17 proposals. Our coders identified a
clear majority of 31 of these 50 DEU issues as discussion points in the videos of
the public deliberations, and 12 of the DEU issues were discussed in more than one
Council meeting. This demonstrates that there is some relationship between the
main legislative issues identified by experts and the issues debated in public sessions
of the Council. Furthermore, for the subset of issues covered by DEU and in the
videos our coders placed actors on the DEU issue scales, based on their DICEU
speeches. This allows us to directly compare information retrieved from experts
and from videos on the same scale (i.e. the DEU issue scales).
Coders identified a total of 209 overlapping positions on 26 DEU issues, that is,
in 209 cases, actors could be placed on the predefined DEU scales from 0 to 100
based on video footage and DEU experts had reported their positions in inter-
views. The correlation of the DEU expert positions and the positions from the
videos coded on the DEU scale is 0.60 with an average absolute distance between
the positions of 23 points on the DEU scale. Hence, the positional information
contained in DEU and DICEU is similar but clearly not identical. Yet differences
between DICEU and DEU could stem from violations of assumption (3) as well as
measurement error in both datasets instead of violations of assumption (2). Hence,
we investigate whether convergence increases when public deliberations are held in
close temporal proximity to the introduction of a proposal, and for particularly
salient issues, where it should be easier for DEU experts and DICEU coders to
accurately identify actors’ positions.
Table 2 shows the results of a linear regression of the absolute distance between
the DEU and the DICEU placements on our 209 overlapping positions on a var-
iable indicating the time lag between the introduction of the proposal and the
public deliberation (within six months, within 6–12 months, after 12 months) as
well as the salience of an issue to the actor as reported by DEU experts. Two-way
cluster-robust standard errors, accounting for the nesting of positions within
actors and issues, are shown in parentheses. As expected, the results demonstrate
that DEU and DICEU placements are much closer when DICEU codings are from
public deliberations held within the first year of the introduction of the legislative
proposal, and for salient issues.
In fact, if an issue was discussed within six months of the Commission’s legis-
lative proposal and the issue was highly salient to the actor (actor salience ¼100),
our model predicts DEU and DICEU measures to diverge by only 9.8 DEU scale
points on average. Similarly, the correlation of positions reported by experts and
those coded from videos rises to 0.78 for public deliberations within the first year
after the Commission’s proposal and with actor salience >50. Hence, initial nego-
tiation positions retrieved from expert interviews and positions coded from videos
of public deliberations strongly converge when the issue is salient and the time lag
12 European Union Politics 0(0)
between the introduction of the proposal and the public deliberation is short. The
DICEU approach can therefore provide similar information to the DEU approach
about initial negotiation positions.
Predictive validity: Forecasting opposition votes
In addition to face and convergent validity, we also assess predictive validity, i.e.
the question of whether variation in DICEU measures can forecast events; specif-
ically, governments’ voting behavior in the Council. This allows us to assess wheth-
er governments practice what they preach: are governments’ statements in public
deliberations an indication of their subsequent voting behavior on these acts in
the Council?
For this purpose, we collected the voting sheets for a total of 95 legislative acts
that were discussed in our 89 debates.
Out of these, six acts had not been decided
on at the time of writing this article, five acts the Commission had withdrawn
before voting, one act was non-legislative and not voted on, and for three acts
we were unable to obtain voting results from the Council’s register. Of the remain-
ing 80 acts we observed 2194 votes by governments with 53 “No” and 30
“Abstain” votes. In line with common practice in the field, we pool no votes
and abstentions as “opposition votes” (e.g. Bailer et al., 2015; Hagemann et al.,
2017). Note that with 3.8% opposition votes, voting on ECOFIN files in this
period is clearly more contentious than voting has been historically across
Council configurations (averaging around 1–2% opposition votes). To assess the
predictive validity of DICEU data for opposition votes, we use the DICEU general
approval variable introduced above, which was coded at least from one debate for
1745 of our 2194 votes. Recall that this variable indicates the tendency to approve
Table 2. Predicting the distance between DEU
and DICEU positions.
Within 6–12 months 5.03
After 12 months 13.41
Salience –0.16
Constant 26.13
Note: Linear regression; cluster-robust standard errors
by actors and issues in parentheses; *p <0.05.
Wratil and Hobolt 13
of the state of negotiations. In Figure 3, we show the association between oppo-
sition/no votes and the general approval variable.
This descriptive overview reveals a strong association between governments’
approval of an act in the public deliberations and their later voting behavior on
this act. Goodman and Kruskal’s gamma is 0.72 for the association between oppo-
sition votes and approval and 0.78 for the association between no votes and
approval. If governments expressed more approval than disapproval or even full
approval, 1% or less of the votes were opposition or no votes. However, if govern-
ments on balance expressed more disapproval or even full disapproval, the fraction
of subsequent no votes rose to 9% and that of opposition votes to 18%. This
illustrates the basic predictive validity of DICEU measures for governments’
voting behavior.
