Cash Transfer and Political Accountability

Economic voting is the process by which individuals’ economic perceptions affect their views about candidates and parties and inform their vote choice to maximize their prospective economic well-being (Fiorina Reference Fiorina1981; Duch and Stevenson Reference Duch and Stevenson2008). These perceptions of the economy can refer either to voters’ evaluations of their general economic situation (sociotropic) or to their own personal economic situation (egotropic or pocketbook) (Lewis-Beck Reference Lewis-Beck1988).

Theory and evidence suggest that pocketbook economic voting is uncommon. Public opinion studies show that sociotropic evaluations have a stronger association with voting behavior and presidential approval compared to pocketbook evaluations (Fiorina Reference Fiorina1981; Lewis-Beck Reference Lewis-Beck1988; Lewis-Beck and Stegmaier Reference Lewis-Beck and Stegmaier2013). The connection between changes in personal financial conditions and governmental actions involves two distant sets of stimuli in the political environment. The task of connecting those two events requires voters to establish distal attitudinal congruence to vote economically, which makes pocketbook voting a complex task that is best suited for the more sophisticated (Gómez and Wilson Reference Gómez and Matthew Wilson2003, Reference Gómez and Matthew Wilson2006).

However, while sociotropic economic evaluations are more systematically associated with voting behavior, the literature also suggests that the task of connecting general economic outcomes with governmental action is not much simpler than pocketbook voting. Studies show that economic evaluations often have myopic foundations; that is, voters tend to consider limited time frames when assessing the economic performance of governments (Healy and Malhotra Reference Healy and Malhotra2009; Bechtel and Hainmueller Reference Bechtel and Hainmueller2011). Also, voters may often form evaluations based on random events or events that public officials cannot control (Achen and Bartels Reference Achen and Bartels2017; Healy et al. Reference Healy, Malhotra and Hyunjung Mo2010). Moreover, the commonly observed association between sociotropic economic perceptions and political choices can also indicate that voters adjust their views about the economy on the basis of their opinions about the electoral options, and not the contrary (Tilley and Hobolt Reference Tilley and Hobolt2011; Visconti Reference Visconti2019).

Additionally, even if voters form strong and stable opinions about the economy based on reliable economic indicators, one of the key components in economic voting models is the accuracy of the attribution of responsibility for those outcomes. The literature shows that the task of assigning responsibility for economic outcomes to political actors depends heavily on the extent to which the policymaking process is shared by institutions and parties (Powell and Whitten Reference Powell and Whitten1993; Duch and Stevenson Reference Duch and Stevenson2008). Economic voting becomes more common in contexts in which the executive concentrates responsibilities and centralizes power relative to opposition parties, as well as other agencies and levels of government. In the same vein, more recent studies find that factors such as decentralization (León Reference León2011; Guiteras and Mobarak Reference Guiteras and Mushfiq Mobarak2015), globalization (Hellwig and Samuels Reference Hellwig and Samuels2007; Cruz and Schneider 2016), characteristics of the local economy (Campello and Zucco Reference Campello and Zucco2020), and features of presidential systems (Hellwig and Samuels Reference Hellwig and Samuels2008; Samuels Reference Samuels2004; Samuels and Hellwig Reference Samuels and Hellwig2010; Valdini and Lewis-Beck Reference Valdini and Lewis-Beck2018) are key determinants of the extent to which voters attribute responsibility for economic outcomes to the executive.

Surprisingly, a specific type of pocket evaluation seems to overcome many of the limitations associated with economic voting models. An emerging body of scholarship provides evidence that voters tend to reward incumbents who implement large-scale antipoverty cash transfer programs, especially in developing economies. Cash transfer programs (or targeted social assistance) are government policies that seek to help the poor through targeted, means-tested financial aid (Barrientos and Santibáñez Reference Barrientos and Santibáñez2009; Layton and Smith Reference Layton and Erica Smith2015). Since these programs generate immediate and distinguishable changes in recipients’ finances and are usually associated with specific procedures, labels, and locations that signal their connection with government spending, they constitute a more powerful version of pocketbook voting (Tilley et al. Reference Tilley, Neundorf and Hobolt2018). Therefore, cash transfer programs provide a clear and reliable shortcut, in which voters assign political responsibility for the program’s outcome to the executive, especially since cash transfer initiatives usually entail credit claiming.

