Do Corruption Scandals Erode Trust in
Government?
Evidence from a Matched Sample of Local
Governments
Albert Solé-Ollé and Pilar Sorribas-Navarro
Do corruption scandals erode trust in government?
Evidence from a matched sample of local governments
Albert Solé-Ollé & Pilar Sorribas-Navarro
Universitat de Barcelona &
Institut d’Economia de Barcelona (IEB)
ABSTRACT:
We examine whether the occurrence of a corruption scandal involving
the incumbent reduces the level of trust in local politicians, and whether
this effect is larger when voters are more exposed to media information
about the scandal. We use a novel data set, with information on press
reporting of local corruption scandals that broke in Spain during the
period 1999-2009, and data on trust in local politicians coming from a
new survey undertook in late 2009. We use matching methods to
improve the identification of the effect of corruption scandals on trust,
comparing municipalities affected by a scandal with municipalities with
similar traits but where no scandal broke out. Our results suggest that
scandals have on average a negative but modest effect on trust in local
politicians. The effect is stronger for individuals obtaining information
from the media and for scandals widely covered by the newspapers.
Keywords: corruption, trust
JEL Classification: P16, D72
* Corresponding author: Albert Solé-Ollé ([email protected]).This research has received funding from
projects ECO2009-12680/ECON (Ministerio de Educación y Ciencia) and 2009SGR102 (Generalitat
de Catalunya).
1. INTRODUCTION
Both scholars and international organizations see rampant corruption as one of the most
evident impediments to development (Tanzi, 1998). There already many studies showing
how corruption directly harms foreign investment (Wei, 2000), trade (de Jong and Bogmans,
2010), and growth (Shleifer and Vishny, 1993; Mauro, 1995). Corruption might also
indirectly affect development, by eroding the legitimacy of democratic institutions and, as a
result, the level of social capital and the quality of government policies (Knack and Keefer,
1997; La Porta et al., 1999)1.
Widespread political corruption is commonly viewed as a brutal threat to public trust on
political institutions. A principle underlying democratic political systems is the presumption
that governments are accountable to citizens (Przeworski et al., 1999). So, the abuse of
entrusted public power for private gain by elected government officials is said to destroy this
accountability process (Bardhan, 1997). Corruption would thus systematically erode
democratic principles and diminish citizens’ faith in the political process. As a result,
disaffected citizens might withdraw from the electoral process (Chong et al., 2012)., use
other less legitimate means of protest, and push for radical changes in the system
(Kostadinova, 2009). There are already several papers that show corruption does indeed have
a negative effect on trust in government (e.g., Anderson and Tverdova, 2003; Chang and
Chu, 2005; Bowler and Karp, 2004; Kumlin and Esaiasson, 2012). This approach even
suggests that this distrust in government might be in the long run detrimental to the very
survival of a democratic regime (Mihsler and Rose, 1997). However, this view can be
challenged, both on theoretical and on empirical grounds.
From a theoretical point of view, corruption should have a substantial effect on political
trust only is some necessary conditions are met. First, voters should be sceptical about the
ability of democratic institutions to fix the problem. For instance, they may believe that
voting will not help to improve the situation. This might occur if they have very strong
partisan preferences, so they are not willing to vote for the opposition (Anderson and
Tverdova, 2003; and Anduiza et al., 2012), or if they think that even if they switch their vote
others will not, due to the prevalence of these ideological preferences in the population.
Voters might also think that all politicians are created equal, so by voting out a bad guy they
will bring in just a worst one. Additionally, citizens may judge other democratic institutions
(e.g., the media, the judiciary, political parties, etc.) as generally unable to punish corrupt
1
There is also a literature that warns about the effect of social capital and/or trust on government
performance (see La Porta et al., 1997; Boix and Posner, 1998; Tavits, 2006; Nannicini et al., 2012).
1
politicians. It is in this type of context that one expects a growth in disaffection, trust in
politicians to plummet, and citizens withdrawing from the political process, and looking for
other forms of protest.
Second, obviously, for this to happen, voters should be aware of how widespread is
corruption. Here, the literature is ambivalent about the role of information. One the one hand,
there is plenty of evidence that media coverage improves accountability (Boix and Stokes,
2003), and that informed voters are more willing to punish corruption at the polls (Anduiza et
al., 2013; Klasnja, 2011, and Costas et al., 2012). On the other hand, other authors have
suggested that in some cases more information might be bad for accountability. This would
happen, for example, if voters get the impression that corruption is widespread among all
parties, and/or that the democratic institution is not able to handle the problem. If this is the
case, then exposure to information about corruption scandals might erode trust in
government. There are virtually no studies focusing on the effect of information on the link
between corruption and trust (see Kumlin and Esaiasson, 2012, for an exception) or with
some related concept, as for example turnout (see Karlan et al., 2012).
From an empirical point of view, there are some doubts that the negative effects of
corruption on trust can actually be interpreted as a causal effect. Most of the empirical papers
rely on measures of ‘perception of corruption’ and ‘statement of trust’ obtained form the
same survey (e.g., Chang and Chu, 2006), and it is thus probable that the answers to both
questions reflect the same underlying individuals characteristics. Even the papers that use
other more appropriate measures of corruption face serious identification issues, as are
typically comparing units (usually countries) that have different levels of corruption but that
are also very different in many other dimensions. There are already some experimental
studies (either on the field ore on the Internet) of the effects of corruption on several
outcomes as turnout (Karlan et al., 2012) or statements of tolerance to corruption (Anduiza et
al., 2012). These approaches are however either costly or with its own drawbacks.
The purpose of the present paper is to newly examine whether the occurrence of a
corruption scandal involving the incumbent reduces the level of trust in government, trying to
improve on the above-mentioned aspects. With this purpose, we use a novel data set
providing information on a recent wave of local corruption scandals in Spain and data on a
new survey of trust in local government. Our database indicates whether there has been at
least a corruption scandal (defined as a “public allegation of corruption brought to light by a
newspaper”) during the three terms-of-office previous to our survey, and how intensive has
been the coverage of the scandal by Spanish newspapers (in terms of number of news articles
2
published). We make three main contributions to the existing literature. First of all, as in
other papers (see, e.g., Bowler and Karp, 2004; Kumlin and Esaiasson, 2012) we analyze the
impact of an average corruption scandal (i.e., without accounting for the intensity of
coverage), but contrary to this previous literature, we are able to estimate the impact of
scandals widely-covered by the press. Moreover, we also take into account whether citizens
obtain or not their information from the media. The second contribution consists of the use of
an improved identification strategy, consisting on the use of a control group of municipalities
not affected by any corruption scandal which has been selected using matching techniques.
Previous papers using contextual-level measures of corruption (Anderson and Tverdova,
2003; Bowler and Karp, 2004) don’t take adequately into account the fact that corruptionridden units need not have the same traits than the corruption-free ones. The ability of
regression analyses (the method typically used in these papers) in adjusting for differences in
covariates between the two samples is limited when the between-group differences in these
covariates are substantial.
The third contribution is that this is the first paper to our knowledge analyzing the
effects the effect of corruption scandals involving local incumbents on the level of trust in the
local government. Other studies have already focused on the determinants of trust in local
government (Rahn and Rudolph, 2005) but they do not address the effect of corruption. There
are also papers analyzing the effect of local performance and corruption on the level of trust
in politicians in general, but they do not study the effect on trust in local politicians (WeitzShapiro, 2008). There are some papers studying the effect of corruption on local turnout (see,
e.g., Karlan et al., 2012) but none on trust. Of course, focusing on trust in local government is
interesting in its own way, given the implications of the findings for the evaluation of the
decentralization reforms2, and also because local governments can be considered as an
essential part of the workings of any democratic system3.
2
Local governments are the closest ones to citizens and its evaluation depend to a great deal on how
accountable they are to local residents (Bardhan, 1997). Local corruption scandals might erode the
confidence in the workings of local governments and generate demands for greater centralization of
government functions.
