Corruption and Political Support: The Case of Peru ’ s Vacuna-gate Scandal

Classic theories of public opinion suggest that negative shocks can undermine system support in weak democracies, but scant work has systematically assessed this thesis. We identify Peru ’ s explosive Vacuna-gate scandal as a most-likely case for finding a connection between corruption and political support. The scandal ’ s unexpected revelation in the middle of the 2021 AmericasBarometer Peru survey created conditions for a natural experiment. Applying an unexpected-event-during-survey design, we consider the consequences of the scandal for perceptions of corruption, system support, and support for democracy. We find robust evidence that the scandal increased even already high perceptions of corruption and lowered system support. Contrary to expectations derived from prior theories, we find no effect on explicit support for democracy. In the conclusion, we discuss the nuanced ways that scandal may shape democratic stability.

of corruptiona high-level scandal involving illicit early access to COVID-19 vaccineson public opinion.
Peru's weakly consolidated democracy represents what Gerring (2007) calls a "most-likely" case for finding a connection between corruption and support for democracy. Under an Eastonian framework (Easton 1975), allegiance to democracy is comparatively more malleable in systems that have lower "reservoirs of goodwill" toward the system. These days, only one-in-two Peruvians express support for democracy, among the very lowest rates in the region (Castorena and Rosario 2021). Simmering discontent coupled with crises that "jolt" public opinion are ripe conditions for eroding beliefs in the validity of democracy (Easton 1975).
We contrast explicit support for democracy with attitudes that are consequential for the indirect pathway through which corruption is theorized to diminish democracy: perceptions of corruption and/or political legitimacy. Extending Easton's logic to theorize that political support is more mutable in weakly institutionalized contexts, we expect scandal will boost perceptions of political corruption and diminish system support in Peru. And, in fact, extant evidence linking corruption to lower levels of political trust and system legitimacy is comparatively more robust (Ares and Hernández 2017;Seligson 2002). Yet, not all research supports this connection: some find that scandal fatigue or countervailing conditions, such as economic upturns, mitigate the opinion consequences of corruption and scandal (Carlin, Love, and Martínez-Gallardo 2015;Kumlin and Esaisson 2012;Zechmeister and Zizumbo-Colunga 2013). Others find little or mixed evidence that diminished trust in politicians and/or parties spills over into evaluations of the system more generally (Halmburger, Baumert, and Rothmund 2019;Maier 2011).
To assess the opinion consequences of exposure to a high-level political corruption scandal, we use data from the 2021 AmericasBarometer. The Vacuna-gate (Vaccine-gate) scandalin which the public learned that former president Martín Vizcarra, his wife, and other elites had clandestinely subverted the rules in order to be vaccinated against COVID-19 earlywas revealed in the middle of fieldwork for the Peru survey. Because the survey used random digit dialing, we apply an unexpected-event-during-survey design to assess differences in public opinion expressed by those who are treated with the scandal (those interviewed after the scandal broke) and by those who are not (those interviewed before the scandal broke) (Muñoz, Falcó-Gimeno, and Hernández 2020). 1 We find that the scandal substantially increased public views of corruption among politicians and reduced political legitimacythat is, support for the Peruvian political system. We find no significant or substantive effect on individuals' explicit support for democracy. Given the popularity of Easton's framework, it is noteworthy to find that even in most-likely cases, the public decouples political malfeasance from their direct demand for democracy. Yet, democracy may still be harmed when scandal decreases other dimensions of diffuse support. Our analyses reveal opinion dynamics consistent with a more indirect path through which 1 As we note later, in the last stage of the survey, the fieldwork team filtered out Lima area residents to balance the sample; in one of our models, we apply a matching approach to account for that potential confound. scandal can fuel democratic decay from the top down: when perceptions of a corrupt elite and illegitimate system are elevated, the public may become more willing to support illiberal and/or ineffective leaders who then subvert the system in their pursuit of power (Hawkins 2009(Hawkins , pp. 1044(Hawkins , 1064Levitsky 2018;Levitsky and Ziblatt 2018;Mudde and Kaltwasser 2018). In brief, we conclude that high-level scandals may not directly undermine explicit support for democracy but they do hold potential to more indirectly threaten democracy by sowing seeds of discontent. Political corruption, system support, and Vacuna-gate Can a major political corruption scandal like Peru's Vacuna-gate have broader implications for the political system at large, and to what extent? Scholarship on political evaluations distinguishes broadly between two types of attitudes: specific support for incumbent politicians and diffuse support for the system. Specific support encompasses a wide array of offices that can be easily delineated. We focus on diffuse support: belief in the authority of the system's core institutions and processes or an ideological belief in the validity of the regime type (Easton 1975). The former type of diffuse support is referred to as political legitimacy or system support and the latter as support for democracy (Booth and Seligson 2009).
