Politics and Governance (ISSN: 2183–2463)
2019, Volume 7, Issue 2, Pages 306–333
DOI: 10.17645/pag.v7i2.1917
Article
Economic Voting in Direct Democracy: A Case Study of the 2016 Italian
Constitutional Referendum
Arndt Leininger
OttoSuhr InstituteofPoliticalScience, FreieUniversitätBerlin,14195 Berlin, Germany;E-Mail:arndt.leininger@fu-berlin.de
Submitted: 15 December 2018 | Accepted: 21 March 2019 | Published: 27 June 2019
Abstract
Referendums provide citizens with more control over policy. At the same time, they often entail choices over highly complex
policies and are politicised along partisan lines, suggesting that partisan rather than policy considerations will guide
voters’ choices. I look to the 2016 Italian constitutional referendum, which was particularly complex and polarised, as
an opportunity to test for mechanisms of government accountability in a referendum. Using a national survey of voters,
I show that the more negative a respondent’s evaluation of the state of the economy, the lower their likelihood to vote ‘yes’
on the government’s reform proposal. This relationship is remarkably strong: an average respondent with a very positive
evaluation of the state of the economy has an 88% probability of supporting the government’s reform proposal compared
to only 12% for a respondent with a very negative evaluation. The fact that economic evaluations are a strong determinant
of vote choice provides evidence for the existence of an economic vote in a referendum. This further suggests that voters
may treat referendums as a sort of second-order election.
Keywords
direct democracy; economic voting; Italy; referendums; second-order election; voting
Issue
This article is part of the issue “The Politics, Promise and Peril of Direct Democracy”, edited by Todd Donovan (Western
Washington University, USA).
© 2019 by the author; licensee Cogitatio (Lisbon, Portugal). This article is licensed under a Creative Commons Attribution
4.0 International License (CC BY).
1. Introduction
Referendums often entail choices over complex policies
and become politicised along partisan lines which suggests
that often partisan rather than policy considerations
guide voters’ choices. The 2016 Italian national referendum
on constitutional reform provides a unique opportunity
to test this hypothesis. Using a national survey
of voters, I show that the more negative a respondent’s
evaluation of the state of the economy, the lower
their likelihood to vote ‘yes’ on the government’s reform
proposal. These findings provide evidence for the
existence of an economic vote in a direct-democratic
setting. I suggest that voters may treat referendums
as a sort of second-order election, with referendums
acting as an exercise in government accountability inbetween
elections.
Referendums provide citizens with more control over
policy. While proponents of referendums believe that
they can increase the legitimacy of policies, opponents
doubt that voters possess the willingness and capacity to
inform themselves and make sensible judgements about
sometimes complicated and far-reaching decisions. Referendums,
especially those concerning substantive issues,
effectively allow voters to punish the government
in-between elections. One of the most prominent and
widely studied determinants of voting for or against the
government is the state of the economy. The so-called
economic vote implies that voters punish a government
for a bad economy and reward it for a good economy
(Lewis-Beck & Stegmaier, 2007).
I embrace the 2016 Italian constitutional referendum,
which was particularly complex and polarised, as an opportunity
to test for mechanisms of government accountability
in a referendum. It is distinguished as a subject for
investigation by two advantages: first, it is one of only a
few national level referendums to date which have been
subject to survey, and second, it shares many common
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 306
characteristics of other national referendums. The government
deliberately triggered the referendum and parties
took clear stances on the highly complex referendum
issue. The polarisation of the campaign between government
and opposition along with Renzi’s vow to resign if
the reform failed provided a clear opportunity for economic
voting. Specifically, I ask in this article whether
citizens’ perceptions of the economy influence how they
vote in a referendum.
My analysis is based on data from a survey conducted
by the Italian National Election Studies Association
(ITANES) project both before and after the 2016
referendum which link respondents’ evaluations of the
national (as well as their own) economic situation to
their vote choice. Results of various specifications and
robustness checks consistently show that economic evaluations
correlate positively with vote choice: a respondent
with a positive evaluation of the economy is on average
more likely to vote ‘yes’ than somebody who has a
less favourable view of the economy. This relationship is
surprisingly stable and strong. While an average respondent
with a very positive evaluation of the economy had
an 88% probability of supporting the government, it is
only 12% for respondents who evaluated the economy
very negatively.
The empirical results provide strong evidence for an
economic vote in what is an admittedly most-likely case,
because the campaign was strongly polarised, and the
prime minister tied his political future to the outcome
of the vote. Most-likely cases are strong tests of a theory
in the sense that a failure to find the hypothesised
relationship would cast strong doubts on the theory in
general (George & Bennett, 2004). While this article provides
strong evidence for an economic vote in the 2016
Italian constitutional referendum, it remains to be seen
whether it extends to other cases as well. For that matter,
future research should also incorporate least-likely cases,
referendums where parties do not play such a dominant
role or are internally divided.
2. Referendums as Second-Order Elections
In a referendum, there are no parties or candidates to
elect but only an issue to decide. Therefore, Schoen
(2012), for instance, cautions that while determinants of
participation in elections and referendums may be similar,
determinants of vote choices may differ drastically
between elections and referendums. Referendums are
thought to be on average more volatile than elections because
in some referendums “voters must choose among
alternatives that are sometimes unfamiliar and [are] perhaps
lacking in reliable voting cues”, while at other times
referendums can be “highly partisan contests, even without
the appearance of party or candidate names on the
ballot” (LeDuc, 2002, p. 711).
There are many reasons that referendums can be
considered more cognitively demanding than elections.
They often involve complex and sometimes technical
matters and confront citizens with issues that may be
new and unfamiliar to them. Additionally, because less
is at stake in a referendum, citizens may exert less effort
to inform themselves about the issue at hand. Not surprisingly
then, much of the research on voting behaviour
in referendums has focused on the issue of voter competence.
One of the main findings in this line of research
is that low-information voters can mimic informed voters’
choices by relying on cues from governments, parties
and interest groups (Christin, Hug, & Sciarini, 2002;
Clarke, Elliott, & Stewart, 2017; Lupia, 1994).
Work in this vein assumes that voters who possess
factual information about politics in general or a particular
referendum, take sensible decisions, or less normatively
charged decisions, in their own interest. However,
it is not clear whether possessing information about
a ballot proposition translates into an informed vote.
Knowledgeable voters also tend to be more ambivalent
about the referendum issues they have to decide upon
(Lanz & Nai, 2014; Nai, 2014)—more information does
not necessarily make it easier to decide. These papers
and the literature they stand for are all based on the assumption
that voters do vote on the issues on the ballot.
Either they try to weigh the arguments themselves, or
they rely on cues to mimic informed decision-making. In
the following section, I will provide an alternative viewpoint,
which suggests that voters treat referendums, particularly
those that are polarised along partisan lines as in
the Italian case, quite similar to second-order elections.
Second-order election, a term popularised by Reif
and Schmitt (1980) to describe the elections to the European
Parliament, denotes any election that can be considered
subordinate to the electoral contest that serves
to form the national government—in parliamentary systems,
the national parliamentary election would be the
first-order election, and in presidential systems it would
be the presidential election that is first-order. The second
order elections model postulates that issues and cleavages
of the national first-order contest will largely determine
the vote in the second-order contest as well. Specifically,
governing parties will lose votes relative to the preceding
first-order election.
European elections, despite an increase in the powers
of the EU, and in particular the European Parliament,
continue to be characterised as second-order to national
elections (Schmitt, 2005; Schmitt & Teperoglou, 2015).
Coming to the Italian case then, it is reasonable to assume
that in spite of Italy’s relatively extensive experience
with national referendums, that these are secondorder
to (first-order) national elections. The outcome of
the 2016 referendum mirrored the characteristics of a
typical second-order election in the sense that turnout
was lowerthan in the preceding national election and the
governing parties lost the vote.
The point of this article is not to show that electoral
politics dominate referendum policy but rather that
they are sufficiently important to have an impact on the
vote—similar to what one would see in a second-order
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 307
electoral contest. The specific argument I make in this article
is that we should also be able to see an economic
vote in a referendum. Just as Okolikj and Quinlan (2016)
who have shown that an economic vote can also be observed
in European parliament elections. The next section
discusses the concept of economic voting and how
it might apply to referendums more generally and the
2016 Italian constitutional referendum in particular.
3. Economic Voting in Referendums
The so-called economic vote is one of the most important
paradigms in electoral research. While the economic
vote has developed into a literature of considerable
breadth and variety (see, for instance, Duch & Stevenson,
2008; Kayser, 2014), so far it has only rarely been considered
in referendum research. To the best of my knowledge,
only two contributions to date have estimated an
economic vote in referendums; however, with entirely
different interpretations and contradictory expectations.
