lying about corruption in surveys: evidence from a joint … · 2018. 1. 9. · lying about...
TRANSCRIPT
-
Lying About Corruption in Surveys: Evidence from a Joint Response Model
Virginia Oliveros Tulane University
Daniel W. Gingerich University of Virginia [email protected]
Word count: 5471
Abstract As is the case with other social undesirable behaviors, survey respondents tend to underreport corrupt behavior. Survey-based estimates utilizing direct questions about corruption are thus likely to underestimate the incidence of this phenomenon. More importantly, bias in reporting may be systematically related to personal characteristics, which makes deriving inferences from such data highly challenging. Recent studies have tried to address this issue by using sensitive survey techniques such as the list experiment or the randomized response technique. This paper studies how individual characteristics shape untruthful reporting in surveys about corruption. It does so by using a novel technique, the joint response model, which allows us to generate unbiased estimates of the extent of corruption among respondents as well as the proclivity towards untruthful reporting among bearers of the sensitive trait. The model permits one to estimate a parameter that captures the probability of a truthful response under direct questioning, which can be used to estimate the proportion of “liars” across groups. Examining the value of this parameter across a variety of different demographic groups, we find that truthfulness under direct questioning seems to differ by citizenship status, income, and to some degree by education, but not by gender and age.
-
1
The literature on the causes and consequences of corruption is voluminous. Work on the topic
has shown how its incidence and form relates to factors such as government regulation, natural
resource abundance, democratic institutional design, bureaucratic structures, and even culture
(e.g., Ades and Di Tella 1999; Barr and Serra 2010; Charron et al. 2017; Dahlström, Lapuente,
and Teorell 2012; Djankov et al. 2002; Fisman and Miguel 2007; Gingerich 2013; Oliveros and
Schuster Forthcoming; Persson, Tabellini, and Trebbi 2003; Ross 2001). Equally well
documented are the negative impacts corruption has on societal wellbeing, be these in terms of
economic growth, public health, education, or democratic legitimacy (e.g., Aidt 2009; Gingerich
2009; Gupta and Abed 2002; Machado, Scartascini, and Tommasi 2011; Mauro 1995; Seligson
2002). Yet, in spite of the progress that has been made in understanding how and why corruption
manifests itself in different ways around the globe, there is still a great deal that needs to be
learned about why certain individuals and not others engage in the practice. One of the obstacles
to such understanding is misrepresentation in survey studies of this phenomenon.
As is the case with other social undesirable behaviors, survey respondents tend to
underreport corrupt behavior, a problem that has long being recognized by scholars working on
corruption (e.g., Treisman 2007). Survey-based estimates utilizing direct questions about
corruption are thus likely to underestimate the incidence of this phenomenon. More importantly,
bias in reporting may be systematically related to personal characteristics, which makes deriving
inferences from such data highly challenging. Recent studies on corruption and other sensitive
behaviors have tried to address this issue by using sensitive survey techniques such as the list
experiment (e.g., Gonzalez‐Ocantos et al. 2012; Gonzalez-Ocantos, Kiewiet de Jonge, and
Nickerson 2015; Malesky, Gueorguiev, and Jensen 2015; Oliveros 2016), endorsement
experiments (e.g., Lyall, Blair, and Imai 2013), or the randomized response technique (e.g.,
-
2
Corbacho et al. 2016; Gingerich 2013). At the same time, scholars have devoted more attention
to the development of new methods and improvements on old ones to study these sensitive topics
(e.g., Corstange 2009; Gingerich 2010; Gingerich et al. 2016; Glynn 2013; Imai 2011; Imai,
Park, and Greene 2015; Blair and Imai 2012; Blair, Imai, and Zhou 2015; Blair, Imai, and Lyall
2014). Yet we still know very little about the characteristics of respondents who are most prone
to lie. In particular, we know very little about the characteristics of respondents who are most
prone to lie when asked about corruption.
Using original survey data from Costa Rica, this paper studies how individual
characteristics shape untruthful reporting in surveys about corruption. It does so by using a novel
technique, the joint response model, which allows us to generate unbiased estimates of the extent
of corruption among respondents as well as the proclivity towards untruthful reporting among
bearers of the sensitive trait (Gingerich et al. 2016). The model permits one to estimate a
parameter that captures the probability of a truthful response under direct questioning, a
parameter which can be used to estimate the proportion of “liars” across groups.1 Examining the
value of this parameter across a variety of different demographic groups, we find that
1 De Jonge (2015) uses list experiments to conduct a similar exercise on another sensitive topic
(vote buying); Eady (2017), in turn, develops a new method to analyze misreporting on survey
responses using the list experiment, which is applied to the study of misreporting of prejudice
across the ideological spectrum.
-
3
truthfulness under direct questioning seems to differ by citizenship status, income, and to some
degree by education, but not by gender and age.2
1. Who lies about corruption?
Several studies examining the effect of gender on crime and illicit behavior find that women are
less inclined towards these activities than men. For instance, in a large tax compliance
experiment conducted in Italy, Sweden, the United Kingdom, and the United States, D’Attoma et
el. (2017) find that women are significantly more compliant than men across all countries.
Similar results have been found with laboratory experiments (Kastlunger et al. 2010; Cadsby,
Maynes, and Trivedi 2006; Torgler and Schaltegger 2005) and survey data (Hasseldine 2002).
Women are also less likely to approve of tax evasion (Torgler and Valev 2010; Torgler and
Schneider 2007). In terms of delinquent behavior more broadly, the gender gap is such that
criminology scholars have argued that: “Males commit more offenses than females at every age,
2 In this paper, we focus strictly on the individual characteristics of the respondents, treating
interviewer characteristics as fixed. For a recent study that focuses on interviewer effects, see
Simpser (2017). He shows that a respondent’s second-order beliefs about the interviewer’s priors
affect the likelihood of obtaining a truthful response for questions about non voting and cheating.
Other characteristics of interviewers that seem to affect responses are ethnicity and race (e.g.
Adida L. et al. 2016; Campbell 1981; Davis 1997), the existence of a personal relationship with
the respondent (e.g. Lazarev, n.d.), perceived religiosity (e.g. Benstead 2014; Blaydes and
Gillum 2013) and gender (e.g Huddy et al. 1997; Kane and Macaulay 1993).