But do DICEU data simply capture factors that we know from the literature
predict opposition votes, or do they contain novel information that contributes to
predicting opposition votes while controlling for other factors? To answer this
question, we run logistic regression models with the binary indicator of opposition
votes as dependent variable, the DICEU disapproval measure (running from 1 “full
approval” to 5 “full disapproval”) as main independent variable, and controls
from a standard Council voting model as in Hagemann et al. (2017). Following
Hagemann et al. (2017), we include a measure of Public opinion on European
integration from the Eurobarometer series, specifically the “EU image” question,
to detect signaling to domestic audiences (see also Wratil, 2018b). To capture the
Figure 3. DICEU general approval and opposition/no votes.
14 European Union Politics 0(0)
“party politics” thesis, we use measures of Government parties’ positions on EU
integration and left–right at the time of the last election from the Comparative
Manifesto Project’s (CMP) coding of election manifestos (Volkens et al., 2018;
Wratil, 2018a). Further, we account for economic interest explanations with meas-
ures of the Receipts from the EU budget (in % of national gross domestic product)
as well as the national Unemployment and Inflation rates (Bailer et al., 2015). Last,
we use a dummy variable for votes on Budget acts (e.g. draft amending annual
budgets), since opposition votes on budget acts are much more frequent than on
other legislative files (74 out of the 83 opposition votes occurred on acts concerning
the EU budget). We impute missing values using chained equations with predictive
mean matching (see the Online appendix for full details on variables
and imputation).
In Table 3, we report the results from a baseline model (Model 1) and a model
adding country fixed effects (Model 2), which reduces our sample to just 18 coun-
tries that casted at least one opposition vote. The results demonstrate that the
DICEU approval variable is a strong predictor of opposition votes even when
controlling for other factors from the literature. The more disapproval
Table 3. Logit models predicting opposition votes in the Council.
Model 1 Model 2
DICEU disapproval 0.54 0.38
(0.13)* (0.18)*
Public opinion –2.74 –4.84
(0.66)* (1.63)*
Government position EU integration –0.17 0.10
(0.08)* (0.19)
Government position left–right 0.02 –0.01
(0.01) (0.01)
Receipts from EU budget –0.25 0.15
(0.15) (0.63)
Unemployment rate –0.13 –0.42
(0.05)* (0.17)*
Inflation rate 0.10 0.25
(0.10) (0.13)
Budget acts 0.88 1.18
(0.40)* (0.44)*
Constant 4.03 12.05
(2.19) (6.15)
Number of acts 80 80
Number of countries 28 18
Country fixed effects No Yes
N2194 1436
Note: Logistic regressions; standard errors in parentheses; *p <0.05.
Wratil and Hobolt 15
governments show in public deliberations, the more likely they are to cast an
opposition vote. Hence, DICEU positions do not only capture the effect of
other variables but provide us with new information to predict Council voting.
In addition, we find support for the “signaling” explanation of Council votes, since
the coefficient on public opinion is a statistically significant predictor, indicating
that governments facing more Eurosceptic publics cast more opposition votes. In
contrast, we find little support for the “party politics” explanation, except for one
significant coefficient on the government parties’ position on EU integration.
In sum, our results demonstrate that DICEU positions have strong predictive
validity with regard to Council voting and a significant part of the information
they provide is novel and cannot be substituted by common other predictors of
Council voting from the literature.
In this paper, we have introduced and validated a new data source and approach
for studying politics in the Council of the European Union. The DICEU approach
extracts and codes textual data from videos of public deliberations in the Council,
a data source not used in any publication in political science so far. We have
addressed the face, convergent, and predictive validity of data gathered with
this approach.
Studying the Council from videos of its public deliberations has obvious advan-
tages. Most importantly, it is the most direct way of measuring government posi-
tions during actual negotiations, since it does not rely on the recollection or
interpretation by third parties (e.g. experts, presidencies). This approach can
also be applied by researchers across the world at any time and the analyses are
fully replicable.
Video data also bridge the gap between the well-established
expert data projects (such as DEU) that measure initial negotiation positions
before bargaining takes place and voting data that reveal government positions
after negotiations are concluded. Video data essentially open up the black box of
Council negotiations, although – as with any other forum for political negotiation
– we can never be certain what is happening informally behind closed doors.
Our pilot dataset on public deliberations in ECOFIN can be easily extended in
the future. The finding that QTA estimates (e.g. from Wordfish and Wordshoal)
correlate highly with estimates from models based on human coding suggests that
collecting transcriptions of debates without extensive and costly human coding
may be sufficient for many research projects. Proksch et al. (2019) have recently
validated the use of automatic speech recognition systems to transcribe video and
audio material for QTA. Based on these findings, future projects can analyze the
abundance of available video data from further Council configurations and other
time frames by simply feeding it into an automatic speech recognition system
(e.g. Google’s Speech application programming interface or YouTube) and ana-
lyzing retrieved transcriptions through QTA models. Our results indicate that
Poisson text scaling models will retrieve a conflict dimension that largely reflects
16 European Union Politics 0(0)
the approval of EU legislation by governments. The interpretation of quantities of
interest from any other text models should be validated by researchers (e.g. on the
basis of hand-coded sub-samples).
DICEU data will enable researchers to address new research questions about
Council politics and multilevel governance. Crucially, many questions about
dynamics can now be addressed, as changes in actors’ positions can be analyzed.