A good deal of scholarship has considered the intuitive connection between corruption and evaluations of specific, implicated politicians, candidates, and political parties. Experimental work has shown that corruption scandals damage evaluations of candidates (e.g., Miller 2011, 2014;Funk 1996;McDermott, Schwartz, and Vallejo 2015) and erode vote intention for and attachments toward the implicated party (Klašnja, Lupu, and Tucker 2021;Winters and Weitz-Shapiro 2013;Wolsky 2022). Others have found that such scandals can damage the approval of the government or president (Solaz, de Vries, and de Geus 2019;but see Fischle 2000;Zaller 1998), though this may be conditioned by partisanship (Solaz, de Vries, and de Geus 2019), gender (Reyes-Housholder 2019), or the economy (Carlin, Love, and Martínez-Gallardo 2015). Effects can be quite modest (Bågenholm 2013), and many affected politicians go on to win elections (Bågenholm and Charron 2020), but it is still common to find that corruption scandals shape election outcomes (Basinger 2013;Costas-Pérez, Solé-Ollé, and Sorribas 2012;Peters and Welch 1980).
To what extent does a major corruption scandal affect diffuse political support? Easton (1975) provides a classic framework for considering this question. In general, in well-consolidated systems, diffuse support is likely to be inelastic in response to specific outcomes and events. As Easton (p. 444) puts it, "this type of evaluation tends to be more difficult : : : to weaken once it is strong." Yet, where diffuse support is weak, where "discontent with perceived performance continues over a long enough time," and/or at times marked by "the sudden frustration of expectations," diffuse support may decline (Easton 1975, p. 445). The case of Peru and the Vacunagate scandal shares all three of these characteristics.
First, diffuse support is comparatively low in the country. According to the 2021 AmericasBarometer, Peru ranks highest in the Latin America and Caribbean region on the extent to which the public views politicians as corrupt and in the bottom quarter with respect to expressing respect for the country's political institutions. Furthermore, only 50% of Peruvians express support for democracy, compared to the regional average of 62% (Castorena and Rosario 2021). While low, the attitude is far from immutable: in the 2012 AmericasBarometer, support for democracy in Peru was 10 percentage points higher.
Second, discontent with political performance was simmering for years before the Vacuna-gate scandal. Peru has experienced a constant state of political crisis since 2017, when former president Pedro Pablo Kuczynski was first implicated in a massive corruption scheme involving Brazilian construction conglomerate Odebrecht (see, e.g., Zarate and Casey 2019). These days, only two-fifths of the public expresses satisfaction with how democracy is working in the country; the only country with a lower value in the region is Haiti, at 11% (Castorena and Rosario 2021). Following a peak in satisfaction in 2012, at 52%, this figure decreased steadily over the years to 21% in 2021.
Third, when the Vacuna-gate scandal broke on February 11, 2021, it jolted public opinion. On that day, Vizcarra claimed in a press conference that he had received the Sinopharm vaccine as a volunteer in the clinical trial (De La Quintana 2021). 2 In fact, he had obtained it through irregular means while it was still going through phase III trials in October 2020. On February 13, several cabinet members responded to the news by tendering their resignations to then-president Francisco Sagasti; this included the health minister, who was herself implicated in the scandal. That same day, researchers in charge of the clinical trials released a statement noting that the former president had in fact not been a volunteer in the trial. By February 17, prosecutors had opened an investigation into nearly 500 public officials, including Vizcarra, the health minister, and the foreign minister (Osborne 2021).
