1 “Can Citizens Discern? Information Credibility, Political Sophistication, and the
Punishment of Corruption in Brazil” Online Appendix
Table of Contents
Vignettes ... 2
Main Follow Up Questions ... 3
Sampling Procedure ... 4
Randomization Procedure and Balance Checks ... 5
Correlations among Measures of Political Sophistication ... 8
Replication of Results using Regression Analysis ... 11
Replication of Main Results using Only Vignettes that Specifically Accuse the Mayor ... 13
Party Identification and Responses to Credibility ... 15
Full Measures of Sophistication and Index ... 23
Sophistication Results Including Unsourced Accusations ... 26
Sophistication Results Including Control Conditions ... 27
2
Vignettes
Pure Control
Imagine que você vive num bairro como o seu, mas numa cidade diferente do Brasil. Vamos chamar o Prefeito dessa cidade em que você mora de Carlos. Agora imagine que o Prefeito Carlos está concorrendo à reeleição. Durante os quatro anos em que foi Prefeito a cidade teve várias melhorias, com crescimento econômico e melhores serviços públicos de saúde e transporte.
No Corruption
[Pure control plus] Também nessa cidade, todo mundo diz que o Prefeito Carlos não aceitou suborno para fechar contratos com fornecedores da Prefeitura.
Corruption without Source
[Pure control plus] Também nessa cidade, todo mundo diz que o Prefeito Carlos aceitou suborno para fechar contratos com fornecedores da Prefeitura.
Credible Source / Specific Accusations
[Pure control plus] Também nessa cidade, uma auditoria do governo federal diz que o Prefeito Carlos aceitou suborno para fechar contratos com fornecedores da Prefeitura.
Less Credible Source / Specific Accusations
[Pure control plus] Também nessa cidade, o partido de oposição diz que o Prefeito Carlos aceitou suborno para fechar contratos com fornecedores da Prefeitura.
Credible Source / Less Specific Accusations
[Pure control plus] Também nessa cidade, uma auditoria do governo federal diz que ocupantes de cargos na Prefeitura aceitaram suborno para fechar contratos com fornecedores da Prefeitura.
Less Credible Source / Less Specific Accusations
3
Main Follow Up Questions
Na sua opinião, qual a chance de você votar para o Prefeito Carlos? Grande chance; alguma chance; pouca chance; ou nenhuma chance?
4
Sampling Procedure
Our survey was included as the second module of the May 2013 IBOPEBus survey.
The IBOPEBus is a monthly omnibus survey that uses a probabilistic sample of geographic areas to obtain a representative sample of the over-16-years-old Brazilian population. The sampling frame is based on the 2010 census, the 2011 Pesquisa Nacional por Amostra de Domicilios
(National Household Survey), and 2012 data from the Tribunal Superior Eleitoral (National Electoral Tribunal).
140 cities were sampled using a probability-proportional-to-size (PPS) method within 25 strata that are defined by 25 of Brazil‘s 27 states. (The survey rotates on a monthly basis among three small states in the northern region of the country.) Census tracts were selected using PPS with stratification across zones of major metropolitan areas. Enumerators recruited individual respondents in public or semi-public places according to a quota scheme designed to produce a representative sample of the national population in terms of age, gender, and employment characteristics (sector of the economy and employment status).
5
Randomization Procedure and Balance Checks
The seven vignettes were to be randomly assigned to survey respondents within each sampling strata. Since seven respondents were sampled from each census tract, each vignette was to be assigned once per census tract.
Unfortunately, rather than assigning the vignettes in random order, they were assigned in the same order – from the first through the seventh – within each of the sampled census tracts. If different types of respondents were recruited earlier in the day as compared to later in the day, this failed randomization could imply a correlation between observable or unobservable characteristics of respondents and their treatment status.
While we cannot comment on correlations between treatment status and unobservable characteristics, we examine here whether or not any observable pre-treatment characteristics predict selection into the treatment categories. To do this, we use two methods. Both indicate a degree of variation in observed characteristics across treatment groups that is consistent with what could be generated by chance. We nonetheless replicate results reported in Tables 3 and 4 in the paper using regression analyses controlling for possible confounding covariates and report those below. In both cases, the results are substantively unchanged from the simpler difference-in-means tests reported in the text.
Here, we explain our two methods for checking balance. First, we run two multinomial logit models where the seven categories of treatment assignment defined the outcome variable.1 We compare a null model with no predictors to a model with predictors for gender, age, education, social class, income, an indicator for whether or not the respondent is catholic, a variable measuring how often the respondent talks about politics, a variable measuring how often the respondent reads the news, a variable representing the respondent’s score on a two-question measure of political knowledge, and a set of indicators for whether or not the respondent
identifies with one of the three major political parties in Brazil, the PT, the PSDB, or the PMDB.
The table below presents the results of the multinomial logit model. The chi-squared statistic indicates that the model with the set of predictor variables is not statistically distinguishable from a null model without any predictor variables at all (p < 0.85).