Bowler and Donovan (1998) see the economic vote
rooted in risk aversion; they suggest that voters may ask
themselves: is this a good time for the state to adopt a
new policy? Their analysis is based on survey data on
three 1990 California and a 1992 Colorado state referendum
and shows a significant correlation between economic
evaluations and vote choice, but only for less educated
voters and only for some ballot propositions. They
take these findings to mean that less-educated voters are
more likely to engage in economic voting to make up for
a lack of political sophistication. Uncertainty about the
consequences of a proposal may lead voters to stick with
the status quo rather than voting for a change, which may
be why proposals to change the status quo are often defeated
(LeDuc, 2003).
In contrast, Jenssen (1998, p. 195), seeking to understand
divergent outcomes of three EU membership
referendums in the Nordic countries Sweden, Finland
and Norway, hypothesizes that “economic hardship led
a majority in Sweden and Finland to vote for change—
in this case EU membership—whereas better-off Norwegian
voters preferred things as they were”. However,
in almost perfect contradiction of that hypothesis, economic
evaluations correlated positively with a vote for
EU membership in all countries surveyed.
Although these authors do estimate an economic
vote in a sense, their approach is divorced from the original
paradigm as applied in electoral research. Their interpretation
of the economic vote is seemingly inspired
by the literature on cue-taking: voters use external information
to make up their mind on an issue on the ballot.
However, voting in line with one’s preferred party recommendation
might just be voting for that party. If political
parties take visible and differing positions on a referendum,
it might take on the character of a second-order
election. Such a perspective takes seriously the possibility
that voters treat referendums as an opportunity to
vent dissatisfaction with the government.
Economic voting is one of the most important mechanisms
of government accountability. I argue in this article
that it can also be at work in a referendum. Economic
voting is in a sense also more broadly applicable
than cue-taking because those without a partisan attachment
cannot rely on cues. However, they are easily able
to engage in economic voting. In the Italian case analysed
here, about a quarter of voters have no partisan affiliation
but almost all of them voice an opinion on the state
of the economy.
The economy is the most important issue for a large
majority of voters in most, if not all, elections. Rather
than being positional, it is a valence issue. This means
voters agree that they all prefer a good economic situation,
stable economic growth and low unemployment
and inflation rates, to a bad one. The basic idea underlying
the paradigm is that voters hold the reigning government
responsible for the economic situation. If it is
good, they will reward the government by voting for the
governing party or parties, and if it is bad, they will punish
the government by choosing one of the opposition
parties (Lewis-Beck & Stegmaier, 2000, p. 138).
The economic vote is largely retrospective (LewisBeck
& Stegmaier, 2013; Wlezien, 2015), the process being
as follows: first, voters evaluate the economic situation;
they then decide whether to attribute the responsibility
to the government to the economic situation and,
finally, they punish or reward it with their vote. In this
way, voters in elections in large part evaluate government
performance in retrospect. This perspective does
not apply in the same way in a referendum. Here, the
choice at hand seems to be more prospective in nature:
what are the anticipated consequences of the proposed
policy? Are these consequences deemed desirable? This
might explain why the economic vote has been reinterpreted
in the two contributions discussed above not as
a retrospective evaluation of government performance,
but as a prospective calculation of the merits of changing
the status quo.
Nevertheless, one can still apply the concept of the
economic vote to referendums without engaging in conceptual
stretching. Voters can easily identify both a government’s
position in a referendum as well as the state
of the economy. Parties often take sides in referendum
campaigns, and the sides they take often mirror
a government-opposition divide. Sometimes, even the
head of government’s fate rests implicitly or explicitly, as
in the Italian case, on the outcome of the referendum.
Therefore, voters may, as they do in elections, choose
to reject the government’s position in economically bad
times and, vice versa, approve it in good times. Alternatively,
we might think that if a citizen is dissatisfied with
the government’s handling of the economy that they do
not put trust in the quality of a constitutional reform formulated
and propagated by that same government.
Adapting the economic voting paradigm to direct
democracy, I hypothesise that the more positive a voters
evaluation of the economic situation, the more likely
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 308
they are to vote for the government’s position. Formulated
differently with the Italian case in mind: the more
positive a respondent’s evaluation of the economic situation
of the country, the more likely it is that they vote
‘yes’ (in favour of the constitutional reform) in the referendum.
Having laid out the expectations for the empirical
analysis, in the following section, I briefly present
the reform proposal put to a vote, the political situation
leading up to the vote and the campaign accompanying
it, before going on to describe the research design.
4. The Italian Constitutional Referendum 2016
On 4 December 2016 eligible Italian citizens were asked
to vote on a constitutional law which would amend Italy’s
Constitution to change the composition and powers of
the parliament as well as the division of powers between
the federal state, the regions and other administrative
units. Besides the rather pragmatic reason of data availability,
the referendum is a particularly useful case for
study for three reasons.1
First, the subject of the referendum was a constitutional
reform. Such referendums are a common occurrence
at the national level. Secondly, the referendum
was triggered by the government, as are most national
referendums. Finally, the referendum campaign became
polarised along partisan lines, pitting the governing parties
against essentially all opposition parties.
Unlike other West European states, Italy does not
have a differentiated system of two chambers. The Italian
Constitution stipulates that both houses of parliament
are equal in rights. This institutional arrangement has
long been regarded as a problem because policymaking
is hampered by laws circulating endlessly between the
two chambers with neither chamber having the ability to
override the other. This also implies that the government
is dependent on the trust of both chambers. Hence, after
the parliamentary elections in Italy in 2013, which led to
differing majorities in the two parliamentary chambers—
the social-democratic Partito Democratico (PD) won an
absolute majority in the House, but no party won a majority
in the Senate—a political crisis ensued. President
Giorgio Napolitano, urged by almost all party leaders to
stay on for another term to moderate the crisis (Pasquino
& Valbruzzi, 2017, p. 148), appointed Enrico Letta of the
PD to lead a grand coalition between the PD and the
Popolo della Libertà (PdL), a merger of Silvio Berlusconi’s
Forza Italia and the radical right party Alleanza Nazionale.
He tasked that coalition with reforming bicameralism
and the electoral system.
Only a year after his appointment Letta was replaced
as prime minister by the PD’s then party secretary
Matteo Renzi after an internal party leadership
challenge. Renzi inherited Letta’s commitment to reform
and would later reference Napolitano’s call for reform to
justify the need for and the content of his reform proposal
(Pasquino & Valbruzzi, 2017, p. 148). Earlier, the
PD’s coalition partner PdL had split with some ministers
and MPs leaving the government and re-joining Forza
Italia, while others founded the Nuovo Centrodestra
(NCD) and continued in government. Just before becoming
prime minister, Renzi, then still party secretary of the
PD, reached an agreement with Berlusconi’s Forza Italia,
the so-called Nazarene Pact, that assured Forza Italia’s
support. However, Berlusconi withdrew his party from
the agreement in February 2015 when Renzi unilaterally
put forward Sergio Mattarella as a candidate for President
of the Republic who was subsequently elected. The
government, and in particular Prime Minister Matteo
Renzi, nevertheless pressed on with its plans for reform.
The reform package aimed at simplifying the institutional
architecture, speeding up the legislative process
and reducing the cost of government. It foresaw a reorganisation
of territorial structure, foremost the elimination
of the provinces and the National Council for Economics
and Labour (CNEL), an advisory body to the government.
It would abolish Italy’s symmetric bicameralism
by reducing the Senate’s power and size and making it an
indirectly elected body.1 Besides the institutional goal of
streamlining and centralising the political system, Renzi
also pursued a political goal. Having acquired the prime
ministership through an intra-party leadership challenge,
his “goal was to show to the opposition and, perhaps
even more so, to those within the PD who were challenging
his reforms and his policies, that he enjoyed the support
of a large majority of the Italian people” (Pasquino
& Valbruzzi, 2017, p. 153).
The constitutional reform, also named Riforma RenziBoschi,
after its initiators Prime Minister Matteo Renzi,
and the responsible minister for constitutional reforms
and relations with the parliament Maria Elena Boschi,
was passed by the Senate on 13 October 2015 and by
the lower house, the Chamber of Deputies, on 11 January
2016. Both houses confirmed their votes on 20 January
2016 and 12 April 2016 respectively. The two parliamentary
chambers approved the reform with an absolute majority
which, however, fell short of a two-thirds majority.
This created a situation where the reform failed in parliament,
but the parliamentary vote opened a path to a
referendum according to article 138 of the Italian Constitution.
The referendum was held on 4 December 2016,
and the following question was put to the eligible population
of Italy:
Do you approve the text of the Constitutional Law
concerning ‘Provisions for overcoming equal bicameralism,
reducing the number of Members of Parliament,
limiting the operating costs of the institutions,
the suppression of the CNEL and the revision of Title V
of Part II of the Constitution’ approved by Parliament
1 As outlined above, few referendums are subject to scientific survey and even fewer of these include questions on the state of the economy (for an
exception, see Curtis, Jupille, & Leblang, 2014). The ITANES survey used in this paper is unique because of the breadth of items related to the economy
included in the questionnaire.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 309
and published in the Official Gazette no. 88 of 15 April
2016? (Wikipedia, n.d.)2
Therefore, if voters wanted to approve the reform they
had to vote ‘yes’, and ‘no’ to reject it. Voter turnout
was 65.5%, which is relatively high for a constitutional
referendum—only 9.7 points less than the last national
election in 2013 and about 13 points more than the last
constitutional referendum in 2006. Voters rejected the
reform by a substantial margin. 59.1% voted against the
reform and only 40.9% in favour. In almost all regions,
except for Emilia-Romagna, Toscana and Trentino-Alto
Adige, a majority of voters voted against the referendum.
