information source credibility and political ... · despite long-running attention to economic...
TRANSCRIPT
INFORMATION, SOURCE CREDIBILITY AND
POLITICAL SOPHISTICATION:EXPERIMENTAL EVIDENCE ON ECONOMIC VOTING∗
JAMES E. ALT† DAVID D. LASSEN‡ JOHN MARSHALL§
APRIL 2014
How does the source of politically-relevant economic information affect voter beliefsand ultimately political preferences? This paper randomly varies whether voters re-ceive an aggregate unemployment projection from the central bank, government oropposition party using a survey experiment in Denmark with unique access to detailedpanel and administrative data. All sources induce voters to update their unemploymentexpectations. While all voters regard the Danish Central Bank as the most crediblesource, only sophisticated voters update more after receiving information from a partywith political incentives to state otherwise. However, belief updating is no greaterwhen the source is aligned with the voter’s previously expressed political preferences.After decreasing unemployment expectations, we find clear evidence of intended eco-nomic voting, without voters changing their policy preferences: the average respon-dent is 3.5 percentage points more likely to vote for the government. Such economicvoting is driven by politically sophisticated rather than swing voters.
∗We wish to thank Alberto Abadie, Charlotte Cavaille, Alex Fouirnaies, Torben Iversen, Horacio Larreguy andVictoria Shineman for valuable advice and comments, as well as participants at the Harvard Political Economy andComparative Politics workshops, NYU Center for Experimental Social Science Conference 2014, Midwest PoliticalScience Association 2014, and MIT Political Economy Breakfast. Lassen thanks the Danish Council for IndependentResearch under its Sapere Aude program for financial assistance.†Department of Government, Harvard University, james [email protected].‡Department of Economics, University of Copenhagen, [email protected].§Department of Government, Harvard University. [email protected]. (Corresponding author.)
1
1 Introduction
Possessing and processing politically-relevant information is a central feature of how voters hold
governments to account and express their preferences over policies. However, most evidence sug-
gests that voters lack basic information about their political or economic contexts (see Anderson
2007). Thus, the provision of credible information has the potential to ensure politicians are more
accountable to voters.1 This is particularly true for economic voting, where aggregate economic
information updates voter beliefs about a government’s competence in office (Anderson 1995; Ro-
goff and Sibert 1988).
However, providing voters with credible information is not straight-forward in practice. In-
formation is rarely provided by independent sources and without an accompanying slant. Rather,
economic and political information is typically communicated by actors with incentives to deceive
or persuade recipients (Baron 2006; Besley and Prat 2006; Larcinese, Puglisi and Snyder 2011).
Recognizing that much of the information available to voters is biased,2 new information may not
affect the beliefs of skeptical voters (Gentzkow and Shapiro 2006).
This raises the question of when sources of political information affect voter beliefs and polit-
ical preferences. We address this important and unanswered question by examining the conditions
under which the source of messages conveying information about future aggregate unemployment—
probably the most important indicator of government performance for voters (Anderson 1995)—
affect voter beliefs and political preferences using a survey experiment. Our experiment is con-
ducted in Denmark, an open economy where macroeconomic concerns have been highly salient in
1For example, information about corruption (Chong et al. 2011; Ferraz and Finan 2008), eco-nomic performance (Bartels 2008; Healy and Lenz 2014) and politician activity (Banerjee et al.2011), and transparent chains of accountability (Powell Jr. and Whitten 1993) have helped holdgovernments to account at the polls.
2Goidel and Langley (1995) and Nadeau et al. (1999) document that voters do understand thatsources of information may be biased. Similarly, many studies note significant differences in trustacross political and media institutions (e.g. Dalton 2008).
2
the aftermath of the financial crisis and where left-right political divisions remain entrenched. The
combination of a panel political survey and access to extremely detailed administrative govern-
ment data provides a unique opportunity to understand in detail which voters update their beliefs
and when such beliefs translate into economic voting.
We first examine how the source of unemployment projections affect unemployment expecta-
tions. We find that the objective credibility of the information source matters: an unemployment
projection from the DCB, which is highly trusted among citizens, causes voters to update their
belief more than receiving information from government or opposition political parties. While in-
formation from both governing and opposition political parties do still affect voter beliefs, a more
sophisticated subset of voters also recognize that a projection from a source with electoral incen-
tives to say the opposite is more credible. However, we find no evidence of subjective credibility
such that voters update more in response to unemployment projections from the party they favor.
Our instrumental variable analysis shows that a percentage point decrease in unemployment
expectations increases the probability that the average complier intends to vote for Denmark’s
coalition government by 3.5 percentage points. This large effect, which only helped the parties of
the Prime Minister and Minister for the Economy and Interior, would have been more than enough
to have altered the outcome of Denmark’s recent knife-edge elections. Supporting the economic
voting interpretation, we observe a large increase in confidence in the government and no change in
support for non-government left-wing parties. Although these results could still reflect unemploy-
ment expectations changing voter policy preferences, rather than beliefs regarding the competence
of the government, unemployment expectations do not affect attitudes toward redistributive or un-
employment insurance policies.
Since assigning responsibility for policy outcomes is especially challenging in Denmark’s com-
plex political system and very open economy, it is not surprising to find that providing new infor-
mation only induces a subset of voters to vote economically. In particular, we find that economic
voting is neither concentrated among swing voters nor ideologues. Rather, economic voting is only
3
observed among sophisticated—better informed, more educated and politically-engaged—voters
and those who already believe the economy is improving. These results show that politically-
relevant information can support democratic accountability, even in political environments where
the clarity of responsibility is low, but is not sufficient to induce all voters to reward good per-
formance. This finding may explain why parties tend to target target their messages at more
politically-engaged voters who appear to be more sensitive to new information (Adams and Ezrow
2009; Gilens 2005).
The paper is structured as follows. Section 2 distinguishes the objective and subjective cred-
ibility of a source of political messages, and considers how economic information might affect
political preferences. Section 3 details our experiments designed to parse out these effects. Section
4 examines how beliefs change, before Section 5 maps these beliefs to vote intention and welfare
policy preferences. Section 6 concludes.
2 Theoretical motivation
This section first considers how voters may differ in their responses to receiving politically-relevant
information from different sources. Focusing on aggregate unemployment expectations, we then
consider how such information could affect economic voting.
2.1 Information sources
Despite long-running attention to economic voting and growing interest in political information,
it remains unclear what types of new information will change the beliefs and behavior of voters.
Many researchers treat information as an unbiased resource helping voters to make the right deci-
sion (e.g. Feddersen and Pesendorfer 1996), or assume that voters start from a common prior (e.g.
Rogoff and Sibert 1988; Rogoff 1990). In experimental work, information is frequently provided
without a source and consequently relies upon the experimenter’s credibility.
4
However, in the real world, most politically-relevant information is conveyed by agents with
distinct and often well-understood ideological biases and incentives to distort perceptions of the
true state of the world (e.g. Baron 2006; Besley and Prat 2006; Gentzkow and Shapiro 2006; Zaller
1999).3 Empirically, Larcinese, Puglisi and Snyder (2011) have shown that pro-Democrat newspa-
pers in the U.S. are more likely to report high unemployment under Republican Presidents, while
Durante and Knight (2012) point to significant biases in television coverage in Italy. Accordingly,
voters must evaluate the information they receive in terms of the credibility of the source.
We distinguish two forms of source credibility that could affect belief updating after receiving
new information. Objective credibility reflects beliefs about the source’s credibility that depend
upon institutional characteristics of the source that are extrinsic to the voter (see also Ansolabehere,
Meredith and Snowberg forthcoming; Zaller 1999). Two important characteristics are institutional
expertise and incentives to deceive. Independent central banks are typically relatively credible
because they have few political incentives to deceive voters,4 and often successfully establish a
reputation for sending accurate messages by virtue of employing highly-trained economists and
providing convincing technical data. Conversely, political parties (and certain media channels)
have widely-understood biases (e.g. Prior 2013): governments have strong incentives to play up
their performance in office, while opposition parties may do the reverse. These features of objective
credibility imply the following hypotheses:
H1. (Institutional expertise) Fixing message content, voters change their beliefs more after re-
ceiving information from an expert source.
H2. (Institutional incentives) Fixing message content, voters change their beliefs more after re-
3Voters receiving biased information is also a demand side phenomenon as well (see Mul-lainathan and Shleifer 2005). We focus on supply by experimentally varying the sources voters areprovided with.
4An influential literature begun in the 1980s persuaded politicians that independent centralbanks could credibly commit countries to sound monetary policies that politicians would otherwisehave incentives to renege on after winning elections (see Barro and Gordon 1983).
5
ceiving information from a source with political incentives to conceal such information.
To be precise, greater belief updating entails larger shifts in the mean of an individual’s probability
distribution over the future unemployment rate. Whether an expert source affects beliefs more than
receiving information going against the expected bias of a less expert source is an empirical ques-
tion that our experiment can answer, but H1 on its own implies that a central bank is regarded as
more credible than political parties while H2 implies that positive information from the opposition
is more credible.
Subjective credibility, on the other hand, depends upon characteristics intrinsic to the receiver
of the information. One critical basis for difference among voters in their response to a given
source is their political sophistication (see Gomez and Wilson 2001, 2006).5 The most sophis-
ticated voters—those that are both politically informed and able to comprehend and assess more
technical information—are unlikely to significantly update their beliefs, since their prior is likely
to be tighter (Ansolabehere, Meredith and Snowberg forthcoming). Nevertheless, given the best
informed voters are generally fairly imperfectly informed (Anderson 2007; Duch and Stevenson
2008), we still expect most voters to respond to specific information. The least sophisticated voters
are likely to have the most diffuse priors. Consequently, we expect the least politically sophisti-
cated voters to update most:
H3. (Political sophistication) Fixing message content, sophisticated voters change update their
beliefs less after receiving new information.
However, while the prior beliefs of the least politically sophisticated are likely to be least accurate,
such voters are less likely to discern biases in the source. This points to an important interaction
between objective and subjective credibility.
5For example, Duch and Stevenson (2010) and Imai, Hayes and Shelton (2014) find that bettereducated and more informed voters are better able to disentangle domestic from imported sourcesof growth.
6
A second dimension of subjective credibility is differences among respondents in their “accu-
racy goals” and “directional goals”, where the former types seek to make decisions based on the
most accurate information (akin to objective credibility) while the latter only seek information that
confirms their prior beliefs (Taber and Lodge 2006). Similarly, Mullainathan and Shleifer (2005)
show that with heterogeneity in priors over politically divisive issues, newspapers separate in their
reporting of the news and cater to a segmented market where the credibility of information in the
eyes of the consumer varies considerably. A large literature in the U.S. has suggested that knowing
the position of a political party on an issue strongly conditions a voter’s beliefs and preferences
(see Boudreau and MacKenzie 2014; Bullock 2011; Malhotra and Kuo 2008; Healy and Malhotra
2013). Given the U.S. is currently experiencing high political polarization and has only two politi-
cal parties, it is not obvious that the partisans in other contexts will respond similarly. Accordingly,
we consider:
H4. (Partisanship) Fixing message content, voters change their beliefs more after receiving in-
formation from a source the voter is politically close to.
Of course, aspects of both objective and subjective credibility could simultaneously affect vot-
ers. By randomly varying sources with differing levels of objective credibility, and comparing
responses to a given source across different types of voter, our empirical design separates differ-
ences in credibility.
2.2 Political implications for economic voting
The idea that governments may be rewarded or sanctioned by voters on the basis of their economic
performance is well-established (see Anderson 2007; Lewis-Beck and Paldam 2000; Lewis-Beck
and Stegmaier 2000). The logic underlying this argument is that voters impose sanctions on the
basis of economic outcomes to deter re-election seeking politicians from choosing suboptimal
policies (Barro 1973; Ferejohn 1986), or looking forward use the available information to select the
7
most competent candidate (Fearon 1999; Rogoff and Sibert 1988; Rogoff 1990).6 Both backward-
and forward-looking information can help to evaluate the competence of office-holders.
To the extent that economic performance is a key election issue and is deemed to possess the
capacity to affect the economy (Duch and Stevenson 2010), information about macroeconomic
performance is expected to increase economic voting. The empirical evidence assessing whether
economic success translates into higher likelihoods of an incumbent being re-elected has been
mixed (Anderson 2007), and has struggled to provide compelling evidence of a causal relationship
(Healy and Malhotra 2013). To the extent that voting is economic, most studies conclude that it is
macroeconomic “sociotropic” aggregates rather than individual-specific “pocketbook” calculations
that drive this relationship (e.g. Kiewiet 1983; Lewis-Beck and Stegmaier 2000).