Moreover, in the ideal point framework, changes in positions over time can be
identified by disentangling debate- or period-specific effects from genuine move-
ments on the latent dimension (see Martin and Quinn, 2002). Dynamic ideal point
models are either available already (e.g. Martin and Quinn, 2002; Schnakenberg
and Fariss, 2013) or can be easily developed from their existing static versions
(Lauderdale and Herzog, 2016: 392). We can thus use DICEU data to address
questions such as: when and why are governments changing positions in the
Council and what are the effects on the bargaining process and outcomes?
Beyond the dynamic perspective, the linguistic characteristics of text from video
data provide new opportunities for innovative projects. These are not limited to
the application of standard scaling models like Wordfish and Wordshoal, but
could also include dictionary-based approaches to sentiment, complexity, or tech-
nicality of speech, as well as topic models to measure issue attention or arguments
(e.g. Grimmer, 2010; Roberts et al., 2014). Moreover, the growing methodological
literature on the analysis of audio-visual features (e.g. Dietrich et al., 2018; Knox
and Lucas, 2018; Schonhardt-Bailey, 2017), such as facial expressions or vocal
pitch, suggests that DICEU data could be used in the future to systematically
measure these novel aspects of political communication in the Council.
As any data source, videos of public deliberations have their limitations. One
weakness of these data compared to expert data is that it will always be finite. Not
all national delegations speak in all debates and on all topics. Hence, if researchers
are interested in coding government positions on very specific issues, they may
encounter missing data. The underlying missing data process is likely ignorable
(see the Online appendix), and for many analyses that are prone to missing data,
the use of multiple imputation techniques will suffice to obtain unbiased estimates.
However, where analysis models are very sensitive to missing data, such as tests of
competing bargaining models (e.g. Slapin, 2014; Thomson, 2011), this may present
a limitation of the DICEU approach.
Our choice of data sources and collection strategies should be guided by the
questions we ask. We believe that many key questions in European politics can be
addressed by studying videos of public deliberations in the Council, and by making
the DICEU dataset publicly available, we hope that researchers will be able to
answer some of these.
The authors would like to thank Isabelle Brusselmans and her team in the Council
Secretariat for support in obtaining video footage from the Council, including archival
material that was re-digitized for this project. The authors are also grateful to Rebecca J
Wratil and Hobolt 17
Brooks, Sarah Ciaglia, Carolyn Konopka, Jason Krstic, Lukas Lehner, Ida Popovski, and
Tim Rogers for their excellent research assistance and to Felix Bethke, Sven-Oliver Proksch,
and Gerald Schneider for helpful comments on earlier versions of this article. The entire
DICEU dataset used in this article is included in the replication files and more information
on the data is available online at
The authors would like to acknowledge the generous funding by The Leverhulme Trust
(RF-2013-245) and the Fritz Thyssen Foundation (
Christopher Wratil
1. Note that the Council can still deliberate in closed session over non-legislative issues
such as implementation of legislation or preparations of international events
(e.g. G7, G20).
2. As with any other legislature, we do not know what happens behind the scenes and to
what extent public deliberations are representative of the broader process of negotiations
(e.g. in COREPER, “Comite
´des repre
´sentants permanents”, and the working groups).
3. When the speaker does not speak in English, we use the simultaneous translation into
English provided by the Council’s interpreters. Note that representatives can freely
choose the language they use, since all speech is simultaneously translated into all offi-
cial EU languages. Eighty-three percent of representatives choose to address the
Council in English. Pooling speeches by English native speakers, non-native speakers
using English as well as speakers translated into English best reflects the actual conduct
of negotiations. De Vries et al. (2018) demonstrate that texts translated into other
languages yield almost identical estimates in quantitative text models as the original
texts. In the Online appendix, we also show that language choice made no difference to
the human codings we obtained.
4. Note that according to our data about 70% of all speeches are given by national
ministers. However, this information is primarily based on participant lists included
in press releases and we found these to be inaccurate in many cases, without an obvi-
ous bias.
5. Note that we do not code any discussions with less than two national delegations
(excluding the presidency) participating. Seventeen of such discussions occurred in
our period of investigation. Hence, our data covers around 84% (89 out of 106) of
all discussions in the Council, and the remaining 16% are by definition very short
discussions with only a few participating actors.
6. In the ideal point models described below, we also estimate the positions of the
European Commission and “other” speakers (usually the ECB). But we do not report
their positions. It turns out that due to specificities of the human coding instructions,
these actors have very different ideal points when scaled from human coding as opposed
to speeches. Re-estimating all models without these actors yields substantively the same
high correlations between governments’ ideal points.
18 European Union Politics 0(0)
7. Obviously, this increased relevance can be due to changing actor alignments, or alter-
natively, to a changing political agenda, e.g. new legislative issues that more strongly
mobilize the north–west versus south–east divide. We therefore hand-coded debates and
separated them into five different issue areas. This reveals that the debate loadings
showed upward trends over time in four out of these five areas, suggesting that the
increased relevance of the latent conflict dimension is not due to certain issues reaching
the political agenda.
8. Note that a single debate can deal with no (e.g. on the presidency work program), one,
or several specific legislative acts (e.g. debate on banking union).
9. If more than one debate was held on an act and approval was coded for more than one
debate, we take its coded value from the chronologically first debate on an act. In the
Online appendix, we report results using the last debate on an act.