1
6 Outcome Vignette 1 Vignette 3 Vignette 4 Vignette 5 Vignette 6 Vignette 7 Male (0/1) 0.07
(0.18) -0.11 (0.18) 0.13 (0.18) 0.11 (0.18) 0.15 (0.18) 0.15 (0.18)
Age Category 0.01
(0.06) -0.06 (0.06) 0.00 (0.06) -0.09 (0.06) -0.04 (0.06) -0.04 (0.06)
Education -0.05 (0.10) -0.05 (0.10) 0.01 (0.10) -0.08 (0.10) 0.09 (0.10) 0.00 (0.10)
Social Class 0.44**
(0.17) 0.09 (0.17) 0.05 (0.17) 0.19 (0.17) 0.10 (0.17) 0.10 (0.17)
Income -0.13 (0.12) 0.03 (0.13) -0.03 (0.12) -0.16 (0.13) -0.16 (0.13) -0.13 (0.13)
Catholic (0/1) -0.29 (0.19) -0.27 (0.19) -0.33* (0.19) -0.18 (0.19) -0.14 (0.19) -0.12 (0.19)
Talk about Politics 0.18 (0.13) 0.19 (0.13) -0.04 (0.13) 0.12 (0.13) -0.12 (0.13) -0.03 (0.13)
Read the News -0.00 (0.11) -0.02 (0.11) -0.01 (0.11) -0.05 (0.11) -0.01 (0.11) 0.02 (0.11) Political Knowledge Index -0.04 (0.14) 0.09 (0.14) 0.02 (0.14) 0.21 (0.14) 0.03 (0.14) 0.14 (0.14)
PT Identifier (0/1) 0.10 (0.21) 0.02 (0.21) 0.31 (0.20) -0.05 (0.21) -0.16 (0.21) -0.02 (0.21)
PSDB Identifier (0/1) -0.35 (0.42) -0.27 (0.41) -0.02 (0.40) -0.66 (0.46) -0.28 (0.42) -0.26 (0.42) PMDB Identifier (0/1) 0.46 (0.34) 0.16 (0.37) -0.15 (0.40) 0.17 (0.36) 0.37 (0.35) 0.47 (0.35)
Constant -0.77 (0.48) -0.14 (0.48) 0.06 (0.48) 0.29 (0.48) 0.37 (0.48) 0.13 (0.48)
N 1,795
Pseudo R2 0.01
Chi-Squared 59.65
p-value for H0: No
difference with null model
0.85
8
Correlations among Measures of Political Sophistication
Education: Exact educational attainment was recorded and then collapsed into five categories: Illiterate / less than primary (0); Complete primary or incomplete middle (1); Complete middle or incomplete secondary (2); Complete secondary (3); At least some tertiary (4). The table below describes the distribution.
Category N Percent
Illiterate / Less Than Primary 254 12.7 Complete Primary –
Incomplete Middle
481 24.0
Complete Middle – Incomplete Secondary
468 23.4
Complete Secondary 499 24.9 At Least Some Tertiary 300 15.0
Political Knowledge: Knowledge was measured with two factual, open-ended questions that asked respondents to supply the number of states in Brazil and the name of Argentina’s president. We accepted either 26 or 27 as the correct answer for the number of states (accounting for the federal district) and any variant of then-president Cristina Fernández de Kirchner’s name was counted as correct. We classify respondents according to whether they answered 0, 1, or 2 questions correctly.
Category N Percent
No Questions Correct 1,249 62.4 One Question Correct 412 20.6 Both Questions Correct 341 17.0
Political Discussion: Political discussion is a measure of how frequently a respondent reports discussing politics with her family and friends. Respondents had the option of answering “very frequently,” “frequently,” “rarely,” or “never.” Only 5% of respondents answered “very
frequently,” so for the purpose of analysis, we group together “very frequently” and “frequently” responses into a single high interest category.
Category N Percent
Never 692 35.0
Rarely 840 42.5
9 Cross-tabulations between measures of sophistication
In all cases, a chi-squared test allows us to reject the null hypothesis of no correlation between the variables at better than the p < 0.01 level.
Education and Political Knowledge Education
Political Knowledge: Questions correct
Illiterate / less than
primary
Complete primary; incomplete
middle
Complete middle; incomplete
secondary
Complete secondary
At least some tertiary
0 238
(94%)
377 (78%)
293 (63%)
259 (52%)
82 (27%)
1 14
(5%)
76 (16%)
109 (23%)
137 (27%)
76 (25%)
2 2
(1%)
28 (6%)
66 (14%)
103 (21%)
142 (47%)
Total N 254 481 468 499 300
Numbers of respondents in each category; column percentages are presented in parentheses.
Education and Political Discussion Education
Discuss Politics
Illiterate / less than
primary
Complete primary; incomplete
middle
Complete middle; incomplete
secondary
Complete secondary
At least some tertiary
Never 118
(48%)
208 (44%)
176 (38%)
137 (28%)
53 (18%)
Rarely 82
(34%)
190 (40%)
205 (44%)
231 (47%)
132 (44%)
Fairly/Very 44
(18%)
78 (16%)
84 (18%)
127 (26%)
113 (38%)
Total N 244 476 465 495 298
10
Political Knowledge and Political Discussion Political Knowledge
(Questions correct)
Discuss Politics
0 1 2
Never 537
(44%)
111 (27%)
44 (13%)
Rarely 501
(41%)
191 (46%)
148 (44%)
Fairly/Very 191
(16%)
109 (27%)
146 (43%)
Total N 1229 411 338
11
Replication of Results using Regression Analysis
Our examination of variation on observable covariates detailed above suggests that, in spite of the failure of our randomization procedure, differences on observable characteristics across treatment groups are no greater than we would expect by random chance. Nonetheless, we replicate the analysis in Tables 2 and 3 from the paper using regression analyses that control for covariates. As can be seen below, the results are extremely similar and substantively equivalent in all three cases.