Economic voting models require that voters both obtain and process sufficient information
about policy choices—or at least their (expected) outcomes—to attribute responsibility and evalu-
ate incumbent performance. These assumptions are now receiving greater scrutiny (see Anderson
2007; Healy and Malhotra 2013). Research has shown that voters often lack even the minimal
information required to vote according to economic performance (e.g. Campbell et al. 1960;
Delli Carpini and Keeter 1996) or suffer partisan biases in attribution (Fiorina 1981; Rudolph
2003a, 2003b, 2006; Malhotra and Kuo 2008; Tilley and Hobolt 2011), while informed voters
have lacked the motivation or cognitive capacity to translate information into responsibility desig-
nation (e.g. Bartels 1996; Delli Carpini and Keeter 1996; Krause 1997). These problems are multi-
plied in institutional contexts characterized by multiple loci of decision-making power, where even
the most willing economic voter may struggle to assign responsibility for economic performance
(Anderson 1995; Duch and Stevenson 2008; Nadeau, Niemi and Yoshinaka 2002; Powell Jr. and
Whitten 1993). Furthermore, information about performance in office may not persuade extreme
or especially partisan voters to act upon it (Ansolabehere and Snyder Jr. 2000).
6The motives underpinning this approach could be either sociotropic or self-interested. AsAnsolabehere, Meredith and Snowberg (forthcoming) have shown, parsing out these effects ischallenging.
8
Combining these insights, our economic voting hypothesis is stated with significant condition-
ality:
H5. (Economic voting) If an individual’s unemployment expectations decrease, the likelihood
that they vote for a party in government (in any given institutional context) increases only
if the individual has the cognitive capacity and will to assign government responsibility to
economic performance.
In this light, economic voting is not the inevitable by-product of providing economic information
for all voters.
3 Research design
3.1 Danish political context
Left-right differences over economic policy remain the salient division in Danish politics, with
governments oscillating between center-left and center-right coalitions. In 2011, Social Democrat
Helle Thorning-Schmidt became Denmark’s first female Prime Minister, having narrowly led the
left bloc—containing the Social Democratic, Social Liberal and Socialist People’s parties as coali-
tion partners, and supported by the Red-Green Alliance—to victory over a center-right coalition
led by the Liberals that had held office since 2001.
Dissatisfaction with the government’s economic performance was the major issue in the 2011
election.7 Having sustained very low levels of aggregate unemployment throughout the 2000s, the
financial crisis hit Denmark’s trade-dependent economy badly. In early 2008 unemployment hit
7E.g. this Economist article. The Danish Election Study polls, available here, show that theeconomy was definitively the most importance issue for voters, while nearly 20% specificallycited unemployment. The study also shows that left-wing voters thought the labor market was thebiggest issue, while right-wing voters thought the economy in general was the biggest issue. Voterssimilarly divided over whether a left or right coalition would best fight unemployment.
9
new lows of nearly 3%, but had increased to around 8% by the 2011 election.8 The budget deficit
also ballooned, leaving Denmark with hard fiscal choices regarding welfare and pension reform.
The center-right’s austerity policies were widely blamed for the failure to produce a stronger eco-
nomic recovery.9 Despite this, the left only just achieved a parliamentary majority, as shown by the
seat distribution for Denmark’s legislative assembly (the Folketing) in Figure 1. In fact, the Social
Democrats actually lost one seat relative to the 2007 election, while the Liberals gained one seat.
The shift in political power particularly reflected the rise of the Social Liberals at the expense of
the Conservative People’s Party.
Although the Danish economy has improved since the 2011 election, left-right economic dif-
ferences have become more politically salient. In January 2013, gross unemployment had officially
fallen to 7.4%.10 Importantly for our study, the DCB expected this rate to fall to just below 7%
by January 2014 (which turned out to be exactly right). Nevertheless, the share of Danes regard-
ing unemployment as the biggest political problem rose from 18% at the 2011 election to 20% by
November 2012, and 36% by late 2013.11 Moreover, within-coalition tensions between the eco-
nomically liberal Social Liberals and the socialist Socialist People’s parties increased. The Social
Liberals only joined the coalition after agreeing a significant conservative welfare reform with the
center-right before the election, and these differences culminated in the Socialist People’s Party
leaving the coalition in January 2014 over unpopular plans to privatize the country’s state-owned
energy company. Economic policy has been contentious throughout the government’s tenure.12
8Unemployment data from Eurostat here. Although Eurostat computes unemployment usingsurveys to ensure cross-national comparability, the Danish government uses administrative recordsto calculate gross unemployment (which is very similar).
9Even though the financial crisis itself was not the fault of Denmark’s government at the time,governments can still be held responsible for exogenous shocks (see Duch and Stevenson 2008),or for failing to respond effectively.
10Gross unemployment is the official unemployment figure used by the government, and is cal-culated using administrative register data. Gross unemployment differs from net unemployment inthat participants in active labor market programs are included in the unemployment rate.
11The November 2012 poll was taken from DR Nyheder here, while the December 2013 pollwas taken from Jyllands-Posten here.
12Another example is the reduction of the maximum length of unemployment benefits from four
10
Left Right
Danish People's Party Liberal PartyConservative People's Party Liberal AllianceDanish Social Liberal Party Social Democrat PartySocialist People's Party Red-Green Alliance
Figure 1: Folketing seat distribution after 2011 election
Notes: Left bloc shaded in red, right bloc shaded in blue. Intensity of color roughly indicates strength of ideologyaccording to the 2011 Danish Election Study.
3.2 Experimental design
To examine the hypotheses derived above, we embedded a survey experiment in the 2013 wave
of the Danish Panel Study of Income and Asset Expectations (Kreiner, Lassen and Leth-Peterson
2013), an annual panel survey of around 6,000 broadly nationally representative Danes conducted
to two years, which was subsequently repealed following an agreement to instead reduce benefitsin the final two years to 60% of their initial level.
11
every January/February.13 The panel, which has been conducted by telephone since 2010, asks
wide-ranging question about the respondent’s financial position as well as their political prefer-
ences. Furthermore, the survey data has been linked by Statistics Denmark, using unique per-
sonal identifiers from the Danish Central Person Registry, to an extraordinarily rich administrative
dataset containing official government register data containing wide-ranging information about all
Danes. The final data set made available for research was anonymized. The combination of panel
political data and detailed respondent histories permits unprecedented detail in our analysis of
differential responses to politically-relevant information.
The central goal of our experiment is to evaluate the conditions under which the provision of
economic information affects individual beliefs and political preferences. We designed our treat-
ments to differentiate the effects of political sources by providing “factual” content in an apolitical
manner.
3.2.1 Treatments
We examine source credibility by varying the source of simple unemployment forecasts, as well
as the forecast itself. After being asked what they estimate the current unemployment rate is, re-
spondents were randomly assigned to one of eight treatment conditions with around 700 members
each. The control group received no information, while six treated groups were read the following
statement:
“Assume that that the [DCB/government/Liberals] estimates that unemployment in
2013 will be [almost 7%/around 5%].”14
13The first wave randomly chose around 6,000 respondents from the Central Person Registry.Annual attrition is around 20-30%. The sample has been replenished with randomly chosen re-spondents from the Registry.
14Survey treatments and questions are translated from Danish; see Online Appendix for Danishphrasing. It is important to emphasize that in Danish the prime translates as a prospective estimate.
12
Respondents were therefore informed that the DCB, the government or main opposition party
project that unemployment over the next year will be “almost 7%” or “around 5%”. The true DCB
projection for gross unemployment was almost 7%. However, because only the DCB has publicly
stated this, ethical considerations required that our other primes begin with “assume that...”. In
order to examine the extent to which such wording weakens the treatment, our final treatment
group was truthfully told “The DCB estimates unemployment in 2013 to be almost 7%.” We
compare this treatment to the analogous “assume” version, and will show no statistical difference
in the distribution of unemployment expectations.
These sources vary considerably in their credibility among voters of all political stripes. Unlike
some other central banks, the DCB is highly regarded by voters, and is not seen as having a right-
wing agenda or being an instrument of government. Asking respondents how much trust they place
in each source, 67% of respondents trusted or greatly trusted the DCB while only 17% and 27%
trusted or greatly trusted the government and Liberals respectively.15 Eurobarometer data indicates
that trust in Denmark’s political parties is very similar to the European Union mean (European
Commission 2011).
3.2.2 Outcome variables
We consider two types of outcome variables: unemployment expectations and preferences over
political parties. To capture unemployment expectations we asked respondents “What is your best
estimate of what unemployment will be in 2013? We would like your best estimate, even if you
are not entirely sure.”16 This question was asked immediately after respondents received their
treatment, and the 20 respondents who answered that the unemployment rate would exceed 50%
15Only the control group responses were used because this question followed the treatment, andthus including post-treatment responses could bias our estimates. These numbers are in line withmass surveys conducted by Statistics Denmark: in 2011, they found that while 82% trusted theDCB, only 59% trusted Parliament. See report summary here.
16From a Bayesian perspective (see Online Appendix), this response can be thought of as anindividual’s posterior unemployment belief (updated after receiving new information).
13
were removed.17
Political preferences are primarily measured by voting intentions and evaluations of the govern-
ment, although we also consider various placebo tests. We code indicator variables for intending
to vote for Denmark’s main political parties, as well as groups for the governing coalition (Social
Democrats, Social Liberals and Socialist People’s parties) and right-wing parties. Vote intention
was elicited 18 questions after the treatment was administered. Because turnout in Denmark regu-
larly exceeds 85%,18 and 72% of respondents ultimately reported voting for the party they intended
to vote for eight months prior to the 2011 election, vote intention represents a good approximation
for what would happen if an election was held immediately. To assess voter perceptions of gov-
ernment competence, we asked respondents how much confidence they have in the government.
Respondents were provided a five-point scale ranging from little great mistrust (1) to great trust
(5) in the government.19
3.3 Identification and estimation
Treatment status is well balanced across pre-treatment covariates. Tables 8 and 9 in the Online
Appendix confirm balance across 16 political and socioeconomic variables frequently included in
observational studies regressing political preferences on a set of covariates. Given random assign-
ment, our empirical analysis can straight-forwardly identify the causal effects of the treatments.
To estimate the average treatment effect on the treated (ATT) for each information treatment
on unemployment expectations U expecti, we estimate the following equation using OLS:
U expecti = Ziα + εi, (1)
17These individuals were very evenly spread across treatment conditions, with between 2 and 4omitted respondents in each group. Removing these observations does not affect the results.
18See Institute for Democracy and Electoral Assistance.19This question was asked 11 questions after the treatment was administered.
14
where Zi is the vector of treatment assignments. Interaction terms are added to allow for hetero-
geneous responses to treatments, and thus aid characterization of which types of individual the
treatments affect. Robust standard errors are reported throughout.
To identify our ultimate quantity of interest—the causal effects of unemployment expectations
on political preferences—we use our information treatments as instruments for unemployment
expectations. Instrumenting overcomes the obvious concern that economic expectations may be
correlated with omitted variables that also affect political preferences. Taking equation (1) as the
first stage, we estimate the local average causal response (LACR) (Angrist and Imbens 1995),
averaging the causal effects for compliers—individuals for whom our randomly-assigned informa-
tion treatments induced respondents to change their unemployment expectations—across different
unemployment expectation levels.20 Accordingly, we estimate the following structural equation
using 2SLS:
Yi = τU expecti + δU nowi + ξi, (2)
where Yi is vote intention, confidence in the government, or a policy preference placebo test. The
respondent’s estimate of the current unemployment rate (U nowi), a good approximation for an
individual’s prior unemployment expectation, is included to enhance efficiency.21
Consistent estimation of the LACR requires two assumptions beyond the randomization of our
instruments: monotonicity and an exclusion restriction (Angrist, Imbens and Rubin 1996; Imbens
and Angrist 1994). Monotonicity entails that each individual would update their unemployment
expectations in the same direction upon receipt of the treatment. Although it is hard to imagine
20The LACR here is the linearized causal effect of unemployment expectations, weighted towardareas where the density function of complier responses is greatest.
21The (average) prior belief is most accurately estimated using the control group’s unemploy-ment expectation. However, the estimate of the current unemployment rate is also an excellentproxy for the prior over future the unemployment rate: among the control group, there is a 0.93correlation between current and future estimates. Our results are almost identical using differencebetween the current and future unemployment estimate as the endogenous variable.
15
when prominent public sources would induce voters to update their beliefs against the information
provided, respondents with low prior unemployment expectations may increase their unemploy-
ment expectations, especially after the 7% treatments.
Fortunately, the monotonicity assumption can be weakened in ways consistent with our data.