-
4
within all racial or ethnic groups examined to date, and for all but a handful of offense types that
are peculiarly female” (Mears, Ploeger, and Warr 1998, 251).
In relation to corrupt behavior and attitudes towards corruption more specifically, the
literature is also quite vast. Using data from the World Values Survey, Swamy et al. (2001) show
that women are less likely to condone corruption (see also, Dollar, Fisman, and Gatti 2001). In
the same paper, but using survey data from Georgia, these authors also show that officials in
firms owned or managed by men are significantly more likely to be involved in bribe-giving than
those managed or owned by women. Similarly, using data from the World Values Survey and the
European Values Survey from eight Western European countries, Torgler and Valev (2010)
show the existence of significantly greater aversion to corruption among women. Using the same
data set but including 64 countries, Blake (2009) find the same results. Several laboratory
experiments have also found that women are less corrupt than men (e.g., Rivas 2013; Frank and
Schulze 2000).3 In the Latin American context, using survey data covering around 30 years
(1987-2015), Senter et al. (forthcoming) find that in Brazil men are more likely than women to
name corruption as an important problem facing the country. Using a conjoint experiment in the
Dominican Republic public sector, Oliveros and Schuster (Forthcoming) find that female public
employees are perceived by their colleagues to be less corrupt that their fellow male colleagues.
Gingerich et al. (2016) find that in Costa Rica men are substantially more likely than women to
indicate a willingness to bribe as well as to indicate that they have done so in the past. Following 3 Armantier and Boly (2011), in contrast, conducted a field experiment in Burkina Faso
reproducing a corruption scenario in which a grader is offered a bribe for a better grade and find
no effect of gender on the decision to accept the bribe. For an overview of the experimental
evidence, see Chaudhuri (2012).
-
5
the existing literature, we expect social desirability bias to be positively associated with being a
woman. We therefore expect women to be more likely to lie about willingness to engage in
corrupt behavior than men (H1).
Less studied than gender is the effect of age on corruption and attitudes towards
corruption. A few studies, however, have found significant differences between older and
younger people. In the paper mentioned above, Torgler and Valev (2010) find that greater age is
associated with both lower tolerance for corruption and for tax evasion. Similarly, with data from
the World Values Survey, Swamy et al. (2001), also find lower tolerance for corruption (bribery
in their case) among older people. In the aforementioned paper that includes 64 countries, Blake
(2009) also finds that older people are less tolerant of bribe-taking. In a field experiment in
Burkina Faso in which a grader is offered a bribe and asked for a better grade, Armantier and
Boly (2011) find that older graders are significantly less likely to accept bribes. Finally, in a
study conducted in Costa Rica, younger survey respondents were found to be more inclined to
indicate a willingness to bribe than older ones (Corbacho et al. 2016; Gingerich et al. 2016).4 We
thus expect social desirability bias to be positively associated with age. We expect older
respondents to be more likely to lie about willingness to engage in corrupt behavior than
younger respondents (H2).
Another personal characteristic that has been found to be related to corrupt behavior and
attitudes about corruption is educational attainment. In the study by Torgler and Valev (2010)
4 Younger respondents were not, however, more likely to indicate that they had bribed in the past
(Gingerich et al. 2016).
-
6
mentioned above, they find that education is negatively correlated with tolerance for corruption.5
In the Dominican Republic, Oliveros and Schuster (Forthcoming) find that more educated public
sector employees are perceived to be less corrupt than less educated employees. Similarly, in
Costa Rica, individuals with incomplete secondary school education were found to be more
inclined to bribe a police officer than individuals with some exposure to college (Corbacho et al.
2016). Using cross national survey data from the World Justice Project (WJP), Botero et al.
(2013) find that more educated people complain more about government misconduct (when it
happens).6 Interestingly, Morris and Klesner (2010) find that while the more educated are more
likely to experience corruption, the less educated are more likely to perceive it. Following the
literature, we expect social desirability bias to be higher among more educated respondents. We
then expect more educated respondents to be more likely to lie about willingness to engage
in corrupt behavior than less educated respondents (H3).
A number of studies have found an effect of income on corruption and attitudes towards
corruption. Yet, empirical findings remain mixed and contradictory. In the study of Brazilian
voters by Senters et al. (forthcoming) mentioned above, they find that wealthier respondents are
more likely to list corruption as the country’s most important problem. Using household data on
bribery of public officials in Uganda and Peru, Hunt and Laszlo (2012) find that the rich are
5 They find, however, no education effect on the justifiability of tax evasion (Torgler and Valev
2010).
6 Using the International Crime Victims Survey (ICVS), they find a similar effect of education
on the reporting of crime in general (Botero, Ponce, and Shleifer 2013). Similarly, Gonzalez-
Ocantos et al. (2014) find that more educated citizens are less tolerant of clientelism.
-
7
more likely to bribe. Torgler and Valev (2010), in turn, find that wealthier individuals are less
tolerant of corruption.7 Similarly, using data from the World Values Survey but restricting the
sample to only Latin American countries, Blake (2009) finds a negative relationship between
income and tolerance for corruption: wealthier respondents in his sample of 10 Latin American
countries were 19 percent less likely to be tolerant of corruption than the baseline respondent.8
Winters and Weitz-Shapiro (2013), in contrast, find greater tolerance of corruption among the
upper class.9 This finding might be related to the fact that wealthier individuals appear to be less
affected by corruption, at least the type of petty corruption we study here. Indeed, in a field
experiment conducted in Mexico City, Fried, Lagunes, and Venkataramani (2010) found that
police officers engaging in traffic stops were more likely to demand bribes from poorer drivers.
Based on the mixed existing empirical evidence, we are agnostic about the direction of the effect
of wealth on the veracity of reporting. We expect wealthier respondents to be more (less)
likely to lie about willingness to engage in corrupt behavior than poorer respondents (H4,
H4’).
Finally, there are good reasons to suspect that vulnerable groups like immigrants would
be less likely to openly express a willingness to engage in illegal behavior or to admit to having
7 However, they find no effect of wealth on tolerance towards tax evasion (Torgler and Valev
2010).
8 In line with these findings, Weitz-Shapiro (2014) has found that wealthier voters are less likely
to support politicians who engage in clientelism.
9 Note that, in their case, class is constructed with an index that includes different household
goods and services, but also education.