The former two regions are strongholds of the PD while
the latter is a stronghold of the CD, the junior member
of the governing coalition (Pasquino & Valbruzzi, 2017,
p. 154). This aggregate pattern provides a first indication
that this vote was not just about constitutional reform
but also about the current government of the time.
The referendum campaign was described as “unusually
bitter and protracted” (Pasquino & Valbruzzi, 2017,
p. 153). It pitted the governing parties against essentially
all opposition parties. In addition to the authors of the reforms,
Matteo Renzi’s centre-left PD, their coalition partner
the NCD supported the reforms as well as the leftist
Christian Democrats of the Centro Democratico (CD). So
did the liberal party Scelta Civica and the anti-corruption
parties Italia dei Valori and Radicali Italiani. The reforms
were most strongly opposed by the MoVimento 5 Stelle
(five-star movement) led by comedian Beppe Grillo. Further
opponents of the reform included regional and rightwing
parties such as the Lega Nord or Forza Italia, but
also some left-wing parties such as Sinistra Italiana or the
Partito della Rifondazione Comunista.
Many associations and interest groups partook in
the campaign. Early on the National Association of Italian
Partisans announced its opposition to the reform.
They were joined by the leftist trade union CGIL and a
wide range of committees for a ‘no’-vote that sprang up,
most importantly “Libertà e Giustizia”. On the other side,
the national association of employers Confindustria, the
farmers’ association Coldiretti and the Christian democratic
trade union CISL campaigned for a ‘yes’. Unusually,
the President of the Italian Republic, Giorgio Napolitano,
took a public stance in favour of the reform. Experts were
divided on the issue. As a testament to the complexity of
the issue, in the months leading up to the vote, a dozen
books were published in which scientists and other experts
on legal and institutional matters argued for or
against the reforms, while others, in turn, took on a more
neutral tone seeking to inform rather than influence the
general public (Ceccarini & Bordignon, 2017, p. 290).
In short, the referendum was preceded by a monthslong
intensive campaign, which was politicised along partisan
lines. The referendum turned into a vote on the
population’s satisfaction with the government, not least
because Renzi promised to resign if he were to lose the
referendum. Early in the campaign, Renzi tied his political
future to the outcome of the vote in the hope
that his (initial) popularity would carry the reform to success.
While his approval was already down to 40% (from
initially more than 70%) at the time he made the announcement
in April 2016, he still was the most popular
leader. Consequently, after losing the referendum, Renzi
announced his resignation immediately after the result
became known. At the request of the Italian President
Sergio Mattarella, Renzi postponed his resignation until
after the following year’s budget was passed on 7 December
2016. In a sense, Renzi lost not a parliamentary, but
a popular vote of confidence in his government.
5. Research Design
The 2016 Italian constitutional referendum was covered
by a scientific survey fielded by the ITANES project. The
data were collected in a two-wave panel survey (pre- and
post-referendum) of the Italian population. The survey
comprised 3050 respondents in total. My analysis will
focus on voters in the referendum only, therefore disregarding
non-voters as well as voters casting a blank ballot.
I use list-wise deletion in case of missing values on any of
the dependent or independent variables.
The dependent variable is a respondent’s vote choice,
measured as vote intention in the pre-referendum wave
and as recall question in the post-referendum wave. Both
measures are not without their problems. As far as the
item in the pre-referendum questionnaire is concerned:
vote intentions may change, and some respondents may
voice a vote intention although in the end they will not
participate. As far as the post-referendum questionnaire
is concerned: there is strong over-reporting of participation
and likely a bandwagoning effect. It is, however,
hard to gauge to what extent this creates bias. A comparison
of the aggregated survey data with the actual referendum
outcome indicates that vote intentions, while
still predicting ‘no’ to win, predict a much closer outcome
than actually occurred. However, a substantial
number of voters reported not having made up their
minds yet. While strongly overstating turnout, reported
vote choices are more accurate, overestimating the ‘no’vote
only slightly (see Table 1).
Since vote intentions are more inaccurate than reported
vote choices, I rely on reported vote choices from
the post-referendum wave. In alternative specifications,
vote intentions instead of choices are used as robustness
check reported in the Appendix I (see Table A2). Both variables
are coded as dummy variables which are 1 if a respondent
voted or intended to vote ‘yes’ (for the consti-
2 English translation from Wikipedia and accuracy confirmed by an Italian colleague. Italian original obtained from the Ministry of the Interior
(http://www.interno.gov.it/it/italiani-voto-referendum-costituzionale): “Approvate il testo della legge costituzionale concernente ‘disposizioni per il
superamento del bicameralismo paritario, la riduzione del numero dei parlamentari, il contenimento dei costi di funzionamento delle istituzioni, la
soppressione del CNEL e la revisione del Titolo V della parte II della Costituzione’, approvato dal Parlamento e pubblicato nella Gazzetta Ufficiale n. 88
del 15 aprile 2016?”
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 310
Table 1. Vote intentions obtained from the pre-referendum wave, the actual result of the vote and reported vote choices
obtained from the post-referendum wave.
Yes No Undecided Turnout
Pre-referendum wave 36.4% 39.7% 23.9%
Actual result 40.9% 59.1% 65.6%
Post-referendum wave 38.9% 61.1% 89%
tutional reform) and 0 if they voted or intended to vote
‘no’ (against the constitutional reform).
The key independent variable is a respondent’s retrospective
evaluation of the national economy. Respondents
were asked to complete the following sentence
‘do you think the economic situation in Italy in the last
year…’ using one of the following answers: ‘improved
a lot’, ‘improved somewhat’, ‘remained equal’, ‘little
worse’, ‘much worse’ or ‘don’t know?’ (see Appendix II
for details). This question was asked in both the preand
post-referendum waves. I use the answers from the
pre-referendum wave to avoid that evaluations may be
coloured by the referendum result or a respondent’s own
vote choice.
In Appendix I, I present results for the same specifications
replacing economic evaluations from the first
wave with evaluations from the second wave. Respondents’
evaluations of their economic situation are only
available in the post-referendum questionnaire. The literature
on economic voting in elections almost unanimously
suggests that sociotropic voting, voting based on
assessments of the national economy, is stronger than
egotropic voting, voting based on one’s own economic
situation (Lewis-Beck & Stegmaier, 2013). Nevertheless,
it is interesting to explore whether this holds for referendums
as well. Hence, I include variations of my main models
testing for egotropic economic voting in Appendix I.
This study is based on cross-sectional survey data,
as are most other individual levels studies of economic
voting. I correlate differences in the perception of the
economic situation between individuals with their vote
choice or intention. However, this must mean that some
of the interviewees are wrong in their assessment of the
economic situation. For the national economy cannot be
both good and bad at the same time. One prime reason
for these differences in the assessment of the overall economic
situation may be a respondent’s party identification.
Early on, Kramer (1983), a pioneer of economic voting,
pointed out that supporters of the governing party
or parties tend to view the economic situation more positively
than adherents of the opposition party or parties.
Thus, the question is whether a greater consistency of results
in individual-level studies, vis-à-vis studies based on
aggregate data, is not ultimately an artefact of the endogeneity
of perceptions of the economic situation.
The concern that political preferences colour economic
evaluations is at least partly alleviated by the fact
that I use economic perceptions measured before the
referendum. Therefore, the outcome of the referendum
could not have influenced respondents’ economic perceptions.
In addition, I control for partisanship as well
in additional specifications, such as vote choice in the
last national election, government approval, and the popularity
of the prime minister (Table 3). I recoded party
identification and vote choice in the last national election
into dummy variables that indicate whether a respondent
identifies with or voted for one of the governing
parties at the time of the referendum.
Respondents’ factual knowledge about the referendum
was measured by their answers to three questions
asking about details of the constitutional reform and
what the quorum for the referendum was. Respondents
are presented with three possible answers per question.
I take the sum of correct answers (0 to 3) to derive an
index of referendum knowledge. I interact this variable
with respondents’ economic evaluations in the second
model to test hypothesis 2 (see Table 3). I also include approval
of the government in general and the Prime Minister
Matteo Renzi more specifically in some specifications.
Here, respondents were asked to rate Renzi or his government
on a scale from 0 to 10 (0 = completely negative
judgement; 10 = completely positive view). I also
include a respondent’s gender, their age and education
level. The latter is captured by a dummy variable indicating
whether a respondent has attended university or is
currently doing so. Table 2 presents summary statistics
for the key variables.