3
The establishment of elected local governments is a crucial step in the development of new democracies and, in fact, it often precedes the democratization of the country at higher levels (MartínezBravo et al., 2012). Also, the effect of scandals on trust in local government is often a prelude of what
will happen afterwards with the rest of politicians (Weitz-Shapiro, 2008), either because citizens end
up believing that all politicians are created equal, or because local scandals have ramifications
involving politicians at higher layers. This seems to be the case in Spain, where the initial surge of
local corruption scandals during the last years of the boom and the beginning of the crisis has been
followed by an spread of corruption to the other layers of government (see section 3).
3
The results of our analysis suggest that municipalities affected by a corruption scandal
have a lower level of trust in local politicians than other similar municipalities. The average
effect is, however, quite moderate. The effect is much stronger for scandals widely covered
by the newspapers and for individuals obtaining information from the media. Overall, the
results suggest that, although corruption has the potential of creating disaffection amongst
citizens, its real impact depends on the amount and type of information citizens really get.
The paper is organized as follows. The next section reviews the previous literature on
corruption on trust, focusing first on the few papers that analyse the role of information, and
second on the different empirical approaches used. The third section provides some background for the analysis by describing the recent surge of local corruption scandals in Spain
and the current discussion regarding how this is affecting trust in government and the legitimacy of political institutions. Section fourth describes the data (i.e., corruption data-base and
trust survey) and the methods. Section four presents the results. The last section concludes.
2. CORRUPTION AND TRUST: PREVIOUS LITERATURE
2.1 The role of information
Most of the papers studying the interaction between corruption and media information focus
on the effects of information on government performance and on vote decisions. There are
papers showing how media coverage and readership improves the responsiveness citizen’s
needs (see, e.g., Besley and Burguess, 2002), improves government performance (Boix and
Stokes, 2003, Snyder and Strömberg, 2010), and reduces corruption (Svaleryd and Vlachos,
2011; Ferraz and Finan, 2012). There are also some studies documenting that the electoral
punishment of corrupt politicians is stronger the more informed voters are (see, e.g., Anduiza
et al., 2013; Klasnja, 2011, and Costas et al., 2012). Other authors have analysed the effect of
biased media information. Some of these papers suggest biased information do interfere with
voters evaluations of politicians (Kaplan and DellaVigna, 2003; Zhu et al., 2013) while other
suggest voters are able to ‘filter out’ the messages coming from biased media sources (Chang
and Knight, 2010). In general, these results corroborate the hypothesis that having well
informed voters seems to be crucial to ensure that the accountability process works
(Przeworski et al., 1999).
There are just a few studies that point to the other direction: information can in some
cases be bad for accountability. The more information about corruption voters has the more
they may be convinced that the democratic institutions are not able to handle the problem.
The work by Karlan et al. (2012) finds that exposing Mexican voters to information on in4
cumbent’s corruption just before local elections increases both the probability of voting
against the incumbent, but also the probability of not turning out to vote. The resulting effect
is troubling: although the wrongdoing incumbent looses votes, the challenger loses votes too
(since some of his supporters decide to stay home on Election Day) and the percentage of
vote for the incumbent and the probability of re-election might even increase. The authors
suggest that the reason voters behave this way is that they interpret that scandal reveal that
corruption is widespread and that replacing the incumbent by another corrupt politician will
not improve the situation.
Although the authors analyze the effects on turnout, the arguments they use suggest that
the effect would also hold for trust in local politicians. We expect that trust in local
politicians should be the highest in places where individuals don’t get information about
corruption. For individuals exposed to corruption scandals we expect two countervailing
effects. One the one hand, the mere realization that someone has betrayed the confidence of
voters might generate some distrust, but on other hand the belief that the vote act might help
to fix the problem sustains the faith in the democratic system. In places where there is huge
accumulation of corruption scandals with substantial coverage by the press, the second effect
might prevail. If information about the scandals is biased and noisy (media aligned with
different parties supplying contradictory information), the effect can be even more
detrimental to trust in government.
To our knowledge, the paper by Kumlin and Esaiasson (2012) is the only one studying
the effects of scandals on trust in government. They focus on the effects of three decades of
scandals that broke out during election time in Europe. They do find that scandals erode trust
in government. However, they also find that the marginal effect of a scandal is lower the
more scandals accumulate over time, suggesting that the electorate gets eventually tired of the
topic and stops paying attention. They attribute this effect to the opportunistic behaviour of
the press, which is prompt to publish sensationalistic information on scandals to sells
newspapers, but that does not follow up with attention the development of the story in the
future. The fact that the paper covers all sorts of scandals (e.g., sexual affairs included) might
contribute to this result. It could also be that the fatigue theory is consistent with an
increasing negative effect as scandals accumulate in the short run. In any case, the study by
Kumlin and Esaiasson (2012) also suggest that media bias might have an influence on the
effect of information on trust. If voters perceive that newspapers cover corruption scandals
only to sell, they might not believe the accusations are serious enough and, also, readers will
eventually lose the interest in the story and the coverage will be reduced. If the voters
5
perceive newspapers are ideologically biased in the coverage of corruption scandals, they
might try to filter out the information (Chang and Knight, 2010), but if there is too much
noise (they receive contradicting information, see Anduiza et al., 2012), they might end up
being confused and will tend to think that all politicians are engaged in corrupt deals.
2.2 Empirical approaches
Perceptions of corruption. Most papers studying the effects of corruption on trust in
government use individual responses to questions on ‘perception of corruption’ and
‘statements of trust’ in government (e.g., Della Porta, 2000; Seligson, 2002; Chang and Chu,
2008; Morris and Klesner, 2010)4. The results of most of these studies suggest that countries
with higher levels of corruption do indeed show lower levels of trust in government. It is not
clear, however, whether this result is indicative of a mere correlation between variables or
can indeed be interpreted as a causal effect. The main problem with this approach is that it is
affected by the well-known ‘chicken-and-egg’ problem, due to the fact that both variables are
measured from survey data and, most of the times, they even come from the same survey5.
Thus, it is conceivable that individuals respond in the same way to two questions they feel are
quite similar. Some authors have deal with this problem with a simultaneous equations
system (e.g., Chang and Chu, 2008), but face the obvious difficulty of justifying the
exogeneity of the instruments. Another solution recently proposed relies on using a question,
included in some of the surveys, related to the ‘real experience with corruption’ (e.g., Have
you been asked for a bribe by a public official in the previous year?). This information is
supposed to “reflect better the decision of a public official to solicit a bribe, rather than just
the respondent’s traits” (Clausen et al., 2012, p. 214). This approach might be appropriate to
get a proxy for ‘petty’ or bureaucratic corruption’ but not about corruption involving high
rank politicians, which is the nearly unique type of corruption found in Spain (as is also the
case in other developing countries).
Contextual-level corruption. There are just a few papers mixing individual-level trust
variables with contextual-level corruption measures. The most cited paper is by Anderson
and Tverdova (2003), which also find a negative and statistically significant effect of
corruption on trust in government. This study relies on country-level information on
corruption perceptions from Transparency International. The information comes from surveys
4
Other studies in this group are: Pharr (2000), Rose et al., (1998), Mishler and Rose (2001), Cho and
Kirwin (2007), Lavallée et al. (2000) and Bratton (2007).
5
Another problem with the use of ‘perceptions of corruption’ is that they are often biased (see Olken,
2004; and Donchev and Uhjely, 2011).
6
to experts and businessmen, and so the source is not the same than the one for the trust
variable, overcoming the above-mentioned ‘chicken-an-egg’ problem. However, the
aggregate nature of this index, which mixes opinions by different agents on different kinds of
corruption, means that it is not that evident how these evaluations of corruption are linked to
the citizen’s statements of trust. There are also some papers using information on corruption
scandals (e.g., Chanley et al., 2000; Bowler and Karp, 2004; Kumlin and Esaiasson, 2012),
which is the kind of information we use in this paper. Scandals are defined as ‘accusations of
corruption that have reached the general public’, guaranteeing a more clear link between the
corruption acts and citizens evaluations of trust. However, the procedure is not free from
problems, since the exposure to information about the scandal might vary across individuals
and contexts. Nevertheless, this might in practice constitute an advantage, since it opens the
door to developing some complementary analyses that might help reassure the validity and
interpretation of the results.