The scandal marked the abrupt downfall of a popular politician. Following Kuczynski's resignation amid the Odebrecht scandal, Vizcarra's presidency was marked by high levels of approval as he championed anti-corruption reform and successfully evoked a constitutional right to dissolve an unpopular Congress, leading to early congressional elections in January 2020. Vizcarra's attacks on Congress placed him on the defensive against that institution, which successfully removed him from office in a second impeachment after a first one failed. Because many believed his removal was in response to his anti-corruption efforts, Vizcarra left office as an admired political figure, with an overwhelming majority of Peruvians approving of his presidency and expressing disagreement with his removal (La Vanguardia 2020). That made it all the more shocking for the public to see him revealed as the figurehead of a large scandal that involved sidestepping rules to access the COVID-19 vaccine before its public release. The scandal made frontline news in a country that has been one of the hardest hit by the pandemic (see Figure A2; Schwalb and Seas 2021).
In brief, the case of Peru's Vacuna-gate scandal is a most-likely case for finding a connection between corruption and diffuse support: levels of support were low, 2 News of the scandal began to percolate the day prior, but we take February 11 as the scandal's official start date given that we find a large spike in Google searches of "vizcarra" after February 11 (see Figure A1). discontent had been brewing for years, and the scandal was a shock. Yet, other findings in extant research provide reason to question whether the thesis will hold across all indicators of diffuse support. There is some evidence of spillover effects in which corruption scandals involving one or multiple individual politicians lead to worse evaluations for the institution to which the politician belongs (Bowler and Karp 2004;Chong et al. 2015;Lee 2018), politicians in general (Ares and Hernández 2017;Bowler and Karp 2004;Von Sikorksi, Heiss, and Matthes 2020), and overall perceptions of political corruption (Villora, Van Ryzin, and Lavena 2013). Other studies find limited spillover effects on diffuse values (Maier 2011;Ruderman and Nevitte 2015). In correlational analysis of Latin American public opinion, some find no evidence of a connection between corruption and support for democracy (Bailey and Paras 2006;Canache and Allison 2005). Such an outcome may be most common under conditions of scandal fatigue, when the public becomes dulled to news of another scandal following waves of scandals from administration to administration (Kumlin and Essaisson 2012;Waisbord 2004). Arguably, the Peruvian context is ripe for scandal fatigue: since the transition to democracy, all of Peru's former presidents, multiple members of the judiciary, and many members of Congress have been implicated in major corruption schemes.

Measures and models
Our data come from the 2021 AmericasBarometer survey of Peru, fielded between January 22 and March 26. The survey was designed to collect a national sample of voting-age respondents, who are 18 years or older, are citizens or permanent residents of Peru, and have access to a mobile phone. Additional information about the sample design can be found in Appendix B. Mobile phone coverage in Peru is approximately 87% (Montalvo, Pizzolitto, and Plutowski 2022). Participation in the AmericasBarometer was voluntary and not compensated. All interviewers were trained and certified to work on the project, and all interviews were audited using voice recording, timing checks, and metadata collected alongside the project. The final dataset consists of 3,038 interviews; of these, 1,332 were collected before the scandal broke on February 11 and 1,706 were collected after. We consider three measures of diffuse support: perceptions of corruption among politicians, system support (regime legitimacy), and support for democracy. For ease of comparison, all are linearly transformed to range from 0 to 100 (see Appendix A for full wording and coding). Perceptions of corruption are skewed: 89% of Peruvians express that half or more of the country's politicians are corrupt. Previous research has operationalized system support as an additive index of four items that ask about respect for political institutions, protection of basic rights, pride in the system, and support for the system (see Booth and Seligson 2009). We opt to use principal component analysis instead because it makes fewer assumptions about the relationships among the items and omits respondents who did not answer all four questions. The four items load highly onto a single factor, with an eigenvalue of 2.14 (Cronbach's alpha is 0.70). The mean value is 48.2. Support for democracy is measured by the respondent's level of agreement with the statement, "Democracy may have problems, but it is better than any other form of government." The mean is 57.8.