In general, 2SLS estimation recovers a very similar quantity of interest to the LACR when “few
subjects are defiers, or if defiers and compliers have reasonably similar distributions of potential
outcomes” (de Chaisemartin 2013: 7)—and is identical under constant causal effects (see Angrist,
Imbens and Rubin 1996).22 In this application, 27% of respondents upwardly update their unem-
ployment expectations relative to their current estimate. Since upward and downward effects may
be very similar, and given that compliers significantly outnumber defiers, the presence of defiers is
relatively unproblematic. Nevertheless, our results are very similar when we restrict the sample to
the 5% treatments with almost no defiers. Our analysis also considers a variety of other subgroup
analyses, based on respondents’ current estimates of unemployment, where monotonicity almost
certainly holds.
The exclusion restriction, which requires that the instrument only affects Yi through U expecti,
is usually more problematic in empirical studies. Although such violations are unlikely in this
application, perhaps the most plausible violation arises where information treatments prime re-
spondents to think more carefully about government performance and policies (beyond the effect
of changing beliefs about unemployment expectations), inducing bias if such thinking systemati-
cally affects support for the government. We assess this possibility by looking at whether belief
in the importance of political information for either private economic decisions or as part of the
22Rather than recover the local average treatment effect, the Wald estimator (with no covariates)recovers the average treatment effect for a smaller group of compliers (precisely those not canceledout by the defiers). A sufficient assumption for this to equate to the case with no defiers is that thereare more compliers than defiers for any combination of potential outcomes (Assumption (2.4) andequation (2.2) in de Chaisemartin 2013), while a weaker condition requires only that some subsetof compliers has the same size and marginal distribution over potential outcomes as defiers, or thatthere are more compliers and defiers at each potential outcome (de Chaisemartin 2013).
16
respondent’s job differs across treatments groups (or comparing the control to all treated respon-
dents), and find no difference.
4 Effects of information source on economic expectations
We first show that the information treatments substantially change unemployment expectations.
While we find evidence for both forms of objective credibility, and differential responses by prior
knowledge of the current unemployment rate, there is no evidence that political preferences cause
voters to update differentially. We first examine the distribution of the data, before proceeding to
regressions identifying average effects and then heterogeneity in voter responses.
4.1 Results
Figure 2 plots the distribution of unemployment expectation responses by treatment condition.
Before turning to our main results, it is clear from Panel A that the “assume” wording does not
affect the distribution of the DCB 7% projection responses.23 This suggests that the statement
wording is not biasing the results. Henceforth we pool the DCB 7% treatment groups. Although
this similarity may not necessarily extend to other treatments, it suggests that any differences are
likely to be small, while if anything our treatment effects are lower bounds.
The leftward shift in density associated with all treatments indicates that all information sources
reduce average unemployment expectations. This reduction reflects systematic pessimism in a
population where the average member of the control group expected an unemployment rate of
9.0%. Despite its optimism relative to the true DCB claim, the 5% treatments dragged expectations
below those receiving the 7% treatments. In all cases, the information treatments reduced the
variance of the distributions, providing further evidence that the treatments affected voters.24 We
23Tests comparing the mean and variance of the distributions cannot reject the null hypothesisof identical sample moments.
24Distributional tests confirm that the variance reduction is statistically significant. Although
17
0 5 10 15 20
0.00
0.10
0.20
0.30
Panel A
2013 unemployment expectation (%)
Den
sity
0 5 10 15 20
0.00
0.10
0.20
0.30
0 5 10 15 20
0.00
0.10
0.20
0.30
0 5 10 15 20
0.00
0.10
0.20
0.30
ControlDCB 7%Assume DCB 7%Assume DCB 5%
0 5 10 15 20
0.00
0.10
0.20
0.30
Panel B
2013 unemployment expectation (%)
Den
sity
0 5 10 15 20
0.00
0.10
0.20
0.30
0 5 10 15 20
0.00
0.10
0.20
0.30
0 5 10 15 20
0.00
0.10
0.20
0.30
0 5 10 15 20
0.00
0.10
0.20
0.30
ControlAssume govt. 7%Assume govt. 5%Assume opp. 7%Assume opp. 5%
Figure 2: Unemployment expectations by DCB treatments
Notes: For graphical exposition, the x-axis is truncated so that the 1% of the sample with expectations above 20%are not visible.
18
now turn to our source credibility hypotheses.
Consistent with differences in expertise (H1), receiving information from political parties caused
the average voter to update their beliefs less than receiving information from the DCB. The DCB
treatments also induced more similar responses from voters (i.e. a smaller standard deviation in re-
sponses), especially compared to the opposition treatments. Although it could have been the case
that simply being primed by a source increased confidence in the source, the Online Appendix
shows that receiving a treatment does not affect trust in either political party.25
Partisan sources also reduced unemployment expectations. Panel B clearly shows a downward
shift in modal unemployment expectations for both the government and opposition treatments.
Surprisingly, given that the opposition has a political incentive to criticize government economic
performance, the Liberal projections did not cause voters to differentially update their beliefs rela-
tive to the predictably optimistic government message. We therefore find little support for H2, on
average.
Table 1 confirms our graphical analysis by estimating equation (1). Receiving a 7% treatment
reduces unemployment expectations by around 1 percentage point, while a 5% treatment subtracts
a further 0.5 percentage points. For both levels, the DCB has a larger effect on unemployment
expectations. Supporting the importance of differences in institutional expertise (H1), the p-values
associated with F-tests comparing the DCB source coefficients to the party source coefficients
generally show a credibility difference for both the 7% and 5% treatments. Contrary to H2, there
is no discernible difference between the government and opposition 7% or 5% treatments in the
full sample.
these belief shifts could in part reflect anchoring biases (Tversky and Kahneman 1974), it is hardto see how such explanations could explain the changes in political preferences we documentbelow.
25There is a slight increase in trust of the DCB, but the change cannot explain the large responseto the DCB treatments.
19
Table 1: Effect of information treatments on unemployment expectations (%)
Unemployment expectations (%)
Control 9.012***(0.185)
DCB 7% treatment (combined) -1.123***(0.197)
DCB 5% treatment -1.663***(0.230)
Government 7% treatment -0.848***(0.213)
Government 5% treatment -1.218***(0.233)
Opposition 7% treatment -0.923***(0.223)
Opposition 5% treatment -1.335***(0.236)
Test: DCB 7% = Government 7% p = 0.03**Test: DCB 7% = Opposition 7% p = 0.16Test: Government 7% = Opposition 7% p = 0.65
Test: DCB 5% = Government 5% p = 0.02**Test: DCB 5% = Opposition 5% p = 0.10Test: Government 5% = Opposition 5% p = 0.57
Observations 5,705Outcome mean 7.98Outcome standard deviation 3.55
Notes: Estimated using OLS. Robust standard errors in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. Thecoefficient tests at the foot of the table report the p value from a two-sided F test of coefficient equality.
20
4.2 Heterogeneous effects: when and how do voters update their beliefs?
The potential impact of different information sources depends on which voters update their beliefs.
We explore this issue using heterogeneous effects and sub-samples as a means of identifying how
new information affects voter beliefs.
We first test for partisan subjective credibility (H4). The bottom row of Table 2 shows that there
is no evidence for differential updating: respondents who voted for a government (right) party
at the 2011 election did not differentially update their beliefs when provided with information
from the government (opposition).26 Given these results are surprising from the perspective of
previous findings in the U.S., we examined various alternative definitions of political disposition.
We similarly found no difference when defining left and right-wing supporters as respondents
who intended to vote for the same left or right party in the 2011 and 2012 surveys. Looking for
differences within education groupings and alternatives measures of political ideology all yielded
no differential response. The results therefore strongly suggest that the political beliefs of Danes
do not affect their views on politically-relevant information.
While prior political dispositions do not cause voters to respond differentially, there are system-
atic differences by voter political sophistication (H3). Table 2 shows that men and respondents with
greater education, higher wage income, and faith that the Danish economy will improve relative to
the previous year update less in response to unemployment information.27 However, a respondent’s
current unemployment estimate effectively serves as a “sufficient statistic” for these characteristics
representing political sophistication: the Online Appendix shows that once a respondent’s prior
is included as an interaction with the treatments, the interaction coefficients in Table 2 dramati-
cally decline in magnitude and leave only the interaction with perceptions of national economic
prospects as statistically significant.
26There is similarly no difference if we examine only the interaction between previous votingbehavior and our treatments.
27The respondent’s subjective probability of being without a job in the forthcoming year alsohad no interaction effect, but substantially reduced the sample size.
21
Table 2: Heterogeneous effects of information treatments on unemployment expectations (%), byconditional marginal effect
DCB 7% DCB 5% Govt. 7% Govt. 5% Opp. 7% Opp. 5%
Linear effect -5.006*** -5.14*** -4.203*** -4.778*** -4.263*** -4.364***(1.239) (1.492) (1.379) (1.519) (1.386) (1.489)
× News every day 0.798* 0.501 0.016 0.605 0.342 0.306(0.442) (0.538) (0.509) (0.526) (0.514) (0.53)
× Denmark economic prospects 0.615** 0.611* 0.764** 0.47 0.637** 0.388(0.279) (0.325) (0.299) (0.389) (0.308) (0.337)
×Wage income (log) 0.105** 0.049 0.099** 0.161*** 0.072 0.127**(0.044) (0.053) (0.045) (0.047) (0.051) (0.05)
×Medium education 1.448*** 1.638** 1.019 0.9 1.443** 1.183*(0.549) (0.642) (0.624) (0.645) (0.61) (0.645)
× High education 1.748*** 1.534** 1.02 0.703 1.93*** 1.593*(0.631) (0.722) (0.692) (0.731) (0.689) (0.844)
×Woman -1.6*** -1.133** -1.39*** -1.038** -1.092** -1.492***(0.384) (0.445) (0.416) (0.463) (0.438) (0.469)
× Voted left at last election 0.084 0.095 -0.134 -0.083 -0.34 0.193(0.386) (0.451) (0.42) (0.465) (0.437) (0.467)
Notes: All coefficients are estimated from a single OLS equation interacting all treatments conditions with thevariables on the left hand side of the table (see Online Appendix for their definitions). The coefficient for thecontrol group is 15,429***(1.173). The sample size is 5,446. Robust standard errors in parentheses. ∗p <
0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
22
Simply linearly interacting a respondent’s current unemployment estimate with the informa-
tion shocks suggests that better informed voters are not affected by our treatments. However,
this approach could miss important variation in responses by political sophistication, or ignore
voters with low current unemployment estimates who may revise up their posterior beliefs. To
provide a clearer picture, we first split the sample between under- and over-estimators according to
whether a respondent’s current unemployment estimate exceeds the true 7.4% rate, before restrict-
ing attention to respondents with initial estimates between the treatment rate bounds and within 2
percentage points of the truth. The Online Appendix confirms that the latter three samples include
more sophisticated voters with more education, who discuss politics more often and watch the
news more often. The distribution of political preferences, however, is very similar across these
samples.
The results in columns (1) and (2) of Table 3 reveal that under- and over-estimators respond
quite differently. Unsurprisingly, over-estimators experience the largest declines in unemploy-
ment expectations. Coefficient comparison tests show that the DCB reduces expectations further
than political parties, but again identifies no difference between political parties. However, under-
estimators do exhibit an understanding of differential political incentives: supporting H2, under-
estimators increase their expectations after receiving the government 7% projection relatively more
than after receiving the opposition 7% projection. The greater institutional credibility of the DCB
cannot be distinguished from the credibility arising from the government providing information
that goes against their political incentives. Although the relative magnitudes support similar con-
clusions for the 5% projections, we cannot reject coefficient similarity.28
Turning to columns (3) and (4), a similar picture of sophisticated voting emerges among re-
spondents for whom the treatments imply belief updating in opposing directions and respondents
whose current unemployment estimate was within 2 percentage point of the truth. Column (3)
28Controlling for current unemployment guess, there is a statistically significant difference suchthat under-estimators trust the government less.
23
Tabl
e3:
Eff
ects
ofin
form
atio
ntr
eatm
ents
onun
empl
oym
ente
xpec
tatio
ns(%
),by
curr
entu
nem
ploy
men
test
imat
e
Est
imat
e>7.
4%E
stim
ate≤
7.4%
Est
imat
e∈[5
%,7
%]
Est
imat
e∈[5
.4%
,9.4
%]
(1)
(2)
(3)
(4)
Con
trol
11.7
03**
*6.
099*
**6.
505*
**7.
536*
**(0
.282
)(0
.083
)(0
.071
)(0
.080
)D
CB
7%tr
eatm
ent(
com
bine
d)-2
.582
***
0.50
1***
0.29
5***
-0.2
73**
*(0
.303
)(0
.092
)(0
.079
)(0
.087
)D
CB
5%tr
eatm
ent
-2.7
34**
*-0
.361
***
-0.6
23**
*-1
.224
***
(0.3
72)
(0.1
00)
(0.0
91)
(0.1
01)
Gov
ernm
ent7
%tr
eatm
ent
-2.1
83**
*0.