The paper most similar to ours is the one by Bowler and Karp (2004), who rely on
corruption cases related to the famous U.S. House Bank Scandal. This is the only paper that
links specific corruption scandals with measures of trust at the level of the electoral district of
the politicians involved in the scandal. This is also our choice, since we analyze the effect of
a corruption scandal involving a local incumbent with the statements of trust made by
residents in the municipality with respect to the politicians in the local government in general.
Bowler and Karp (2004) argue that this type of design helps isolating the impact of the
scandal from other potentially confounding factors. They state, for example, that it would be
important to see “if voters in those districts whose legislators have engaged in scandals have
higher awareness of the scandal and lower regard for politicians and legislative institutions
than voters who live in districts whose representatives have not been caught by scandals.”
Measuring both corruption and trust using small electoral districts is indeed an improvement
over previous studies, although as we explain below further improvements are possible.
Matching. A problem with the papers using contextual measures of corruption is that
the corruption-ridden units need not have the same traits than the corruption-free ones. Most
papers try to deal with this issue by controlling for other contextual level factors in a
regression framework. However, it is well-know that the ability of regression analysis in
adjusting for differences in observed covariates is severely reduced when the between-group
differences in these covariates are substantial (e.g., Cochran, 1965; Rubin, 2001), as it is
probably the case in most of the aforementioned studies. In this situation, using matching
methods to balance the distribution of covariates of the two subsamples helps reducing the
7
bias of the estimates. The idea is to compare cases where all other causal variables are as
similar as possible so that any difference between cases can be attributed to the treatment
(see, e.g., Rubin, 1979; Rosenbaum and Rubin, 1984, and Ho et al., 2007). In the paper we
use Propensity Score Matching to construct our matched sample. This matching is the
observational analog to randomization in ideal experiments (see, Rosenbaum and Rubin,
1983, and Rubin and Thomas, 2000), although less complete because it can only balance the
distribution of observed covariates6.
A further advantage of matching is its complete transparency. The matching algorithm
is applied before the estimation of the treatment with the purpose of balancing as much as
possible the covariates of the two groups. This ensures that the choices made by the
researcher at this stage are not contaminated by the knowledge of the outcome variable or by
how this choice impact on the estimation results.
As Ho et al. (2007) note, by using
matching the researcher is forced to specify a priori the research design she is going to use. In
our case, this effect is further enhanced by the fact that we are using matching to select the
municipalities where to run our survey. Budget considerations mean that once the matching
has been performed and the survey run, there is no way to come back and change the initial
research design. In this sense, our design completely guarantees that the matched sample was
selected before having any information regarding the outcome variable (i.e., trust, which we
obtain after running the survey), an ideal trait of a well-designed observational study (see
Rubin, 2001) and a task suited for the use of Propensity Score Matching (Rosenbaum and
Rubin, 1985).
3. CORRUPTION SCANDALS IN SPAIN
The recent wave of corruption scandals. In the first two decades following the restoration of
Spain’s democratic local governments (1979-99) not much concern was expressed in the
media, among the political elite, or the population in general to the lack of accountability or
possible cases of corruption (see Jimenez and Caínzos, 2003). The situation began to change
after 1995, with the switch in the housing market situation, but it did not really take off until
1999. Before 1999, there were just 46 corruption scandals, and this number jumped to 288
during the term 1999-2003 (see next section for a discussion of the data sources). Corruption
did not go into decline after this, with 408 scandals breaking out during the 2003-07. During
6
This method has been used for a long time in medicine (see, e.g., Rubin, 2001) or economics (e.g.,
Dehejia and Waha, 1999), and more recently in political science (e.g., Ladd and Lenz, 2008; Gilligan
and Sergenti, 2008).
8
the period that runs from the June 2007 elections to November 2009, 72 new cases appeared.
The reduction in the rate of new cases might be related to the crisis in the real estate market,
which began in 2007 and became more intense in the years that followed.
Corruption in land use regulations. Most local corruption scandals that broke out in
Spain in the past years involve bribes received by local politicians in exchange for amendments to the land use plans. Land use regulations in Spain adhere to an extremely
interventionist and highly rigid system (Riera et al., 1991). Town planning responsibilities are
mostly in the hands of local governments. Municipalities draw up a ‘General Plan’, which
provides a three-way land classification: built-up land, developable land, and nondevelopable land. The existence of a ‘development border’, a line between plots of land on
which developers are allowed to build and plots where development is prohibited, is a key
feature of Spain’s land regulation system. In periods of high demand this border creates a rent
differential which might fuel the bribes developers are willing to pay to local politicians in
exchange for shifting this border to their advantage. There are a large number of cases
involving local officials that wrongfully allowed huge tracts of land to be developed, that
allowed building in places where it had been previously prohibited, or that amended the land
use plan so as to permit higher construction densities in already developed land (Fundación
Alternativas, 2007).
Corruption, voting, and disaffection. In Spain, there is the view that corrupt politicians
are not punished at the polls. The press has followed intensively some very prominent scandals that ended with the re-election of the incumbent accused of corruption. The conclusions
of several studies (see Fundación Alternativas, 2008, and Barberá et al., 2012) corroborate the
idea that the average punishment is quite low (i.e., around 4-5% of the vote), although recent
studies have appeared that indicate that this effect can be occasionally larger (Costas et al.,
2012; Anduiza et al., 2012), depending on factors as the quality of media information, the
intervention of the judiciary, the existence of clientelistic networks, or the degree of
ideological polarization. There has been also considerable discussion regarding the possible
adverse consequences of corruption on disaffection (which we take as a synonym of trust in
govern-ment). For example, in 2009 a prominent think tank entitled its annual report The
erosion of confidence and well-being. Against citizens’ disaffection (see Fundación
Alternativas, 2010). Amongst the motives of this increases disaffection cited in this report
were: (i) the partisan-ship of voters, which make them tolerant to scandals affecting their own
party and/or make them unwilling to vote for the opposition even if they know about
incumbent’s wrongdoing, (ii) the bias of the media, publishing news attacking non-aligned
9
parties and ignoring scandals that affect aligned politicians, and so creating noisy that makes
really difficult to disentangle the corrupt politicians from the clean ones, and (iii) the lengthy
judiciary process, which prevents voters from having prompt access to useful information
(whether the politicians is indeed guilty or not).
4. EMPIRICAL ANALYSIS
4.1 Measuring local corruption
We had access to a database of corruption scandals compiled by the Fundación Alternativas
(2007), a Spanish think-tank. In 2007, shortly after the upsurge in corruption scandals that
occurred in 2006, this organization commissioned a survey of local corruption in order to
gauge quantitatively the relevance of the phenomenon. They hired a journalist in each Spanish
province with the task of compiling all corruption related stories involving municipalities in
the province between 1 January 2000 and 1 February 2007 that appeared in national, regional
or local newspapers, and related to this period or to the past. Overall, a total of 426 corruption
scandals broke out during this period.
Since our survey was carried out in late 2009, we decided to complete the database for
this late period with internet-guided searches in MyNews (http://mynews.es), a paid digital
information management service covering all national and many of the regional newspapers.
We screened the period that runs from 1 February 2007 to 1 November 2009 (the day this
search was performed). We conducted a search for news reports containing the word
‘corrupción urbanística’ (i.e. corruption related to land planning) and each of the more than
8,000 names of the Spanish municipalities. We found 131 additional scandals breaking out
during this late period7. At the end, the total number of scandals in our database is 557.
The database also contains the number of news reports related to each of these scandals.
The number of news reports totals 5144, with an average of ten news items per scandal. For
nearly 30% of the cases there is only one news report, for 33% the number of reports is
greater than one but less than five, 12% of them were mentioned in between five and ten news
items, while 25% of the cases were written about in more than ten news stories.
4.2. Measuring trust in local politicians
To obtain a measure of trust in local politicians at the municipal level, we designed a special
survey. We interviewed a sample of residents in a fraction of the municipalities where a
corruption scandal broke out during the period 1999-2009, and also in a number of municipal7
As a robustness check, we also searched for news reports containing just the word “corrupción”, but we did not
find additional cases.
10
ities with similar traits to the ones affected by corruption. The survey was undertaken in
November 2009 and so the information gathered about trust in government will be a reflection
of the mood about politics of Spanish citizens at that time8. Below, we describe the, first, the
Questionnaire used in the survey, the selection of the sample of Treated municipalities, and
the construction of the Matched sample used as a control group.