Due to space constraints on the questionnaire, perceptions of corruption and the system support module were by design assigned to two random splits of the questionnairesuch that only half the sample was asked these questions. The support for democracy question and all demographic questions were asked of all respondents, so our sample for analyses of this dependent variable is larger. For all three variables, item nonresponse is low, ranging from 0.1% for system support to 2.5% for perceptions of corruption. Our analyses omit cases of nonresponse. 3 To model the effect of corruption scandal, we use a linear regression with an indicator for treatment (interview conducted after the scandal broke) versus control (interview conducted before the scandal broke) as our main independent variable. The unexpected nature of the event along with the survey's random digit dialing sampling method allows us to treat scandal exposure as exogenous (see Table A2 for balance tests). Yet, one feature of the survey design needs to be accounted for: under the adaptive sampling approach applied by the survey team, respondents from the Lima area (the city as well as the provinces of Lima and Callao) were filtered out towards the end of data collection to achieve a final sample balanced on geographic region. 4 To address any potential confounding due to regional imbalance, our main specification employs coarsened exact matching (CEM), which balances the treatment and control groups on the Lima area dummy variable (we also balance on age, gender, education, and urban/rural residence), at the cost of only slightly reducing the sample (see Table A3).
We also report specifications without CEM weights as well as models that employ the survey weights provided by LAPOP to calibrate the sample to national benchmark data from the 2019 AmericasBarometer survey. In both cases, we report these results with and without controls for age, gender, education, urban/rural residence, and Lima area residence.
In an effort to evaluate the substantive implications of our results (following Rainey 2014), we define a minimal substantively interesting effect as an effect that changes a variable by more than 15% of its standard deviation and examine whether the 90% confidence interval falls outside that margin. While this threshold is necessarily arbitrary, we can also give it a more intuitive interpretation. The AmericasBarometer fields its surveys across the Latin America and Caribbean region and reports how countries compare on these measures. Those reports regularly highlight the three countries where averages are most alarming (lowest on system support and support for democracy; highest for perceptions of corruption). If we take the difference between the country average of the most alarming country and that of the fourth most alarming countrythat is, the aggregate change that would take a country out of the most alarming threethose values turn out to be very close to 15% of the standard deviation for system support and support for 3 Table A4 reports a comparison of item nonresponse rates before and after the scandal. We do not find any statistically significant differences (although the difference for nonresponse on system support is close to the threshold), and all the estimated differences are substantively very small. 4 The filter was applied beginning February 26. Interviews were assigned at random to a version of the questionnaire that included the filter or one that did not; when that mixed approach did not fully balance the sample, beginning on March 9 all interviews filtered out Lima area residents until the fieldwork ended on March 26. The proportion of the sample in the pre-treatment condition that is a resident of the Lima area is 40% while it is 29% in the post-treatment condition. democracy. Peru is itself the outlier country in the region with the highest rate of perceptions of corruption; setting Peru aside, the same pattern holds for countrylevel comparisons on this measure. Figure 1 plots the estimated effects of the scandal on our outcome variables of interest using our five model specifications. The figure shows results from models that include an indicator for treatment with the corruption scandal but in which we apply different weights and do or do not include controls. Dotted vertical lines also show the minimal threshold for substantively meaningful effects. 5 For perceptions of corruption, the treatment indicator is consistently positive, statistically significant, and substantively meaningful. Respondents interviewed in the post-scandal period believed that corruption was more prevalent among politicians compared to those interviewed in the pre-scandal period. The 2.5-point magnitude of the effect with CEM weights is substantial and impressive when considering the already high average level of perceptions of corruption in the sample (about 82 points on the 0 to 100 scale). After exposure to the scandal, an additional 6.6% of those in the survey said that all politicians were corrupt (42.6% versus 36%).

Findings
Our results also show that scandal exposure appears to have a negative effect on system support. The estimated effect is statistically significant at either the 90% or 95% level for all but one model specification, and it consistently crosses the threshold for substantively meaningful effects. When we apply CEM weights, the effect represents a drop by about 2 points on the 0 to 100 scale, rising to 2.8 points in Notes: Values report the effect of corruption scandal on corruption perceptions, system support, and support for democracy, for each model specification. Thick and thin lines report the 90% and 95% confidence intervals, respectively. Dotted vertical lines represent the minimal threshold for substantively meaningful effects. Regression results are reported in Tables A5, A6, and A7.