10. Video material is rarely used to systematically measure government positions in inter-
national negotiations, but one exception is McKibben’s (2016) study of multilateral
climate change negotiations using archived webcasts.
Armstrong DA, Bakker R, Carroll R, et al. (2014) Analyzing Spatial Models of Choice and
Judgment in R. Boca Raton: CRC Press.
Bailer S (2004) Bargaining success in the European Union: The impact of exogenous and
endogenous power resources. European Union Politics 5(1): 99–123.
Bailer S, Mattila M and Schneider G (2015) Money makes the EU go round: The objective
foundations of conflict in the Council of Ministers. Journal of Common Market Studies
53(3): 437–456.
Bueno de Mesquita B (2004) Decision-making models, rigor and new puzzles. European
Union Politics 5(1): 125–138.
Costello R and Thomson R (2013) The distribution of power among EU institutions:
Who wins under codecision and why? Journal of European Public Policy
20(7): 1025–1039.
de Vries E, Schoonvelde M and Schumacher G (2018) No longer lost in translation.
Evidence that Google translate works for comparative bag-of-words text applications.
Political Analysis 26(4): 417–430.
Dietrich BJ, Enos RD and Sen M (2018) Emotional arousal predicts voting on the U.S.
Supreme Court. Political Analysis 1–7.
European Council (2006) An overall policy on transparency. Available at: www.consilium. (accessed 13
March 2019).
Franchino F and Mariotto C (2012) Explaining negotiations in the conciliation committee.
European Union Politics 14(3): 345–365.
Golub J (2012) How the European Union does not work: National bargaining success in the
Council of Ministers. Journal of European Public Policy 19(9): 1294–1315.
Grimmer J (2010) A Bayesian hierarchical topic model for political texts: Measuring
expressed agendas in Senate press releases. Political Analysis 18(1): 1–35.
age FM (2014) Scaling policy preferences from negotiation positions. In: 7th Pan-
European Conference on EU politics, 5–7 June 2014, The Hague.
Hagemann S (2007) Applying ideal point estimation methods to the Council of Ministers.
European Union Politics 8(2): 279–296.
Wratil and Hobolt 19
Hagemann S, Bailer S and Herzog A (2019) Signals to their parliaments? Governments’ use
of votes and policy statements in the EU Council. Journal of Common Market Studies.
Epub ahead of print 8 March 2019. DOI:
Hagemann S, Hobolt SB and Wratil C (2017) Government responsiveness in the European
Union: Evidence from Council voting. Comparative Political Studies 50(6): 850–876.
Hagemann S and Høyland B (2008) Parties in the Council? Journal of European Public
Policy 15(8): 1205–1221.
Heisenberg D (2005) The institution of ‘consensus’ in the European Union: Formal versus
informal decision-making in the Council. European Journal of Political Research
44(1): 65–90.
Hillebrandt MZ, Curtin D and Meijer A (2014) Transparency in the Council of Ministers of
the EU: An institutional approach. European Law Journal 20(1): 1–20.
Kleine M (2013) Informal Governance in the European Union. How Governments Make
International Organizations Work. Ithaca: Cornell University Press.
Knox D and Lucas C (2018) A dynamic model of speech for the social sciences.
Working Paper.
Lauderdale BE and Herzog A (2016) Measuring political positions from legislative speech.
Political Analysis 24(3): 374–394.
Martin AD and Quinn KM (2002) Dynamic ideal point estimation via Markov Chain
Monte Carlo for the US Supreme Court, 1953–1999. Political Analysis 10(2): 134–153.
Mattila M (2004) Constested decisions: Empirical analysis of voting in the European
Council of Ministers. European Journal of Political Research 43(1): 29–50.
Mattila M (2009) Roll call analysis of voting in the European Union Council of Ministers
after the 2004 enlargement. European Journal of Political Research 48(6): 840–857.
McKibben HE (2016) To link or not to link? Agenda change in international bargaining.
British Journal of Political Science 46(2): 371–393.
Novak S (2013) The silence of ministers: Consensus and blame avoidance in the Council of
the European Union. Journal of Common Market Studies 51(6): 1091–1107.
Obholzer L (2014) Bicameral Coalition Formation: Dynamics of Legislative Cooperation in
the European Union. London: London School of Economis and Political Science.
Plechanovova B (2011) The EU Council enlarged: North-south-east or core-periphery?
European Union Politics 12(1): 87–106.
Proksch S-O, Wratil C and W
ackerle J (2019) Testing the validity of automatic speech
recognition for political text analysis. Political Analysis. Epub ahead of print 19
February 2019. DOI:
Quinn KM (2004) Bayesian factor analysis for mixed ordinal and continuous responses.
Political Analysis 12(4): 338–353.
Rasmussen A and Reh C (2013) The consequences of concluding codecision early: Trilogues
and intra-institutional bargaining success. Journal of European Public Policy
20(7): 1006–1024.
Roberts ME, Stewart BM, Tingley D, et al. (2014) Structural topic models for open-ended
survey responses. American Journal of Political Science 58(4): 1064–1082.
Schnakenberg KE and Fariss CJ (2013) Dynamic patterns of human rights practices.
Political Science Research and Methods 2(1): 1–31.