Binary logistic regression will be used throughout.
The dependent variable is a dummy variable indicating
whether respondents voted in favour of the proposed
constitutional reform (‘yes’ = 1) or against it (‘no’ = 0).
In a first specification, I regress a respondent’s vote
choice on their retrospective sociotropic economic evaluation
to test whether economic evaluations predict vote
choice. Further specifications, included in Table 3, add
measures of partisanship as control variables to assess
the robustness of these findings. In a second set of models,
reported in Table 4, I test for an economic vote in
three separate groups of respondents: voters who identify
with one of the governing parties, voters who identify
with one of the opposition parties and voters who
do not identify with any party. This probes whether the
economic vote is conditioned by partisanship.
6. Results
My empirical results indicate a strong and persistent economic
vote in the 2016 Italian constitutional referendum.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 311
Table 2. Summary statistics for the variables included in the models presented in Table 3.
Variable N Mean St. Dev. Min Max
Vote choice 2716 0.39 0.49 0 1
Sociotropic economy 3020 2.58 0.88 1 5
PID (Government) 3050 0.23 0.42 0 1
Vote (Government) 2907 0.13 0.34 0 1
Government approval 2977 3.83 2.76 0 10
Approval (PM) 2864 3.59 3.08 0 10
Referendum knowledge 3049 1.59 1.05 0 3
Age 3027 48.06 17.12 18 88
Female 3027 0.48 0.50 0 1
University education 3027 0.34 0.47 0 1
Table 3. Model (1) correlates a respondent’s vote choice with their evaluation of the national economy; further models
add partisanship (2), (3) vote choice in the last national election, (4) government approval, or (5) approval of the prime
minister as control variable.
Dependent variable:
Vote choice (Yes)
(1) (2) (3) (4) (5)
Sociotropic economy 1.00*** .73*** .99*** .17** .36***
(.06) (.06) (.06) (.07) (.07)
Referendum knowledge .07* −.02 .05 .15*** .15***
(.04) (.05) (.05) (.05) (.05)
PID (Government) 2.66***
(.13)
Vote (Government) 1.14***
(.13)
Government approval .57***
(.03)
Approval PM .44***
(.02)
Female −.01 −.04 −.01 −.05 −.07
(.09) (.10) (.09) (.10) (.11)
Age .02*** .01** .01*** .01*** .01***
(.003) (.003) (.003) (.003) (.003)
University education .20** .07 .17* .07 .10
(.09) (.10) (.09) (.10) (.11)
Constant −4.07*** −3.35*** −3.98*** −4.19*** −3.84***
(.23) (.25) (.24) (.26) (.26)
Observations 2,682 2,682 2,572 2,656 2,567
Log Likelihood −1,573.14 −1,299.71 −1,472.11 −1,227.44 −1,221.48
Notes: * p < 0.1; ** p < 0.05; *** p < 0.01.
A respondent’s evaluation of the economy shows a significant
positive association with their probability of supporting
the constitutional reform (Table 3). The more positively
a respondent evaluated the national economic situation
in the past year, the more likely they were to vote
‘yes’. Because coefficients from a logistic regression do
not lend themselves to straightforward interpretation,
I also plot the predicted probabilities (Figure 1). A respondent
who evaluates the national economy very negatively
has only a 12% probability of supporting the constitutional
reform, while somebody who rates the state
of the national economy to be very good has a probability
of 88% to vote ‘yes’—a 76-point difference. A regression
model including respondents’ evaluations of the national
economy fares considerably better in predicting a
respondent’s vote choice than a model which does not,
leading to a proportional reduction in error of around
21% compared to the simpler model not including economic
evaluations (see Tables A13 and A14).
Because economic evaluations can be considered
both a cause and a consequence of partisan preferences,
in models (2) to (5) I add a range of measures of a votPolitics
and Governance, 2019, Volume 7, Issue 2, Pages 306–333 312
Table 4. Models correlate a respondent’s vote choice with their evaluation of the national economy, separately for subgroups
of respondents with a party identification with one of the governing parties (column 1), with no party identification
(2) and with a party identification with one of the parties in opposition (3).
Subgroups:
PID (Government) PID (None) PID (Opposition)
(1) (2) (3)
Sociotropic economy .54*** (.15) .83*** (.13) .72*** (.09)
Knowledge −.05 (.13) −.05 (.10) −.08 (.07)
Female −.47* (.25) −.21 (.20) −.10 (.15)
Age .03*** (.01) .01* (.01) .002 (.004)
University education .09 (.24) −.22 (.20) .22 (.16)
Constant −1.01 (.62) −2.78*** (.51) −3.43*** (.37)
Observations 647 559 1,393
Log Likelihood −238.66 −347.97 −598.26
Notes: * p < 0.1; ** p < 0.05; *** p < 0.01.
Sociotropic economic evaluaƟon
Probabilitytovoteyes
1
0.1
0.3
0.5
0.7
0.9
2 3 4 5
Figure 1. Predicted probabilities for a woman of median age (48) with a university degree to vote ‘yes’ in the referendum
dependent on whether she thinks the national economy (1) got much worse, (2) got a little worse, (3) remained equal,
(4) improved somewhat or (5) improved a lot, based on model 1, Table 3.
ers’ partisan disposition. When these variables are included
as additional explanatory variables the coefficient
on economic evaluations is substantially reduced but
remains significant. These results should alleviate concerns
that economic evaluations are simply a proxy for
respondents’ partisan predispositions. Economic evaluations
correlate with vote choices even when controlling
for government approval. Results are also robust to
using vote choice in the last national election, government
approval, or approval of Prime Minister Matteo
Renzi (see Table 3). This is also reflected in the predicted
probabilities associated with these models. As shown in
Appendix I, a hypothetical female voter of median age
with a university degree who, for instance, neither approves
nor disapproves of the government is more likely
to vote for or against the reform depending on her evaluation
of the state of the economy (see Figure A2).
These results hold up in alternative specifications,
where I use vote intentions instead of reported vote
choices, sociotropic economic evaluations measured after
the referendum instead of before the referendum, replace
sociotropic with egotropic economic evaluations,
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 313
and add a voter’s distance on the left-right scale to
Renzi’s PD (see Tables A2, A3, A4 and A5). For egotropic
evaluations, the partial correlation with vote choices is
weaker, but not by much (see also Table A6). This is in line
with the findings of economic voting literature, which
generally finds assessments of the national economy to
be more predictive of vote choices than respondents’
personal economic situation.
Including government approval as an additional explanatory
variable leads to the most substantial reduction
in the coefficient on economic evaluations.3 However,
if one assumes that economic evaluations are
causally prior to government approval, then one can consider
the latter as a mediating factor. Put otherwise, this
reflects the view that government approval is an influence
on an individual’s vote choice, which is itself influenced
by economic evaluations. A simple back-of-theenvelope
calculation based on the mediation procedure
proposed by Baron and Kenny (1986) allows me to differentiate
between a direct and indirect effect of economic
evaluations on vote choice and to calculate the relative
importance of these two channels. As it turns out, both
channels contribute about equally to the total economic
vote. 47% of the correlation between economic evaluations
and vote choice, captured in model 1 in Table 3, is a
direct effect and the remaining 53% an indirect effect mediated
through government approval. I will discuss these
results further in the discussion.
The inclusion of partisanship in model 2 in Table 3
implies a comparison between, on the one hand, voters
who indicated identifying with one of the governing parties
and, on the other hand, voters who either identified
with one of the opposition parties or had no partisan attachment
at all. In total, 24% of respondents who voted
in the referendum did not identify with a political party.
The crucial difference between the latter two groups of
voters is that those without partisan attachment are not
able to rely on partisan cues. However, 99% of those nonpartisans
were able to provide an assessment of the national
economy. Hence, while they are not able to follow
partisan cues, they can cast an economic vote.
Therefore, rather than merely treating partisanship
as a control variable, I also estimate separate models
for partisans of the government, partisans of the opposition
and non-partisans to look into the possibility that
the strength of the economic vote varies between these
groups (Table 4). These results show that all groups of
voters engaged in economic voting, meaning that within
those three groups, respondents with a more negative
assessment of the economic situation were less likely to
vote for the reform than those with a more positive assessment.
The difference in voting behaviour between
those with negative and positive assessments was greatest
among the group of non-partisans, suggesting that
those lacking partisan cues most strongly relied on an
economic vote. These results also hold up in alternative
specifications, which replace reported vote choices with
vote intentions, sociotropic economic evaluations measured
before the referendum with sociotropic economic
evaluations measured after the referendum, and which
replace sociotropic with egotropic economic evaluations
(see Tables A8 to A11).