The evidence in favor of the positive effect of targeted social assistance on presidential approval is overwhelming in Latin America, with examples from countries such as Brazil (Zucco Reference Zucco2008; Licio et al. Reference Licio and Rennó2009; Zucco Reference Zucco and Power2013), Mexico (De La O 2013; but see Imai et al. Reference Imai, King and Velasco Rivera2020), Honduras (Linos Reference Linos2013), Ecuador (Winters Reference Winters2010), and Uruguay (Manacorda et al. Reference Manacorda, Miguel and Vigorito2011).Footnote ¹ All in all, while some researchers are inclined to see cash transfer programs as clientelistic practices, or at least as policies that can potentially be –hijacked for electoral purposes, economic voting based on cash transfer programs is generally seen as a type of pocketbook voting that can serve as positive reinforcement for incumbents and, after all, as a mechanism of accountability (Ashworth Reference Ashworth2012; Pavão Reference Pavão2016; Tilley et al. Reference Tilley, Neundorf and Hobolt2018).Footnote ²

Cash transfer programs tend to center on the executive branch throughout the different stages—agenda setting, policy formulation, selection, and implementation (Hoefer Reference Hoefer2021)—of their policy process. In the case of Brazil, different versions of such programs followed an executive-centered logic, with the federal government proposing the program, gathering support in the legislative arena, and carrying out implementation (Sugiyama Reference Sugiyama2011; Zucco Reference Zucco2013). It is particularly important that the federal government was able to design Bolsa Família, Brazil’s largest CCT program to date, while bypassing state-level authorities that would otherwise have undermined the program’s effectiveness (Fenwick Reference Fenwick2009). In this sense, the findings from the literature on the electoral effects of cash transfer programs rescue the connection between economic voting and democratic accountability by looking at those programs as instances where standard barriers to correct assessment and attribution—such as the multitude of actors participating in decisionmaking or the scarcity of information—are mostly absent.

However, little research has been conducted in more complex contexts in which the policy process makes it harder for voters to discern political responsibility for cash transfer programs. More specifically, a puzzling situation emerges when the executive participates in the policy process without exercising proposal power; that is, without being the agent that sets the agenda and initiates the legislative process. In this context, the executive participates in the policy process as an agent with formal veto power or when legally bound with the task of delivering the policy once it becomes law. From the policy process standpoint, an executive that does not initiate the policy process but “goes along” by not exercising full veto power and abiding by the law in implementation has limited causal bearing on the policy.

While proposal power signals commitment to vertical accountability, failing to exercise the veto or carrying on a mandate to implement the policy come because of horizontal accountability, in which the executive cooperates due to its political and legal liability to the other branches of government. Evidence from the scholarship, however, shows that voters tend to concentrate the rewards for collective decisionmaking on political actors with proposal power, rather than on those with veto power (Duch and Stevenson Reference Duch and Stevenson2013; Duch et al. Reference Duch, Przepiorka and Stevenson2015). Moreover, when policy delivery is largely influenced and performed by “non–electorally accountable officials” (Duch and Stevenson Reference Duch and Stevenson2008), such as the federal and local bureaucracies that operate as intermediaries in cash transfer programs, it is rational for voters to discount the reward they would offer to “electorally accountable officials” at the executive level.

All in all, when voters reward the executive for a policy that it did not propose and that it even opposed at the legislative level, they largely misattribute political responsibility. In other words, when the executive cooperates by not using its veto power and by carrying on implementation as mandated by law, it is responding to horizontal accountability pressures rather than to voter preferences. Therefore, it is not clear if cash transfer programs are immune to the conventional shortcomings of economic voting or if they, too, can be subject to “contingency dilemmas” that would make them a less effective mechanism of accountability. By examining a case in which the executive is not responsible for the design and approval of the program, this study sheds light on this question.