-
8
been engaged in such behavior in the past. Given their more precarious status, non-citizens might
be more worried about the potential legal consequences of their answers. Indeed, in Costa Rica,
non-citizens were found to be more reluctant to respond honestly when asked directly about
paying a bribe (Gingerich et al. 2016). We expect social desirability bias and potential legal
concerns to be higher among non-citizens. We then expect non-citizens to be more likely to lie
about willingness to engage in corrupt behavior than citizens (H5).
2. Empirical Strategy
In order to estimate untruthful responses about corruption across groups, we use the joint
response model that combines individual responses about a sensitive issue based on both direct
questioning and a sensitive survey technique (SST) (Gingerich et al. 2016). Two different
questions about corruption were included in the survey. The first one (Q1) asks respondents
about their willingness to pay a bribe to a police officer in order to avoid paying a traffic ticket;
the second one (Q2) asks respondents whether they had actually bribed a police officer in the
past to avoid paying a traffic ticket. The questions were presented twice to the respondents. First,
they were presented in the format of a particular type of SST called the crosswise model (Tan,
Tian, and Tang 2009). This technique provides full anonymity to respondents by combining the
response to an innocuous question (in our case, the birth month of the respondent) with the
response to the sensitive item (in our case, corruption). At the end of the survey, the same
questions were asked directly, with the explicit option of choosing “prefer not to respond” also
offered to respondents. Responses about willingness to bribe were thus a discrete combination of
responses under the protection afforded by the SST and the absence of protection under direct
-
9
questioning.10
Figure 1 presents the question about willingness to bribe a police officer based on the
crosswise model. The respondent were presented with two statements and asked how many were
true. The first statement (the innocuous one) refers to the respondent’s mother birth month,
stating that the respondent’s mother was born in October, November, or December.11 The second
statement (the sensitive one) denotes a willingness to pay a bribe in order to avoid paying a
traffic ticket. The privacy of the responses was protected by constraining the response options to
only two potential responses. Response A, indicating that either both statements are true or
neither statement is true, and response B, indicating that only one of the two statements is true,
without specifying which is true. The respondents’ anonymity is guaranteed since neither of the
two responses necessarily indicates willingness to bribe.
Figure 1: Crosswise Survey Question on Willingness to Bribe a Police officer
10 For a detailed description of the technique, see Gingerich et al. (2016).
11 In order to calculate the probability of having one’s mother born in October, November, or
December, we conducted a nationally representative telephone survey of 1,200 respondents. The
survey queried respondents directly about the birthday of their mother. These responses were
then checked against data from Costa Rica’s National Institute for Statistics and Censuses
(INEC) on month of birth for newborns for the 2000–11 period. The figures from the phone
survey self-reports and census data were essentially identical. See Online Appendix for more
details.
-
10
If we denote the outcome when respondent i is queried directly about the sensitive trait or
behavior as 𝑦!! ∈ {0 (absent),1 (present),∅ (prefer not to respond )} and the outcome when i is queried
about the sensitive trait or behavior using the crosswise technique as 𝑦!! = 0 ("A"), 1 ("B") ,
then the total observed outcomes for each respondent is captured by the vector Yi=(yiA, yiD). Any
realization of this vector belongs to the total observed response set, which is an array with six
elements, 𝒴 ∈ { 0,0 , 0,1 , 1,0 , 1,1 , ∅, 0 , (∅, 1)}. The joint response framework provides a
model for which of the six elements in this set will obtain.
The modeling strategy rests on two key assumptions. The first is called honesty given
protection: given the protection afforded by the sensitive question technique (the crosswise
model), all respondents are assumed to respond as prompted by the technique (cf. Gingerich
2010; Blair and Imai 2012). Thus, if lying occurs in the survey responses, it is assumed to occur
only when respondents are prompted to respond directly about the sensitive trait. The second
-
11
assumption is called one-sided lying: individuals who bear the sensitive trait may either lie about
their status or refuse to respond when queried directly but those who do not bear the sensitive
trait always either tell the truth or refuse to respond, they never falsely claim to bear the sensitive
trait.
Exploiting these assumptions, the joint response model permits one to estimate a number
of relevant parameters. Most fundamentally, the model permits one to estimate the prevalence of
the sensitive trait or behavior, which we denote by 𝜋. In addition to the prevalence rate, the
model permits one to estimate the value of a set of key diagnostic parameters capturing
respondent evasiveness under direct questioning. These parameters are 𝜆!!, which represents the
probability of a truthful response under direct questioning when the respondent bears the
sensitive trait, 𝜆!!, which represents the probability of an untruthful response under direct
questioning when the respondent bears the sensitive trait, and 𝜆!!, which represents the
probability of a truthful response (as opposed to non-response) when the respondent does not
bear the sensitive trait. In examining who lies about corruption under direct questioning, we will
be examining the values of these diagnostic parameters.
To explore how individual characteristics and experiences shape untruthful reporting
about corruption, we analyze original survey data from Costa Rica.12 The survey consisted of
face-to-face interviews of 4,200 residents 18 years or older in what is known as the Gran Área
Metropolitana (GAM). This area is the principal urban center in Costa Rica (2.6 million
12 Costa Rica scores relatively well on corruption, when compared to other countries. According
to the 2016 Corruption Perception Index by Transparency International, Costa Rica is ranked
41/176 (same as Brunei and Spain).
-
12
residents), accounting for 60% of the country’s population.13 The survey was administered
between October 2013 and April 2014, by the firm Borge y Asociados.14
3. Results
We begin our analysis by showing that respondents, in general, do lie about corruption. After
estimating the level of truthful responses for the full sample, the rest of the analysis is focused on
how individual characteristics shape untruthful reporting. In particular, we focus on gender (H1),
age (H2), education (H3), wealth (H4, H4’), and citizenship (H5). Figure 2 presents the
prevalence estimates from direct questioning, SST questioning only, and the parameter estimates
provided by the joint response for the entire sample for both corruption questions (Q1 & Q2).15
Figure 2: Willingness to Bribe and Past Bribing, Whole Sample
13 The GAM includes 30 cantons in the provinces of Alajuela, Cartago, Heredia, and San José.
14 See the Online Appendix A for more details on the survey methodology and execution.
15 The exact numeric values are reported in Table A2 in the Online Appendix.
-
13
Notes: Parameter estimates for questions on corruption (whole sample). Black circles indicate
point estimates for model parameters. Ninety-five percent confidence intervals denoted by
horizontal black lines.