Comparison Credible
Accusations versus Pure Control Credible Accusations versus Control with Clean Mayor Credible Accusations versus Unsourced Accusations Credible Accusations versus Less Credible Accusations Difference-in-Means Estimate
from Table 2
-1.30 (p < 0.01) [p < 0.01]
-1.31 (p < 0.01) [p < 0.01]
-0.10 (p < 0.21) [p < 0.22]
-0.29 (p < 0.01) [p < 0.01] Regression-Based Estimate
with Controls
-1.27 (0.08) [p < 0.01]
-1.30 (0.08) [p < 0.01]
-0.05 (0.09) [p < 0.57]
-0.24 (0.07) [p < 0.01] Comparison Less Credible versus Pure Control Less Credible Accusations versus Control with Clean Mayor Less Credible Accusations versus Unsourced Accusations Difference-in-Means Estimate
from Table 2
-1.01 (p < 0.01) [p < 0.01]
-1.03 (p < 0.01) [p < 0.01]
0.18 (p < 0.03) [p < 0.03] Regression-Based Estimate
with Controls
-1.00 (0.08) [p < 0.01]
-1.05 (0.08) [p < 0.01]
0.21 (0.09) [p < 0.02] Comparison Unsourced Accusations versus Pure Control Unsourced Accusations versus Control with Clean Mayor Control with Clean Mayor versus Pure Control Difference-in-Means Estimate from Table 2
-1.19 (p < 0.01) [p < 0.01]
-1.21 (p < 0.01) [p < 0.01]
-0.02 (p < 0.83) [p < 0.74] Regression-Based Estimate
with Controls
-1.19 (0.09) [p < 0.01]
-1.24 (0.09) [p < 0.01]
0.03 (0.08) [p < 0.73]
12
How likely are you to vote for the mayor?
Completed High School or
Less (1)
Some Tertiary Education
or More (2)
Less Politically Knowledg
e-able (3)
Most Politically Knowledg
e-able (4)
Less Political Discussion
(5)
Most Political Discussion
(6)
Difference-in-Means Estimate from Table 3
0.25 (0.07) (p < 0.01) [p < 0.01]
0.47 (0.17) (p < 0.01) [p < 0.01]
0.23 (0.07) (p < 0.01) [p < 0.01]
0.51 (0.16) (p < 0.01) [p < 0.01]
0.24 (0.07) (p < 0.01) [p < 0.01]
0.43 (0.15) (p < 0.01) [p < 0.01] Coefficient on
Treatment Indicator for Credible versus Less Credible Accusations
0.22 (0.08) [p < 0.01]
0.51 (0.17) [p < 0.01]
0.22 (0.08) [p < 0.01]
0.49 (0.17) [p < 0.01]
0.19 (0.08) [p < 0.01]
0.46 (0.15) [p < 0.01]
13
Replication of Main Results using Only Vignettes that Specifically Accuse the Mayor
As reported in the text, for the analysis presented there, the credible and less credible conditions include vignettes that explicitly mention the mayor (specific vignettes) as well as those that mention municipal officials (less specific vignettes). Here, we replicate the results from Tables 2, 3, and 4 using only those vignettes that mention the mayor.
The results from the replicated Table 2 are almost identical for many comparisons. For the credible accusations treatment, the effects are all somewhat stronger than those reported in the text.
In the replicated Table 3, treatment effects are as strong or stronger for both non-sophisticates and sophisticates. For sophisticates, in particular, they are often substantially larger than in the text. As a result of this, we can reject the null hypothesis of no difference in treatment effects between the more and less sophisticated groups at the 90 percent confidence level or greater for all three operationalizations of sophistication. For political discussion, the difference in CATEs is, in fact, highly statistically significant.
Treatment Condition Credible Accusations
Less Credible Accusations
Unsourced Accusations
Control with Clean
Mayor
Pure Control
N 279 278 278 279 280
Average Response to “How likely are you to vote for the mayor?” (Standard Error)
1.97 (0.06)
2.36 (0.07)
2.18 (0.07)
3.39 (0.06)
3.38 (0.06)
Estimated Difference from Pure Control
-1.40 (p < 0.01) [p < 0.01]
-1.02 (p < 0.01) [p < 0.01]
-1.19 (p < 0.01) [p < 0.01]
0.02 (p < 0.83) [p < 0.73]
--
Estimated Difference from Control with Clean Mayor
-1.42 (p < 0.01) [p < 0.01]
-1.03 (p < 0.01) [p < 0.01]
-1.21 (p < 0.01) [p < 0.01]
-- --
Estimated Difference from Unsourced Accusations
-0.21 (p < 0.03) [p < 0.04]
0.18 (p < 0.07) [p < 0.07]
-- -- --
Estimated Difference from Less Credible Accusations
-0.38 (p < 0.01) [p < 0.01]
-- -- -- --
14
How likely are you to vote for the mayor?