The Case of Brazil’s Auxílio Emergencial

The implementation of the Auxílio Emergencial during the COVID-19 pandemic in Brazil is an example of a cash transfer program with complex attribution of responsibility. In addition to Brazil’s large territory and population, the management of a collective response to a disease outbreak poses challenges that are magnified by the country’s high levels of income inequality and poverty (Aquino et al. Reference Aquino and Ismael Henrique Silveira2020). More specifically, the collective measures of social distancing recommended by health experts, such as sheltering in place, closing schools, and reducing economic activity, have disproportionately negative effects on the most vulnerable populations. In this sense, the country’s existing cash transfer program, Bolsa Família, was deemed by specialists as insufficient to provide a safety net for the wider group of the population affected by growing rates of unemployment and diminished economic activity during the pandemic.

Although experts, legislators, and state and local-level authorities concurred about the need to implement large-scale social distancing measures combined with financial relief to citizens since late February 2020, the executive sent ambiguous signals about the course of action. In the first half of March, while Health Minister Luiz Henrique Mandetta seemed aligned with most forces in the political system in favor of following the guidelines from the World Health Organization, President Bolsonaro’s stance became increasingly more explicit against social distancing measures and favorable to the idea that the regular functioning of the economy was preferable to providing citizens with extra government aid. Health Minister Mandetta was discharged less than a month after Bolsonaro gave a nationally televised address on March 24, in which he trivialized the outbreak and spoke openly against social distancing measures. Due to his handling of the pandemic, Bolsonaro was vehemently criticized by the media and the majority of political elites.

At that stage, the executive’s ambivalent response to the pandemic, in combination with constant political turmoil—which peaked with the resignation of popular Minister of Justice Sergio Moro—led to a decrease in Bolsonaro’s approval ratings. These crises reinforced the weakness of Bolsonaro’s fragile legislative coalition within Brazil’s highly fragmented lower house. After weeks of pressure from different sectors of society, and especially from members of both chambers of Congress, the executive proposed a financial assistance payment of 200 reais (about 39 US dollars at the time) to be paid to eligible groups of Brazilian citizens over three months. The proposal produced an immediate negative reaction from Congress, governors, and mayors, who campaigned for a higher monthly payment. Later in March, the bill written by Congress member Eduardo Barbosa and introduced by rapporteur Marcelo Aro proposed a total payment of 500 reais for three months, which received overwhelming support from different sectors of society. Given the imminent defeat, the executive reopened negotiations and, to change perceptions about its resistance to a widely popular measure, proposed a last-minute change to increase the monthly payment to 600 reais.

After both the Chamber of Deputies and the Senate approved the bill, the executive further delayed the implementation of the policy by negotiating funding sources. The first payment was available for withdrawal on April 27, nearly two months after the first confirmed case of COVID-19 in Brazil and four weeks after the bill was approved by Congress. Nearly one month later, the federal government launched a campaign on social media that attributed credits from the program to Bolsonaro and was followed by similar messages from the president and those close to him. Table 1 presents a timeline of these events.

Table 1.

Timeline of Events Involving the Creation of the Auxílio Emergencial

The events from the initial negotiation to the later credit claiming battle for the Auxílio Emergencial impose a particularly difficult task in attributing responsibility for government policy. Although the executive has, by design, an active role in shaping the policy during the legislative process and holds primary administrative responsibility for its delivery, it was Congress, operating through nearly unanimous legislative support, that exercised proposal power that realized the Auxílio Emergencial as a policy. While legislators were proactive in pressuring in favor of the program, as well as for a larger amount to be paid, the executive was passive and mostly responsive to horizontal pressures rather than to voter preferences.