625*
**0.
340*
**-0
.134
(0.3
29)
(0.1
03)
(0.0
90)
(0.0
98)
Gov
ernm
ent5
%tr
eatm
ent
-2.4
29**
*-0
.290
***
-0.4
26**
*-0
.773
***
(0.3
55)
(0.1
06)
(0.0
93)
(0.1
00)
Opp
ositi
on7%
trea
tmen
t-1
.963
***
0.30
5***
0.07
6-0
.311
***
(0.3
46)
(0.1
06)
(0.0
95)
(0.1
02)
Opp
ositi
on5%
trea
tmen
t-2
.145
***
-0.3
96**
*-0
.496
***
-0.9
61**
*(0
.371
)(0
.109
)(0
.098
)(0
.106
)
Test
:DC
B7%
=G
over
nmen
t7%
p=
0.05
**p=
0.09
*p=
0.50
p=
0.04
**Te
st:D
CB
7%=
Opp
ositi
on7%
p=
0.01
***
p=
0.01
***
p=
0.00
***
p=
0.60
Test
:Gov
ernm
ent7
%=
Opp
ositi
on7%
p=
0.41
p=
0.00
***
p=
0.00
***
p=
0.04
**
Test
:DC
B5%
=G
over
nmen
t5%
p=
0.35
p=
0.41
p=
0.02
**p=
0.00
***
Test
:DC
B5%
=O
ppos
ition
5%p=
0.08
*p=
0.70
p=
0.15
p=
0.00
***
Test
:Gov
ernm
ent5
%=
Opp
ositi
on5%
p=
0.38
p=
0.27
p=
0.44
p=
0.04
**
Obs
erva
tions
2,95
52,
750
2,31
73,
132
Not
es:S
eeTa
ble
1.
24
further confirms that voters respond to our treatments, increasing their expectations following the
7% treatment and decreasing expectations following the 5% treatment. The coefficient tests again
show that the 7% projection from the DCB and government are equally credible, and both signifi-
cantly exceed the opposition treatment. For the 5% treatment, only institutional credibility appears
to matter, although the reduction in expectations is again larger for the opposition than the govern-
ment source. Column (4), which considers the voters with the most accurate current unemployment
estimates, demonstrates that DCB and opposition information reduce unemployment expectations
more than claims by the government—for both the 7% and 5% treatments. It again shows that new
information affects even the voters with the most accurate prior assessments, demonstrating that
all types of respondent can be considered compliers for our instrumental variable analysis. We thus
find significant support for H2, but only among a subset of more politically sophisticated voters.
5 Effects on political preferences
The preceding analysis has shown that information about aggregate unemployment projections
affects voter beliefs about the economy’s prospects. However, does this matter for political pref-
erences? This section shows that exogenously changing expectations causes informed and cogni-
tively able voters to change their vote intentions in accordance with economic voting motivations,
but does not affect their policy opinions. By showing that lowering unemployment expectations
increases confidence in the government without affecting policy preferences, these results imply
that aggregate unemployment expectations are principally used to evaluate the competence of the
government.
5.1 Results
Table 4 reports estimates of equation (2), identifying the LACR of a percentage point increase
in unemployment expectations on political preferences for individuals affected by the instruments.
25
The outcomes in columns (1)-(6) are indicators for supporting a particular party or group of parties.
The large F statistic unsurprisingly indicates a very strong first stage.29
The results are highly consistent with a significant proportion of citizens engaging intending to
engage in economic voting. The decrease in unemployment expectations induced by the informa-
tion treatments causes compliers to increase their support for the parties of government on average
by 3.5 percentage points for each percentage point decrease in aggregate unemployment expecta-
tions.30 Increased government support is almost exactly mirrored by the decrease in support for
right-wing parties in column (5), with the majority of votes coming from the main right-wing Lib-
eral party shown in column (6). In the context of coalition politics, and especially the extremely
close recent Danish elections, information about aggregate unemployment could easily have al-
tered the composition of government. Even by the standards of countries with greater clarity of
responsibility, the effect is very substantial—in spite of vote intention being asked 18 questions
after the treatment.
While the allocation of credit and blame for the economy’s progress is usually relatively clear
when there is a single-party government, voter sanctioning is not obvious among coalition partners
(Anderson 1995; Duch and Stevenson 2008). Columns (2)-(4) disaggregate the government vote
share by the three parties in the governing coalition. The results clearly indicate that the two largest
coalition partners—the Social Democrats and the Social Liberal Party, who had 44 and 17 seats
and 10 and 6 cabinet positions respectively—are the sole beneficiaries, both gaining 1.6 percent-
age point increases in the probability of a respondent voting for them for each percentage point
decrease in unemployment expectations. This represents a relatively larger gain for the smaller
Social Liberal party. In line with the findings of Anderson (1995) and Duch, Przepiorka and
Stevenson (forthcoming), responsibility is assigned to the parties with greatest control over eco-
29The Online Appendix provides the first stages estimated, which are very similar to Table 1.30The reduced form estimates show similar results in the Online Appendix. Examining the DCB,
government and opposition treatments as separate groups, the LACR magnitudes are consistentacross information sources rather than being driven by particular sources.
26
Table 4: Effect of unemployment expectations on political preferences
(1) (2) (3) (4) (5) (6)Govt. Soc. Dem. Soc. Lib. Soc. Peop. Right Liberals
Unemployment expectations (%) -0.035** -0.016 -0.016* -0.003 0.034** 0.024*(0.014) (0.011) (0.009) (0.007) (0.015) (0.014)
First stage F statistic 32.64 32.64 32.64 32.64 32.64 32.64Observations 5,705 5,705 5,705 5,705 5,705 5,705Outcome mean 0.32 0.17 0.09 0.06 0.41 0.28Outcome standard deviation 0.47 0.37 0.29 0.24 0.49 0.45
Notes: All specifications estimated using 2SLS, and control for current unemployment expectations. Robuststandard errors in parentheses.
nomic policy: while the Social Democrats led the coalition and held the Premiership, the leader of
the Social Liberals—who campaigned on their centrist economic agenda—became Minister for the
Economy and Interior. The intended vote share of the more extreme left-wing Socialist People’s
Party, which held 16 seats and 6 cabinet positions, is essentially unaffected.
After observing macroeconomic performance, the key theoretical claim underpinning eco-
nomic voting is that unemployment expectations affect vote choice through voter perceptions of
government competence. Strongly supporting this mechanism, column (1) in Table 5 shows that
lower unemployment expectations significantly increase confidence in the government.
Nevertheless, a potentially confounding explanation of our results is that evaluations of govern-
ment competence are not changing, but rather that lower unemployment expectations have shifted
policy preferences toward those associated with left-wing parties (e.g. Meltzer and Richard 1981;
Moene and Wallerstein 2001). Self-interested voters maximizing their expected income should
decrease their support for redistribution and unemployment insurance to the extent that higher ag-
gregate unemployment expectations are taken as a signal of economy-wide, rather than individual-
specific, economic prospects. If aggregate unemployment expectations instead primarily update a
27
Table 5: Mechanism and placebo tests
(1) (2) (3) (4)Conf. govt. Redist. U. insurance Red-Green
Unemployment expectations (%) -0.100*** 0.032 -0.011 0.004(0.029) (0.030) (0.018) (0.008)
First stage F statistic 28.65 32.64 33.51 32.64Observations 5,688 5,705 5,614 5,705Outcome mean 2.69 3.20 2.23 0.06Outcome standard deviation 1.00 1.02 0.61 0.25
Notes: All specifications estimated using 2SLS, and control for current unemployment expectations. Robuststandard errors in parentheses.
voter’s subjective probability of being unemployed, support for redistribution and unemployment
insurance should increase. We show these predictions formally in the Online Appendix.
However, changes in policy preferences cannot account for the results observed here. First,
we examine five- and three-point scales that respectively increase with general support for redis-
tribution and specific support for unemployment benefits. The precisely estimated null effects in
columns (2) and (3) of Table 5 show no support for either claim, despite the question about un-
employment insurance being asked one question after the treatment was administered.31 Second,
the existence of left-wing parties outside the government provide a further placebo test for our
economic voting interpretation. The Red-Green Alliance—the most left-wing party represented in
the Danish Parliament—might expect to pick up votes if the information treatments were inducing
a change in preferences. Column (4) shows that changes in unemployment expectations do not
affect the probability of voting for the Red-Green Alliance. Together, this evidence reinforces the
conclusion that economic voting is the principal political manifestation of changes in aggregate
unemployment expectations.
31Unreported results show that the effect does not differ by income level.
28
As noted above, the monotonicity assumption is violated for the 7% information treatments.
We confirm that defiers are not biasing the results by restricting the sample to cases where mono-
tonicity almost certainly holds. Focusing only on the 5% treatments where the cumulative distri-
bution of unemployment expectations lies almost everywhere to the left of the control group, the
Online Appendix shows very similar LACR estimates.
5.2 Heterogeneous effects: who are the economic voters?
To better understand how economic voting works, we investigate which types of voters act po-
litically on their unemployment expectations. Our detailed data provides significant leverage to
examine the heterogeneous effects implied by existing theories.
Political economy models typically regard swing voters as the most likely to transfer their
votes to a party on the basis of competence, while the vote choices of partisans are unaffected (e.g.
Ansolabehere and Snyder Jr. 2000; Persson and Tabellini 2000). However, in practice it is hard to
empirically differentiate such swing voters from capricious disengaged voters. Furthermore, swing
voters may lack the cognitive capacity or political engagement required to link unemployment
expectations to government accountability for economic policy (Campbell et al. 1960).
We test for whether swing voters are driving the changes in vote intention by exploiting the
panel structure of the dataset. We define an indicator for the 43% of respondents who reported
voting for different parties at the 2007 and 2011 elections. Figure 3 demonstrates that such swing
voters are not driving changes in government support. Rather, the effect of unemployment expec-
tations among swing voters is indistinguishable from zero. Given the first stage for swing voters
is especially strong, this result does not reflect swing voters failing to update their unemployment
expectations. To ensure our definition of swing voters is not picking up shifts to parties offering
similar platforms, we also calculated measures for left and right party groupings and examined
swings to the left and swing to the right and in each case found similar results. The results are sim-
ilarly robust to defining swing voters as individuals whose 2011 and 2012 survey vote intentions
29
differed.
That economic voting is concentrated among respondents who have expressed consistent re-
cent political preferences may at first seem surprising. However, assigning responsibility over
economic policy to different parties is complicated in Denmark, where coalition governments are
the inevitable outcome of a PR electoral system with many parties and unstable alliances in the po-
litical center (Anderson 1995; Powell Jr. and Whitten 1993). This is particularly challenging if, as
in the U.S., swing voters are less politically engaged (Campbell et al. 1960) and less likely to link
their voting decisions to government actions or retrospective economic assessments (Delli Carpini
and Keeter 1996). We similarly find that swing voters in Denmark are characterized by low polit-
ical sophistication: swing voters discuss politics less with friends, family and neighbors, are less
educated and have lower math test scores, and follow economics and politics in the news less regu-
larly. Given this lack of political engagement and cognitive capacity, our results suggest that swing
voters are unable or unwilling to link economic performance to evaluations of the government.
Although the respondents whose vote intention was affected were not swing voters, they are not
ideological extremists. Coding the 17% of the sample who provided the most extreme responses
(from either end) to the redistribution question in the 2012 survey, Figure 3 shows that the response
of such voters is statistically insignificant and significantly below non-extreme voters.
Our data permit more detailed tests of the claim that political sophistication is essential for
economic voting (H5). We measure political engagement by defining an indicator for the 72% of
respondents who read or watch economics or politics on the news every day.32 To capture cognitive
capacity, we define an indicator for the 77% of respondents with education beyond high school.
Finally, an individual’s initial view of the Danish economy’s prospects could induce subjective
experience biases. We measure this with an indicator for the 34% of the sample who expected that
32Although this question was asked after the treatment was administered, regressing this vari-able on all information treatments provided no evidence to suggest that the treatments influencedresponses. We find similar effect for discussion of politics, aggregating the indicators for dis-cussing politics with friends, family, neighbors, workers and others.
30
Non-swing voter
Swing voter
Non-extreme voter
Extreme voter
News less than every day
News every day
High school only
Beyond high school education
Non-improving economic prospects
Improving economic prospects
-.15 -.1 -.05 0 .05
Marginal effect of unemployment expectations on voting for government party
Figure 3: Heterogeneous effect of unemployment expectations on intending to vote for agovernment party (95% confidence intervals)
Notes: Estimates are from separate 2SLS regressions instrumenting for unemployment expectations and its inter-action. Regression coefficients are provided in the Online Appendix.