Completed High School or
Less (1)
Some Tertiary Education
or More (2)
Less Politically Knowledge
-able (3)
Most Politically Knowledge
-able (4)
Less Political Discussion
(5)
Most Political Discussion
(6)
Less Credible Accusations
2.35 (0.07) N=235
2.40 (0.18) N=43
2.31 (0.08) N=224
2.56 (0.14)
N=54
2.27 (0.08) N=210
2.68 (0.14) N=66 More Credible
Accusations
2.03 (0.07) N=246
1.58 (0.15) N=33
2.00 (0.07) N=230
1.86 (0.14)
N=49
2.03 (0.07) N=224
1.78 (0.13) N=50 Difference
between Credible and Less Credible
0.32 (0.10)
0.82 (0.24)
0.31 (0.10)
0.79 (0.20)
0.24 (0.10)
0.90 (0.20)
p-value on H0: No Difference
0.01 0.01 0.01 0.01 0.03 0.01
p-value on H0: No Difference between CATEs
0.08 0.10 0.01
15
Party Identification and Responses to Credibility
In this section of the appendix, we address two alternative explanations for the patterns we observe in the data that stem from the possibility that partisanship, although not explicitly a component of our experimental vignettes, may nonetheless affect responses.
First, the more credible information source included in the vignettes is “federal government audits.” Since the federal government at the time of the survey was controlled by the Partido dos Trabalhadores (PT), it might be the case that PT sympathizers are especially likely to believe the results of these audits. If the reaction of PT sympathizers is sufficiently strong, it is possible that our main results showing discernment (H1) are driven by PT sympathizers only. Separately, if sophisticated respondents are particularly likely to be PT supporters, our results showing greater discernment among sophisticates (H2) might also be confounded by PT identity, although this seems unlikely since our results in support of H2 are driven by sophisticates’ skepticism of the opposition party accusations, rather than by giving greater credence to accusations made by the federal audit.
Second, we do not provide information about the hypothetical mayor’s partisan identity in the experimental vignette. In the absence of such information, respondents might impute the partisan identity of the mayor from their own city (since we have asked them to think about a neighborhood that is like their own, although we explicitly say they should imagine a different city in Brazil), and this may influence their reactions to the information in the vignettes. Given the expectation that partisans of any given party should be evenly distributed across the treatment groups, we do not expect this to have any systematic effect on our estimates of treatment effects (except to bias against finding any result by introducing more noise into the data). Nonetheless, it is possible that examining our hypotheses disaggregated by partisan groups will reveal
interesting patterns, and we do so below. Furthermore, if political sophisticates are more likely to project their own mayor’s partisanship, that might drive some of our findings with regard to sophistication (e.g., if the reason that sophisticates discount opposition party accusations more is because they do so when thinking about particular political parties making those accusations).
We investigate each of these possibilities in great detail below. In all cases, we find some evidence that partisanship matters for responses to corruption information. At the same time, we find that our main results continue to hold. Even accounting for partisanship in different ways, it is quite clear that Brazilian respondents discern between sources of variable credibility and that discernment is greater among more politically sophisticated voters.
16 consistent with other recent work on partisanship in Brazil (Samuels and Zucco 2014; Winters and Weitz-Shapiro 2014). This points to the general weakness of partisanship in Brazil.
PT Sympathizers and Responses to Credible Information Sources
Our credible accusations come from a federal audit. Given that the federal government of Brazil was under the control of the Partido dos Trabalhadores (PT) at the time that we conducted the survey, it may be the case that the federal audits treatment particularly resonated with PT supporters, who would then be especially likely to report that they would vote against a mayor accused of corruption by a federal audit.
The table below compares PT supporters (by far the largest group of partisans in our whole sample, yet still only about 27 percent of the total sample) with respondents who sympathize with another political party and non-partisans (the final group is the largest in the survey and among the Brazilian population). It shows that PT supporters do, in fact, respond more strongly to the accusations of corruption coming from the federal audit. In contrast, PT and non-PT identifiers respond very similarly to opposition party accusations. As a result of the greater punishment of corruption allegations made by a federal audit among PT sympathizers, our estimate of discernment for PT identifiers between the more- and less-credible conditions is nearly twice the size of that for non-PT identifiers. Importantly, however, our estimate of
17
How likely are you to vote for the mayor?
Other Party Sympathizers
and Non-Partisans
PT Sympathizers
Difference p-value on H0: No
Difference
Less Credible Accusations
2.36 (0.06) N=404
2.38 (0.09) N=143
0.03 (0.11)
0.82 [0.81]
Credible Accusations
2.12 (0.04) N=405
1.97 (0.08) N=148
0.15 (0.10)
0.14 [0.19]
Difference 0.24 0.42
p-value on H0:
No Difference
0.01 [0.01]
0.01 [0.01] p-value for H0:
No difference in CATE between PT and non-PT respondents
0.22
Differences in Conditional Average Treatment Effects (CATEs) by PT Partisans and Others. Note: initial p-values are from a t-test of the null hypothesis of no difference in means between the two groups; p-p-values in brackets are from a Wilcoxon rank sum test of the null hypothesis of no difference in the distribution of the outcome variable between the two groups; p-value in last row is a p-value for differences across the CATEs from
randomization inference tests described in Gerber and Green (2012).