This proposition is supported by an exercise of counterfactual reasoning. If Congress had not exercised proposal power in the policy process, the likely outcome would have been either no policy promoted or a cash amount much lower than the one delivered, as the executive first signaled. Alternatively, if the executive had exercised proposal power and made efforts to push it through the legislative process, the result might have been a very similar outcome to the one observed. Therefore, members of Congress were politically responsible for the existence of the program, while the executive was legally bound to carry through its delivery to the public.

Nevertheless, Bolsonaro’s approval ratings increased after the start of the Auxílio Emergencial, particularly among poorer respondents. After a couple of months of payments, the rate of respondents who disapproved of his administration dropped from 44 percent to 34 percent, while his approval rating rose from 32 percent to 37 percent in August 2020 and remained steady in December of the same year, which represented the highest level of approval since Bolsonaro took office.Footnote ³

This variation in approval is striking not simply because of its rapid growth despite the poor handling of the pandemic, but also because of larger shifts within income groups. While wealthier voters became disenchanted with Bolsonaro’s handling of the public health crisis, his approval ratings among the larger share of poorer individuals increased substantially. Specifically, the percentage of poorer respondents (i.e., family income was up to twice the minimum wage) who disapproved of the president dropped from 45 percent before payments were made to 27 percent in the December poll. In contrast, among the middle class (i.e., those who earn between three and five times the minimum wage), the disapproval rate only dropped from 47 to 41 percent in the same period, and increased from 29 percent to 41 percent in the entire period from December 2019 to December 2020. These trends in presidential approval indicate that despite evidence that the federal government mismanaged the several aspects of the pandemic (Ferigato et al. Reference Ferigato, Michelle Fernández and Ilana Ambrogi2020; Smith Reference Smith2020), Bolsonaro enjoyed a rise in popularity due to a policy that his government initially had opposed. Overall, Bolsonaro’s levels of support slightly improved while his base shifted from being composed largely of wealthier voters to a more diverse socioeconomic profile.

If the Auxílio Emergencial program contributed to the bump in presidential approval, we should observe recipients of the benefit to be more approving than nonrecipients, ceteris paribus. But because of the unique dynamics, in which the lack of proposal power by the executive created a scenario of unclear responsibilities, we also expect that the general effect of receiving the benefit should be small, given that individuals may attribute credit for the policy to Congress rather than to the president. That is, the increase in approval associated with the program should be driven by those who believe Bolsonaro was responsible for the Auxílio Emergencial.

Again, attribution of responsibility for this cash transfer program was marked by an unusually complex scenario for this type of policy, which means that the challenges voters usually face when assigning responsibility to other economic outcomesalso apply here. For one thing, extensive research shows that levels of information shape individuals’ attitudes and how they process information (Zaller Reference Zaller1992; Bartels Reference Bartels1996; Krause Reference Krause1997), which means that those who were not aware of the process should have had more difficulty in correctly attributing responsibility. In addition, we know that citizens’ evaluations of policies are often formed to be consistent with previously held beliefs and attachments (e.g., attitudes toward parties) (Anderson et al. Reference Anderson, Mendes and Tverdova2004; Evans and Andersen Reference Evans and Andersen2006; James and Van Ryzin Reference James and Van Ryzin2017; Wlezien et al. Reference Wlezien, Franklin and Twiggs1997); this means that attribution of responsibility for the program also was probably influenced by predisposed attitudes toward the president and Congress. Furthermore, politicians may be particularly incentivized to claim credit for a policy even if they are not responsible in scenarios of unclear attribution of responsibility.