31
Table 6: Effect of unemployment expectations on political preferences, by current unemploymentestimate
Estimate>7.4% Estimate≤7.4% Estimate∈[5%,7%] Estimate∈[5.4%,9.4%](1) (2) (3) (4)
Unemployment expectations (%) -0.014 -0.050** -0.065** -0.047**(0.011) (0.023) (0.028) (0.022)
First stage F statistic 33.74 74.00 64.34 66.38Observations 2,955 2,750 2,317 3,132Outcome mean 0.30 0.34 0.34 0.34Outcome standard deviation 0.46 0.47 0.48 0.47
Notes: See Table 4.
the Danish economy would improve in 2013 relative to 2012.
Figure 3 shows the conditional economic voting estimates. While differences between types
of voters are not always quite statistically significant, the effect of unemployment expectations
accords with economic voting only among voters who regularly watch the news, completed higher
levels of education, and expect to experience improved aggregate economic performance. As
noted above, the current unemployment estimate is a good proxy for political sophistication. The
results in Table 6 reinforce our preceding findings: economic voting is only detected in the more
sophisticated subsamples where respondents processed the political incentives of different sources
and disproportionately comprise better educated and more politically engaged voters.
An alternative explanation for swing and less sophisticated voters not engaging in economic
voting is that economic competence is not a salient issue among these voters. Rising immigration
in Denmark has become a second political cleavage in recent years, so it is possible that such voters
are instead principally concerned with this issue. However, voter opinions and contextual data find
no support for this possibility. Results in the Online Appendix show that if anything economic
voting is more prevalent among those supporting the reinstatement of separate and lower state
32
benefits for immigrants,33 and in parishes (or municipalities) with higher shares of immigrants.
An apparent tension underlying our results is that those with the least accurate beliefs about
current unemployment update their beliefs most, but economic voting is concentrated among better
educated, more politically engaged and non-extreme voters. However, these results are consistent
with H5 and a large psychological literature pointing to the importance of cognitive awareness
and subjective biases (see Healy and Malhotra 2013). Furthermore, Lassen and Serritzlew (2011)
similarly find that the merging of Danish municipalities only decreased the political efficacy of
well-educated and politically informed voters. We argue that our results highlight an important
limit on the provision of political information: only a subset of those who update their beliefs
translate them into actions, and those who update the most are not necessarily most likely to act on
the updates.
6 Conclusion
Given that politically-relevant information cannot be transmitted to voters in a vacuum, a key
question for democratic accountability is when different sources cause voters to update and act
politically on their beliefs. We find that providing voters with unemployment forecasts causes all
types of voters—regardless of prior partisan affiliations—to update their unemployment expecta-
tions, particularly when faced with an expert source like the DCB. However, only a subset of more
sophisticated voters, which excludes voters that have recently switched their votes, understand that
political parties differ in their incentives to portray the state of the economy. Ultimately, although
all voters update their beliefs, only among non-ideologically extreme sophisticated voters does this
affect political preferences. We find clear evidence of a large economic voting response, which re-
flects changes in evaluations of government competence rather than changes in policy preferences.
At least in the case of Denmark, we conclude that it is primarily the objective credibility of a source
33We use 2012 survey responses here because the 2013 question is post-treatment.
33
and the sophistication of voters, not prior partisanship, that matters most for explaining when new
information will affect political behavior.
The democratic implications of these results are somewhat mixed. While economic voting is
generally regarded as a positive for democratic accountability (Anderson 2007), our results show
that information about aggregate unemployment is insufficient to induce non-sophisticated voters
to link their unemployment expectations to government performance. Nevertheless, finding any
effect in Denmark’s complex institutional environment and open economy is an important result
because it may represent a lower bound cross-nationally.
Our results also illuminate the behavior of political parties. That the least politically engaged
voters do not translate their information into political action could explain why political parties in
developed polities target their platforms toward prominent and well-informed voters (Adams and
Ezrow 2009; Gilens 2005). Furthermore, our results suggest that parties can benefit electorally
from providing specific macroeconomic information, and this is of course prevalent among suc-
cessful governments. However, given incorrect information also affects voter beliefs, our results
question why parties do not distort the facts more often. While this may entail losing credibility in
some instances (see Druckman 2001), the line between proclaiming truths and falsehoods is often
unclear if multiple numbers are available. Further research should explore these issues in greater
detail.
While this paper provides a first step toward understanding how voter beliefs are formed and
map to political preferences, there are important further steps to take. First, outside experimen-
tal intervention, it is critical to understand how individuals can be induced to acquire politically
relevant information. Second, the media represent a crucial mediator and communicator of in-
formation, and future work should also enlighten the black box of how voters respond to media
exposure. Third, the belief updating process itself deserves further attention as political scientists
know little about how political information is processed or what a distribution of beliefs looks like
at either a particular point in time or over multiple horizons. Finally, although electoral politics is
34
often based on short-term responses to stimuli, an important next step in this research agenda is to
assess the durability of our results over time and in the face of repeated interventions.
35
References
Adams, James and Lawrence Ezrow. 2009. “Who do European parties represent? How West-
ern European parties represent the policy preferences of opinion leaders.” Journal of Politics
71(01):206–223.
Anderson, Christopher. 1995. Blaming the Government: Citizens and the Economy in Five Euro-
pean Democracies. ME Sharpe.
Anderson, Christopher J. 2007. “The End of Economic Voting? Contingency Dilemmas and the
Limits of Democratic Accountability.” Annual Review of Political Science 10:271–296.
Angrist, Joshua D. and Guido W. Imbens. 1995. “Two-Stage Least Squares Estimation of Average
Causal Effects in Models With Variable Treatment Intensity.” Journal of the American Statistical
Association 90(430):431–442.
Angrist, Joshua D., Guido W. Imbens and Donald B. Rubin. 1996. “Identification of Causal Effects
Using Instrumental Variables.” Journal of the American Statistical Association 91(June):444–
455.
Ansolabehere, Stephen and James M. Snyder Jr. 2000. “Valence politics and equilibrium in spatial
election models.” Public Choice 103(3-4):327–336.
Ansolabehere, Stephen, Marc Meredith and Erik Snowberg. forthcoming. “Mecro-Economic Vot-
ing: Local Information and Micro-Perceptions of the Macro-Economy.” Economics and Politics
.
Banerjee, Abhijit V., Selvan Kumar, Rohini Pande and Felix Su. 2011. “Do Informed Voters Make
Better Choices? Experimental Evidence from Urban India.” Working paper.
Baron, David P. 2006. “Persistent Media Bias.” Journal of Public Economics 90(1):1–36.
36
Barro, Robert J. 1973. “The Control of Politicians: An Economic Model.” Public Choice 14:19–
42.
Barro, Robert J. and David B. Gordon. 1983. “Rules, discretion and reputation in a model of
monetary policy.” Journal of Monetary Economics 12(1):101–121.
Bartels, Larry M. 1996. “Uninformed Votes: Information Effects in Presidential Elections.” Amer-
ican Journal of Political Science 40(1):194–230.
Bartels, Larry M. 2008. Unequal Democracy: The Political Economy of the New Gilded Age.
Princeton, NJ: Princeton University Press.
Besley, Timothy and Andrea Prat. 2006. “Handcuffs for the Grabbing Hand? Media Capture and
Government Accountability.” American Economic Review 96(3):720–736.
Boudreau, Cheryl and Scott A. MacKenzie. 2014. “Informing the Electorate? How Party Cues
and Policy Information Affect Public Opinion about Initiatives.” American Journal of Political
Science 58(1):48–62.
Bullock, John. 2011. “Elite Influence on Public Opinion in an Informed Electorate.” American
Political Science Review 105(3):496–515.
Campbell, Angus, Philip E. Converse, Warren E. Miller and Donald E. Stokes. 1960. The American
Voter. New York: Wiley.
Chong, Alberto, Ana L. De La O, Dean Karlan and Leonard Wantchekon. 2011. “Looking beyond
the incumbent: The effects of exposing corruption on electoral outcomes.”.
Dalton, Russell J. 2008. Citizen Politics: Public Opinion and Political Parties in Advanced Indus-
trial Democracies. Sage.
37
de Chaisemartin, Clement. 2013. “Defying the LATE? Identification of local treatment effects
when the instrument violates monotonicity.” Working paper.
Delli Carpini, Michael X. and Scott Keeter. 1996. What Americans Know about Politics and Why
It Matters. New Haven, CT: Yale University Press.
Druckman, James N. 2001. “On the limits of framing effects: who can frame?” Journal of Politics
63(4):1041–1066.
Duch, Raymond M. and Randolph T. Stevenson. 2008. The Economic Vote: How Political and
Economic Institutions Condition Election Results. Cambridge University Press.
Duch, Raymond M. and Randolph T. Stevenson. 2010. “The global economy, competency, and the
economic vote.” Journal of Politics 72(01):105–123.
Duch, Raymond M., Wojtek Przepiorka and Randolph T. Stevenson. forthcoming. “Responsibility
Attribution for Collective Decision Makers.” American Journal of Political Science .
Durante, Ruben and Brian Knight. 2012. “Partisan control, media bias, and viewer responses:
Evidence from Berlusconi’s Italy.” Journal of the European Economic Association 10(3):451–
481.
European Commission. 2011. “Eurobarometer 74, Autumn 2010: Public Opinion in the European
Union.”.
Fearon, James D. 1999. Electoral accountability and the control of politicians: Selecting good
types versus sanctioning poor performance. In Democracy, Accountability, and Representation,
ed. Adam Przeworski, Susan C. Stokes and Bernard Manin. New York: Cambridge University
Press pp. 55–97.
Feddersen, Timothy J. and Wolfgang Pesendorfer. 1996. “The Swing Voter’s Curse.” American
Economic Review 86(3):408–424.
38
Ferejohn, John. 1986. “Incumbent Performance and Electoral Control.” Public Choice 50(1/3):5–
25.
Ferraz, Claudio and Frederico Finan. 2008. “Exposing Corrupt Politicians: The Effects of Brazil’s
Publicly Released Audits on Electoral Outcomes.” Quarterly Journal of Economics 123(2):703–
745.
Fiorina, Morris P. 1981. Retrospective Voting in American National Elections. New Haven, CT:
Yale University Press.
Gentzkow, Matthew and Jesse M. Shapiro. 2006. “Media Bias and Reputation.” Journal of Political
Economy 114(2):280–316.
Gilens, Martin. 2005. “Inequality and Democratic Responsiveness.” Public Opinion Quarterly
69(5):778–796.
Goidel, Robert K. and Ronald E. Langley. 1995. “Media Coverage of the Economy and Aggregate
Economic Evaluations: Uncovering Evidence of Indirect Media Effects.” Political Research
Quarterly pp. 313–328.
Gomez, Brad T. and J. Matthew Wilson. 2001. “Political Sophistication and Economic Voting in
the American Electorate: A Theory of Heterogeneous Attribution.” American Journal of Politi-
cal Science 45(4):899–914.
Gomez, Brad T. and J. Matthew Wilson. 2006. “Cognitive heterogeneity and economic voting:
A comparative analysis of four democratic electorates.” American Journal of Political Science
50(1):127–145.
Healy, Andrew and Gabriel S. Lenz. 2014. “Substituting the End for the Whole: Why Vot-
ers Respond Primarily to the Election-Year Economy.” American Journal of Political Science
58(1):31–47.
39
Healy, Andrew and Neil Malhotra. 2013. “Retrospective Voting Reconsidered.” Annual Review of
Political Science 16:285–306.
Imai, Masami, Rosa C. Hayes and Cameron A. Shelton. 2014. “Attribution Error in Economic
Voting: Evidence from Trade Shocks.”.
Imbens, Guido W. and Joshua D. Angrist. 1994. “Identification and Estimation of Local Average
Treatment Effects.” Econometrica 62(2):467–475.
Kiewiet, D. Roderick. 1983. Macroeconomics and Micropolitics: The Electoral Effects of Eco-
nomic Issues. Chicago, IL: University of Chicago Press.
Krause, George A. 1997. “Voters, information heterogeneity, and the dynamics of aggregate eco-
nomic expectations.” American Journal of Political Science 41(4):1170–1200.
Kreiner, Claus T., David D. Lassen and Soren Leth-Peterson. 2013. “The Danish Panel Study of
Income and Asset Expectations.”.
Larcinese, Valentino, Riccardo Puglisi and James M. Snyder. 2011. “Partisan bias in economic
news: Evidence on the agenda-setting behavior of US newspapers.” Journal of Public Eco-
nomics 95(9):1178–1189.
Lassen, David D. and Søren Serritzlew. 2011. “Size and Equal Opportunity in the Democratic
Process: The Effect of the Danish Local Government Reform on Inequality in Internal Political
Efficacy.” World Political Science Review 7(1):1–15.
Lewis-Beck, Michael S. and Martin Paldam. 2000. “Economic voting: an introduction.” Electoral
studies 19(2):113–121.