Questionnaire. In that survey, we asked to individuals the following question: ‘With
respect to your city, do you think politicians in the local council can be trusted?’. The
interviewees could answer by selecting one of the following four categories: 1 (‘Local
politicians can never be trusted), 2 (‘Almost never can a local politician be trusted), 3 (‘Local
politicians can be trusted most of the time’), and 4 (‘Local politicians can be always
trusted’)9. The survey also includes questions regarding the degree of media exposure (e.g.,
whether the media is the main source of information regarding the activities of the local
government), political preferences (e.g., self-assessed ideology), and information on a bunch
of socio-economic controls (e.g., unemployed, type of job, marital status, etc.). Below, we
will be more precise in the description of those variables that we end up using the empirical
analysis. The technical details of the survey are presented in the Annex; the questionnaire
used in the survey is available upon request.
Treated municipalities. Data limitations forced us to focus on municipalities larger than
1.000 residents10; 495 out of the 557 municipalities affected by corruption scandals are that
big. Because of the budget constraint we had to select a subsample of these municipalities11.
We selected 160 municipalities where corruption broke out as our treatment group and 130
similar municipalities as the control group. The number of control municipalities is a little
lower because some of the municipalities in this group are used as controls for more than one
treated municipalities (see the reason why below). In each of these municipalities, we
interviewed between 20 and 50 residents, this number growing with population size (see also
Box A.1 in the Annex). Our treated municipalities are selected taking into account the
proportions of corruption scandals that broke during each of the three terms-of-office we
analyze (i.e., 1999-2003, 2004-2007, 2008-2009) and also across different population sizes
8
Certainly, this mood was deteriorating because of the economic crisis, but had yet to reach the levels
of discontent of some years later.
9
There is also an additional category 5 (Don’t know – Don’t answer), but following standard procedures we do not use these responses in our analysis.
10
We lacked information on the municipal-level variables needed to implement the matching for
municipalities smaller than 1,000 inhabitants. Spain has 8,114 municipalities, and 3,252 of them are
larger than 1,000 inhabitants. These are the municipalities that potentially belong to the control group.
11
Because of the fixed budget of the project, we faced a trade off between the number of
municipalities surveyed, the number of individuals interviewed, and the length of the questionnaire.
11
(i.e., less than 10.000, between 10.000 and 100.000, between 100.000 and 500.000 and larger
than 500.000).
Matched sample. As for the control municipalities, these were selected using matching
techniques. We constructed the matched sample using the ‘propensity score’. We estimated a
logit model, using as a dependent variable a dummy equal to one if a corruption scandal broke
out in the municipality (and zero otherwise) and as regressors variables deemed to have a
influence both on corruption and on the level of trust in local politician (see below)12. Then,
the ‘propensity score’ was computed and control municipalities were matched to the treatment
ones based on having a similar value of the ‘propensity score’13. The method used was the
‘nearest neighbour matching with replacement’. This method allowed for more a given
control unit matching more than one treatment unit, which increases the average quality of
matching and reduces the bias14. At the same time, the method had the side benefit of
allowing us to reduce the number of municipalities in the control group which, in turn, given a
fixed budget, permitted us to increase a little bit the sample of treated municipalities and/or
the number of interviews per municipality.
The matching strategy builds on the ‘conditional independence assumption’, requiring
that the outcome variable (i.e., trust) must be independent of treatment (i.e., corruption)
conditional on the ‘propensity score’. Hence, implementing the matching procedure requires
choosing a set of variables that credibly satisfy this condition (see Heckman et al, 1997). Only
variables that influence simultaneously the participation decision and the outcome variable
should be included. More concretely, the municipal-level variables used to estimate the logit
equation are: log(Population), % Unemployed, Ethnic diversity, Income per capita, %
Graduate studies, % Divorced, % Right voters (historical average), and % Vote turnout
(historical average). The information these variables comes from sources dated close to the
starting year of corruption scandals in our database, so we can consider them as predetermined (see Table A.1 in the Annex for the definition and sources of these variables).
12
The Logit equation was estimated with information of all the corrupt municipalities with more than
1,000 inhabitants plus all the non-corrupt municipalities with the same size. The random selection of
the municipalities included in the survey –stratified according term of corruption and population size–
was done afterwards.
13
Just 17 municipalities fell outside the common support. They were also included in the survey. We
performed a robustness check by excluding them from the analysis, and the results did not change.
14
The main risk of this matching procedure is the generation of bad matches, i.e. the distance to the
nearest neighbor is too large. This problem can be solved by specifying the caliper, i.e. the maximum
distance in the propensity scored allowed in any matching. In our case, however, the matching is quite
good, with a 95% of the matches having an absolute distance in the ‘propensity score’ which lower
than 0.01. We also tried other matching options (e.g., ‘without replacement’) but these did not work
that well for some of the larger municipalities, so we decided not to use them.
12
Following the advice in Ho et al. (2007) we opt for a parsimonious specification where all the
variables used are statistically significant and help predicting the outcome of interest. The use
of this parsimonious specification produced a good balance of covariates and good matches.
We also tried two more specifications, adding some additional covariates with low
explanatory power (% age over 65, % immigrants, % electoral margin of victory (historical
average), and coalition government (historical average)), and also higher order terms of the
main covariates and interactions amongst them (as proposed by Rubin, 2001) but the balance
did not improve in either of the two cases, so we choose the simpler option.
The choice of variables was determined by data availability and by searching the
literature on the determinants of corruption and trust. First, regarding corruption, we know
that there is some evidence that this phenomenon is more prevalent in uneducated polities,
and that it is also related to income, unemployment or ethnic diversity (Glaeser and Saks,
2006). Corruption is also said to be more prevalent in places with structurally low levels of
social capital and/or low trust in government (Nannicini et al., 2013). We also know that
turnout is a good proxy for social capital and trust in government, hence the idea that places
where turnout has been historically high display lower corruption levels. Ideology might also
be related to corruption; it seems that right-wing voters are more tolerant with corruption and
that right-wing politicians have stronger connections with private firms (Hessami, 2012).
There is also some literature suggesting that the proximitiy of government to citizens (which
is presumably higher in smaller municipalities) fosters political participation and
accountability (Lassen and Serritzlew 2011) and thus might result in reduced levels of
corruption.
Second, there is also evidence that trust is negatively affected by: belonging to a
minority, living in racially mixed community, having experienced a recent traumatic
experience (e.g., divorce, unemployment), and being economically unsuccessful in terms of
income or education (Alesina and La Ferrara, 2002; Gustavsson and Jordahl, 2008). Trust is
also a persistent variable, so if historical trust is closely correlated with historical turnout,
historical turnout should also bear a considerable over the actual levels of trust. Regarding
ideology, the level of support for democracy as shown in several surveys is lower in the case
of right-wing voters (although still very high), the reason being that the main national rightwing parties were filled with the high-ranks of Franco’s regime. Finally, some authors
document that trust in government is higher the smaller is the polity (Rahn and Rudolph,
2005).
13
Using the aforementioned variables we are able to balance the covariates in the two
subsamples. We have performed several tests to test whether we achieved or not a good
matching. First, we performed a comparison of means between treated and control units in the
unmatched and matched samples (see Rosenbaum and Rubin, 1985). These tests are shown in
Table A.2 in the Annex. In the unmatched sample, the treated group (the corruption-ridden
municipalities) has lower levels of historical turnout, lower % of divorced people, and of
people with graduate studies, and are more ethnically diverse and have larger population
sizes, than the control group (the corruption free municipalities). In the matched sample, none
of the differences in means between the treated and the control group are statistically
significant. Second, we also looked ate the percentage reduction in the standardized bias as
the result of the matching procedure. The reduction in bias is substantial in all the variables
that showed a statistically significant bias before the matching: historical turnout (79% drop),
ethnic fragmentation (86% drop), log(Population), % graduate and % divorce (98% drop
each. Third, we also re-estimated the propensity score on the matched sample and compare
the pseudo-R2’s before and after matching, which actually were 0.237 and 0.002, respectively. LR tests of joint significance of the regressors before and after the matching have values
of 1871.77 and 2.32, with p-values of 0.000 and 0.941. Finally, we also use the stratification
test proposed by Dehejia and Wahba (1999); to implement this test, we divide the observation
into strata (actually, seven strata were used) based on the estimated propensity score, such that
no statistically significant difference between the mean of the estimated propensity score in
both treatment and control group remain. This test also suggests that the matching was of
good quality (complete results of all these tests available upon request).