Misattribution of responsibility by voters is a problem for democracy to the extent that it provides politicians with incentives to be less concerned with their performance and responsiveness and instead to act more nearly how they please (Achen and Bartels Reference Achen and Bartels2017; Campello and Zucco Reference Campello and Zucco2020). The remedy for this bias seems to be to correct voters’ misattribution. However, existing works are unclear on whether debiasing efforts work. On the one hand, we should expect citizens to update their political views based on new policy information (Fiorina Reference Fiorina1978; Nicholson, Reference Nicholson2011; Bullock Reference Bullock2009; Boudreau and MacKenzie Reference Boudreau and MacKenzie2014), and efforts to debias evaluations can potentially work (Healy et al. Reference Healy, Malhotra and Hyunjung Mo2010). On the other hand, individuals often employ motivated reasoning in political assessments (Bolsen et al. Reference Bolsen, Druckman and Lomax Cook2014; Taber and Lodge Reference Taber and Lodge2006) and often disregard information on incumbent responsibility (Huber et al. Reference Huber, Hill and Lenz2012; Campello and Zucco Reference Campello and Zucco2020).

We contribute to answering this question by testing whether information about the political negotiations for the Auxílio Emergencial influenced views toward the president and Congress, with a survey experiment. We expect that more information about the negotiations in which Congress, rather than the president, had proposal power over the program should harm views on the latter while improving views on the former. Yet the effect of more information should be especially significant among those who did not have strong feelings toward these actors, since they would be less prone to motivated reasoning, and among respondents who were ambivalent or did not know who deserved credit for the program. The experiment was based on an analysis of the relationship between receiving the benefit, attribution of responsibility, and approval.

Observational Analysis

We assessed the impact of the program on views toward the president and Congress with an online survey of one thousand respondents conducted by the polling firm Quaest Consultoria e Pesquisa. Given the infeasibility of conducting face-to-face surveys at the time, the study relied on a sample designed to mirror the online population in terms of income, age, gender, education, and region.Footnote ⁴ Data collection was carried out between June 14 and 17, 2020, weeks after the federal government made the first two payments of the 600 reais benefit.Footnote ⁵

The main independent variable of this first analysis is whether respondents received the Auxílio Emergencial, which was measured as a question that briefly described the policy and asked respondents if they had requested and received it. Response options included “request approved and received payments,” “request approved but still waiting for first payment,” “request denied,” and an option for respondents who had not requested the benefit.Footnote ⁶ The percentage of respondents in the sample with requests approved is 38 percent, the proportion with requests denied is 16 percent, and the proportion who had not requested it is 46 percent.Footnote ⁷ Benefit recipient status was coded as a binary variable, with those who had the request approved as 1 and all others as 0.

The second key variable in the analysis is who respondents believed to be responsible for the Auxílio Emergencial. The question includes six response options: President Bolsonaro, the National Congress (deputies and senators), governors, and mayors, along with a “don’t know” option. A total of 47.5 percent of respondents indicated that Congress was responsible for the policy, while 38 percent indicated the president. This split in responses reflects the high degree of confusion among citizens about who was politically responsible for the implementation of the policy. Less than 4 percent of respondents indicated that either governors or mayors were responsible for the benefit, while 11 percent did not know. Responsibility for the program was coded in the analyses as a binary variable: those who believed that Bolsonaro or Congress were responsible for the program were coded 1 and other responses 0.

The dependent variables refer to evaluations of the president and Congress. Respondents were asked, “In general, how do you evaluate Jair Bolsonaro’s performance as president of Brazil?,” and “In general, how do you evaluate Congress’s performance?” “Don’t know” responses were coded as middle categories, resulting in a seven-point scale that ranged from “terrible” to “excellent.”

We evaluated the relationship between receiving the benefit and evaluations of the president and Congress by first balancing respondents with respect to pretreatment covariates, using coarsened exact matching (Blackwell et al. Reference Blackwell, Iacus, King and Porro2009). The covariates used were family income, sex, age, region, and religious affiliation.Footnote ⁸ Respondents from higher income brackets (above ten minimum ages) that did not have any recipients were not included in the models.Footnote ⁹ Moreover, although the survey included a question on the vote choice for president in the 2018 election, we did not attempt to balance the sample based on prior voting, since the vote recall question cannot be considered a pretreatment covariate because it is affected by short-term factors (Van Elsas et al. Reference Van Elsas, Rozemarijn Lubbe, Van Der and Van Der Brug2014; Van Elsas et al. Reference Van Elsas, Miltenburg and van der Meer2016). As the vote choice question asked respondents to recall an event that occurred almost two years before receiving the benefit, we cannot rule out the possibility that those who received the benefit became less likely to declare a past vote for Fernando Haddad, Bolsonaro’s challenger in 2018. Moreover, further examination shows that recipients were not more likely than nonrecipients to consider either Congress or the president as responsible for the benefit.Footnote ¹⁰