Next, we consider the possibility that, if PT supporters are more likely to score highly on our measures of sophistication, then this fact may account for the greater discernment we detect among sophisticates.2 There is no statistically significant relationship between PT identification and sophistication, however; if anything, there is a slight (but not significant) decrease in rates of PT identification as education increases. When looking at the two major opposition parties, sophisticates are somewhat more likely to identify with the PSDB and somewhat less likely to identify with the PMDB. Given the PSDB’s historical status as the more relevant opposition party in national politics, this would suggest that, if respondents have particular political parties in mind when listening to the vignette, then more sophisticated respondents would give more credence to corruption information provided by the opposition. But this is precisely the opposite of what we find in the data.
2
A related hypothesis would be that Brazilian sophisticates have a greater affinity for the federal
18 In spite of the lack of any relationship between sophistication and PT sympathy, as an additional check, we rerun our analysis of the sophistication hypothesis excluding PT supporters. The results are essentially indistinguishable from the main results in the paper. By any of our three definitions of sophistication, political sophisticates engage in more discernment than our sophisticated respondents. If anything, the differences become somewhat starker, since less-sophisticated PT sympathizers react somewhat more strongly to the credible treatment than do their non-PT counterparts. As with the main results in the text, the greater discernment among non-PT political sophisticates is driven almost entirely by their greater skepticism of the opposition party accusations.
How likely are you to vote for the mayor? Completed High School or Less (1) Some Tertiary Education or More (2) Less Politically Knowledge -able (3) Most Politically Knowledge -able (4) Less Political Discussion (5) Most Political Discussion (6) Less Credible Accusations 2.30 (0.06) N=338 2.64 (0.14) N=66 2.30 (0.06) N=329 2.63 (0.12) N=75 2.30 (0.06) N=316 2.61 (0.12) N=84 Credible Accusations 2.13 (0.06) N=344 2.07 (0.13) N=61 2.12 (0.06) N=332 2.14 (0.13) N=73 2.12 (0.06) N=322 2.12 (0.13) N=76 Difference 0.17
(0.09) 0.57 (0.19) 0.18 (0.09) 0.49 (0.18) 0.18 (0.09) 0.49 (0.18)
p-value on H0: No Difference 0.05 [0.04] 0.01 [0.01] 0.04 [0.04] 0.01 [0.01] 0.05 [0.04] 0.01 [0.01]
Differences between More and Less Sophisticated Respondents (Non-PT Supporters Only). Note: p-values in parentheses are from a t-test of the null hypothesis of no difference in means between the two groups; p-values in brackets are from a Wilcoxon rank sum test of the null hypothesis of no difference in the distribution of the outcome variable between the two groups.
Projection of Local Mayoral Partisan Identity onto Hypothetical Mayor
Next, we turn to the possibility that some respondents may project the partisan identification of their local mayor onto the hypothetical “Mayor Carlos” mentioned in the vignette. If this is the case, respondents may furthermore react differently to corruption information when they share a partisan identity with their local mayor.
19 This is, in fact, what we find. Pooling across all treatment conditions, we see in the first table below that respondents who share the partisan identification of their real-world mayor express a higher vote intention for the hypothetical mayor than respondents who express a partisan
identification that is different from their real-world mayor or who do not express any partisan identification. The difference between these respondents with a shared partisanship and
respondents who do not express any partisanship is statistically significant (δ = 0.24, p < 0.01), while the difference between these respondents with a shared partisanship and respondents who have a different partisan identification from their local mayor is marginally statistically
significant (δ = 0.15, p < 0.10).
Respondents who Share Party ID of
Their Mayor
Respondents who Have a Party ID that is Not Shared with Their Mayor
Respondents who Do Not Express a
Party ID
Average Vote Intention for Mayor
2.73 (0.08) N=227
2.58 (0.05) N=713
2.49 (0.04) N=931 Difference from respondents
who do not express a party id
0.24 (p < 0.01) [p < 0.01]
0.09 (p < 0.13) [p < 0.12]
-
Difference from respondents who express a party id different from their real-world mayor
0.15 (p < 0.10) [p < 0.11]
- -
Average Vote Intention for Hypothetical Mayor by Relationship with Real-World Mayor. Note: initial p-values are from a t-test of the null hypothesis of no difference in means between the two groups; p-p-values in brackets are from a Wilcoxon rank sum test of the null hypothesis of no difference in the distribution of the outcome variable between the two groups.
20 the respondents who share a partisan identity with their real-world mayor are somewhat less enthusiastic about the hypothetical mayor as compared to those who have a partisan identity different from their real world mayor—this result runs counter to expectations (although it is not statistically significant). It is important to note that across all categories of respondents,
punishment of corruption (as judged through a comparison to the control conditions) is large and significant.
How likely are you to vote for the mayor?