20 European Union Politics 0(0)
Schonhardt-Bailey C (2017) Nonverbal contention and contempt in U.K. parliamentary
oversight hearings on fiscal and monetary policy. Politics and the Life Sciences
36(1): 27–46.
Slapin JB (2014) Measurement, model testing, and legislative influence in the European
Union. European Union Politics 15(1): 24–42.
Slapin JB and Proksch S-O (2008) A scaling model for estimating time-series party positions
from texts. American Journal of Political Science 52(3): 705–722.
Sullivan J and Selck TJ (2007) Political preferences, revealed positions and strategic votes:
Explaining decision-making in the EU Council. Journal of European Public Policy
14(7): 1150–1161.
Thomson R (2008) The Council presidency in the European Union: Responsibility with
power. Journal of Common Market Studies 46(3): 593–617.
Thomson R (2009) Actor alignments in the European Union before and after enlargement.
European Journal of Political Research 48(6): 756–781.
Thomson R (2011) Resolving Controversy in the European Union. Legislative Decision-
Making before and after Enlargement. Cambridge: Cambridge University Press.
Thomson R, Arregui J, Leuffen D, et al. (2012) A new dataset on decision-making in the
European Union before and after the 2004 and 2007 enlargements (DEUII). Journal of
European Public Policy 19(4): 604–622.
Thomson R, Boerefijn J and Stokman FN (2004) Actor alignments in European Union
decision making. European Journal of Political Research 43(2): 237–261.
Thomson R, Stokman FN, Achen CH, et al. (eds) (2006) The European Union Decides.
Cambridge: Cambridge University Press.
Veen T (2011) The dimensionality and nature of conflict in European Union politics: On the
characteristics of intergovernmental decision-making. European Union Politics
12(1): 65–86.
Volkens A, Lehmann P, Matthieß T, et al. (2018) The Manifesto Data Collection. Manifesto
Project (MRG/CMP/MARPOR). Version 2018a. Berlin.
Wasserfallen F, Leuffen D, Kudrna Z, et al. (2019) Analysing European Union decision-
making during the Eurozone crisis with new data. European Union Politics 20(1): 3–23.
Wratil C (2018a) CMP Government Positions Data. Version 1.1. Cambridge.
Wratil C (2018b) Modes of government responsiveness in the European Union: Evidence
from Council negotiation positions. European Union Politics 19(1): 52–74.
Young L and Soroka S (2012) Affective news: The automated coding of sentiment in polit-
ical texts. Political Communication 29(2): 205–231.
Zimmer C, Schneider G and Dobbins M (2005) The contested Council: Conflict dimensions
of an intergovernmental EU institution. Political Studies 53(2): 403–422.
Wratil and Hobolt 21
... To examine these questions, we adopt a novel empirical approach. Specifically, we analyse negotiation positions of governments in the Council by coding the official video recordings of public deliberations in the Council, the so-called Debates in the Council of the European Union (DICEU) database (see Wratil and Hobolt 2019). This database allows us to systematically measure conflict in the Council during the different stages of legislative negotiations, and show that intergovernmental contestation is highest at the early stages of negotiations. ...
... Moreover, we investigate a different stage in the legislative process by looking at the entire process of official intergovernmental negotiations at the ministerial level instead of pre-or post-negotiation stages. Specifically, we make use of an original dataset of the public deliberations of legislative packages in the ECOFIN Council between 2010 and 2015, which covers virtually all official deliberations between national ministers and their deputies (see also Wratil and Hobolt 2019). ...
... In order to measure government positions and contestation in the Council, we rely on the DICEU dataset (Wratil and Hobolt 2019). DICEU is based on the analysis of video footage of public deliberations of the Council and is the first dataset that systematically makes available data on these deliberations. ...
Full-text available
Decision-making in the Council of the European Union appears highly consensual at the voting stage. However, we focus on Council deliberations, where we find higher levels of contestation. What drives government opposition in the Council? Using a novel approach of studying the Council through video footage of its public deliberations (DICEU – Debates in the Council of the European Union), we demonstrate that contestation between governments is, at least in part, driven by their responsiveness to domestic public opinion. Analysing deliberations on legislative packages in the Economic and Financial Affairs Council between 2010 and 2015, we show that governments are responsive to public opinion when setting out their policy positions, but primarily when the policy issues are salient domestically. Our study thus contributes to our understanding of government responsiveness in the EU.
... On the one hand, decision-making in the Council has long been characterised by a depoliticised consensus culture (Hayes-Renshaw and Wallace, 2006: 24, 56;Heisenberg, 2005: 67-79;Thomson, 2009: 760). On the other hand, the crises have enhanced politicisation in the member states (de Wilde, 2011;de Wilde and Lord, 2016;Zu¨rn, 2016), which places pressure on governments (Bressanelli et al., 2020) to adjust their policy position (Hagemann et al., 2017(Hagemann et al., , 2019Hobolt and Wratil, 2020;Wratil, 2018;Wratil and Hobolt, 2019). ...