Figure 2 shows that, in line with our expectations and previous findings, people respond
differently when asked about corruption under a format that provides anonymity: respondents do
lie about corruption. In both cases, the estimates of willingness to bribe based on direct responses
are below those based on the joint response model and crosswise responses only. According to
the estimates based on the joint response model for Q1, the proportion of respondents who would
0.0 0.1 0.2 0.3 0.4 0.5
Prevalence estimate (π̂)
To avoid paying a traffic ticket, I would be willing to pay a bribe to a police officer
Direct only
SST only
Joint
response type
0.0 0.2 0.4 0.6 0.8 1.0
Diagnostic parameters (joint response model)
λ1T
λ1L
λ0T
0.0 0.1 0.2 0.3 0.4 0.5
Prevalence estimate (π̂)
I have paid, at least once, a bribe to a police officer to avoid a traffic ticket
Direct only
SST only
Joint
response type
0.0 0.2 0.4 0.6 0.8 1.0
Diagnostic parameters (joint response model)
λ1T
λ1L
λ0T
-
14
be willing to bribe a police officer was 0.29 [0.26, 0.32], whereas according to the crosswise
model and direct question the proportion was 0.22 [0.18, 0.25] and 0.18 [0.17, 0.19],
respectively.16 For Q2 the estimated proportion of respondents that admitted having paid a bribe
in the past was 0.16 [0.13, 0.19] based on the joint response model, compared to 0.13 [0.10,
0.16] with the crosswise model and 0.09 [0.08, 0.10] with direct questioning. Evasiveness under
direct questioning was higher for Q2, in which respondents were asked to implicate themselves
in having engaged in an illicit act. Indeed, the value of the parameter capturing the probability of
a truthful response under direct questioning was lower for Q2 (𝜆!!= 0.54 [0.45, 0.65]) than for
Q1 (𝜆!!= 0.61 [0.55, 0.69]). Respondents appeared more inclined to respond truthfully about
willingness to bribe a police officer than about having actually done so (although the difference
between the two estimates is not significant).
We now move on to focus on how personal characteristics may affect a respondent’s
willingness to provide (or not) a truthful response when asked directly. Figures 3 to 7 present the
prevalence estimates from direct questioning, SST questioning only, and the parameter estimates
provided by the joint response model for gender (female, male), age (less than 28, 29-42, 43 and
older), education (some university education, secondary education or less), wealth (low, middle,
high), and citizenship (citizens, non-citizens).17 We focus here only on the question about
willingness to bribe (Q1). Results for Q2 are similar and can be found on Online Appendix B.
16 Ninety-five percent confidence intervals in square brackets.
17 In order to measure wealth, a two-factor item response theory (IRT) model was estimated
utilizing respondent information on ownership of a cellphone, laptop, tablet, car, widescreen
television, cable television, and internet access in the home. The IRT model was estimated using
the R package ltm (Rizopoulos 2006). After estimation, factor scores for each respondent were
-
15
Figure 3: Willingness to Bribe, by Gender
Notes: Parameter estimates for willingness to bribe (by gender). Black circles indicate point
estimates for model parameters. Ninety-five percent confidence intervals denoted by horizontal
black lines.
First, as in the previous Figure reporting the results for the whole sample, Figure 3 shows that
both men and women respond quite differently when asked under the sensitive survey technique.
In both the case of men and women, the estimates of willingness to bribe based on direct
questioning (0.24 [0.22, 0.26] for men, 0.12 [0.11, 0.14] for women) are below those based on
crosswise model (0.28 [0.24, 0.33] for men, 0.15 [0.10, 0.19] for women) and the joint response
model (0.37 [0.33, 0.41] for men, 0.20 [0.16, 0.25] for women). At the same time, and in line
with the existing literature (e.g. Blake 2009; Gingerich et al. 2016; Swamy et al. 2001; Torgler
and Valev 2010), the Figure clearly shows that men are significantly more likely to express a
willingness to bribe across all response types. Men simply seem to be more corrupt than women. calculated using the empirical Bayes method. A three-value, ordinal indicator of wealth (low,
middle, high) was then created based on the terciles of the factor scores.
0.0 0.1 0.2 0.3 0.4 0.5
Prevalence estimate (π̂)
To avoid paying a traffic ticket, I would be willing to pay a bribe to a police officer
Direct onlySST only
Joint
Direct onlySST only
Joint
response type
Men (n=2,096)
Women (n=2,097)
0.0 0.2 0.4 0.6 0.8 1.0
Diagnostic parameters (joint response model)
Men (n=2,096)
Women (n=2,097)
λ1T
λ1L
λ0T
λ1T
λ1L
λ0T
0.0 0.1 0.2 0.3 0.4 0.5
Prevalence estimate (π̂)
I have paid, at least once, a bribe to a police officer to avoid a traffic ticket
Direct onlySST only
Joint
Direct onlySST only
Joint
response type
Men (n=2,096)
Women (n=2,091)
0.0 0.2 0.4 0.6 0.8 1.0
Diagnostic parameters (joint response model)
Men (n=2,096)
Women (n=2,091)
λ1T
λ1L
λ0T
λ1T
λ1L
λ0T
-
16
Finally, we find no support for H1: women are not more likely to lie about engaging in corrupt
behavior than men. Indeed, the diagnostic parameters are quite similar for men and women. The
value of the parameter that captures the probability of a truthful response under direct
questioning by those who are actually willing to bribe, 𝜆!!, is 0.63 [0.56, 0.71] for men 0.58
[0.48, 0.72] for women. We find no support for H1 either when asked about having paid a bribe
in the past (Q2) (see Table A3 in Online Appendix).
Figure 4: Willingness to Bribe, by Age
Notes: Parameter estimates for willingness to bribe (by age). Black circles indicate point
estimates for model parameters. Ninety-five percent confidence intervals denoted by horizontal
black lines.
In line with the existing literature (e.g. Armantier and Boly 2011; Blake 2009; Swamy et al.