Partisans who Share Party ID of
Mayor
Partisans Who Don’t Share Party ID of Mayor
Estimated Difference from Partisan Match Group Non-Partisans Estimated Difference from Partisan Match Group Credible Accusations 2.22 (0.14) N=60 2.03 (0.08) N=200 -0.19 (p < 0.24) [p < 0.21]
2.08 (0.07) N=274
-0.13 (p < 0.40) [p < 0.37] Less Credible Accusations 2.41 (0.16) N=54 2.42 (0.08) N=202 0.01 (p < 0.97) [p < 0.95]
2.30 (0.07) N=273
-0.11 (p < 0.51) [p < 0.52] No Source of
Corruption Accusations 2.64 (0.19) N=36 2.22 (0.11) N=105 -0.42 (p < 0.07) [p < 0.06]
2.06 (0.09) N=126
-0.58 (p < 0.01) [p < 0.01] Control with Clean Mayor 3.50 (0.13) N=42 3.51 (0.10) N=82 0.01 (p < 0.95) [p < 0.61]
3.30 (0.08) N=146
-0.20 (p < 0.25) [p < 0.34] Pure Control 3.29
(0.18) N=35 3.43 (0.08) N=124 0.14 (p < 0.45) [p < 0.59]
3.37 (0.09) N=112
0.08 (p < 0.68) [p < 0.99]
Average Vote Intention for Mayor within Treatment Conditions by Partisan Relationship with Real-World Mayor. Note: p-values in parentheses are from a t-test of the null hypothesis of no difference in means between the two groups; p-values in brackets are from a Wilcoxon rank sum test of the null hypothesis of no difference in the distribution of the outcome variable between the two groups.
21 addition, there is no case where the difference in CATEs is statistically significant, and the CATE for respondents who share a partisan identity with their real-world mayor remains negative (although because of the small number of respondents with such a shared identity, it is not statistically significantly different from zero). In particular, the CATE for respondents who share a partisan identity with their mayor is essentially indistinguishable from the CATE for non-partisan respondents. We therefore conclude that, to the extent that respondents project mayoral partisanship onto the mayor named in the prompt, this does not alter our conclusions with regard to H1 in the text. All groups of respondents in the survey show evidence of discerning between more and less credible accusations of corruption.
Respondents who Share Party ID of
Their Mayor
Respondents who Have a Party ID that is Not Shared with Their Mayor
Respondents who Do Not Express a
Party ID
Conditional Average Treatment Effect for Credible versus Less Credible Corruption
Accusations
-0.19 (p < 0.37) [p < 0.38] N=114
-0.39 (p < 0.01) [p < 0.01] N=402
-0.21 (p < 0.03) [p < 0.03] N=547 p-value for H0: No difference in
CATE from respondents who do not express a party id
0.93 0.22
-
p-value for H0: No difference in
CATE from respondents who express a party id different from their real-world mayor
0.40
- -
Differences in Conditional Average Treatment Effects (CATEs) by Partisan Relationship with Real-World Mayor. Note: p-values in parentheses are from a t-test of the null hypothesis of no difference in means between the two groups; p-values in brackets are from a Wilcoxon rank sum test of the null hypothesis of no difference in the distribution of the outcome variable between the two groups; p-values in last two rows are p-values for differences across the CATEs are from randomization inference tests described in Gerber and Green (2012).
Even if our overall results for H1 are not threatened by any projection of respondents’ real-world mayor’s partisan identity onto the hypothetical mayor in the vignette, if this behavior is
particularly pronounced among political sophisticates, it is possible that our inferences about H2 are driven by this propensity for partisan projection rather than by political sophistication.
22 sophisticates, consistent with H2. These results parallel the findings presented in Table 3 of the main text, which includes both partisan and nonpartisan respondents.
How likely are you to vote for the mayor?
Completed High School or
Less (1)
Some Tertiary Education
or More (2)
Less Politically Knowledge
-able (3)
Most Politically Knowledge
-able (4)
Less Political Discussion
(5)
Most Political Discussion
(6)
Less Credible Accusations
2.22 (0.07) N=231
2.71 (0.17) N=42
2.20 (0.07) N=219
2.69 (0.15)
N=54
2.25 (0.07) N=229
2.56 (0.16) N=41 Credible
Accusations
2.06 (0.07) N=236
2.24 (0.17) N=38
2.02 (0.07) N=228
2.41 (0.17)
N=46
2.08 (0.07) N=238
2.16 (0.21) N=31 Difference 0.16
(0.10)
0.48 (0.24)
0.18 (0.10)
0.27 (0.22)
0.18 (0.10)
0.40 (0.26)
p-value on H0: No Difference
0.12 [0.11]
0.06 [0.06]
0.08 [0.08]
0.23 [0.23]
0.09 [0.09]
0.14 [0.13]
23
Full Measures of Sophistication and Index
In Table 3 in the text, we use dichotomized measures of our three indicators of political sophistication (education, political knowledge, and political discussion). In this section, we show the results using the full range of the variables. We also show the results using an additive index of the three variables.
How likely are you to vote for the mayor?