To facilitate an interpretation of the substantive significance
of these results in Figure 2 plots the predicted
probabilities for the three models in Table 4. When comparing
the plots, two patterns become apparent: the preSociotropic
economic evaluaƟon
Probabilitytovoteyes
0.9
0.7
0.5
0.3
0.1
PID (Government)
1 2 3 4 5
PID (None)
1 2 3 4 5
PID (OpposiƟon)
1 2 3 4 5
Figure 2. Predicted probabilities for three hypothetical women of median age (48) with a university degree to vote ‘yes’
in the referendum dependent on whether they think the national economy (1) got much worse, (2) got a little worse,
(3) remained equal, (4) improved somewhat or (5) improved a lot, based on models 1 to 3 in Table 4.
3 Given the cross-sectional nature of the data, this is nothing more than assumption made to support the calculations that follow. These calculations
should be treated as merely descriptive and illustrative.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 314
dicted voting patterns differ in levels and slopes. Firstly,
independent of their views of the economy, supporters
of the government are of course on average much more
likely to vote for the reform than non-partisans and supporters
of the opposition. Secondly, the relationship between
economic evaluations and vote choice is strongest
in the group of non-partisans. In this group, the difference
between very negative and very positive evaluations
amounts to a full 67 points, while the difference
is 49 points among opposition and only 31 points among
government supporters. However, these differences in
the correlation between economic evaluations and vote
choice between groups are not statistically significant
(see Table A7 in Appendix I, in which I present the relevant
formal test in form of a model interacting economic
evaluations with the categorical PID variable, which is
functionally equivalent to the subgroup analyses, presented
here. The interaction terms are not statistically
significant). The main take-away here is that economic
voting occurs within all groups.
To summarise, these results are in agreement with
the theoretical argument made earlier in this article.
However, that argument is admittedly not the only possible
interpretation of these results, as I will discuss in the
next section.
7. Discussion
In the previous section, I have demonstrated a persistent
and robust correlation between economic perceptions
and vote choice. This is not to say that economic
evaluations were the only or even the most important
determinant of vote choices; however, the results presented
here provide robust evidence in line with my argument
that referendums can turn into a sort of secondorder
election if they become polarised along partisan
lines. Following the government-versus-opposition logic
of the campaign, voters showed a marked tendency towards
economic voting.
However, one must acknowledge that the empirical
results do not dictate the exact interpretation of this
economic vote. Specifically, Bowler and Donovan (1998,
chapter 4) suggest a different frame for interpreting findings
such as my own. They argue that when faced with
a choice between an uncertain outcome—in this case,
ballot initiatives—and a more certain outcome—keeping
the status quo—voters tend to be risk-averse. The authors
connect this reasoning to the literature on economic
voting by arguing that the state of the economy
provides information that influences risk aversion. In line
with their argument, their findings based on multiple referendum
surveys show that voters who hold more positive
evaluations of the economic situation are more likely
to vote to change the status quo.
Following Bowler and Donovan (1998), one could
interpret the results presented in this article as being
(partially) driven by status quo bias. Uncertainty about
the consequences of the reform package in combination
with a negative economic evaluation may have led voters
to stick with the status quo. In that case, my interpretation
of having witnessed an economic vote in an
unusual second-order election would be mistaken. However,
these two interpretations are neither theoretically
nor empirically in contradiction with each other.
Theoretically, it is perfectly possible that some voters
look to the state of the economy to judge whether
it is a good time to risk changing the status quo, while
others treat the referendum as a second-order election,
an opportunity to punish the government. It is also possible
that both considerations play a role in an individual
voter’s mind. Empirically, I have shown that economic
evaluations are predictive of vote choice even when controlling
for measures of a respondent’s orientation towards
the government. One could take this to imply that
both mechanisms played a role in the 2016 Italian referendum.
Specifically, one could argue that the correlation
between economic evaluations and vote choice obtained
after controlling for government approval, which
is substantially reduced but still significant, represents a
heuristic economic vote as Bowler and Donovan (1998,
chapter 4) frame it.
Adjudicating more conclusively between the two arguments
is, unfortunately, beyond the scope of this single
case study. Therefore, it is impossible to disentangle
whether voters who rejected the reform rejected it
based on a change in the status quo or the government’s
position. Given the specifics of the Italian case and the
empirical results presented in this article, however, it
seems safe to say that the referendum took on the character
of a second-order election in which voters exercised
an economic vote, although this does not rule out
risk aversion as an alternative mechanism. Concerning
the Italian case, risk aversion was probably not an essential
factor. Two surveys carried out after the referendum
indicated that more than four-fifths of respondents were
open to changing the constitution and four-fifths of ‘no’voters
were open to reforms to make the system more
efficient (Ceccarini & Bordignon, 2017, p. 295).
8. Conclusions
In this article, I analysed the relationship between individuals’
assessments of the national as well as their personal
economic situation and their vote choice in the
2016 Italian constitutional reform based on a survey conducted
by the ITANES project. I have chosen to adopt the
economic voting paradigm to study voting in this referendum
because the campaign was strongly polarised along
partisan lines, not least because Prime Minister Matteo
Renzi, the architect of the reform, promised to resign if
he lost the referendum.4
As expected, the more positive a respondent’s evaluation
of the economy, the higher their likelihood to
vote ‘yes’ on the referendum proposed by the govern-
4 One can of course only speculate as to whether the economic vote would have been as strong as it was if Renzi had not tied his future to the referendum.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 315
ment. This relationship is surprisingly strong and stable.
These results concur with Pasquino and Valbruzzi’s (2017,
p. 155) verdict that ‘a significant portion of the electorate
decided to punish the government’. When thinking
about referendums more broadly, the Italian constitutional
referendum of 2016 can be considered a mostlikely
case for an economic vote—the polarisation of the
campaign between government and opposition along
with Renzi’s commitment to resign if the reform failed
provided a clear opportunity for economic voting. In the
logic of case studies, a most-likely case exhibits characteristics
that make a theoretically predicted outcome very
likely. Most-likely cases are strong tests of a theory in
the sense that a failure to find the hypothesised relationship
would cast strong doubts on the theory in general
(George & Bennett, 2004).
Having established the relevance of an economic
vote in this particular case, the question now is whether
it can also be found in least-likely cases, referendums
where parties do not take a dominant role, are internally
divided or both of these situations are present. Future
research should, in cases studies of least-likely cases
or in comparative studies, extend the economic voting
paradigm to other referendums and test whether it replicates
in other contexts. This could be complemented
by aggregate data research, building on early research
by Bowler and Donovan (1998) that seeks to establish
whether referendums are more likely to pass in economically
good times. One important question such research
could address is whether an economic vote also occurs
when a government takes the position to uphold the status
quo in a referendum. That question can only be answered
by also looking at referendums where the government
defends the status quo (against an initiative, for instance).
Ideally, one would study multiple referendums
that provide variation in the government’s position in a
referendum to understand more fully whether an economic
vote in a referendum is a form of status quo bias,
or government accountability. Lastly, having likened referendums
to second-order election raises the question
whether the outcome of a referendum also depends on
its relative position in the (first-order) electoral cycle, as
is the case for second-order elections.
If future research finds the economy to be a relevant
factor in referendum voting more generally, then
where in the business cycle a referendum is held could
be a determinant of its outcome. In the Italian case,
a majority of respondents believed that the economy
had not improved over the past year and only about
every tenth respondent thought the economy had gotten
better. Clearly, given that voters’ perceptions of the
economy were such a strong predictor of vote choice
this was not a good time for Renzi to achieve institutional
reform by referendum. In a sense, this poses a
catch-22 for any political leader pursuing reform: Renzi
advocated reforms for a more efficient government as
a means of addressing Italy’s economic woes. However,
it was precisely the economic situation, which he hoped
to alleviate, that kept him from implementing his plans
for reform.
Acknowledgments
I thank Kathrin Ackermann, Christopher Anderson, JeanBenoit
Pilet, Gil Pradeau and Denise Traber as well as participants
at the Empirical Political Science colloquium at
the University of Mainz, and the “Democratic Anxieties”
workshop, also held in Mainz, NIG Annual Work Conference
2017, MPSA 2018, ECPR Joint Sessions 2018 and
EPSA 2018 for very helpful comments and suggestions.
Conflict of Interests
The author declares no conflict of interests.
References
Baron, R. M., & Kenny, D. A. (1986). The moderatormediator
variable distinction in social psychological
research: Conceptual, strategic, and statistical considerations.
Journal of Personality and Social Psychology,
51(6), 1173–1182.
Bowler, S., & Donovan, T. (1998). Demanding choices:
Opinion, voting, and direct democracy. Ann Arbor,
MI: University of Michigan Press.
Ceccarini, L., & Bordignon, F. (2017). Referendum on
Renzi: The 2016 vote on the Italian constitutional
revision. South European Society and Politics, 22(3),
281–302.
Christin, T., Hug, S., & Sciarini, P. (2002). Interests and information
in referendum voting: An analysis of Swiss
voters. European Journal of Political Research, 41(6),
759–776.