4.2. Estimation method
We follow the recommendation by Ho et al. (2007) and we estimate a parametric model with
the data of our final matched sample. Other authors, as Rubin (2001) and Crump et al. (2009),
also recommend this procedure, suggesting that the propensity score should be used only for
systematic sample selection as a precursor to regression estimation (or to more complex
parametric methods). In most studies using matched techniques, the analysis performed to
obtain the treatment effect is a simple difference in means (or the equivalent to a bivariate
regression between the treatment indicator and the outcome, in the parametric case).
However, it is well-know that if the matching is not exact, this procedure can be much
improved by adjusting for covariates. There are several ways to perform this adjustment in a
non-parametric way (Rubin, 2001, and Dehejia and Wahba, 1999), but in the parametric case
an obvious one is just to run a multivariate regression with the matched sample and the
14
covariates used in the estimation of the propensity score. Ho et al. (2007) recommend this
procedure and suggest treating the predetermined covariates as fixed, meaning that standard
errors and confidence intervals should be computed as in normal regression framework15.
In our case, the multivariate regression has two additional advantages. First, it will allow
us to use as additional covariates the individual-level information extracted from the survey16.
The individual variables we will use as additional controls are income, education, gender,
age, divorced, unemployed, student, retired, and immigrant. Controlling for individual-level
variables is standard in the empirical analysis of trust (Alesina and La Ferrara, 2002;
Anderson and Tverdova, 2003; Chang and Chu, 2005). By doing so we are purging the trust
variable from a bunch of individual traits, which means that at the end we will be comparing
the level of trust for similar individuals living in municipalities with and without corruption
which are also similar (by virtue of the matching procedure). Second, the use of a parametric
framework allows us to choose the most appropriate estimation method. In our case, the fact
that our dependent variable is a categorical one means that we should use an Ordered Logit
model. An alternative would be to estimate a Logit model by collapsing the four categories
into two. Actually, it is possible from the results of the Ordered Logit to test whether it is
feasible to reduce the number of categories. In our case, it turns out that this option can be
rejected, so we have to use the Ordered Logit model. The problem with logistic models is that
the quantitative interpretation of the coefficients is not straightforward, so we will also
provide information on the marginal probabilities.
5. RESULTS
5.1 Main results
The results of the estimation of the Ordered Logit model using the matched sample selected
as we explained in the previous section are presented in Table 2. The results suggest that the
occurrence of corruption scandal has a negative and statistically significant effect (at the 1%
level) on trust in local politicians. This results hold whenever we adjust or not for historical
turnout (a proxy of the previous level of trust in the municipality), for all the contextual-level
variables, and for the individual characteristics. The estimated coefficient drops when we
15
In some types of matching, the parametric analysis might require some adjustment. For instance,
when using ‘matching with replacement’, weights must be used to ensure that the parametric analysis
reflects the actual observations (see Ho et al., 2005; and Dehejia and Wahba, 1999). We take this into
account in our estimation.
16
We will deal with the multilevel structure of the dataset, with individuals belonging to different
municipalities, by clustering standard errors at the municipality level.
15
control for the contextual variables, but not when we adjust only for historical turnout or
when we add the individual characteristics to the equation.
[Insert Table 2]
Regarding the control variables, historical turnout has a positive and statistically (also at
the 1% level) significant effect on trust when included alone in the equation. This effect
vanishes when we add the remaining contextual variables. Although these variables are
jointly statistically significant, the only variable with a coefficient which is statistically
different from zero is log(Population). These results suggest that big municipalities and/or
places with low historical turnout display low levels of political trust. In contrast, all the
individual-level variables have the expected sign and most of them are statistically. Also their
explanatory capacity is much higher. It seems that the main determinants of trust (aside from
a few exceptions) are to be found at the individual level. The individual-level results suggest
that people that are rich and old, and also students and immigrants tent to trust more local
politicians. People divorced or living with a partner without being married tends to trust local
politicians less. These effects are statistically significant. Unemployment and education have
the expected negative and positive sign, but the coefficients are not statistically significant at
conventional level.
5.2 The role of information
So far, the results of Table 2 tell us about the effect of an average scandal. However, some of
the scandals are covered more widely by the press and, as a result, might have a more profound effect, especially on people exposed to this information. Table 3 presents the results
obtaining when adding to the equation interactions with the degree of press coverage (i.e., a
dummy called Wide coverage, equal to one if more than ten news articles have appeared
related to scandals breaking out in the municipality), the degree of exposure to media
information (i.e., a dummy called Media exposure, equal to one is the individual declares to
obtain information mainly from media sources), and also with an interaction with both
situations at the same time.
[Insert Table 3]
The results in column (1) replicate the one of column (6) of Table 2 for the reduced
sample used in this analysis. The reduction of the sample is due to the fact that the survey
questions used to construct the two interaction dummies have a residual category (Don’t
know – Don’t answer) that is not being used in the analysis. The coefficient estimated for the
Corruption dummy in column (1) is virtually the same than the one in Table 2, so the
16
reduction in the sample does not seem to have any impact on the estimates. In column (2), the
coefficient of the interaction between Corruption and Wide coverage is, as expected negative,
meaning that scandals widely reported by the press have a larger impact on trust on
politicians. Note, however, that the coefficient is not statistically significant. In column (3),
the coefficient of the interaction between Corruption and Media exposure is negative and
highly significant, but the coefficient of the Corruption dummy (which indicates the effect
for the individuals that not obtain the information from the media) is negative but not
statistically significant at conventional levels.
The conclusion is thus that the effect of corruption scandals on trust in local politicians
is restricted to the case of citizens obtaining their information from the media. This
conclusion still holds when the two interactions are included at the same time (see column
(4)). Finally, in column (5) we include the two interactions plus a triple interaction between
Corruption, Wide Coverage and Media exposure. Note that in this case the coefficient of the
triple interaction is the only one which is statistically significant at the 5% level, and the size
of the coefficient is sizeable. The coefficient of the interaction with Media exposure is
significant at the 10% level and its size is also considerable. The size of the other two
coefficients is also not negligible, but the standard errors are quite large. Overall, the joint
statistical significance of all the interacted terms can not be rejected. In any case, the
interpretation of the results is that the effect is much stronger in the case of individual
exposed to the media and when the flow of information is substantial. In the other cases (i.e.,
individuals obtaining information from grapevine sources, with or without extensive media
coverage and individuals informed though the media when the coverage is low) the effects
seem to be much smaller and the statistical significance is quite dubious.
Several robustness checks have been performed in order to check the sensitivity of these
results to different research design choices. For instance, we have also estimated the equation
with dummies defining the wide coverage threshold at different numbers of news articles.
The ten news articles thresholds performs much better than the other options. Also, we have
estimated the equation by adding extra interactions depending on which type of media supplied the information to individuals, national, regional or local. It seems, for instance, that the
effect of local media readership is not different from that of national or regional media, and
that the effect of regional media is the same than that of national media.
Finally, we performed some additional checks to see whether the interaction effects
have or not the same causal interpretation than those in Table 2. The selection of the matched
sample did not take into account the possibility of multiple treatments (i.e., widely vs. loosely
17
covered scandals). We did not do this because at the moment of designing the survey we just
had access to the list of scandals, and still had to compile the information on number of news
articles. It means, therefore, that even if the covariates are on average balanced between the
same of municipalities with and without any scandal, it is not clear that there is balance with
respect to each of the two heterogeneous treatment subsamples17. In order to guess how much
are we departing from this benchmark, we repeated the t-test of equality of means between
treated and controls in the matched sample, but now comparing the means of the two
treatment subsamples. The results (available upon request) say that although it seems that
there are some differences, these are in general not large, and the t-test indicates that they are
not statistically significant at the 5% level. We also run a logit with a dummy equal to one in
the case of the wide coverage case and zero in the case of the loose coverage one, against all
these variables. None was statistically significant at conventional levels. So, it seems that the
validity our basic matching estimation can be extended to the case of these two treatments.