Illiterate / less than
primary
Complete primary; incomplete
middle
Complete middle; incomplete
secondary
Complete secondary
At least some tertiary
Less Credible Accusations
2.13 (0.15) N=58
2.26 (0.09) N=137
2.46 (0.10) N=126
2.41 (0.09) N=136
2.44 (0.12) N=90
Credible Accusations
2.18 (0.15) N=59
2.01 (0.10) N=129
2.16 (0.10) N=143
2.07 (0.09) N=146
1.97 (0.12) N=76
Difference -0.05 0.25 0.31 0.34 0.47
p-value on H0: No Difference
0.82 0.06 0.02 0.01 0.01
p-value on H0: No Difference between CATE and CATE for Lowest Education Group
-- 0.23 0.15 0.13 0.07
24
Political Knowledge
No Questions
Right
One Question
Right
Both Questions
Right
Less Credible Accusations
2.29 (0.06) N=321
2.34 (0.10) N=122
2.60 (0.11) N=104 Credible
Accusations
2.04 (0.06) N=352
2.21 (0.11) N=107
2.09 (0.11)
N=95
Difference 0.26 0.14 0.51
p-value on H0: No Difference
0.01 [0.01]
0.36 [0.37]
0.01 [0.01]
p-value on H0: No Difference between CATE and CATE for no questions right group
-- 0.49 0.16
Discernment by Levels of Political Knowledge. Note: p-values for the null hypothesis on the conditional average treatment effect (CATE) for each group are based on difference-in-means t-tests and (in brackets) Wilcoxon rank sum tests. p-values for differences across the CATEs are based on the randomization inference tests described in Gerber and Green (2012).
How likely are you to vote for the mayor?
Never Discuss Politics
Rarely Discuss Politics
Frequently Discuss Politics Less Credible Accusations 2.18
(0.09) N=187
2.40 (0.07) N=232
2.60 (0.10) N=124 Credible Accusations 1.98
(0.07) N=205
2.13 (0.07) N=235
2.16 (0.11) N=104
Difference 0.20 0.27 0.43
p-value on H0: No Difference 0.07 0.00 0.00
p-value on H0: No Difference
between CATE and CATE for Lowest Discussion Group
-- 0.62 0.22
25 Additive Index
We create an index that is the simple sum of education (running from 0 to 4), political
knowledge (running from 0 to 2) and frequency of discussion of politics (running from 1 to 2). The index runs from 1 to 9, with a median value of 4. We divide respondents into 3 categories: least sophisticates (1-3 on the index); middle sophisticates (4-6 on the index), and high
sophisticates (7-9 on the index). The table below then replicates results from Table 3 in the main text and the tables above to show, once again, that discernment between more and less credible accusations is increasing in sophistication.
Political Sophistication (Index)
1-3 4-6 7-9
Less Credible Accusations
2.24 (0.08) N=189
2.40 (0.07) N=245
2.54 (0.11) N=109 Credible
Accusations
2.04 (0.08) N=193
2.12 (0.07) N=263
2.07 (0.11)
N=88
Difference 0.20 0.27 0.47
p-value on H0: No Difference
26
Sophistication Results Including Unsourced Accusations
In Table 3 in the text, we omit the treatments involving unsourced accusations for the sake of clarity and since we do not have any hypotheses about how individuals will interpret the
unsourced accusations. In the table below, we present the point estimates for voting intention in the unsourced accusations category and calculate the differences between voting intention in that category and the two main categories of interest across levels of sophistication.
How likely are you to vote for the mayor? Completed High School or Less (1) Some Tertiary Education or More (2) Less Politically Knowledge -able (3) Most Politically Knowledge -able (4) Less Political Discussion (5) Most Political Discussion (6) Less Credible Accusations 2.35 (0.05) N=457 2.44 (0.12) N=90 2.31 (0.05) N=443 2.60 (0.11) N=104 2.30 (0.05) N=419 2.60 (0.10) N=124 Credible Accusations 2.10 (0.05) N=477 1.97 (0.12) N=76 2.08 (0.05) N=459 2.09 (0.11) N=94 2.06 (0.05) N=440 2.16 (0.11) N=104 Unsourced Accusations 2.17 (0.07) N=237 2.24 (0.17) N=41 2.21 (0.07) N=231 2.06 (0.17) N=47 2.08 (0.08) N=196 2.48 (0.14) N=78 Difference between Less Credible and Unsourced 0.18 (0.09) 0.20 (0.21) 0.10 (0.09) 0.53 (0.20) 0.23 (0.09) 0.11 (0.17)
p-value on H0: No Difference 0.05 [0.05] 0.35 [0.35] 0.26 [0.25] 0.01 [0.01] 0.02 [0.02] 0.51 [0.56] Difference between Credible and Unsourced -0.08 (0.09) -0.27 (0.20) -0.13 (0.09) 0.02 (0.20) -0.01 (0.09) -0.32 (0.18)
p-value on H0: No Difference 0.39 [0.42] 0.19 [0.20] 0.15 [0.14] 0.92 [0.79] 0.89 [0.88] 0.07 [0.07]
Respondent Sophistication and Responsiveness to Source Credibility (Including Unsourced Accusations).
27
Sophistication Results Including Control Conditions
In Table 3 in the text, we omit the control conditions for the sake of clarity. The table provides a test of the claim that sophisticated citizens discern more based on source credibility than less sophisticated citizens. In the text, we also say that sophisticates punish politicians more in the case of credible corruption information than non-sophisticates. This claim depends on how sophisticates react to the politicians about whom we provide no information about corruption status or who we specifically describe as not corrupt. If sophisticates, as compared to non-sophisticates, react differently to these politicians, then our claim of increased punishment may be incorrect.