Clarke, H., Elliott, E., & Stewart, M. (2017). Heuristics, heterogeneity
and green choices: Voting on California’s
proposition 23. Political Science Research and Methods,
5(4), 755–774.
Curtis, K. A., Jupille, J., & Leblang, D. (2014). Iceland on
the rocks: The mass political economy of sovereign
debt resettlement. International Organization, 68(3),
721–740.
Duch, R. M., & Stevenson, R. T. (2008). The economic
vote. How political and economic institutions condition
election results (political economy of institutions
and decisions). Cambridge: Cambridge University
Press.
George, A. L., & Bennett, A. (2004). Case studies and theory
development in the social sciences. Cambridge,
MA: MIT Press.
Jenssen, A. T. (1998). Personal economies and economic
expectations. In M. Gilljam, P. Pesonen, & A. Jenssen
Todal (Eds.), To join or not to join: Three Nordic referendums
on membership in the European Union (pp.
194–214). Oslo: Scandinavian University Press.
Kayser, M. A. (2014). The elusive economic vote. In L.
LeDuc, R. G. Niemi, & P. Norris (Eds.), Comparing
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 316
democracies (Vol. 4, pp. 112–132). Thousand Oaks,
CA: Sage.
Kramer, G. H. (1983). The ecological fallacy revisited:
Aggregate- versus individual-level findings on economics
and elections, and sociotropic voting. American
Political Science Review, 77(1), 92–111.
Lanz, S., & Nai, A. (2014). Vote as you think: Determinants
of consistent decision making in direct democracy.
Swiss Political Science Review, 21(1), 119–139.
LeDuc, L. (2002). Opinion change and voting behaviour in
referendums. European Journal of Political Research,
41(6), 711–732.
LeDuc, L. (2003). The politics of direct democracy: Referendums
in global perspective. Peterborough, ON:
Broadview Press.
Lewis-Beck, M. S., & Stegmaier, M. (2000). Economic determinants
of electoral outcomes. Annual Review of
Political Science, 3(1), 183–219.
Lewis-Beck, M. S., & Stegmaier, M. (2007). Economic
models of voting. In R. J. Dalton & H. Klingemann
(Eds.), The Oxford handbook of political behavior. Oxford:
Oxford University Press.
Lewis-Beck, M., & Stegmaier, M. (2013). The VP-function
revisited: A survey of the literature on vote and popularity
functions after over 40 years. Public Choice,
157(3/4), 367–385.
Lupia, A. (1994). Shortcuts versus encyclopedias: Information
and voting behavior in California insurance
reform elections. American Political Science Review,
88(1), 63–76.
Nai, A. (2014). The Cadillac, the mother-in-law, and the
ballot: Individual and contextual roots of ambivalence
in Swiss direct democracy. Electoral Studies, 33,
292–306.
Okolikj, M., & Quinlan, S. (2016). Context matters:
Economic voting in the 2009 and 2014 European
parliament elections. Politics and Governance, 4(1),
145–166.
Pasquino, G., & Valbruzzi, M. (2017). Italy says no:
The 2016 constitutional referendum and its consequences.
Journal of Modern Italian Studies, 22(2),
145–162.
Reif, K., & Schmitt, H. (1980). Nine second-order national
elections: A conceptual framework for the analysis of
European election results. European Journal of Political
Research, 8(1), 3–44.
Schmitt, H. (2005). The European parliament elections
of June 2004: Still second-order? West European Politics,
28(3), 650–679.
Schmitt, H., & Teperoglou, E. (2015). The 2014 European
parliament elections in Southern Europe: secondorder
or critical elections? South European Society
and Politics, 20(3), 287–309.
Schoen, H. (2012). Wählen und abstimmen: Zwei Seiten
einer Medaille? Eine Analyse am Beispiel des Volksentscheids
zum Nichtraucherschutz in Bayern [Voting
in elections and referendums: Two sides of the
same coin? An analysis using the example of the
referendum on the protection of non-smokers in
Bavaria]. Politische Vierteljahresschrift Sonderheft,
45, 514–535.
Wikipedia. (n.d.). 2016 Italian constitutional referendum.
Wikipedia. Retrieved from https://en.wikipedia.org/
wiki/2016_Italian_constitutional_referendum
Wlezien, C. (2015). The myopic voter? The economy
and US presidential elections. Electoral Studies, 39,
195–204.
About the Author
Arndt Leininger is a Research Fellow at the Chair for Political Sociology of Germany at Freie Universität
Berlin. His areas of research are comparative politics and applied quantitative methods. Arndt is interested
in direct democracy, turnout, election forecasting, economic voting and electoral studies more
broadly. His work has been published in Political Science Research and Methods, Electoral Studies and
Research and Politics, amongst others.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 317
Appendix I
0%
60%
40%
20%
No
Vote intenƟon
Yes No
Actual result
Yes No
Vote choice
Yes
Figure A1. Aggregated share of vote intentions for and against the constitutional reform (ignoring undecided voters and
non-voters), the actual result of the referendum, and the aggregate reported vote choices (ignoring those who casted a
blank ballot or abstained).
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 318
Sociotropic economic evaluaƟon
Sociotropic economic evaluaƟon
Probabilitytovoteyes
1
0.1
0.3
0.5
0.7
2 3 4 5
Probabilitytovoteyes
1
0.4
0.5
0.6
0.7
2 3 4 5
Sociotropic economic evaluaƟon
Sociotropic economic evaluaƟon
Probabilitytovoteyes
1
0.1
0.3
0.5
0.9
0.7
2 3 4 5
Probabilitytovoteyes
1
0.4
0.3
0.5
0.6
0.7
0.8
2 3 4 5
Figure A2. Top left: predicted probabilities for a woman of median age (48) with a university degree and no partisan identifiction
with a governing party (5) to vote “Yes” in the referendum dependent on whether she thinks the national economy
(1) got much worse, (2) got a little worse, (3) remained equal, (4) improved somewhat or (5) improved a lot, based on
model (2) in Table 1 in the manuscript. Top right: predicted probabilities for the same hypothetical woman who did not
voted for one of the governing parties in the last national elections, based on model (3). Bottom left: predicted probabilities
for the same hypothetical woman who neither approves nor disapproves of the government, based on model (4).
Bottom right: predicted probabilities for the same hypothetical woman who neither approves nor disapproves of the prime
minister, based on model (5).
50%
40%
30%
20%
10%
0%
1 2 3 4 5
Sociotropic economic evaluaƟons
50%
40%
30%
20%
10%
0%
1 2 3 4 5
Sociotropic economic evaluaƟons
60%
40%
20%
0%
1 2 3 4 5
Egotropic economic evaluaƟons
Figure A3. Distributions of economic evaluations: (from left to right) sociotropic economic evaluations (from the
pre-referendum wave), sociotropic economic evaluations (from the post-referendum wave) and egotropic economic
evaluations.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 319
Table A1. Mediation analysis: model 1 replicates model (1) from Table 1 in the manuscript; model (2) regresses government
approval on economic evaluations and control variables; and model (3) replicates model (4) from Table 1 in the manuscript.
All models are OLS regression models (linear probability models in case of models (1), and (3)).
Dependent variable:
Vote choice (Yes) Government approval Vote choice (Yes)
(1) (2) (3)
Sociotropic economy .20*** 1.80*** .02**
(.01) (.05) (.01)
Government approval .10***
(.003)
Referendum knowledge .01* −.05 .02***
(.01) (.04) (.01)
Female −.003 .03 −.01
(.02) (.09) (.02)
Age .003*** .01*** .002***
(.001) (.003) (.0005)
University education .04** .32*** .01
(.02) (.09) (.02)
Constant −.31*** −1.36*** −.17***
(.04) (.19) (.04)
Observations 2,682 2,943 2,656
R2 .15 .34 .35
Notes: * p < 0.1; ** p < 0.05; *** p < 0.01.
Table A2. Replication of models 1 to 5 from Table 1 in the manuscript. All models replace vote choice with vote intentions.
Dependent variable:
Vote intention (Yes)
(1) (2) (3) (4) (5)
Sociotropic economy 1.15*** .85*** 1.12*** .09 .28***
(.07) (.08) (.07) (.10) (.09)
Referendum Knowledge −.44*** −.61*** −.46*** −.59*** −.62***
(.06) (.06) (.06) (.07) (.07)
PID (Government) 2.62***
(.17)
Vote (Government) .92***
(.17)
Government approval .68***
(.03)
Approval PM .61***
(.03)
Female .23** .22* .20* .22 .23
(.11) (.13) (.12) (.14) (.14)
Age .01* −.003 .004 −.0001 −.01
(.003) (.004) (.003) (.004) (.004)
University education .04 −.12 .05 −.22 −.27*
(.11) (.13) (.11) (.14) (.14)
Constant −2.72*** −1.72*** −2.60*** −1.95*** −1.59***
(.27) (.29) (.28) (.32) (.33)
Observations 1,768 1,768 1,720 1,766 1,745
Log Likelihood −1,027.80 −862.99 −981.01 −724.78 −693.10
Notes: *p < 0.1; **p < 0.05; ***p < 0.01.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 320
Table A3. Replication of models 1 to 5 from Table 1 in the manuscript. All models replace sociotropic evaluations measured
before the referendum with sociotropic evaluations measured after the referendum.