Lewis-Beck, Michael S. and Mary Stegmaier. 2000. “Economic Determinants of Electoral Out-
comes.” Annual Review of Political Science 3:183–219.
40
Malhotra, Neil and Alexander G. Kuo. 2008. “Attributing Blame: The Public’s Response to Hur-
ricane Katrina.” Journal of Politics 70(1):120–135.
Meltzer, Allan H. and Scott F. Richard. 1981. “A rational theory of the size of government.”
Journal of Political Economy 89:914–927.
Moene, Karl O. and Michael Wallerstein. 2001. “Inequality, Social Insurance, and Redistribution.”
American Political Science Review 95(4):859–874.
Mullainathan, Sendhil and Andrei Shleifer. 2005. “The Market for News.” American Economic
Review 95(4):1031–1053.
Nadeau, Richard, Richard G. Niemi and Antoine Yoshinaka. 2002. “A cross-national analysis
of economic voting: taking account of the political context across time and nations.” Electoral
Studies 21(3):403–423.
Nadeau, Richard, Richard G. Niemi, David P. Fan and Timothy Amato. 1999. “Elite Economic
Forecasts, Economic News, Mass Economic Judgments, and Presidential Approval.” Journal of
Politics 61(1):109–135.
Persson, Torsten and Guido Tabellini. 2000. Political Economics: Explaining Economic Policy.
Cambridge, MA: The MIT Press.
Powell Jr., G. Bingham and Guy D. Whitten. 1993. “A Cross-National Analysis of Economic Vot-
ing: Taking Account of the Political Context.” American Journal of Political Science 37(2):391–
414.
Prior, Markus. 2013. “Media and political polarization.” Annual Review of Political Science
16:101–127.
Rogoff, Kenneth. 1990. “Equilibrium Political Budget Cycles.” American Economic Review
80(1):21–36.
41
Rogoff, Kenneth and Anne Sibert. 1988. “Elections and Macroeconomic Policy Cycles.” Review
of Economic Studies 55(1):1–16.
Romer, Thomas. 1975. “Individual welfare, majority voting, and the properties of a linear income
tax.” Journal of Public Economics 4(2):163–185.
Rudolph, Thomas J. 2003a. “Institutional Context and the Assignment of Political Responsibility.”
Journal of Politics 65(1):190–215.
Rudolph, Thomas J. 2003b. “Who’s Responsible for the Economy? The Formation and Conse-
quences of Responsibility Attributions.” American Journal of Political Science 47(4):698–713.
Rudolph, Thomas J. 2006. “Triangulating political responsibility: The motivated formation of
responsibility judgments.” Political Psychology 27(1):99–122.
Taber, Charles S. and Milton Lodge. 2006. “Motivated Skepticism in the Evaluation of Political
Beliefs.” American Journal of Political Science 50(3):755–769.
Tilley, James and Sara B. Hobolt. 2011. “Is the Government to Blame? An Experimental Test of
How Partisanship Shapes Perceptions of Performance and Responsibility.” Journal of Politics
73(02):316–330.
Tversky, Amos and Daniel Kahneman. 1974. “Judgment under uncertainty: Heuristics and biases.”
science 185(4157):1124–1131.
Zaller, John. 1999. “A Theory of Media Politics.” Unpublished manuscript.
42
7 Appendix
7.1 Bayesian interpretation of information updating
Our approach can be clearly shown in a Bayesian updating framework. Specifically, we write
individual i’s conditional posterior belief about future unemployment level, U , as:
P(U = u|Xi,Zi) = P(U = u|Xi)P(Zi|U = u,Xi)
P(Zi|Xi),
where Zi is an information shock received by i, and Xi captures i’s characteristics (e.g. ideology
and sophistication). The location and specificity of i’s prior belief, P(U = u|Xi), depends upon
Xi. The likelihood P(Zi|U = u,Xi) represents i’s interpretation of the informativeness of the signal
they received: P(Zi|U = u,Xi)/P(Zi|Xi) = 1 or P(Zi|U = u,Xi) = P(Zi|Xi) captures i not believing
that receiving signal Zi is related to the likelihood that the state of the world is U = u. H1 and H2
hypothesize that P(Zi|U = u,Xi) (or more simply P(Zi|U = u) because individual characteristics
play a weak role in objective credibility) is large where Zi comes from an expert or surprising
source. H4 instead hypothesizes that updating depends upon the interaction of the source of Zi
and Xi. H3 allows for P(U = u|Xi) to be large, but also implies that Zi does not add much new
information.
7.2 Formal model of information and policy preferences
We extend the Romer (1975) and Meltzer and Richard (1981) framework to include uncertainty
over income in a simple way.
Take a continuum of voters of unit mass, differentiated by their income prospects. Voter i’s
realized income is yi ∈ Y ⊆ R+. We build in uncertainty in a simple fashion. Individual i’s
uncertainty at the time of determining their policy preferences is operationalized as follows: with
43
probability pi(zi) ∈ (0,1) their income is yLi , and with probability 1− pi(zi) their income is yH
i ,
where yHi > yL
i and zi denotes the amount of information i possess about the economy (increases in
zi represent more information). Assume pi(zi) is differentiable and monotonic in zi. To save space,
define i’s expected income as Yi(zi) ≡ pi(zi)yLi +(1− pi(zi))yH
i .
Voters are also uncertain about the aggregate distribution of income. In particular, with voter
i assigns probability q(zi) ∈ (0,1) to economy-wide average income being yL, and probability
1− q(zi) to economy-wide average income being yH , where yH > yL. Assume q is differentiable
and monotonic in zi. Define i’s expected average income in the economy as Y (zi)≡ q(zi)yL+(1−
q(zi))y
The government must choose a tax and benefit policy pair (τ ,T ) to be implemented after
income is realized, where τ ∈ [0,1] is a proportional tax rate levied on y and T ≥ 0 is a lump-sum
transfer made to all citizens. There is a cost φ (τ)y to increasing τ , where∫
y∈Y ydF(y) = y is
the realized mean income and φ : [0,1] 7→ R+ is a convex-increasing function such that φ ′(τ) >
0,φ ′′(τ)> 0 and φ (0) = 0. This cost could be labor supply disincentives, capital misallocation or
the inefficiency of revenue collection. We assume that the tax rate cannot depend upon the realized
state of the world.
We now derive individual i’s preferences over policies before income is realized. From the
perspective of voter i, the ex ante government budget constraint is:
[τ−φ (τ)]Y (zi) ≤ T . (3)
Since the budget constraint will bind in equilibrium, the problem is reduced to a single dimensional
problem in τ .
A voter with income y has the following policy utility function, receiving utility from post-tax
44
income and the lump-sum transfer:
u((1− τ)Yi(zi)+ [τ−φ (τ)]Y (zi)+ (1− τ)Y (zi)+ [τ−φ (τ)]Y (zi)
)= u((1− τ)Yi(zi)+ [τ−φ (τ)]Y (zi)
), (4)
where u : R 7→ R is a concave-increasing function: u′(·) > 0,u′′(·) < 0 and u(0) = 0. Tax rates
have two effects on voter utility: redistribution of income and a (disincentive) cost to increasing
taxation.
Given preferences are strictly concave in τ , they are single-peaked. We can identify the ideal
policy of voter i as:
τ∗i = max
{(φ ′)−1
(1− Yi(zi)
Y (zi)
),0}
. (5)
This reiterates the Romer-Meltzer-Richard logic that i’s preference for taxation is increasing as
their expected income relative to the expected average income falls. Note that all Yi(zi) > Y (zi)
prefer τ∗i = 0.
In the space where τ∗i > 0, or for voters with expected incomes exceeding the average expected
income, the comparative static with respect to new information is:
dτ∗idzi
=p′i(zi)(yH
i − yLi )Y (zi)−q′(zi)(yH− yL)Yi(zi)
φ ′′(τ)[Y (zi)]2. (6)
Given yHi − yL
i > 0 and yH − yL > 0, and expected individual and aggregate income is positive, it
is clear that we have opposing effects when sgn(p′i(zi)) = sgn(q′(zi))—this is the obvious case
for this paper as it is very unlikely that aggregate information would cause voters to differentially
update. Intuitively, the first term in the numerator captures how information affects i’s taxation
preferences associated with their own expected income, while the second term captures how infor-
45
mation affects taxation preferences in the rest of the economy. These are easiest to see by setting
q′(zi) = 0 and p′i(zi) = 0 respectively.
Without loss of generality (the results will just be the opposite), let us focus on the case where
sgn(p′i(zi)) = sgn(q′(zi)) ≤ 0; this turns out to be the most appropriate case for our analysis
because our information treatments cause voters to become more positive about the economy. It
is now clear that new information causing voters to reduce their belief of being unemployed and
reduce their belief of others being unemployed has the expected effects: the fall in the likelihood
of i being unemployed reduces their preference for taxation (first term in (6)), while the fall in
the likelihood of others being unemployed increases their preference for taxation (second term in
(6)). These effects clearly conflict, with the individual income incentive overpowering the general
economy incentive when:
p′i(zi)
q′(zi)>
(yH− yL)Yi(zi)
(yHi − yL
i )Y (zi), (7)
or when the change in pi(zi) is sufficiently large relative to the change in q(zi).
These insights can be easily extended to voter in Denmark, where politics is primarily based
on a left-right axis. Simplifying the analysis such that voters choose between left and right parties,
it is clear that any increase in preferred tax rate should correspond to increased support for the
left-wing government.
7.3 Variable definitions and summary statistics
Information source treatments. Respondents were randomly assigned to a control group receiving
no information, one of six groups receiving the prime “Assume that that the [DCB/government/Liberals]
estimates unemployment in 2013 will be [almost 7%/around 5%]” or “The DCB estimates unem-
ployment in 2013 to be almost 7%.” Respectively, these statements translate as: “Antag at [Na-
tionalbanken/Regeringen/Venstre] vurderer at arbejdsløsheden i 2013 vil være [knap 7%/ca. 5%]”
46
and This translates from: “Nationalbanken vurderer at arbejdsløsheden i 2013 vil være knap 7%.”
In the main text, treatments are denoted, for example, by “DCB 7% treatment”.
Unemployment expectations. The percentage (not restricted to integers) reported by the re-
spondent in response to the question “What is your best estimate of what unemployment will be in
2013? We would like your best estimate, even if you are not entirely sure.” This translates from:
“Hvad er dit bedste bud pa hvad arbejdsløsheden vil blive i 2013? Vi vil gerne have dit bedste bud,
ogsa selvom du ikke er helt sikker.” This question immediately followed the treatment.
Current unemployment estimate. The percentage (not restricted to integers) reported by the
respondent in response to the question “Unemployment in Denmark is typically measured by the
unemployment rate, that is, the share of people who want to work but don’t have a job. Over the
last 25 years, the unemployment rate has been between 1.5 and 12 %. What is your estimate of
the current unemployment rate in Denmark? We would like your best estimate, even if you are not
entirely sure.” This question immediately preceded the treatment.
Vote intention. Respondent stated the party that would vote for in response to the question
“How would you vote tomorrow?” Respondents choose one of the following: Social Democrat
Party, Social Liberal Party, Conservative People’s Party, Socialist People’s Party, Danish People’s
Party, Liberal Party, Liberal Alliance, or Red-Green Alliance. Answers not stating a party were:
blank, no answer, other, would not vote, and don’t know. We counted the Conservative People’s
Party, Danish People’s Party, Liberal Party, and Liberal Alliance as right-wing parties. We counted
the Social Democratic Party, Social Liberal Party and Socialist People’s Party as parties in govern-
ment.
Confidence in the government. Responded were asked, on a scale from great mistrust (1) to
great trust (5), how much they trust the government. This question was asked 11 questions after
the treatment was administered.
Redistribution. This variable measures support for redistribution on a five-point scale ranging
from “every man for himself” (1) to the government “should help the poor a lot” (5). This was
47
in response to the prime “Some think the Government should do all it can to raise the standard of
living for poor Danes: that is 1 on the scale. Others think it is not the responsibility of government,
each should take care of themselves: that is 5.” This question was 19 questions after the treatment.
Unemployment insurance. Three-point “less-same-more” response to the question “The eco-
nomic crisis has meant that many people have lost their job. Do you think that the government
should support the unemployed?” this question was asked immediately after the question eliciting
the respondent’s unemployment expectations.
Lower immigrant benefits. Indicator coded 1 for respondents who responded that separate and
lower benefits for immigrants should be reinstated in 2012.
Denmark economic prospects. Five point scale, from “worsen considerably” to “improve con-
siderably, response to the question of how the Danish economy overall will do in 2013.
Improving economic prospects. Indicator coded 1 for individuals responding that the Danish
economy for 2013 will be “much better” or “better” than 2012. This variable is re-coded from
Denmark economic prospects.