In the table below, we provide the point estimates for the combined pure control and clean control conditions, and estimate the difference between that point estimate and each of the types of corruption accusations for both sophisticates and non-sophisticates. As with Table 2 in the text, we see that both more and less credible accusations of corruption result in expressed voting intents that are statistically significantly different from those obtained in the case of either explicitly clean mayors or else mayors for whom we do not describe corruption.
28
How likely are you to vote for the mayor? Completed High School or Less (1) Some Tertiary Education or More (2) Less Politically Knowledge -able (3) Most Politically Knowledge -able (4) Less Political Discussion (5) Most Political Discussion (6) Less Credible Accusations 2.35 (0.05) N=457 2.44 (0.12) N=90 2.31 (0.05) N=443 2.60 (0.11) N=104 2.30 (0.05) N=419 2.60 (0.10) N=124 Credible Accusations 2.10 (0.05) N=477 1.97 (0.12) N=76 2.08 (0.05) N=459 2.09 (0.11) N=94 2.06 (0.05) N=440 2.16 (0.11) N=104 Control with
Clean Mayor / Pure Control 3.37 (0.04) 3.46 (0.10) 3.37 (0.04) 3.48 (0.09) 3.33 (0.05) 3.54 (0.07) Difference between Less Credible and Control 1.02 (0.07) 1.02 (0.16) 1.05 (0.07) 0.89 (0.14) 1.03 (0.07) 0.94 (0.12)
p-value on H0: No Difference 0.01 [0.01] 0.01 [0.01] 0.01 [0.01] 0.01 [0.01] 0.01 [0.01] 0.01 [0.01] Difference between More Credible and Control 1.27 (0.07) 1.49 (0.15) 1.29 (0.07) 1.40 (0.15) 1.27 (0.07) 1.38 (0.13)
p-value on H0: No Difference 0.01 [0.01] 0.01 [0.01] 0.01 [0.01] 0.01 [0.01] 0.01 [0.01] 0.01 [0.01]
29
Replication of Results Using Feeling Thermometer Outcome
As an alternative outcome variable, we measured respondent’s attitudes toward the mayor using a feeling thermometer. In English, the question asked, “What grade, on a scale from 1 to 7, would you give to Mayor Carlos, where 1 means that you think he is a terrible mayor, and 7 means that you think he is an excellent mayor?” The original Portuguese question is reported in section 2 above.
The patterns that we observe are the same as with the vote intention variable. Across all of the treatment conditions, respondents express less enthusiasm for a corrupt mayor. All of these differences are statistically significant. The responses on the feeling thermometer between the more and less credible accusations and between the less credible accusations and the unsourced accusations are statistically distinguishable from each other, whereas the responses between the unsourced and the credible treatments are not statistically distinguishable.
For the replication of Table 3, we similarly find that the degree of discernment is greater for our more sophisticated respondents as compared to our less sophisticated respondents, although the difference between the two CATEs does not reach conventional levels of statistical significance for any of the three operationalizations.
Treatment Condition Credible Accusations
Less Credible Accusations
Unsourced Accusations
Control with Clean
Mayor
Pure Control
N 572 572 286 286 286
Average Response to “From 1 to 7, do you think that Mayor Carlos is (1) ‘a very bad mayor’ or (7) ‘an excellent mayor’?” (Standard Error)
3.56 (0.08)
4.01 (0.08)
3.59 (0.11)
5.53 (0.10)
5.45 (0.10)
Estimated Difference from Pure Control
-1.89 (p < 0.01)
-1.44 (p < 0.01)
-1.86 (p < 0.01)
0.09
(p < 0.53) -- Estimated Difference
from Control with Clean Mayor
-1.98 (p < 0.01)
-1.53 (p < 0.01)
-1.94
(p < 0.01) -- --
Estimated Difference from Unsourced Accusations
-0.03 (p < 0.82)
0.42
(p < 0.01) -- -- --
Estimated Difference from Less Credible Accusations
-0.45
(p < 0.01) -- -- -- --
30 Average
Response to “From 1 to 7, do you think that Mayor Carlos is (1) ‘a very bad mayor’ or (7) ‘an excellent mayor’?”
Completed High School or
Less (1)
Some Tertiary Education
or More (2)
Less Politically Knowledge
-able (3)
Most Politically Knowledge
-able (4)
Less Political Discussion
(5)
Most Political Discussion
(6)
Less Credible Accusations
4.01 (0.09) N=479
4.01 (0.18) N=93
3.96 (0.09) N=465
4.21 (0.17) N=107
3.92 (0.09) N=443
4.31 (0.17) N=125 More Credible
Accusations
3.58 (0.08) N=495
3.42 (0.19) N=77
3.59 (0.08) N=478
3.43 (0.19)
N=94
3.54 (0.09) N=454
3.69 (0.19) N=108 Difference
between Credible and Less Credible
0.42 (0.12)
0.60 (0.26)
0.38 (0.12)
0.78 (0.25)
0.38 (0.12)
0.63 (0.25)
p-value on H0: No Difference
0.01 0.03 0.01 0.01 0.01 0.02
p-value on H0: No Difference between CATEs
0.55 0.16 0.37