Dependent variable:
Vote choice (Yes)
(1) (2) (3) (4) (5)
Sociotropic economy(after referendum) 1.29*** .96*** 1.27*** .54*** .71***
(.06) (.07) (.07) (.08) (.07)
Referendum Knowledge .03 −.04 .01 .11** .11**
(.05) (.05) (.05) (.05) (.05)
PID (Government) 2.49***
(.13)
Vote (Government) 1.09***
(.14)
Government approval .52***
(.03)
Approval PM .39***
(.02)
Female −.02 −.05 −.03 −.04 −.07
(.09) (.10) (.10) (.11) (.11)
Age .02*** .01*** .01*** .01*** .01***
(.003) (.003) (.003) (.003) (.003)
University education .22** .10 .17* .07 .10
(.09) (.11) (.10) (.11) (.11)
Constant −4.91*** −4.01*** −4.79*** −5.04*** −4.74***
(.25) (.27) (.26) (.28) (.28)
Observations 2,679 2,679 2,571 2,650 2,564
Log Likelihood −1,474.46 −1,254.61 −1,384.37 −1,198.07 −1,182.76
Notes: **p < 0.05; ***p < 0.01.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 321
Table A4. Models (1) to (5) replicate the models in Table 1 in the manuscript but replace a respondent’s sociotropic economic
evaluation with their evaluation of their own economic situation (egotropic).
Dependent variable:
Vote Choice (Yes)
(1) (2) (3) (4) (5)
Egotropic economy .95*** .70*** .91*** .35*** .50***
(.07) (.08) (.07) (.08) (.08)
Referendum knowledge .06 −.04 .05 .13*** .14***
(.04) (.05) (.04) (.05) (.05)
PID (Government) 2.80***
(.13)
Vote (Government) 1.20***
(.13)
Government approval .58***
(.02)
Approval (PM) .45***
(.02)
Female −.05 −.08 −.06 −.06 −.11
(.09) (.10) (.09) (.10) (.11)
Age .01*** .01** .01*** .01*** .01***
(.003) (.003) (.003) (.003) (.003)
University education .28*** .13 .24*** .07 .11
(.09) (.10) (.09) (.11) (.11)
Constant −3.94*** −3.27*** −3.79*** −4.74*** −4.33***
(.25) (.28) (.26) (.30) (.30)
Observations 2,689 2,689 2,579 2,658 2,572
Log Likelihood −1,647.76 −1,327.45 −1,541.75 −1,217.66 −1,215.44
Notes: **p < 0.05; ***p < 0.01.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 322
Table A5. Models (1) to (5) replicate the models in Table 1 in the manuscript but, additionally, includes a respondents’
distance on the left–right scale to Matteo Renzi’s Partito Democratico (PD). It is calculated as absolute distance, treating
deviations on left and right equally because the referendum was opposed from parties the right and to the left of the PD.
The position of the PD is based on a respondent’s own assessment elicited in the survey.
Dependent variable:
Vote choice (Yes)
(1) (2) (3) (4) (5)
Sociotropic economy 1.00*** .61*** .78*** .08 .23***
(.06) (.07) (.07) (.08) (.08)
Referendum knowledge .07* −.06 −.03 .08 .10*
(.04) (.06) (.05) (.06) (.06)
LR distance −.21*** −.32*** −.23*** −.25***
(.02) (.02) (.03) (.03)
PID (Government) 2.24***
(.14)
Vote (Government) .90***
(.15)
Government approval .53***
(.03)
Approval PM .39***
(.02)
Female −.01 −.13 −.07 −.11 −.15
(.09) (.12) (.11) (.12) (.12)
Age .02*** .01** .01*** .01*** .01***
(.003) (.003) (.003) (.003) (.003)
University education .20** −.01 .05 .04 .06
(.09) (.12) (.11) (.12) (.12)
Constant −4.07*** −2.15*** −2.09*** −2.98*** −2.47***
(.23) (.31) (.29) (.32) (.31)
Observations 2,682 2,168 2,109 2,164 2,121
Log Likelihood −1,573.14 −1,026.03 −1,128.67 −983.83 −988.90
Notes: *p < 0.1; **p < 0.05; ***p < 0.01.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 323
Table A6. Models (1) to (5) replicate the models in Table 1 in the manuscript but includes a respondents’ evaluation of the
nation’s as well as their own economic situation.
Dependent variable:
Vote Choice (Yes)
(1) (2) (3) (4) (5)
Sociotropic economy .82*** .59*** .82*** .04 .22***
(.06) (.07) (.07) (.08) (.08)
Egotropic economy .52*** .41*** .49*** .33*** .41***
(.08) (.08) (.08) (.09) (.09)
Referendum knowledge .05 −.04 .04 .13** .13**
(.05) (.05) (.05) (.05) (.05)
PID (Government) 2.63***
(.13)
Vote (Government) 1.11***
(.13)
Government approval .57***
(.03)
Approval (PM) .43***
(.02)
Female −.01 −.05 −.01 −.05 −.09
(.09) (.10) (.09) (.11) (.11)
Age .02*** .01*** .01*** .01*** .01***
(.003) (.003) (.003) (.003) (.003)
University education .19** .06 .16* .07 .09
(.09) (.10) (.09) (.11) (.11)
Constant −5.02*** −4.08*** −4.86*** −4.79*** −4.59***
(.28) (.30) (.29) (.31) (.31)
Observations 2,675 2,675 2,565 2,649 2,563
Log Likelihood −1,543.73 −1,282.53 −1,447.39 −1,211.59 −1,206.07
Notes: *p < 0.1; **p < 0.05; ***p < 0.01.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 324
Table A7. Model mirrors model 1 in Table 1 in the manuscript but interacts a respondent’s economic evaluation with their
party identification (PID), distinguishing between a PID with one of the governing parties, the base category, with no party
or with one of the opposition parties. This model is functionally equivalent to the subsample analyses presented in Table
2 in the manuscript.
Dependent variable:
Vote
Sociotropic economy .54***
(.15)
Sociotropic economy PID (None) .27
(.19)
Sociotropic economy PID (Opposition) .21
(.18)
PID (None) −2.69***
(.56)
PID (Opposition) −3.71***
(.52)
Knowledge −.07
(.05)
Female −.20*
(.11)
Age .01***
(.003)
University education .05
(.11)
Constant −.13
(.48)
Observations 2,599
Log Likelihood −1,193.78
Notes: *p < 0.1; ***p < 0.01.
Table A8. Models correlate a respondent’s vote intention, instead of vote choice, with their evaluation of the national
economy, separately for subgroups of respondents with a party identification with one of the governing parties (column 1),
with no party identification (2) and with a party identification with one of the parties in opposition (3).
Subgroups:
PID (Government) PID (None) PID (Opposition)
(1) (2) (3)
Sociotropic economy 1.08*** .75*** .78***
(.20) (.17) (.10)
Knowledge −.71*** −.49*** −.70***
(.19) (.15) (.08)
Female −.45 .14 .26
(.31) (.27) (.17)
Age .01* −.001 −.01**
(.01) (.01) (.005)
University education −.40 −.37 .03
(.29) (.27) (.17)
Constant −.13 −.93 −1.43***
(.77) (.66) (.39)
Observations 492 294 957
Log Likelihood −162.51 −183.74 −469.21
Notes: *p < 0.1; **p < 0.05; ***p < 0.01.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 325
Table A9. Models correlate a respondent’s vote choice, with their evaluation of the national economy (measured after
the referendum), separately for subgroups of respondents with a party identification with one of the governing parties
(column 1), with no party identification (2) and with a party identification with one of the parties in opposition (3).
Subgroups:
PID (Government) PID (None) PID (Opposition)
(1) (2) (3)
Sociotropic economy(after referendum) 1.00*** .87*** .97***
(.16) (.13) (.10)
Knowledge −.10 −.12 −.07
(.13) (.10) (.08)
Female −.37 −.33 −.08
(.26) (.20) (.16)
Age .04*** .01** .0004
(.01) (.01) (.005)
University education .12 −.06 .19
(.25) (.20) (.16)
Constant −2.79*** −2.94*** −4.09***
(.70) (.53) (.38)
Observations 647 556 1,394
Log Likelihood −224.48 −344.83 −576.56
Notes: **p < 0.05; ***p < 0.01.
Table A10. Models correlate a respondent’s vote choice, with their evaluation of their own economic situations instead
of the national economy, separately for subgroups of respondents with a party identification with one of the governing
parties (column 1), with no party identification (2) and with a party identification with one of the parties in opposition (3).