Discuss politics. The sum of the set of indicators coded 1 for respondents who answered that
they talk to family, friends, neighbors, work colleagues or others about politics.
News every day. Indicator coded 1 for respondents who state that they watch or read about
politics and economics in the news “every day”.
Wage income (log). Total wage income before taxation; we then added 1 and took the natural
logarithm.
Own job risk. The probability assigned by the respondent to the possibility that they will
experience a period of unemployment in the forthcoming year.
Education. Three-point scale indicating the level of education achieved by the respondent:
1 is less than university education; 2 is some or complete undergraduate university; 3 is further
academic study.
Woman. Indicator coded 1 for women.
48
Voted left at last election. Indicator coded 1 for respondents who voted for one of the Social
Democrats, Social Liberals, Socialist People’s, or Red-Green Alliance parties in the 2011 election.
Swing voter (previous votes). Indicator coded 1 for respondents who provided different re-
sponses to the question asked respondents to recall who they voted for in the 2007 and 2011
elections.
Swing voter (previous intentions). Indicator coded 1 for respondents who provided different
responses to the vote intention questions asked in 2011 and 2012.
Extreme voter. Indicator coded 1 for respondents who answered either “should help the poor a
lot” or “every man for himself” to the redistribution question in 2012.
Municipal immigration share. The share of immigrants in the municipality that the respondent
resides in. Denmark contains 98 municipalities.
Parish immigration share. The share of immigrants in the parish that the respondent resides in.
The average parish contains around 2,500 residents. Parishes are the lowest administrative units in
Denmark.
7.4 Support for the identification assumptions
Tables 8 and 9 look at balance over pre-treatment covariates from both the Register and the survey.
F tests of all treatment coefficients being equal are rejected with regularity consistent with chance:
only in one of 16 tests was the joint test statistically different from zero at the 5% level (and also
once at the 10% level). Even in those cases, the differences between treatment conditions are
small. Accordingly, we do not include controls, although the results are robust to including such
variables.
Figures 4 and 5 show the cumulative density functions plotting the proportion of individuals
for each instrument expecting unemployment below a certain level. The key point to note is that
while the 7% treatments lie almost entirely to the left of the control group, the 5% treatments do
not. This implies that the monotonicity assumption underpinning the instrumental variable does
49
Table 7: Summary statistics
Obs. Mean Std. dev. Min. Max.
Dependent variablesSocial Democrat Party 5,705 0.17 0.37 0 1Social Liberal Party 5,705 0.09 0.29 0 1Socialist People’s Party 5,705 0.06 0.24 0 1Liberal Party 5,705 0.28 0.45 0 1Red-Green Alliance 5,705 0.06 0.25 0 1Confidence in government 5,688 2.69 1.00 1 5Redistribution 5,705 3.20 1.02 1 5Unemployment insurance 5,614 2.23 0.61 1 3
Dependent/endogenous variableUnemployment expectations 5,705 7.98 3.55 0 45
Treatment variablesControl 5,705 0.13 0.33 0 1DCB 7% treatment (combined) 5,705 0.25 0.44 0 1DCB 5% treatment 5,705 0.13 0.33 0 1Government 7% treatment 5,705 0.12 0.33 0 1Government 5% treatment 5,705 0.12 0.33 0 1Opposition 7% treatment 5,705 0.12 0.33 0 1Opposition 5% treatment 5,705 0.12 0.33 0 1
CovariatesCurrent unemployment estimate 5,705 8.58 4.31 0 45Swing voter (previous votes) 3,827 0.43 0.50 0 1Swing voter (previous intentions) 4,566 0.23 0.42 0 1News every day 5,705 0.72 0.45 0 1Improving economic prospects 5,675 0.34 0.47 0 1Municipal immigrant share 5,704 9.74 5.53 3.67 32.75Parish immigrant share 5,704 8.76 7.09 0 69.72Medium education 5,642 0.67 0.47 0 1High education 5,642 0.11 0.31 0 1Woman 5,705 0.49 0.50 0 1Discuss politics 5,705 2.30 1.14 1 4Voted left at last election 5,705 0.50 0.50 0 1Extreme voter 4,566 0.17 0.38 0 1Wage income (log) 5,532 10.19 5.15 0 19.83Lower immigrant benefits 5,705 0.25 0.43 0 1
50
Tabl
e8:
Bal
ance
test
s1
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Cur
rent
UE
st.
Wom
anA
geB
asic
Edu
.M
ed.E
duL
ong
Edu
.Vo
ted
Gov
t.Vo
ted
Lef
t
Con
trol
8.70
4***
0.49
3***
1961
.541
***
0.24
3***
0.66
0***
0.09
7***
0.43
2***
0.49
2***
(0.1
77)
(0.0
19)
(0.4
34)
(0.0
16)
(0.0
18)
(0.0
11)
(0.0
19)
(0.0
19)
DC
B7%
trea
tmen
t-0
.264
-0.0
020.
869
0.00
3-0
.016
0.01
30.
021
0.01
5(0
.230
)(0
.026
)(0
.608
)(0
.023
)(0
.025
)(0
.016
)(0
.026
)(0
.026
)D
CB
7%tr
eatm
ent(
true
)-0
.364
-0.0
090.
171
0.00
20.
012
-0.0
140.
012
-0.0
00(0
.226
)(0
.027
)(0
.617
)(0
.023
)(0
.025
)(0
.015
)(0
.027
)(0
.027
)D
CB
5%tr
eatm
ent
-0.2
490.
025
0.37
4-0
.047
**0.
026
0.02
10.
049*
0.04
2(0
.229
)(0
.026
)(0
.601
)(0
.022
)(0
.025
)(0
.016
)(0
.026
)(0
.026
)G
over
nmen
t7%
trea
tmen
t-0
.086
0.00
90.
234
-0.0
330.
010
0.02
40.
004
-0.0
02(0
.240
)(0
.027
)(0
.601
)(0
.022
)(0
.025
)(0
.017
)(0
.026
)(0
.027
)G
over
nmen
t5%
trea
tmen
t0.
224
-0.0
04-0
.190
-0.0
300.
012
0.01
70.
031
0.01
7(0
.252
)(0
.027
)(0
.626
)(0
.022
)(0
.025
)(0
.016
)(0
.026
)(0
.027
)O
ppos
ition
7%tr
eatm
ent
-0.2
65-0
.024
0.44
3-0
.009
0.01
3-0
.004
-0.0
00-0
.007
(0.2
39)
(0.0
27)
(0.6
07)
(0.0
23)
(0.0
25)
(0.0
16)
(0.0
26)
(0.0
27)
Opp
ositi
on5%
trea
tmen
t0.
011
-0.0
120.
529
-0.0
21-0
.010
0.03
1*0.
050*
0.03
6(0
.249
)(0
.027
)(0
.602
)(0
.023
)(0
.025
)(0
.017
)(0
.026
)(0
.027
)
Obs
erva
tions
5,70
55,
705
5,69
25,
642
5,64
25,
642
5,70
55,
705
Fte
stof
equa
lity
over
trea
tmen
tsp=
0.19
p=
0.75
p=
0.78
p=
0.18
p=
0.74
p=
0.07
p=
0.29
p=
0.47
Not
es:A
llsp
ecifi
catio
nses
timat
edus
ing
OL
S.R
obus
tsta
ndar
der
rors
inpa
rent
hese
s.∗ p
<0.
1,∗∗
p<
0.05
,∗∗∗
p<
0.01
.
51
Tabl
e9:
Bal
ance
test
s2
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Vote
dri
ght
Wag
es(l
og)
Exp
.Inc
ome
(log
)H
omeo
wne
rE
con.
pros
pect
sTe
nure
dJo
bR
isk
Ris
kav
erse
Con
trol
8.70
4***
0.49
3***
1961
.541
***
0.24
3***
0.66
0***
0.09
7***
0.43
2***
0.49
2***
(0.1
77)
(0.0
19)
(0.4
34)
(0.0
16)
(0.0
18)
(0.0
11)
(0.0
19)
(0.0
19)
DC
B7%
trea
tmen
t-0
.264
-0.0
020.
869
0.00
3-0
.016
0.01
30.
021
0.01
5(0
.230
)(0
.026
)(0
.608
)(0
.023
)(0
.025
)(0
.016
)(0
.026
)(0
.026
)D
CB
7%tr
eatm
ent(
true
)-0
.364
-0.0
090.
171
0.00
20.
012
-0.0
140.
012
-0.0
00(0
.226
)(0
.027
)(0
.617
)(0
.023
)(0
.025
)(0
.015
)(0
.027
)(0
.027
)D
CB
5%tr
eatm
ent
-0.2
490.
025
0.37
4-0
.047
**0.
026
0.02
10.
049*
0.04
2(0
.229
)(0
.026
)(0
.601
)(0
.022
)(0
.025
)(0
.016
)(0
.026
)(0
.026
)G
over
nmen
t7%
trea
tmen
t-0
.086
0.00
90.
234
-0.0
330.
010
0.02
40.
004
-0.0
02(0
.240
)(0
.027
)(0
.601
)(0
.022
)(0
.025
)(0
.017
)(0
.026
)(0
.027
)G
over
nmen
t5%
trea
tmen
t0.
224
-0.0
04-0
.190
-0.0
300.
012
0.01
70.
031
0.01
7(0
.252
)(0
.027
)(0
.626
)(0
.022
)(0
.025
)(0
.016
)(0
.026
)(0
.027
)O
ppos
ition
7%tr
eatm
ent
-0.2
65-0
.024
0.44
3-0
.009
0.01
3-0
.004
-0.0
00-0
.007
(0.2
39)
(0.0
27)
(0.6
07)
(0.0
23)
(0.0
25)
(0.0
16)
(0.0
26)
(0.0
27)
Opp
ositi
on5%
trea
tmen
t0.
011
-0.0
120.
529
-0.0
21-0
.010
0.03
1*0.
050*
0.03
6(0
.249
)(0
.027
)(0
.602
)(0
.023
)(0
.025
)(0
.017
)(0
.026
)(0
.027
)
Obs
erva
tion
5,70
55,
532
5,55
45,
705
5,67
54,
540
4,54
05,
580
Fte
stof
equa
lity
over
trea
tmen
tsp=
0.20
p=
0.95
p=
0.75
p=
0.44
p=
0.38
p=
0.68
p=
0.26
p=
0.03
Not
es:S
eeTa
ble
8.
52
020
4060
8010
0
Cum
ulat
ive
dens
ity
0 10 20 30 40 50
Unemployment expectations (%)
Control DCB 7% treatmentDCB 5% treatment DCB 7% treatment (true)
Figure 4: Cumulative density plots of unemployment expectations by information treatment 1
53
020
4060
8010
0
Cum
ulat
ive
dens
ity
0 10 20 30 40
Unemployment expectations (%)
Control Government 7% treatmentGovernment 5% treatment Opposition 7% treatmentOpposition 5% treatment
Figure 5: Cumulative density plots of unemployment expectations by information treatment 2
54
Table 10: Effects of treatments on belief that political information is important
(1) (2)Info. important Info. important
Control 0.742*** 0.742***(0.022) (0.022)
DCB 7% treatment (combined) -0.004(0.021)
DCB 5% treatment -0.005(0.025)
Government 7% treatment 0.018(0.025)
Government 5% treatment 0.022(0.025)
Opposition 7% treatment -0.009(0.025)
Opposition 5% treatment -0.014(0.025)
Any treatment -0.006(0.019)
Observations 5,803 5,803
Notes: Dependent variable is a dummy for whether the respondent believes political information is important foreither private economic decisions or as part of the respondent’s job. Both models control for current unemploy-ment estimate. Robust standard errors in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
not hold in such cases. The implications of this are discussed in the main text.
As noted in the main text, Table 10 shows no treatment affects the respondent’s belief that
political information is important. This serves as an important robustness check for the exclusion
restriction concern that simply receiving the treatment inducing respondents to think about politics
differentially without being affected by the particular unemployment information provided.
Finally, Table 11 provides our first stage estimates for vote intention regressions. The results
are very similar to the coefficients provided in Table 1 of the main paper, but gain precision due to
55
Table 11: Effect of information treatments on unemployment expectations (%)—controlling forcurrent unemployment estimate (first stage)
Unemployment expectations (%)
Control 3.523***(0.216)
DCB 7% treatment (combined) -0.927***(0.104)
DCB 5% treatment -1.501***(0.127)
Government 7% treatment -0.792***(0.122)
Government 5% treatment -1.360***(0.126)
Opposition 7% treatment -0.756***(0.120)
Opposition 5% treatment -1.342***(0.137)
Current unemployment estimate 0.631***(0.025)
Observations 5,705First stage F statistic 32.64
Notes: Estimated using OLS. Robust standard errors in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. Thecoefficient tests at the foot of the table report the p value from a two-sided F test of coefficient equality.
the inclusion of the current unemployment estimate.