2001; Torgler and Valev 2010), Figure 4 shows that, across the different estimation strategies,
older respondents are less likely to express a willingness to bribe a police officer (joint response
estimates: 0.39 [0.33, 0.44] for those less than 28 years old, 0.31 [0.26, 0.36] for those between
0.0 0.1 0.2 0.3 0.4 0.5
Prevalence estimate (π̂)
To avoid paying a traffic ticket, I would be willing to pay a bribe to a police officer
Direct onlySST onlyJoint
Direct onlySST onlyJoint
Direct onlySST onlyJoint
response typeLess than 28 years old (n=1,295)
Between 28 and 43 years old (n=1,463)
Greater than or equal to 43 years old (n=1,434)
0.0 0.2 0.4 0.6 0.8 1.0
Diagnostic parameters (joint response model)
Less than 28 years old (n=1,295)
Between 28 and 43 years old (n=1,463)
Greater than or equal to 43 years old (n=1,434)
λ1T
λ1L
λ0T
λ1T
λ1L
λ0T
λ1T
λ1L
λ0T
-
17
28 and 43 years old, and 0.18 [0.13, 0.23] for those 43 and older). We find, however, no support
for H2. The diagnostic parameters are indeed quite similar across the three age groups. The value
of the parameter that captures the probability of a truthful response under direct questioning by
those who are actually willing to bribe (𝜆!!) is 0.65 [0.58, 0.74] for those less than 28 years old,
0.60 [0.52, 0.71] for those between 28 and 43 years old, and 0.55 [0.42,0.76] for those 43 and
older. Older respondents are not more likely to lie about their willingness to pay a bribe than
younger respondents, nor are they more likely to lie about having paid a bribe in the past (Q2)
(see Table A4 in Online Appendix).
Figure 5: Willingness to Bribe, by Education
Notes: Parameter estimates for willingness to bribe (by education). Black circles indicate point
estimates for model parameters. Ninety-five percent confidence intervals denoted by horizontal
black lines.
Figure 5 reveals that there is a slight difference in willingness to bribe across education groups.
According to the joint response estimates, the prevalence of such willingness was 0.22 [0.16,
0.29] among those with some university education versus 0.30 [0.27, 0.33] among those with a
0.0 0.1 0.2 0.3 0.4 0.5
Prevalence estimate (π̂)
To avoid paying a traffic ticket, I would be willing to pay a bribe to a police officer
Direct onlySST only
Joint
Direct onlySST only
Joint
response type
Some university education (n=731)
Secondary school or less (n=3,308)
0.0 0.2 0.4 0.6 0.8 1.0
Diagnostic parameters (joint response model)
Some university education (n=731)
Secondary school or less (n=3,308)
λ1T
λ1L
λ0T
λ1T
λ1L
λ0T
0.0 0.1 0.2 0.3 0.4 0.5
Prevalence estimate (π̂)
I have paid, at least once, a bribe to a police officer to avoid a traffic ticket
Direct onlySST only
Joint
Direct onlySST only
Joint
response type
Some university education (n=732)
Secondary school or less (n=3,301)
0.0 0.2 0.4 0.6 0.8 1.0
Diagnostic parameters (joint response model)
Some university education (n=732)
Secondary school or less (n=3,301)
λ1T
λ1L
λ0T
λ1T
λ1L
λ0T
-
18
secondary education or less (with some overlap in the confidence intervals of these two
estimates). However, our findings on truthfulness under direct questioning ran counter to
expectations based on the literature. Whereas in his study of vote buying De Jonge (2015, 721)
finds that “the educated, who are more likely to be aware of and understand social norms against
vote buying, are more likely to lie,” we found the opposite: that college educated respondents, if
anything, were more likely to be truthful about corruption than those with lesser levels of
education. The value of the parameter that captures the probability of a truthful response under
direct questioning by those who are actually willing to bribe, 𝜆!!, is 0.67 [0.51, 0.92] for those
with some college education and 0.60 [0.54, 0.67] for those with a secondary education or less.
Yet we would not place too much emphasis on these differences, as the confidence intervals for
the two estimates exhibit substantial overlap. We find a similar pattern when using the
alternative question (Q2) about having bribe in the past (see Table A5 in the Online Appendix).
Figure 6: Willingness to Bribe, by Wealth
Notes: Parameter estimates for willingness to bribe (by wealth). Black circles indicate point
estimates for model parameters. Ninety-five percent confidence intervals denoted by horizontal
black lines.
0.0 0.1 0.2 0.3 0.4 0.5
Prevalence estimate (π̂)
To avoid paying a traffic ticket, I would be willing to pay a bribe to a police officer
Direct onlySST onlyJoint
Direct onlySST onlyJoint
Direct onlySST onlyJoint
response typeLow material wealth (n=1,535)
Moderate material wealth (n=1,286)
High material wealth (n=1,372)
0.0 0.2 0.4 0.6 0.8 1.0
Diagnostic parameters (joint response model)
Low material wealth (n=1,535)
Moderate material wealth (n=1,286)
High material wealth (n=1,372)
λ1T
λ1L
λ0T
λ1T
λ1L
λ0T
λ1T
λ1L
λ0T
-
19
Turning to the analysis of wealth, we find no clear relationship between willingness to bribe a
police officer and wealth. Richer (poorer) respondents are not more (less) likely to express
willingness to engage in corrupt behavior. However, we do find a strong negative relationship
between wealth and evasiveness in responses. When asked directly, richer respondents are more
inclined to respond truthfully about willingness to bribe. Indeed, the value of the parameter that
captures the probability of a truthful response under direct questioning by those who are actually
willing to bribe is significantly higher for the richer group of respondents (𝜆!!= 0.80 [0.67, 0.94])
than it is for the middle group (𝜆!!= 0.55 [0.47, 0.66]) and the poorest group (𝜆!!= 0.51 [0.43,
0.61]). Poorest respondents bearing the sensitive trait were roughly equally likely to tell the truth
and lie under direct questioning (𝜆!!= 0.51, 𝜆!!= 0.45), while the richest respondents were
significantly more likely to tell the truth when asked directly (𝜆!!= 0.80, 𝜆!!= 0.17). We find
similar results when asking about having paid a bribe in the past (Q2) (see Table A6 in Online
Appendix).