5.3 Marginal effects
The fact that the effects of corruption scandals on trust in government are negative and
statistically significant does not mean that these effects are quantitatively meaningful. One
drawback of the Ordinal Logit effect is that the size of the estimated coefficient can not be
directly interpreted. The interpretation requires the computation of the marginal effects of a
corruption scandal in each of the four categories of trust. In Table 4 we present these marginal
effects, computed as the difference in the predicted value of the probability of as the
corruption dummy changes from zero to one, while all others are held constant at their mean
value.
[Insert Table 4]
We do this exercised both for the basic model (column 6 in Table 2) and for the case
with the triple interaction (column 5 in Table 3). In this last case we define four mutually
exclusive categories defined by the interaction of the two different values (one or zero) that
the two dummies used in the interactions can take (Wide coverage and Media exposure).
Looking at the first row of Table 4, we can see that when a corruption scandal breaks out an
additional 3.7% of the population state that ‘Local politicians can never be trusted’ (category
1), and an additional 1.3% states that ‘Almost never can a local politician be trusted’
(category 2) increases 1.3 points. Conversely, after a corruption scandal an additional 3.3% of
17
Note that the ineraction with respect the type of information at the individual level (the Media
exposure dummy) does not pose the same challenge, since the split of the sample is not made using a
contextual-level variable.
18
the population states that ‘Local politicians can be trusted most of the time’ (category 3) and
an additional 1.7% state that ‘Local politicians can be always trusted’ (category 4). Overall, it
seems that around a 5% of the population shift from trusting to not trusting politicians as a
results of a corruption scandal. Rows two to five show the results for each of the four
subcategories. Note that only in the first row (Wide coverage plus Media exposure) are the
effects statistically significant (now at the 5% level); this is just a reflection of the results
displayed in Table 3. Note, in any case, that the effects for this category are nearly twice the
ones found on average; for example, an additional 6.8% of this subsample of individuals (in
the municipalities affected by widely covered examples) states now that ‘Local politicians
can never be trusted’. In this case, the shift from high-trust to low-trust categories is around a
9% of the subpopulation. This is clearly a more sizeable effect.
In order to put this into perspective, one should compare this marginal effect with that of
other variables. In the case of the historical turnout (results from column 6 of Table 2), an
increase in one standard deviation of this variable (moving from say 71%, which is the sample
average, to 80%) shifts around a 4% of the population from the low-trust to the high-trust
categories. In the case of population (results from column 7 of Table 2), an increase of one
standard deviation reduces the level of trust by 5%. For income, measured at the individual
level, increasing one standard deviation shifts around 5.5% of the population from the lowtrust to the high-trust categories. So, the effects of a corruption scandal on the level of trust in
local government are similar to transforming the municipality into a low turnout place
(defined to a place with an historical turnout rate in local elections which is one standard
deviation below the average) or substantially increasing the size of the municipality. So,
compared to other influences on political trust, the effect of a corruption scandal doesn’t seem
to be that small.
6. CONCLUSION
Local corruption scandals can have a very detrimental effect on the level of trust of citizens
on local politicians. This is what we have found in this paper. On average, the effect seems
quite moderate, only around 5% of people shifting form the high-trust to the low-trust
categories. This number is close to the proportion of votes lost by incumbents accused of
corruption (see, e.g., Peter and Welch, 1980; and specifically for local incumbents in Spain,
Barberá et al., 2012, and Costas et al., 2012). However, it is not clear that this is a small effect
once it is compared with the effect of other differences in other variables that have some
influence on trust in local government. Moreover, we also find in the paper that the size of
this effect is much higher for individuals obtaining information from the media, especially
19
when press coverage of the scandal is intensive. In this case, the proportion of people shifting
from high to low categories of trust in the relevant subpopulation is much higher, around a
9%. This suggests that the impact of corruption on trust in local government might be much
higher in places where the citizenship is more exposed to information about corruption.
REFERENCES
Alesina, A. and La Ferrara, E. (2002): “Who trusts others?,” Journal of Public Economics 85,
207-234.
Anduiza, E., Gallego, A., and Muñoz, J. (2012): “Why do voters forgive corrupt politicians?
Implicit exchange, noise, an cynicism,” Mimeo, Universitat Autónoma de Barcelona,
http://llet-131-98.uab.es/recercapol/images/publications/Experiment%20corruption%20
IPSA.pdf
Anduiza, E., Gallego, A., and Muñoz, J. (2013): “Turning a blind eye: experimental evidence
of partisan bias in attitudes towards corruption,” Comparative Political Studies,
forthcoming.
Barberá, P., Fernández-Vázquez, P. and Rivero, G. (2012): “Rooting out corruption or rooting
for corruption? The heterogeneous electoral consequences of scandals,” Mimeo, New
York University, http://pablofernandez.net/Research_files/rooting_out_corruption.pdf.
Bardhan, P. (1997): “Corruption and development: a review of the issues,” Journal of
Economic Literature 35, 1320-1346.
Bowler, S. and Karp, J.A. (2004): “Politicians, scandals, and trust in government,” Political
Behavior 26(3), 271-287.
Chang, Ch. and Knight, B. (2011): “Media bias and influence: evidence from newspaper
endorsements,” Review of Economic Studies 78(3), 795-820.
Chang, E. and Chu, Y. (2006): “Corruption and trust: exceptionalism in Asian democracies?,” Journal of Politics 68(2), 259–271.
Cho, W. and Kirwin, M.F. (2007): “A vicious cycle of corruption and mistrust in institutions
in Sub-Saharan Africa: a micro-level analysis,” Afrobarometer Working Paper 71.
Chong, A., De la O., A., Karlan, D., and Wantchekon, L. (2011): “Looking Beyond the
Incumbent: The Effects of Exposing Corruption on Electoral Outcomes,” NBER
Working paper 17679.
Clausen, B., Kraay, A. and Nyiri1, Z. (2011): “Corruption and confidence in public
institutions: evidence from a Global Survey,” World Bank Economic Review 25 (2),
212-249.
Costas, E.,; Solé-Ollé, A., and Sorribas, P. (2012): “Corruption, voter information, and
accountability,” European Journal of Political Economy 28/4, 469-484.
Dehejia, R.H. and Wahba, S. (1999): “Causal effects in non-experimental studies: reevaluating the evaluation of a training program,” Journal of the American Statistical
Associtation, 94(448), 1053-1062.
DellaVigna, S. and Kaplan, E. (2007): “The Fox News effect: media bias and voting,”
Quarterly Journal of Economics 122(3), 1187-1234.
Ferraz, C. and Finan, F. (2008): “Exposing corrupt politicians. The effects of Brazil's publicly
released audits on electoral outcomes,” Quarterly Journal of Economics 123(2), 703745.
Fundación Alternativas (2007): Urbanismo y democracia. Alternativas para evitar la
corrupción, Madrid. www.falternativas.org.
20
Fundación Alternativas (2010): Democracy in Spain 2010. The erosion of confidence and
well-being. Against citizens’ disaffection , Madrid. www.falternativas.org.
Gilligan, M.J. and Sergenti, E.J. (2008): “Do UN Interventions Cause Peace? Using Matching
to Improve Causal Inference,” Quarterly Journal of Political Science 3(2), 89-122.
Glaeser, E.L. and Saks, R.E. (2006): “Corruption In America,” Journal of Public Economics
90(6-7), 1053-1072.
Gustavsson, M. and Jordahl, H. (2008): “Inequality and trust in Sweden: Some inequalities
are more harmful than others,” Journal of Public Economics 92(1-2), 348–365.
Hessami, Z. (2012): “On the link between government ideology and corruption in the public
sector,” mimeo, University of Konstanz, http://extranet.isnie.org/uploads/ isnie2011/
hessami.pdf.
Ho, D., Imai, K., King, G., and Stuart, E. (2007): “Matching as Nonparametric Preprocessing
for Reducing Model Dependence in Parametric Causal Inference,” Political Analysis 15,
199–236.