7.5 Additional results
7.5.1 Effects of information source on unemployment expectations
Table 12 replicates Table 3 in the main paper with the exception that a respondent’s current unem-
ployment estimate is included as an additional interactive control. The results clearly show that the
measures of political sophistication cease to be significant predictors of updating once the current
56
Table 12: Heterogeneous effects of information treatments on unemployment expectations (%),by conditional marginal effect—controlling for current unemployment estimate
DCB 7% DCB 5% Govt. 7% Govt. 5% Opp. 7% Opp. 5%
Linear effect 1.928*** -0.891 2.525*** 0.693 1.771* 0.31(0.747) (0.934) (0.735) (0.869) (0.911) (1.026)
× Current unemployment estimate -0.498*** -0.265*** -0.517*** -0.348*** -0.384*** -0.367***(0.058) (0.081) (0.076) (0.07) (0.08) (0.083)
× News every day 0.081 0.016 -0.368 -0.004 -0.267 0.072(0.201) (0.274) (0.26) (0.275) (0.249) (0.298)
× Denmark economic prospects 0.398*** 0.505*** 0.538*** 0.242 0.359** 0.317*(0.117) (0.161) (0.143) (0.172) (0.152) (0.177)
×Wage income (log) -0.003 -0.025 -0.022 0.022 -0.005 0.022(0.017) (0.022) (0.02) (0.02) (0.022) (0.026)
×Medium education 0.097 0.328 0.103 0.106 -0.072 0.378(0.208) (0.293) (0.275) (0.297) (0.269) (0.36)
× High education 0.072 0.408 -0.315 0.007 0.017 0.617(0.257) (0.332) (0.313) (0.341) (0.303) (0.457)
×Woman -0.006 0.042 -0.031 -0.156 0.017 -0.233(0.191) (0.257) (0.212) (0.241) (0.228) (0.259)
× Voted left at last election 0.228 0.128 -0.038 0.058 -0.024 0.14(0.172) (0.224) (0.208) (0.231) (0.204) (0.249)
Notes: All coefficients are estimated from a single OLS equation interacting all treatments conditions with thevariables on the left hand side of the table (see Online Appendix for their definitions). The coefficient for thecontrol group is 3.204***(0.450). The sample size is 5,446. Robust standard errors in parentheses. ∗p< 0.1,∗∗ p<0.05,∗∗∗ p < 0.01.
unemployment estimate is included. As noted in the main text, this suggests that the current un-
employment estimate—which is a highly statistically significant interaction for each treatment—is
almost a sufficient statistic for political sophistication in this context.
Table 13 shows that of the treatment sources, only the DCB treatment significantly increases
trust in the source of the information. Trust is a dummy variable for trusting or greatly trusting
the institution. This test was designed to ameliorate the concern that simply hearing the source’s
name, independently of the information, is driving the results. Although this is not quite possible
57
Table 13: Effect of information treatments on confidence in sources
(1) (2) (3)Trust DCB Trust government Trust opposition
Control 0.662*** 0.166*** 0.263***(0.018) (0.014) (0.016)
DCB 7% treatment (combined) 0.070***(0.021)
DCB 5% treatment 0.072***(0.024)
Government 7% treatment 0.008(0.020)
Government 5% treatment 0.032(0.020)
Opposition 7% treatment 0.015(0.023)
Opposition 5% treatment (0.015)(0.023)
Observations 2,980 2,177 2,180
Notes: Estimated using OLS. Robust standard errors in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
for the DCB, it large effects combined with the high level of initial trust, suggest that this should
not be a problem.
Table 14 shows the summary statistics in terms of political disposition for the four subsamples
that we analyze in the main paper.
7.5.2 Effects of information source on political preferences
Table 15 shows the reduced form estimates. Panel A, which fully separates treatments, generally
shows that the more powerful treatment has a larger effect on support for a political party. That
is to say the treatment effects look to be fairly monotonic given the fact that most individuals
over-estimated the future unemployment rate relative to the true projection. Although most rela-
58
Tabl
e14
:Com
pari
son
ofsu
b-sa
mpl
ech
arac
teri
stic
s
Est
imat
e¿7.
4%E
stim
ate≤
7.4%
Est
imat
e∈[5
%,7
%]
Est
imat
e∈[5
.4%
,9.4
%]
Mea
nSt
.dev
.M
ean
St.d
ev.
Mea
nSt
.dev
.M
ean
St.d
ev.
Med
ium
educ
atio
n0.
666
0.47
20.
662
0.47
30.
660
0.47
40.
667
0.47
1H
igh
educ
atio
n0.
087
0.28
20.
134
0.34
10.
135
0.34
20.
125
0.33
0D
iscu
sspo
litic
s2.
211
1.13
82.
383
1.13
52.
385
1.13
72.
370
1.13
8Vo
ted
left
atla
stel
ectio
n0.
509
0.50
00.
499
0.50
00.
502
0.50
00.
515
0.50
0Vo
ted
righ
tatl
aste
lect
ion
0.40
30.
491
0.42
80.
495
0.42
60.
495
0.41
10.
492
New
sev
ery
day
0.66
80.
471
0.76
80.
422
0.77
70.
417
0.75
20.
432
Wom
an0.
568
0.49
50.
410
0.49
20.
407
0.49
10.
430
0.49
5
59
tionships are not statistically significant, this is due to three reasons. First, as noted in the text, the
7% treatments cause updating from both directions and thus average over countervailing effects.
Second, the reduced form averages give greater weight to those with a large first stage, which are
generally the individuals who seem to be those least capable of mapping information to political
preferences. And finally, we use many treatments and thus relatively small sample sizes for each
separate treatment, whereas the 2SLS estimates pool information across treatments. In fact, our
2SLS estimates are highly consistent with these reduced form estimates—it is easy to see this by
noting the monontonic relationship between the treatments. This is particularly the case once we
group together treatment levels: Panel B shows that grouping together the 5% and 7% treatments
across source produces clearly statistically significant results.
Table 16 shows the heterogeneous effect estimates underlying the results shown in Figure 3 in
the main paper, as well as comparable results for intending to vote for a right party.
Table 17 show how the effect of unemployment expectations varies by local immigration ex-
periences and respondent views on immigration policy. The results clearly show that there is no
significant difference in economic voting by either measure of immigration.
60
Tabl
e15
:Red
uced
form
effe
ctof
unem
ploy
men
texp
ecta
tions
onpo
litic
alpr
efer
ence
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
Pane
lA:a
lltr
eatm
ents
Gov
t.So
c.D
em.
Soc.
Lib
.So
c.Pe
op.
Con
f.go
vt.
Red
-Gre
enR
ight
Lib
eral
sR
edis
t.U
.ins
uran
ce
DC
B7%
trea
tmen
t0.
020
0.03
1*-0
.007
-0.0
040.
091*
*-0
.015
-0.0
18-0
.017
-0.0
91**
-0.0
19(0
.021
)(0
.017
)(0
.013
)(0
.011
)(0
.045
)(0
.022
)(0
.021
)(0
.011
)(0
.045
)(0
.028
)D
CB
5%tr
eatm
ent
0.04
9**
0.01
80.
024
0.00
70.
108*
*-0
.050
*-0
.043
*-0
.012
-0.0
750.
009
(0.0
25)
(0.0
19)
(0.0
16)
(0.0
13)
(0.0
52)
(0.0
26)
(0.0
24)
(0.0
13)
(0.0
52)
(0.0
31)
Gov
ernm
ent7
%tr
eatm
ent
0.00
70.
013
-0.0
120.
006
0.05
00.
005
0.00
2-0
.007
-0.0
84-0
.008
(0.0
24)
(0.0
19)
(0.0
15)
(0.0
13)
(0.0
51)
(0.0
26)
(0.0
24)
(0.0
14)
(0.0
53)
(0.0
32)
Gov
ernm
ent5
%tr
eatm
ent
0.02
70.
009
0.01
9-0
.001
0.07
6-0
.032
-0.0
18-0
.007
-0.0
350.
025
(0.0
25)
(0.0
19)
(0.0
16)
(0.0
12)
(0.0
52)
(0.0
26)
(0.0
24)
(0.0
14)
(0.0
53)
(0.0
32)
Opp
ositi
on7%
trea
tmen
t-0
.011
-0.0
02-0
.007
-0.0
03-0
.028
0.01
4-0
.015
-0.0
10-0
.069
0.00
9(0
.024
)(0
.019
)(0
.015
)(0
.012
)(0
.051
)(0
.026
)(0
.024
)(0
.013
)(0
.053
)(0
.033
)O
ppos
ition
5%tr
eatm
ent
0.04
1*0.
040*
*0.
000
0.00
10.
202*
**-0
.027
-0.0
30-0
.008
-0.0
560.
001
(0.0
25)
(0.0
20)
(0.0
15)
(0.0
12)
(0.0
53)
(0.0
26)
(0.0
24)
(0.0
13)
(0.0
53)
(0.0
33)
Obs
erva
tions
5,80
35,
803
5,80
35,
803
5,78
65,
803
5,80
35,
803
5,80
35,
709
(11)
(12)
(13)
(14)
(15)
(16)
(17)
(18)
(19)
(20)
Pane
lB:c
ombi
ned
trea
tmen
tlev
els
Gov
t.So
c.D
em.
Soc.
Lib
.So
c.Pe
op.
Con
f.go
vt.
Red
-Gre
enR
ight
Lib
eral
sR
edis
t.U
.ins
uran
ce
All
7%tr
eatm
ent
0.00
90.
019
-0.0
08-0
.001
0.05
1-0
.003
-0.0
12-0
.013
-0.0
84**
-0.0
09(0
.019
)(0
.015
)(0
.012
)(0
.010
)(0
.041
)(0
.021
)(0
.019
)(0
.011
)(0
.041
)(0
.025
)A
ll5%
trea
tmen
t0.
039*
*0.
022
0.01
50.
002
0.12
9***
-0.0
36*
-0.0
30-0
.009
-0.0
550.
012
(0.0
20)
(0.0
16)
(0.0
13)
(0.0
10)
(0.0
43)
(0.0
21)
(0.0
20)
(0.0
11)
(0.0
43)
(0.0
26)
Obs
erva
tions
5,80
35,
803
5,80
35,
803
5,78
65,
803
5,80
35,
803
5,80
35,
709
Not
es:
All
spec
ifica
tions
estim
ated
usin
gO
LS,
and
cont
rol
for
curr
ent
unem
ploy
men
tex
pect
atio
ns.
Rob
ust
stan
dard
erro
rsin
pare
nthe
ses.∗ p
<
0.1,∗∗
p<
0.05
,∗∗∗
p<
0.01
.
61
Tabl
e16
:Het
erog
eneo
usef
fect
s—2S
LS
estim
ates
(1)
(2)
(3)
(4)
(5)
(6)
Gov
t.G
ovt.
Gov
t.G
ovt.
Gov
t.G
ovt.
Une
mpl
oym
ente
xpec
tatio
ns(%
)-0
.059
**-0
.055
***
-0.0
53**
*-0
.014
-0.0
20-0
.014
(0.0
23)
(0.0
19)
(0.0
19)
(0.0
20)
(0.0
16)
(0.0
15)
×sw
ing
vote
r(pr
evio
usvo
tes)
0.03
3(0
.030
)×
swin
gvo
ter(
prev
ious
inte
ntio
ns)
0.04
6(0
.029
)×
ideo
logi
cally
extr
eme
vote
r0.
058*
(0.0
34)
×ne
ws
ever
yda
y-0
.033
(0.0
26)
×be
yond
high
scho
oled
ucat
ion
-0.0
23(0
.022
)×
impr
ovin
gec
onom
icpr
ospe
cts
-0.0
66**
(0.0
29)
Obs
erva
tions
3,82
74,
566
5,70
55,
642
5,67
5
Not
es:A
llsp
ecifi
catio
nses
timat
edus
ing
2SL
S,an
dco
ntro
lfor
curr
entu
nem
ploy
men
texp
ecta
tions
and
linea
rter
mfo
reac
hin
tera
ctio
n.R
obus
tsta
ndar
der
rors
inpa
rent
hese
s.∗ p
<0.
1,∗∗
p<
0.05
,∗∗∗
p<
0.01
.
62
Table 17: Heterogeneous effect of unemployment expectations by immigration exposure andpreferences
(1) (2) (3)Govt. Govt. Govt.
Unemployment expectations (%) 0.046 0.004 -0.027*(0.103) (0.096)
× parish immigrant share -0.008(0.010)
× municipality immigrant share -0.004(0.009)
× lower immigrant benefits -0.010
Observations 5,704 5,704 5,705
Notes: See Table 16.
63