5
Density plots comparing pre-and post-scandal distributions are provided in Figure A3. other models. 6 Considering Peru's very low baseline level of system support, a further drop of 2-2.8 points suggests the strength of the impact on attitudes about the Peruvian political system.
Finally, we find no statistically or substantively significant effect of the corruption scandal on the extent to which respondents agree that democracy is better than any other form of government. Our estimates of treatment effect for this dependent variable are never statistically significant, and the confidence intervals never cross the threshold to be substantively meaningful. To corroborate this null effect, we also looked at the effect of the corruption scandal on two alternative measures of support for democracy that are regularly included in the AmericasBarometer: whether a military coup would be justified in times of high corruption and whether in difficult times the President would be justified in closing Congress and governing alone. Tables A9 and A10 show that we find similarly null results.
Are these results heterogeneous across different groups? We are limited in our ability to explore this given the cross-sectional nature of the survey, and the fact that in this context, political attitudes (including, for instance, partisanship) are fairly unstable and therefore arguably post-treatment. However, we did look at gender, age, education, and political knowledge as potential moderators in Tables A11,  A12, and A13. In this exploratory set of analyses, we find that the scandal increased perceptions of corruption more among older respondents but reduced system support more among younger ones. We find some evidence that explicit support for democracy increases among those with post-secondary education, suggesting that there are some opinion currents that rise to defend direct demand for democracy when the system is under threat. 7 We did not find any evidence that gender or political knowledge moderated the effects of the corruption scandal on public opinion.

Conclusion
Support for democracy matters: higher levels of explicit support for democracy translate into more stable and higher quality democracy (Claassen 2020). We assess the extent to which a major corruption scandal diminishes support for democracy in a most-likely case: Peru. In a context with low reserves of support for democracy and where discontent with the status quo has been high, the revelation that a popular former president and hundreds of officials and elites had secured early access to the COVID-19 vaccine shook the system. Yet, despite meeting Easton's (1975) criteria for the type of case that ought to experience decline in diffuse support, we find no movement in mean levels of explicit support for democracy following the Vacunagate corruption scandal in early 2021. Peruvians responded only by updating their perceptions of the scope of political corruption and their level of political support for the regime. In a well-functioning system, these results could be interpreted as good news for democracy, to the degree the public monitors democracy and sanctions failings in their role as "critical citizens" (Norris 1999, 2021). 6 Table A8 reports the results of models with the individual items within the system support index. Unsurprisingly, these effects are measured with less precision, but all but perceptions of whether basic rights are protected have negative effects. 7 In the weighted dataset, this group is 36% of the sample.
Our conclusion is not as sanguine for opinion dynamics in less established systems. In these systems, decreases in diffuse support for democracy may erode democracy in a more indirect manner. As we asserted at the start, even when explicit support for democracy holds steady, there is a more indirect path through which opinion shifts can undermine democracy. One branch of this path runs through executive aggrandizement, when discontent galvanizes into support for a strong executive with leeway to centralize power and weaken democratic institutions and processes (Bermeo 2016;Levitsy and Ziblatt 2018). A second branch runs through executive mismanagement, in which a public dismayed by persistent performance failings becomes more likely to elect inept outsiders whose failure results in a rotating executive office.
Indeed, Peru recently took this second branch toward greater democratic instability. Following Vacuna-gate, Peruvians elected an inexperienced political outsider -Pedro Castillowhose administration was fraught with governance challenges. On December 7, 2022, a third attempt by Congress to remove Castillo proved successful, if chaotic. Facing an impeachment vote in Congress, Castillo attempted to illegally dissolve the legislature, but was swiftly impeached, removed, and arrestedmaking him the third Peruvian president in the last five years to have his term cut short. Even when scandals only tarnish perceptions of the integrity of the government and support for it, democracy itself may still be at risk. As one political commentator put it, "Where politicians are the enemy, democracy is collateral damage" (Margolis 2022). And major corruption scandals like Vacuna-gate have the capacity to turn politicians into enemies. Data availability. Support for this research came from the United States Agency for International Development (Cooperative Agreement 7200AA19CA00022). The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The data, code, and any additional materials required to replicate all analyses in this article are available at the Journal of Experimental Political Science Dataverse within the Harvard Dataverse Network, at https://doi.org/10.7910/DVN/ KAPALW.