Jiménez, F., Caínzos, M. (2003): “Political corruption in Spain. Perceptions and problems,” in
M. Bull and J. Newell (eds.). Corruption in contemporary politics. London. Palgrave
Press.
Keele, L. (2005): “The authorities really do matter: party control and trust in government,”
Journal of Politics 67(3), 873–886.
Klasnja, M.. (2011): “Why do malfeasant politicians maintain political support? Testing the
‘uninformed voter’ argument,” Paper presented at the conference of the APSA, http://
papers.ssrn.com/sol3/papers.cfm?abstract_id=1901683.
Kostadinova, T. (2009): “Abstain or rebel: corruption perceptions and voting in East
European Elections,” Politics & Policy 37(4), 691–714.
Kumlin, S. and Esaiasson, P. (2012): “Scandal fatigue? Scandal elections and satisfaction
with Democracy in Western Europe, 1977–2007,” British Journal of Political Science
42(2) 263-282
Ladd, J.MD. and Lenz, G. (2009): “Exploiting a rare communication shift to document the
persuasive power of the News Media,” American Journal of Political Science 53(2),
394-410.
Lassen, D.D. and Serritzlew, S. (2011): “Jurisdiction size and local democracy: evidence on
internal political efficacy from large-scale municipal reform,” American Political
Science Review 105, 238-58.
Lavallee, E., Razafindrakoto, M. and Roubard, F. (2008): “Corruption and trust in political
institutions in Sub-Saharan Africa, Afrobarometer Working Paper 102.
Martínez-Bravo, M., Padró, G. and Qian, N. (2012): “The Effects of Democratization on
Public Goods and Redistribution: Evidence from China,” NBER Working Paper 18101.
Mishler, W. and Rose, R. (1997): “Trust, distrust and skepticism: popular evaluations of civil
and political institutions in Post-Communist societies,” Journal of Politics 59(2):418-51.
Morris, S.D. and Klesner, J.L. (2010): “Corruption and trust: theoretical considerations and
evidence from Mexico,” Comparative Political Studies 43/10, 1258-1285.
Olken, B.A. (2009): “Corruption perceptions vs. corruption reality,” Journal of Public
Economics 93(7-8), 950-964.
Przeworski, A., Stokes, S., and Manin, B. (1999): Democracy, Accountability, and
Representation, Cambridge University Press.
Rahn, W. and Rudolph, T. (2005): “A tale of trust in American cities,” Public Opinion
Quarterly 69(4), 530-556.
Riera, P., Munt, I., Keyes, J. (1991): “The practice of land use planning in Spain,” Planning
Practice and Research 6 (2), 11-18.
21
Rosenbaum, P.R. and Rubin, D.B. (1983): “The central role of the propensity score in
observational studies for causal effects,” Biometrika 70, 41-55.
Rosenbaum, P.R. and Rubin, D.B. (1985): “Constructing and control group using multivariate
matched sampling that incorporate the propensity score,” The American Statistician
39(1), 33-38.
Rubin, D.B. (2001): “Using propensity scores to help design observational studies:
application to Tobacco Litigation,” Health Services & Outcomes Research Methodology
2, 169-188.
Rubin, D.B. and Thomas, N. (1996): “Matching using estimated propensity scores: relating
theory to practice,” Biometrika 52, 249-264.
Rubin, D.B., and Thomas, N. (2000): “Combining propensity score matching with additional
adjustment in prognostic covariates,” Journal of the American Statistical Association
95(450), 573-585.
Seligson, M.A. (2002): “The impact of corruption on regime legitimacy: a comparative study
of four Latin American Countries,” Journal of Politics 64(2), 408-433.
Strömberg, D. and Snyder, J. (2010): “Press coverage and political accountability,”
Journal of Political Economy 118(2), 2010.
Weitz-Shapiro, R. (2008): “The local connection: Local government performance and
satisfaction with democracy in Argentina,” Comparative Political Studies 41/3, 285308.
Weitz-Shapiro, R. and Winters, M.S. (2013): “Lacking information or condoning corruption:
when will voters support corrupt politicians?,” Journal of Comparative Politics,
forthcoming.
Zhu, J., Lu, J., and Shi, T. (2013): “When grapevine news meets mass media: different
information sources and perceptions of government corruption in Mainland China,”
Comparative Political Studies, forthcoming.
22
Table 1: Effects of corruption scandals on trust in local politicians. Ordered Logit results.
Variables
Corruption
(2)
(1)
***
-0.259
(0.067)
(3)
***
-0.247
(0.068)
(4)
***
-0.199
(0.069)
(5)
***
(6)
-0.272
(0.068)
***
-0.260
(0.068)
-0.206***
(0.070)
--.--
0.961***
(0.365)
--.--
--.--
--.--
--.--
--.--
--.--
--.--
--.--
--.--
--.--
--.--
--.--
--.--
--.--
0.225
(0.475)
0.393
(0.491)
-4.297
(3.917)
-0.847
(0.712)
0.359
(0.523)
-0.128
(0.303)
0.199
(0.440)
-0.059**
(0.024)
0.126***
(0.031)
0.012
(0.021)
0.017***
(0.002)
0.011
(0.040)
-0.115**
(0.050)
-0.251**
(0.111)
-0.092
(0.059)
0.215***
(0.082)
0.040
(0.079)
0.187*
(0.105)
0.128***
(0.031)
0.017
(0.020)
0.017***
(0.002)
0.015
(0.040)
-0.120**
(0.050)
-0.251**
(0.111)
-0.092
(0.059)
0.219***
(0.082)
0.039
(0.079)
0.192*
(0.106)
--.--
--.-168.21
[0.000]
2912.23
[0.000]
3442.31
[0.000]
19.48
[0.007]
163.59
[0.000]
2984.61
[0.000]
3483
[0.0000]
8197
8197
Contextual-level variables
% Vote turnout
--.--
0.955***
(0.312)
Income p.c.
--.--
--.--
% Divorced
--.--
--.--
% Graduated
--.--
--.--
% Unemployed
--.--
--.--
Ethnic diversity
--.--
--.--
% Right voters
--.--
--.--
Log(Population)
--.--
--.--
0.237
(0.480)
0.404
(0.497)
-4.500
(3.957)
-0.507
(0.696)
0.305
(0.538)
-0.367
(0.313)
0.263
(0.435)
-0.054**
(0.027)
Individual-level variables
Income
--.--
--.--
--.--
Schooling
--.--
--.--
--.--
Age
--.--
--.--
--.--
Female
--.--
--.--
--.--
Cohabitation
--.--
--.--
--.--
Divorced
--.--
--.--
--.--
Unemployed
--.--
--.--
--.--
Student
--.--
--.--
--.--
Retired
--.--
--.--
--.--
Immigrant
--.--
--.--
--.--
F-test. Contex. cov.=0
--.--
--.--
22.35
[0.005]
--.--
--.--
--.--
2970.32
[0.000]
3256.63
[0.000]
2950.28
[0.000]
3236.22
[0.000]
2030.55
[0.000]
3332.05
[0.000]
174.33
[0.000]
3040.97
[0.000]
3308.1
[0.000]
8197
8197
8197
8197
F-test. Indiv. cov.=0
Cur test: T1= T2
Cut test: T2= T3
Observations
0.140***
(0.031)
0.025
(0.020)
0.017***
(0.002)
0.012
(0.040)
-0.128***
(0.050)
-0.219*
(0.112)
-0.077
(0.059)
0.251***
(0.081)
0.048
(0.079)
0.223**
(0.106)
Notes: (1) Dependent variable: Categorical trust. 4=High trust. 3=Medium-High. 2=Medium-Low. 1=Low.
(2) Standard errors clustered at the municipal level in parentheses; ***: p<0.01. ** p<0.05. *: p<0.1. (3)
Estimation method: Maximum Likelihood. (4) F-test. Contex. cov.=0 is an F-test for the joint significance of
all contextual-level variables; F-test. Indiv. cov.=0 is an F-test for the joint significance of all individual-level
variables. (5) Thresholds tests: test indicating whether the cuts delimiting a category are equal. meaning it is
possible to reduce the number of categories. p-values in brackets.
23
Table 2: Effects of press coverage of political scandals on trust. Ordered Logit results.