... We suggest that one crucial facet of politicisation is the degree of Euroscepticism among parties and the public in a given country. As voting in the Council reflects governments' responsiveness to their domestic constituencies (Hagemann et al., 2017(Hagemann et al., , 2019Hobolt and Wratil, 2020;Roos, 2019;Wratil, 2018;Wratil and Hobolt, 2019), we argue that Euroscepticism, either directly expressed by public opinion or channelled through political parties, is also key in understanding governments' voting in the Council. Given that the mass-elite linkage on European integration is not always clear -either party positions form public opinion or parties change their positions to satisfy public opinion (Steenbergen et al., 2007) -we propose three hypotheses on how Eurosceptic party positions and opinions affect governments' voting in the Council. ...
Full-text available
This article presents a new and previously unchartered dataset on roll call votes for all 28 member states in the Council of the EU between 2010 and 2019 and studies the effects of politicisation on governments' oppositional voting in the different policy areas. We contribute to the literature with two main findings. First, our study provides strong evidence for bottom-up politicisation, where Euroscepticism and the left-right positions of national political parties strongly affect governments' voting in the Council. Second, we provide new evidence for a form of differentiated politicisation where ideological standpoints of political parties in government and opposition have different effects on oppositional voting in the various policy areas.
... Our empirical analysis is based on an automated analysis of over 105,000 plenary debates in Germany (1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017) and Austria (2003-18). Following Wratil and Hobolt (2019), we use three types of validity test for our measurement approach: first, we demonstrate the face validity of the measure for coalition mood; secondly, we show that coalition mood correlates with public opinion on the government parties (concurrent validity); and, thirdly, we illustrate the predictive validity of the measure, testing whether bad coalition mood can forecast delays in the legislative process. In the concluding section, we discuss limitations of our measurement approach and show potential for future research on the coalition mood. ...
The success and longevity of coalition governments depends on the ability to keep conflicts between coalition members at bay. The risk of such conflicts is often assessed by drawing on proxy measures, such as the ideological heterogeneity among government parties. This article presents a new approach to measuring the atmosphere between government parties. The ‘coalition mood’ is a time-varying measure that draws on applause patterns between coalition partners during legislative debates. The article exemplifies the measurement approach based on automated analyses of over 105,000 plenary debates in Germany and Austria. The article then assesses the measure's face, concurrent and predictive validity. It finds the measure well aligned with qualitative evidence, shows that the coalition mood is correlated with poll ratings of the government parties and helps to predict the duration of legislative processes. The conclusion highlights future applications of the coalition mood for research on coalition politics and public policy.
... The Council Presidency and sources of influence Formal procedural rules shape much of the work of the President. The debates in the Council are mainly directed towards the approval of either a Commission proposal, a Presidency compromise (or any other 'state of play' proposal), or a Presidency's mandate to negotiate with the European Parliament (Wratil and Hobolt, 2019). Nevertheless, the discretion the Presidency possesses varies depending on the decision-making procedure. ...
jats:p> Although agenda-setting dynamics in the European Union are a well-studied phenomenon, there is a gap in the literature between the current focus on issue attention dynamics (a policy input) and the impact attention dynamics have on policy outputs. This study examines how the rotating Council Presidency’s stated policy goals for their term in office affect EU decision-making efficiency. We show that the salience the rotating chair of the Council attaches to a given policy area affects legislative efficiency in that policy area. We also demonstrate how this effect is conditioned on government effectiveness in the state of the Presidency in question, and is independent of Commission policy priorities. This suggests that the Presidency can drive EU policy outputs and push for its priorities when it holds the chair, but that its ability to do so is contingent on the domestic organisation and effectiveness of the state. </jats:p
... Specifically, it addresses two key limitations of the traditional Wordfish approach, namely that word discrimination parameters are assumed to be the same over debates or time and that topic variation is not considered. Goet (2019) and Wratil and Hobolt (2019) have used the Wordshoal model to analyse political debates in the UK or Europe but to the best of our knowledge has never been used in political economy contexts and economics. ...
Full-text available
Using corpora of business cycle report sections dealing with monetary and fiscal policy issues from 1999 to 2017 and using methods of unsupervised text scaling (Slapin and Proksch, 2008; Lauderdale and Herzog, 2016), namely Wordfish and Wordshoal we scale the institutions' theo-retical/ideological position over debates. The results are in line with the findings from descriptive textual analysis. For monetary policy, we observe a strong but short-lived consensus in debate-specific positions at the height of the financial crisis in 2008 and a larger polarization after 2008 compared to the sample period before. For the fiscal policy textual corpus, the polarization was similarly high before and after the crisis. For both policy areas, the institutions DIW Berlin and IfW Kiel define the outer bounds of the observed spectrum of latent ideological positions.
We scale theoretical/ideological positions of economic research institutes over debates. Using only parts of German research institutes’ business cycle reports that deal with economic policy commentary and advice as an example, we extract sections from these reports dealing with monetary and fiscal policy issues from 1999 to 2020. To these corpora we apply methods of unsupervised text scaling (Slapin and Proksch, 2008; Lauderdale and Herzog, 2016), namely Wordfish and Wordshoal. Roughly, results are in line with the common sense in the public policy discourse. For monetary policy texts, we observe a strong, but short-lived consensus in debate-specific positions at the height of the financial crisis in 2008 and a larger polarization thereafter compared to the sample period before. For the fiscal policy textual corpus, the polarization was similarly high before and after the crisis and decreases somewhat during the COVID-19 pandemic. For both policy areas, the German Institute of Economic Research (DIW), Berlin, and the Institute for World Economics (IfW), Kiel, tend to be the most diverse institutes within the spectrum of latent ideological positions. We argue that text-mining techniques might be useful to scale underlying ideological positions in policy-related publications.