Subgroups:
PID (Government) PID (None) PID (Opposition)
(1) (2) (3)
Egotropic economy .62*** .71*** .65***
(.19) (.15) (.10)
Knowledge −.03 −.08 −.11
(.13) (.10) (.07)
Female −.46* −.39** −.08
(.25) (.19) (.15)
Age .03*** .01 .001
(.01) (.01) (.004)
University education .11 −.03 .22
(.24) (.19) (.15)
Constant −1.29* −2.39*** −3.23***
(.71) (.54) (.39)
Observations 648 563 1,397
Log likelihood −239.71 −360.55 −612.37
Notes: *p < 0.1; **p < 0.05; ***p < 0.01.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 326
Table A11. Models correlate a respondent’s vote choice, with their evaluation of the national economy and that of their
own economic situation, separately for subgroups of respondents with a party identification with one of the governing
parties (column 1), with no party identification and with a party identification with one of the parties in opposition (3).
Subgroups:
PID (Government) PID (None) PID (Opposition)
(1) (2) (3)
Sociotropic economy .41** .72*** .59***
(.16) (.14) (.10)
Egotropic economy .42** .43** .34***
(.21) (.17) (.12)
Knowledge −.06 −.08 −.10
(.13) (.10) (.07)
Female −.44* −.24 −.11
(.25) (.20) (.15)
Age .03*** .01* .002
(.01) (.01) (.004)
University education .09 −.20 .20
(.24) (.20) (.16)
Constant −1.90** −3.64*** −3.99***
(.76) (.62) (.43)
Observations 647 557 1,391
Log likelihood −236.56 −342.48 −592.36
Notes: *p < 0.1; **p < 0.05; ***p < 0.01.
Table A12. Sociotropic economic evaluations (pre-referendum wave), sociotropic economic evaluations (post-referendum
wave), egotropic economic evaluations (post-referendum wave only), in the rows, and vote choice, in the columns.
0 1 0 1 0 1
0.86 0.14 0.88 0.12 0.87 0.13
0.78 0.22 0.82 0.18 0.80 0.20
0.53 0.47 0.59 0.41 0.54 0.46
0.25 0.75 0.16 0.84 0.38 0.62
0.32 0.68 0.26 0.74 0.29 0.71
Table A13. Number of false and correct predictions when always predicting the “Mode” (“No”), using a simple “Reduced”
regression model without the economy variable, the “Full model” reported in the paper. For predictions based on the
regression models a simple is applied stipulating that prediction above .5 are treated as 1 (“Yes”) and prediction below .5
as 0 (“No”).
False Correct
Mode 1046 1636
Reduced model 1032 1650
Full model 827 1855
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 327
Table A14. Reduced model, model (1) from Table 1 in the manuscript but without the economic evaluations, and the full
model, model (1) from Table 1 in the manuscript with the economic evaluations.
Dependent variable: Vote choice (Yes)
(1) (2)
Sociotropic economy 1.00***
(.06)
Referendum knowledg .11*** .07*
(.04) (.04)
Government approval −.08 −.01
(.08) (.09)
Female .01*** .02***
(.002) (.003)
Age .33*** .20**
(.08) (.09)
University education −1.30*** −4.07***
(.15) (.23)
Observations 2,697 2,682
Log likelihood −1,772.12 −1,573.14
Notes: *p < 0.1; **p < 0.05; ***p < 0.01.
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 328
Appendix II
Pre-referendum wave
S18 (sociotropic economic evaluation)
Original
Parliamo di economia. Secondo Lei la situazione economica in Italia nell’ultimo anno è:
1. Molto migliorata
2. Abbastanza migliorata
3. Rimasta eguale
4. Abbastanza peggiorata
5. Molto peggiorata
6. Non saprei
English
Speaking of the economy, do you think the economic situation in Italy in the last year:
1. Improved a lot
2. Improved somewhat
3. Remained equal
4. Got a little worse
5. Got much worse
6. I do not know
S20 (government approval)
Original
Come valuta l’operato del governo guidato da Matteo Renzi, in una scala da 0 a 10 (dove 0 = ‘giudizio completamente
negativo’ e 10 = ‘giudizio completamente positivo’)?
1. 0 = giudizio completamente negativo
2. 1
3. 2
4. 3
5. 4
6. 5
7. 6
8. 7
9. 8
10. 9
11. 10 = giudizio completamente positivo
12. Non saprei
English
How do you assess the work of the government led by Matteo Renzi, on a scale of 0 to 10 (where 0 = ‘completely negative
view’ and 10 = ‘totally positive view’)?
1. 0 = completely negative judgment
2. 1
3. 2
4. 3
5. 4
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 329
6. 5
7. 6
8. 7
9. 8
10. 9
11. 10 = totally positive view
12. I do not know
S30 (Knowledge question 1)
Original
Qual era il quorum (soglia di partecipazione) necessario affinch´e il referendum fosse valido?
• La meta (50%) degli elettori
• I due terzi (66%) degli elettori
• Non c’‘e una soglia: il referendum sar‘a valido qualunque sia il numero di votanti
English
What is the quorum (participation threshold) necessary for the referendum to be valid?
• Half (50%) of voters
• Two-thirds (66%) of voters
• There is no threshold: the referendum will be valid whatever the number of voters
S31 (Knowledge question 2)
Original
La riforma costituzionale prevede che:
1. I senatori non fossero piu‘ eletti direttamente dagli elettori
2. Una riduzione dei membri della Camera dei Deputati
3. L’abolizione del senato e un Parlamento con una sola Camera
English The constitutional reform provides that:
1. The senators are not elected directly by voters anymore
2. A reduction of the members of the Chamber of Deputies
3. The abolition of the Senate and a Parliament with a single Chamber
S32 (Knowledge question 3)
Original La riforma costituzionale prevede di:
1. Abolire il CNEL
2. Abolire il MIUR
3. Abolire la Cassa Depositi e Prestiti
English
The constitutional reform provides that:
1. Abolition of the CNEL
2. Abolition of the Ministry of Education
3. Abolition of the Cassa Depositi e Prestiti
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 330
Post-referendum wave
D1 (sociotropic economic evaluation)
Original
Parliamo di economia. Secondo Lei la situazione economica in Italia nell’ultimo anno è:
1. Molto migliorata
2. Abbastanza migliorata
3. Rimasta eguale
4. Abbastanza peggiorata
5. Molto peggiorata
6. Non saprei
English
Speaking of the economy, do you think the economic situation in Italy in the last year:
1. improved a lot
2. improved somewhat
3. remained equal
4. got a little worse
5. got much worse
6. I do not know
D2 (egotropic economic evaluation)
Original
La situazione economica della sua famiglia ‘e:
1. Molto migliorata
2. Abbastanza migliorata
3. Rimasta eguale
4. Abbastanza peggiorata
5. Molto peggiorata
6. Non saprei
English
The economic situation of your family:
1. improved a lot
2. improved somewhat
3. remained equal
4. got a little worse
5. got much worse
6. I do not know
D9 (left-right self-placement)
Original
Molta gente quando parla di politica usa i termini ’sinistr‘a e ’destr‘a. Qui sotto ‘e riportata una fila di caselle che vanno da
sinistra a destra. Pensando alle sue opinioni politiche, Lei in quale casella si collocherebbe?
1. Sinistra
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 331
2.
3.
4.
5.
6.
7.
8.
9.
10. Destra
11. Non saprei
12. Preferisco non colocarmi
English
Many people when talking about politics use the words ‘left and right’. Below is a row of boxes that go from left to right.
Thinking about your political opinions, in which box would you place?
1. Left
2.
3.
4.
5.
6.
7.
8.
9.
10. Right
11. I don’t know
12. I prefer to not say
D10 (left–right placement of parties)
Original
Pensando invece ai partiti politici, dove collocherebbe ognuno dei seguenti? Utilizzi sempre le stessa fila di caselle che va
da sinistra a destra. Se non conosce il partito o non sa che risposta dare, indichi ‘Non conosco il partit‘o’ o ‘Non saprei.’
1. Partito Democratico
2. Forza Italia
3. Movimento 5 Stelle
4. Lega Nord
5. Nuovo Centrodestra
6. Fratelli d’Italia
7. Sinistra Italiana Opzioni di risposta:
1. Sinistra
2.
3.
4.
5.
6.
7.
8.
9.
10. Destra
11. Non conosco il partito
12. Non saprei
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 332
Thinking instead of political parties, where would you put each of the following? Always use the same row of boxes that
goes from left to right. If you do not know the party or do not know what answer to give, indicate ‘I do not know the party’
or ‘I do not know.’
1. Partito Democratico
2. Forza Italia
3. Movimento 5 Stelle
4. Lega Nord
5. Nuovo Centrodestra
6. Fratelli d’Italia
7. Sinistra Italiana
1. Left
2.
3.
4.
5.
6.
7.
8.
9.
10. Right
11. I don’t know the party
12. I don’t know
Politics and Governance, 2019, Volume 7, Issue 2, Pages 306–333 333