(1)
Variables
(2)
***
-0.193
(0.074)
***
(4)
(5)
-0.087
(0.084)
Corr. × Media exposure
--.--
--.--
Corr. × Wide coverage × Media exposure
-0.240***
(0.087)
-0.073
(0.089)
-0.060
(0.072)
-0.234***
(0.087)
--.--
--.--
--.--
--.--
Corruption
Corr. × Wide coverage
--.--
--.--
0.196
(0.070)
0.194***
(0.070)
-0.110
(0.082)
-0.060
(0.073)
-0.159
(0.098)
-0.215***
(0.087)
0.193***
(0.070)
27.89
[0.000]
151.00
[0.000]
2842.05
[0.000]
3354.00
[0.000]
7871
28.06
[0.000]
151.62
[0.000]
2881.42
[0.000]
3334.83
[0.000]
7871
17.98
[0.021]
150.43
[0.000]
2855.25
[0.000]
3355.33
[0.000]
7871
18.96
[0.011]
151.43
[0.000]
2892.79
[0.000]
3335.53
[0.000]
7871
18.66
[0.017]
152.11
[0.000]
2898.39
[0.000]
3328.13
[0.000]
7871
--.--
Media exposure
F-test. Contex. cov.=0
F-test. Indiv. cov.=0
Cut test: T1= T2
Cut test: T2= T3
Observations
-0.181
(0.077)
-0.068
(0.092)
(3)
--.--
***
Notes: (1) See Table 4. (2) Wide coverage =scandals with more than ten news articles; Media exposure =
individuals that obtain their information only from media sources (of kinds of national. regional. and local
media).
Table 3: Marginal effects.
(1)
Trust Category:
(2)
(3)
(4)
Corruption
0.037**
0.013**
-0.033**
-0.017**
Corr. × Wide coverage × Media exposure
0.068**
0.021**
-0.061**
-0.028**
Corr. × (1- Wide coverage) × Media exposure
0.029
0.010
-0.026
-0.012
Corr. × Wide coverage × (1- Media exposure)
0.019
0.007
-0.018
-0.009
Corr. × (1- Wide coverage) × (1- Media exposure)
0.008
0.003
-0.008
-0.003
% Vote turnout
-0.030***
-0.010***
0.026***
0.014***
Log(population)
0.037**
0.013**
-0.034**
-0.016**
-0.040***
-0.014***
0.036***
0.018***
Income
Notes: (1) Marginal effects = difference in the predicted probability as the corruption variable changes
from zero to one; in the interacted case. is the change in the probability as we change from having no
corruption to each of the four mutually exclusive cases.
24
Annex:
Box A.1: Description of the survey
The survey was conducted by “Treball de Camp”, a firm mainly devoted to design
and conduct surveys. The interviews where implemented by telephone. They started
in December 2009 and were finished by February 2010. Due to budget constraints,
it was impossible to consider all the municipalities where at least a corruption
scandal broke out during the period 1999-2009 and their matching. Thus, a
representative sample of municipalities was selected. It is composed by 160 corrupt
municipalities and 131 non-corrupt municipalities. This sample is representative in
three dimensions: i) when the corruption scandals broke out; ii) the municipality
size (in terms of population); iii) and their geographical location (by province). The
number of individuals interviewed varies according to the municipality size. 20
individuals were interviewed in municipalities with less than 10.000 inhabitants; 40
if 10.000<Population ≤ 100.000; 50 if 100.000<Population ≤ 500.000; and 100 if
Population>500.000. The final sample is composed by 9060 interviews. This
sample is also representative in terms of individual characteristics (gender and age)
regarding these characteristics for the whole Spanish population and by
municipality size.
In order to have a high response rate, the survey was designed to be answered in
five minutes. In order to avoid conditioning the answer to the questions, it was
organized as follows. First, there were basic filter questions regarding gender, age,
nationality and municipality where the individual is registered. These questions
were used to design a representative sample. Then, the question about trust was
stated. Then, there was a bloc of questions regarding voting decisions and
individuals’ information. Finally, several socio-demographic characteristics where
asked.
25
Table A.1:
Definition of the variables and descriptive statistics
Variable
Definition
Mean
St.Dev.
Individual-level variables
Trust
Question (1-4); 1: Never; 2: Rarely; 3: Commonly; 4: Always
Income
Age
Self-reported socio-economic classification (1-5): 1: Low;
2: Medium-low; 3: Medium; 4: Medium-High; 5: High
Highest level of education completed (1-5)
1: any studies; 2: primary; 3: secondary; 4: graduate
Age in years
Female
2.259
0.925
2.754
0.799
3.232
1.275
45.465
17.220
Dummy variable coded 1 for females
0.499
0.500
Divorced
Dummy variable coded 1 for people who are divorced or separated
0.042
0.200
Unemployed
Dummy variable coded 1 for people who are unemployed
0.135
0.342
Student
Dummy variable coded 1 for students (do not work)
0.084
0.278
Retired
Dummy variable coded 1 for people who are retired
0.205
0.404
Immigrant
Dummy variable coded 1 for people who are not born in Spain
0.043
0.202
Media exposure
Dummy variable coded 1 for people informed only by media
sources (of kinds of national. regional. and local media)
0.498
0.500
Schooling
Contextual-level variables
Corruption
Dummy variable coded 1 for municipalities with at least one
corruption scandal in the period 1999-2009
0.584
0.493
Wide coverage
Dummy variable coded 1 for scandals with more than ten news
articles
0.208
0.406
% Vote turnout
Average vote turnout at the 1987, 1991 and 1995 local elections
0.707
0.091
Income p.c.
Average socio-economic condition. Arithmetic average of the
socio-economic condition according to their employment status
0.941
0.146
% Divorced
Percentage of divorced and separated among all population
0.020
0.011
% Graduate
Percentage of population with third level studies (diploma, degree
and doctorate) among population 16 years and older
0.082
0.048
% Unemployed
Ethnic diversity
Percentage of unemployed among individuals aged 20-59
1- Σk(Popk/Population)2 where Pop_contk is population whose
nationality is from continent k, and k refers to Europe, Africa,
America and others
Average historical vote share that the right wing parties obtained in
1979, 1982, 1986 and 1989 local elections
0.144
0.105
0.039
0.048
0.406
0.096
Log of the registered population
8.428
1.190
% Right voters
Log(Population)
Notes: (1) Source of the individual-level variables: own-designed survey (see Box A.1). (2) Sources of
the contextual-level variables: (i) 2001 Census of Population (National Institute of Statistics,
www.ine.es), for Income p.c., % Divorced, % Graduate, % , % Unemployed, population by continent
used to construct the Ethnic diversity index, and Population. (ii) Database on corruption scandals,
constructed form an initial list of scandals compiled by Fundación Alternativas and own Internet
searches (see section 3 for more details). (iii) Voting data from the Ministry of the Interior, used for the
construction of the % Right voters and % Vote turnout variables.
26
Table A.2:
Differences in means between Treated and Control groups
Mean
Treated
Control
t-test
[p-value]
Unmatched sample
% Vote turnout
Income p.c.
% Divorced
% Graduate
% Unemployment
Ethnic diversity
% Right voters
log(Population)
4.25 [0.000]
0.541
0.654
0.947
0.939
1.09 [0.282]
0.026
0.018
14.09 [0.000]
0.106
0.077
12.57 [0.000]
0.147
0.143
0.88 [0.381]
0.060
0.035
10.83 [0.002]
0.507
0.505
0.36 [0.724]
96.10
81.82
27.31 [0.003]
0.541
0.537
0.54 [0.683]
0.946
0.942
0.47 [0.642]
0.026
0.026
-0.12 [0.915]
0.105
0.105
0.19 [0.853]
0.147
0.150
-0.40 [0.694]
0.060
0.057
0.86 [0.390]
0.507
0.509
-0.29 [0.772]
96.09
95.82
0.31 [0.753]
Matched sample
% Vote turnout
Income p.c.
% Divorced
% Graduate
% Unemployment
Ethnic diversity
% Right voters
log(Population)
Note: Treated group = municipalities where at least one corruption scandal
broke out during the period 1999-2009; Control group = municipalities where
no corruption scandal broke out during the same period.
27