Figure 7: Willingness to Bribe, by Nationality
Notes: Parameter estimates for willingness to bribe (by citizenship). Black circles indicate point
0.0 0.1 0.2 0.3 0.4 0.5
Prevalence estimate (π̂)
To avoid paying a traffic ticket, I would be willing to pay a bribe to a police officer
Direct onlySST onlyJoint
Direct onlySST onlyJoint
response type
Costa Rican Nationals (n=3,799)
Non-Costa Rican Nationals (n=394)
0.0 0.2 0.4 0.6 0.8 1.0
Diagnostic parameters (joint response model)
Costa Rican Nationals (n=3,799)
Non-Costa Rican Nationals (n=394)
λ1T
λ1L
λ0T
λ1T
λ1L
λ0T
0.0 0.1 0.2 0.3 0.4 0.5
Prevalence estimate (π̂)
I have paid, at least once, a bribe to a police officer to avoid a traffic ticket
Direct onlySST onlyJoint
Direct onlySST onlyJoint
response type
Costa Rican Nationals (n=3,799)
Non-Costa Rican Nationals (n=394)
0.0 0.2 0.4 0.6 0.8 1.0
Diagnostic parameters (joint response model)
Costa Rican Nationals (n=3,799)
Non-Costa Rican Nationals (n=394)
λ1T
λ1L
λ0T
λ1T
λ1L
λ0T
-
20
estimates for model parameters. Ninety-five percent confidence intervals denoted by horizontal
black lines.
Finally, Figure 7 shows strong support for our H5. Non-citizens are indeed more likely to lie
about willingness to engage in corrupt behavior than citizens. Under direct questioning, both
Costa Ricans and Non-Costa Ricans report an affirmative response proportion of 0.18 about
willingness to bribe a police officer. However, when asked under the protective formats (both the
SST and the joint response model), Non-Costa Ricans (0.31 and 0.37, respectively) show a
higher affirmative response than Costa Ricans (0.21 and 0.28). The value of the truthful response
parameter (𝜆!!) that captures the probability of an honest response when ask directly by those
who are willing to bribe is significantly lower for Non-Costa Ricans (𝜆!! =0.45 [0.34, 0.60]) than
it is for Costa Ricans (𝜆!!= 0.63 [0.58, 0.72]). Indeed, while many Costa Ricans bearing the
sensitive trait were willing to tell the truth under direct questioning (𝜆!! =0.63, 𝜆!!= 0.33), Non-
Costa Ricans were about equally likely to lie or tell the truth under direct questioning (𝜆!!= 0.45,
𝜆!!= 0.48). For the question that asked respondents to directly implicate themselves in having
paid a bribe in the past (Q2), we also found that the probability of a truthful response under
direct questioning by those who had bribed in the past is lower for Non-Costa Ricans than it is
for Costa Ricans, although this difference is not significant at conventional levels (see Table A7
in Online Appendix).
-
21
4. Discussion
This paper has examined how the demographic characteristics of survey respondents relate to
their inclination to lie about corruption in the context of a survey. Applying a joint response
framework—one that makes use of both protected and direct questions about corrupt activity—to
original survey data from urban Costa Rica, the paper presented estimates of the propensity to lie
about corruption under direct questioning across a wide array of demographic groups. We find
that wealth and citizenship status, and to a lesser extent, education, are relevant predictors of the
propensity to lie, but gender and age are not.
How might one make sense of these findings? We would submit that the relevant
distinction between the demographic categories that demonstrate variation in the propensity to lie
and those that do not is degree of vulnerability. Individuals belonging to vulnerable populations
are likely to believe that they have the greatest to lose from being forthcoming in direct questions
about corruption (or, for that matter, any other illegal or socially stigmatizing behavior). The
patterns of responses we observe seem to reflect this reality. For instance, non-citizens,
especially those who are undocumented, may be more concerned about any potential legal
repercussions from honest reporting than citizens, whose legal rights are almost certainly more
expansive. Similarly, poor citizens may have greater apprehension about how honest reporting in
direct surveys could potentially place them in legal jeopardy, since they are unlikely to have the
resources or contacts to navigate the legal system should the need to do so arise. Less educated
citizens may fear forthrightness about corruption for the same reason. By contrast, gender and
age—as relevant as they are for understanding the incidence of corruption—play a less important
role in variation in misreporting since, in the case of modern day Costa Rica, differences in these
characteristics do not seem to map cleanly onto differences in vulnerability.
-
22
The findings also cast a new light on existing results in the literature on corruption. The
well-documented aversion of women to corruption recorded in direct response surveys likely
reflects real differences in behavioral and attitudinal tendencies across genders. Whether this
difference will persist over time as more and more women populate the realms of political and
bureaucratic power remains to be seen. Nevertheless, our findings give us no reason to doubt the
descriptive snapshot provided by extant work on the link between gender and corruption.
On the other hand, our findings linking low education and wealth to evasiveness about
corruption would seem to offer two potential amendments to existing research. First, the
recorded negative effects of education on corruption may be systematically underestimated due
to the fact that lesser educated respondents are more inclined to lie about this topic. Education
may be an even more powerful antidote to corruption than previously thought. Second, the mixed
findings on the link between wealth and corruption in existing work may be due, at least in part,
to differences across income groups in the willingness to respond honestly to questions about
corruption. It is possible that wealth might have had a more consistent reductive effect on
corruption in the literature were it not for the greater inclination of poorer respondents to be
evasive in responding to direct questions on this issue.
5. References Ades, Alberto, and Rafael Di Tella. 1999. “Rents, Competition, and Corruption.” The American
Economic Review 89 (4):982–93.
Adida L., Claire, Karen E. Ferree, Daniel N. Posner, and Amanda Lea Robinson. 2016. “Who’s
Asking? Interviewer Coethnicity Effects in African Survey Data.” Comparative Political
Studies 49 (12):1630–60. https://doi.org/10.1177/0010414016633487.
-
23
Aidt, Toke S. 2009. “Corruption, Institutions, and Economic Development.” Oxford Review of
Economic Policy 25 (2):271–91.
Armantier, Olivier, and Amadou Boly. 2011. “A Controlled Field Experiment on Corruption.”
European Economic Review 55 (8):1072–82.
Barr, Abigail, and Danila Serra. 2010. “Corruption and Culture: An Experimental Analysis.”
Journal of Public Economics 94 (11):862–69.
Benstead, Lindsay J. 2014. “Does Interviewer Religious Dress Affect Survey Responses?
Evidence from Morocco.” Politics and Religion 7 (4):734–60.
Blair, Graeme, and Kosuke Imai. 2012. “Statistical Analysis of List Experiments.” Political
Analysis 20 (1):47–77.