National governments are accused of being evasive and opportunistic in their presentation of European integration, thereby exacerbating the EU's crisis of legitimacy. Yet empirical evidence on how governments present Europe at home is limited to a small handful of qualitative studies. This thesis provides the first comparative, quantitative study of how governments - and the parties that form them - present Europe in their domestic public spheres, and what these presentational strategies mean for representation and legitimacy in the EU. Inspired by Fenno's 1978 classic, I call this their `home style'. Through innovative text as data methods combining machine translation, automated text analysis, and hand coding, I show that rather than adopting a nationalist home style marked by evasiveness and opportunism, governments have responded to EU politicization by adopting a home style I label technocratic-patriotic. Technocratic, in the sense that gov- ernments actually talk frequently about the EU, but avoid clear position taking on the issue by defusing it with complex language. Patriotic, in the sense that governments extensively claim credit for defending the national interest on the European stage, but in fact rarely blame or criticise the EU directly. I argue that despite not fitting the stereotypical image of evasive, opportunistic blame shifters, this technocratic-patriotic home style still poses deep problems for democratic accountability in Europe's multilevel system of governance. The thesis also contributes two resources to the academic community: EUCOSpeech, an original dataset of over 6,000 statements by national leaders in the aftermath of EU summits, and EUParlspeech, an original dataset of over 1 million references to European integration made in parliamentary speeches between 1989 and 2019.
Full-text available
This paper introduces the updated version of our dataset, which is the third iteration of the Decision-making in the European Union (DEU-III) database. First, it explains what the DEU project is and then embeds it into the broader literature. Next it describes the pursued case-selection criteria, the definitions and the operationalisations of the main constructs. The paper discusses the integration of various components of the DEU dataset and its links with other variables that are commonly used in the contributions to this special issue and other studies. Finally, it introduces a validity and reliability test of the DEU-III database as well as avenues for its future use by scholars.
What role does the rotating Council Presidency maintain a decade after Lisbon? This article argues that, regardless of institutional changes, the rotating Presidency still shapes the Council agenda to a large extent. Based on an original hand-coded dataset of rotating Presidency programmes between 1997 and 2017, I show that some policies are ‘stickier’ on the Council agenda, while the others exhibit significant changes in salience over time. Since the magnitude of these shifts varies from Presidency to Presidency, the analysis focuses on domestic political factors and the country positioning vis-à-vis the European Union to determine their relationship with agenda volatility. By means of a panel model, the examination demonstrates that the government issue salience can best explain the levels of issue salience in the Presidency programmes.
The European Union (EU) is an unlikely case for responsive policy-making. Yet, in recent years scholars have found that the EU’s overall decision-making output is correlated with the average preferences of the European citizens toward European integration. Despite recognizing the value of this systemic approach, we argue in this contribution that studies of EU responsiveness should explicitly acknowledge the multi-dimensionality of responsiveness in the EU by addressing the multiplicity of actors, institutions and publics involved. This actor-oriented perspective directs the focus of responsiveness research to the input stage of EU policy-making. This contribution calls for research that a) theoretically situates the responsiveness of actors in specific institutional venues in a broader perspective of multi-dimensional EU responsiveness and that b) empirically links different forms of input responsiveness to one another and to policy outputs.
Full-text available
The collection of articles in this special issue provides a comprehensive analysis of European Union (EU) decision-making during the Eurozone crisis. We investigate national preference formation and interstate bargaining related to major reforms of the Economic and Monetary Union (EMU). The analyses rely on the new 'EMU Positions' dataset. This dataset includes information about the preferences and saliences of all 28 EU member states and key EU institutions, regarding 47 contested issues negotiated between 2010 and 2015. In this introductory article, we first articulate the motivation behind this special issue and outline its collective contribution. We then briefly summarize each article within this collection; the articles analyse agenda setting, preference formation, coalition building, bargaining dynamics, and bargaining success. Finally, we present and discuss the 'EMU Positions' dataset.
Full-text available
Automated text analysis allows researchers to analyze large quantities of text. Yet, comparative researchers are presented with a big challenge: across countries people speak different languages. To address this issue, some analysts have suggested using Google Translate to convert all texts into English before starting the analysis (Lucas et al. 2015). But in doing so, do we get lost in translation? This paper evaluates the usefulness of machine translation for bag-of-words models—such as topic models. We use the europarl dataset and compare term-document matrices (TDMs) as well as topic model results from gold standard translated text and machine-translated text. We evaluate results at both the document and the corpus level. We first find TDMs for both text corpora to be highly similar, with minor differences across languages. What is more, we find considerable overlap in the set of features generated from human-translated and machine-translated texts. With regard to LDA topic models, we find topical prevalence and topical content to be highly similar with again only small differences across languages. We conclude that Google Translate is a useful tool for comparative researchers when using bag-of-words text models.