Blair, Graeme, Kosuke Imai, and Jason Lyall. 2014. “Comparing and Combining List and
Endorsement Experiments: Evidence from Afghanistan.” American Journal of Political
Science 58 (4):1043–63.
Blair, Graeme, Kosuke Imai, and Yang-Yang Zhou. 2015. “Design and Analysis of the
Randomized Response Technique.” Journal of the American Statistical Association 110
(511):1304–19.
Blake, Charles H. 2009. “Public Attitudes toward Corruption.” Corruption and Democracy in
Latin America, University of Pittsburgh Press: Pittsburgh, 103–16.
Blaydes, Lisa, and Rachel M. Gillum. 2013. “Religiosity-of-Interviewer Effects: Assessing the
Impact of Veiled Enumerators on Survey Response in Egypt.” Politics and Religion 6
(3):459–82.
Botero, Juan, Alejandro Ponce, and Andrei Shleifer. 2013. “Education, Complaints, and
Accountability.” The Journal of Law and Economics 56 (4):959–96.
-
24
Cadsby, C. Bram, Elizabeth Maynes, and Viswanath Umashanker Trivedi. 2006. “Tax
Compliance and Obedience to Authority at Home and in the Lab: A New Experimental
Approach.” Experimental Economics 9 (4):343–59.
Campbell, Bruce A. 1981. “Race-of-Interviewer Effects among Southern Adolescents.” Public
Opinion Quarterly 45 (2):231–44.
Charron, Nicholas, Carl Dahlström, Victor Lapuente, and Mihaly Fazekas. 2017. “Careers,
Connections, and Corruption Risks: Investigating the Impact of Bureaucratic Meritocracy
on Public Procurement Processes.” The Journal of Politics 79 (1):000–000.
Chaudhuri, Ananish. 2012. “Chapter 2 Gender and Corruption: A Survey of the Experimental
Evidence.” In New Advances in Experimental Research on Corruption, 13–49. Emerald
Group Publishing Limited.
Corbacho, Ana, Daniel W. Gingerich, Virginia Oliveros, and Mauricio Ruiz‐Vega. 2016.
“Corruption as a Self‐Fulfilling Prophecy: Evidence from a Survey Experiment in Costa
Rica.” American Journal of Political Science 60 (4):1077–92.
Corstange, Daniel. 2009. “Sensitive Questions, Truthful Answers? Modeling the List Experiment
with LISTIT.” Political Analysis 17 (1):45–63.
Dahlström, Carl, Victor Lapuente, and Jan Teorell. 2012. “The Merit of Meritocratization:
Politics, Bureaucracy, and the Institutional Deterrents of Corruption.” Political Research
Quarterly 65 (3):656–68.
D’Attoma, John, Clara Volintiru, and Sven Steinmo. 2017. “Willing to Share? Tax Compliance
and Gender in Europe and America.” Research & Politics 4 (2):2053168017707151.
Davis, Darren W. 1997. “Nonrandom Measurement Error and Race of Interviewer Effects
among African Americans.” The Public Opinion Quarterly 61 (1):183–207.
-
25
De Jonge, Chad P. Kiewiet. 2015. “Who Lies About Electoral Gifts?” Public Opinion Quarterly
79 (3):710–39. https://doi.org/10.1093/poq/nfv024.
Djankov, Simeon, Rafael La Porta, Florencio Lopez-de-Silanes, and Andrei Shleifer. 2002. “The
Regulation of Entry.” The Quarterly Journal of Economics 117 (1):1–37.
Dollar, David, Raymond Fisman, and Roberta Gatti. 2001. “Are Women Really the ‘Fairer’ Sex?
Corruption and Women in Government.” Journal of Economic Behavior & Organization
46 (4):423–29.
Eady, Gregory. 2017. “The Statistical Analysis of Misreporting on Sensitive Survey Questions.”
Political Analysis, 1–19.
Fisman, Raymond, and Edward Miguel. 2007. “Corruption, Norms, and Legal Enforcement:
Evidence from Diplomatic Parking Tickets.” Journal of Political Economy 115 (6):1020–
48.
Frank, Björn, and Günther G. Schulze. 2000. “Does Economics Make Citizens Corrupt?”
Journal of Economic Behavior & Organization 43 (1):101–13.
Fried, Brian J., Paul Lagunes, and Atheendar Venkataramani. 2010. “Corruption and Inequality
at the Crossroad: A Multimethod Study of Bribery and Discrimination in Latin America.”
Latin American Research Review 45 (1):76–97.
Gingerich, Daniel W. 2009. “Corruption and Political Decay: Evidence from Bolivia.” Quarterly
Journal of Political Science 4 (1):1–34.
———. 2010. “Understanding Off-the-Books Politics: Conducting Inference on the
Determinants of Sensitive Behavior with Randomized Response Surveys.” Political
Analysis 18 (3):349–80.
-
26
———. 2013. Political Institutions and Party-Directed Corruption in South America: Stealing
for the Team. New York: Cambridge University Press.
Gingerich, Daniel W., Virginia Oliveros, Ana Corbacho, and Mauricio Ruiz-Vega. 2016. “When
to Protect? Using the Crosswise Model to Integrate Protected and Direct Responses in
Surveys of Sensitive Behavior.” Political Analysis 24 (2):132–56.
Glynn, Adam N. 2013. “What Can We Learn with Statistical Truth Serum? Design and Analysis
of the List Experiment.” Public Opinion Quarterly 77 (S1):159–72.
Gonzalez Ocantos, Ezequiel, Chad Kiewiet Jonge, and David W. Nickerson. 2014. “The
Conditionality of Vote‐Buying Norms: Experimental Evidence from Latin America.”
American Journal of Political Science 58 (1):197–211.
Gonzalez‐Ocantos, Ezequiel, Chad Kiewiet De Jonge, Carlos Meléndez, Javier Osorio, and
David W. Nickerson. 2012. “Vote Buying and Social Desirability Bias: Experimental
Evidence from Nicaragua.” American Journal of Political Science 56 (1):202–17.
Gonzalez-Ocantos, Ezequiel, Chad Kiewiet de Jonge, and David W. Nickerson. 2015.
“Legitimacy Buying: The Dynamics of Clientelism in the Face of Legitimacy
Challenges.” Comparative Political Studies 48 (9):1127–58.
Gupta, Mr Sanjeev, and Mr George T. Abed. 2002. Governance, Corruption, and Economic
Performance. International Monetary Fund.
Hasseldine, John. 2002. “Gender Differences in Tax Compliance.” Taxation: Critical
Perspectives on the World Economy 3:125–39.
Huddy, Leonie, Joshua Billig, John Bracciodieta, Lois Hoeffler, Patrick J. Moynihan, and
Patricia Pugliani. 1997. “The Effect of Interviewer Gender on the Survey Response.”
Political Behavior 19 (3):197–220.
-
27
Hunt, Jennifer, and Sonia Laszlo. 2012. “Is Bribery Really Regressive? Bribery’s Costs,
Benefits, and Mechanisms.” World Development 40 (2):355–72.
Imai, Kosuke. 2011. “Multivariate Regression Analysis for the Item Count Technique.” Journal
of the American Statistical Association 106 (494):407–16.
Imai, Kosuke, Bethany Park, and Kenneth F. Greene. 2015. “Using the Predicted Responses
from List Experiments as Explanatory Variables in Regression Models.” Political
Analysis 23 (2):180–96.
Kane, Emily W., and Laura J. Macaulay. 1993. “Interviewer Gender and Gender Attitudes.”
Public Opinion Quarterly 57 (1):1–28.
Kastlunger, Barbara, Stefan G. Dressler, Erich Kirchler, Luigi Mittone, and Martin Voracek.
2010. “Sex Differences in Tax Compliance: Differentiating between Demographic Sex,
Gender-Role Orientation, and Prenatal Masculinization (2D: 4D).” Journal of Economic
Psychology 31 (4):542–52.
Lazarev, Egor. n.d. “Surveying in Insecure Environments: A Case Study of Using Local
Interviewers in Chechnya.” Working Paper.
Lyall, Jason, Graeme Blair, and Kosuke Imai. 2013. “Explaining Support for Combatants during
Wartime: A Survey Experiment in Afghanistan.” American Political Science Review 107
(04):679–705.
Machado, Fabiana, Carlos Scartascini, and Mariano Tommasi. 2011. “Political Institutions and
Street Protests in Latin America.” Journal of Conflict Resolution 55 (3):340–65.
Malesky, Edmund J, Dimitar D Gueorguiev, and Nathan M Jensen. 2015. “Monopoly Money:
Foreign Investment and Bribery in Vietnam, a Survey Experiment.” American Journal of
Political Science 59 (2):419–39.
-
28
Mauro, Paolo. 1995. “Corruption and Growth.” The Quarterly Journal of Economics 110
(3):681–712.
Mears, Daniel P., Matthew Ploeger, and Mark Warr. 1998. “Explaining the Gender Gap in
Delinquency: Peer Influence and Moral Evaluations of Behavior.” Journal of Research in
Crime and Delinquency 35 (3):251–66.
Morris, Stephen D., and Joseph L. Klesner. 2010. “Corruption and Trust: Theoretical
Considerations and Evidence from Mexico.” Comparative Political Studies 43
(10):1258–85.
Oliveros, Virginia. 2016. “Making It Personal: Clientelism, Favors, and the Personalization of
Public Administration in Argentina.” Comparative Politics 48 (3):373–91.
Oliveros, Virginia, and Christian Schuster. Forthcoming. “Merit, Tenure, and Bureaucratic
Behaviour: Evidence from a Conjoint Experiment in the Dominican Republic.”
Comparative Political Studies.
Persson, Torsten, Guido Tabellini, and Francesco Trebbi. 2003. “Electoral Rules and
Corruption.” Journal of the European Economic Association 1 (4):958–89.
Rivas, M. Fernanda. 2013. “An Experiment on Corruption and Gender.” Bulletin of Economic
Research 65 (1):10–42.
Rizopoulos, Dimitris. 2006. “Ltm: An R Package for Latent Variable Modeling and Item
Response Theory Analyses.” Journal of Statistical Software 17 (5):1–25.
Ross, Michael L. 2001. Timber Booms and Institutional Breakdown in Southeast Asia.
Cambridge University Press.
Seligson, Mitchell A. 2002. “The Impact of Corruption on Regime Legitimacy: A Comparative
Study of Four Latin American Countries.” Journal of Politics 64 (2):408–33.
-
29
Senters, Kelly, Rebecca Weitz-Shapiro, and Matthew S. Winters. forthcoming. “Continuity and
Change in Public Attitudes toward Corruption.” In Routledge Handbook of Brazilian
Politics, Barry Ames, Ed.
Simpser, Alberto. 2017. “Why Do Sensitive Survey Questions Elicit Truthful Answers? Theory
and Evidence with Application to the RRT and the List Experiment.” Available at SSRN:
Https://Ssrn.Com/Abstract=3032684.
Swamy, Anand, Stephen Knack, Young Lee, and Omar Azfar. 2001. “Gender and Corruption.”
Journal of Development Economics 64 (1):25–55.
Tan, Ming T., Guo-Liang Tian, and Man-Lai Tang. 2009. “Sample Surveys with Sensitive
Questions: A Nonrandomized Response Approach.” The American Statistician 63 (1):9–
16.
Torgler, Benno, and Christoph A. Schaltegger. 2005. “Tax Amnesties and Political
Participation.” Public Finance Review 33 (3):403–31.
Torgler, Benno, and Friedrich Schneider. 2007. “What Shapes Attitudes toward Paying Taxes?
Evidence from Multicultural European Countries.” Social Science Quarterly 88 (2):443–
70.
Torgler, Benno, and Neven T. Valev. 2010. “Gender and Public Attitudes toward Corruption and
Tax Evasion.” Contemporary Economic Policy 28 (4):554–68.
Treisman, Daniel. 2007. “What Have We Learned about the Causes of Corruption from Ten
Years of Cross-National Empirical Research?” Annu. Rev. Polit. Sci. 10:211–44.
Weitz-Shapiro, Rebecca. 2014. Curbing Clientelism in Argentina: Politics, Poverty, and Social
Policy. New York: Cambridge University Press.
-
30
Winters, Matthew S., and Rebecca Weitz-Shapiro. 2013. “Lacking Information or Condoning
Corruption: When Do Voters Support Corrupt Politicians?” Comparative Politics 45
(4):418–36.