Does parliamentary oversight of governments’ decisions in the international arena matter? This article finds that it does: governments with strong parliamentary oversight behave differently when negotiating policies at the EU level compared with governments with less powerful parliaments. Where parliaments have formal powers to oversee and restrict their government's positions we see a significantly higher use of opposing votes and formal policy statements by those governments. This behaviour intensifies depending on the governments' standing vis‐à‐vis other political parties at home. When governments are under pressure in their national parliaments they are more likely to go on record and take a stand against the majority in Brussels. These results make it clear that in EU legislative politics, governments not only consider their policy priorities and negotiation tactics with their European counterparts, but also make use of EU decision records to send signals to domestic audiences, including their national parliaments.
The analysis of political texts from parliamentary speeches, party manifestos, social media, or press releases forms the basis of major and growing fields in political science, not least since advances in “text-as-data” methods have rendered the analysis of large text corpora straightforward. However, a lot of sources of political speech are not regularly transcribed, and their on-demand transcription by humans is prohibitively expensive for research purposes. This class includes political speech in certain legislatures, during political party conferences as well as television interviews and talk shows. We showcase how scholars can use automatic speech recognition systems to analyze such speech with quantitative text analysis models of the “bag-of-words” variety. To probe results for robustness to transcription error, we present an original “word error rate simulation” (WERSIM) procedure implemented in $R$ . We demonstrate the potential of automatic speech recognition to address open questions in political science with two substantive applications and discuss its limitations and practical challenges.
Do judges telegraph their preferences during oral arguments? Using the U.S. Supreme Court as our example, we demonstrate that Justices implicitly reveal their leanings during oral arguments, even before arguments and deliberations have concluded. Specifically, we extract the emotional content of over 3,000 hours of audio recordings spanning 30 years of oral arguments before the Court. We then use the level of emotional arousal, as measured by vocal pitch, in each of the Justices’ voices during these arguments to accurately predict many of their eventual votes on these cases. Our approach yields predictions that are statistically and practically significant and robust to including a range of controls; in turn, this suggests that subconscious vocal inflections carry information that legal, political, and textual information do not.
With recent advances in computing power and the widespread availability of preference, perception and choice data, such as public opinion surveys and legislative voting, the empirical estimation of spatial models using scaling and ideal point estimation methods has never been more accessible.The second edition of Analyzing Spatial Models of Choice and Judgment demonstrates how to estimate and interpret spatial models with a variety of methods using the open-source programming language R. Requiring only basic knowledge of R, the book enables social science researchers to apply the methods to their own data. Also suitable for experienced methodologists, it presents the latest methods for modeling the distances between points. The authors explain the basic theory behind empirical spatial models, then illustrate the estimation technique behind implementing each method, exploring the advantages and limitations while providing visualizations to understand the results. This second edition updates and expands the methods and software discussed in the first edition, including new coverage of methods for ordinal data and anchoring vignettes in surveys, as well as an entire chapter dedicated to Bayesian methods. The second edition is made easier to use by the inclusion of an R package, which provides all data and functions used in the book.
Are national governments responsive to citizens’ opinions when negotiating policies in the Council of the European Union? Conceiving of the Council’s policy-making space as encompassing left-right and pro-anti integration issues, I argue that governments apply different ‘modes of responsiveness’ on these issues. As left-right issues are more reliably and intensely salient in domestic elections than pro-anti integration issues, governments’ responsiveness to left-right public opinion should be more systematic than to pro-anti integration opinion. Statistical analyses of 3700 policy positions of governments in the Council demonstrate that governments highly structure their responsiveness on left-right issues according to electoral cycles and systems (‘systematic mode’). However, they only sporadically respond to public opinion on pro-anti integration issues, when parties and events trigger the public salience of integration (‘sporadic mode’).
In parliamentary committee oversight hearings on fiscal policy, monetary policy and financial stability, where verbal deliberation is the focus, nonverbal communication may be pivotal in the acceptance or rejection of arguments proffered by policymakers. Systematic qualitative coding of these hearings in the 2010-15 UK Parliament finds that: (1) facial expressions, particularly in the form of anger and contempt, are more prevalent in fiscal policy hearings, where backbench parliamentarians hold frontbench parliamentarians to account, than in monetary policy or financial stability hearings, where the witnesses being held to account are unelected policy experts; (2) comparing committees across chambers, hearings in the Lords’ committee yield more reassuring facial expressions relative to hearings in the Commons’ committee, suggesting a more relaxed and less adversarial context in the former; and (3) central bank witnesses appearing before both the Commons’ and Lords’ committee tend towards expressions of appeasement, suggesting a willingness to defer to Parliament
Existing approaches to measuring political disagreement from text data perform poorly except when applied to narrowly selected texts discussing the same issues and written in the same style. We demonstrate the first viable approach for estimating legislator-specific scores from the entire speech corpus of a legislature, while also producing extensive information about the evolution of speech polarization and politically loaded language. In the Irish Dail, we show that the dominant dimension of speech variation is government-opposition, with ministers more extreme on this dimension than backbenchers, and a second dimension distinguishing between the establishment and anti-establishment opposition parties. In the US Senate, we estimate a dimension that has moderate within-party correlations with scales based on roll-call votes and campaign donation patterns, however we observe greater overlap across parties in speech positions than roll call positions and partisan polarization in speeches varies more clearly in response to major political events.