For estimating the effect of receiving the benefit on presidential and congressional approval, we then ran multivariate OLS regressions on the balanced sample (using the matching weights from the procedure described above), in which we controlled for the matching weights and the same covariates used to balance the sample. Table 2 presents results from six models: three for presidential approval and three for congressional approval. The first modelincludes recipient status only as a predictor, aside from the control variables listed above. In the second, we included the binary indicator of responsibility for the program as a control, and in the third, we interacted the two variables to estimate the extent to which the association between receiving the benefit and approval was conditioned by the attribution of responsibility. For the sake of parsimony, we present only coefficient estimates for the interaction term, and place full model specifications in the appendix (table A2).

Table 2.

Impact of Treatment (Recipient) on Presidential and Congressional Approval

* p < 0.05, ** p < 0.01

Results are coefficients from OLS regressions with matching weights.

Notes: Models do not include respondents from income brackets that do not have program recipients. Dependent variable ranges from 0 to 6 where 6 means “Great.” Coefficient estimates for matching weights, gender, region, income, and religious affiliation are omitted.

The first two models indicate a positive association between receiving the benefit and views on the president, even after controlling for whether the respondent believed that the president was responsible for the program. The third model indicates that this association between the benefit and approval is primarily observed among those who believed that Bolsonaro was responsible for the Auxílio Emergencial. That is, among those who did not believe that Bolsonaro was responsible, there is no significant difference in approval between those who received the benefit and those who did not. In contrast, among those who believed that Bolsonaro was responsible, those who received the benefit expressed significantly higher rates of approval than those who did not. This suggests that the effect of the program would have been substantially larger if a higher share of benefit recipients perceived Bolsonaro as responsible, or inversely, that the effect would have been smaller if more individuals had not seen him as responsible.

Results from models 4 and 5 also indicate that those who received the benefit approved of Congress more than those who did not. However, the association between attribution of responsibility to Congress and approval is not as clear, given that those who believed that Congress was responsible did not approve of Congress more than others. Moreover, there is no significant difference in the association between receiving the benefit and approval of Congress among those who believed that Congress was responsible and those who did not. That is, while those who attributed the Auxílio Emergencial to Congress expressed slightly more positive views toward Congress, the regression coefficients do not give us confidence that there are differences in congressional approval associated with attributing responsibility. Furthermore, the coefficients do not show that the association between the benefit and approval is conditional on credit attribution.Footnote ¹¹

Obviously, it is difficult to infer any causal relationships from this type of analysis. We do not have a time series of respondents to properly evaluate the causal effect of receiving the benefit on later approval of the president and Congress. Yet our results are nonetheless consistent with previous studies on Bolsa Família using similar empirical strategies (Zucco Reference Zucco2013) and with a small panel of telephone respondents that corroborates the positive association between receiving the benefit and more positive views toward the president (Pavão et al. Reference Pavão, Meneses and Borba2020).We also know that the attribution of responsibility may be conditioned by attitudes toward the president in the first place, which may impact our estimates in favor of the conditional relationship. With this limitation in mind, we relied on a survey experiment to examine the role of attributions of responsibility when we showed respondents information about the policymaking process that led to the benefit.

Experimental Analysis

Because voters may face a challenging task in accurately attributing responsibility for the Auxílio Emergencial, our experiment was designed to assess whether providing more information about the process that led to the program affected attitudes toward the president and Congress; that is, how voters distribute the reward for the program between the executive and legislative branches. At the end of the same online survey described previously, we provided a random share of respondents with information about the negotiation process that resulted in the program to potentially debias views on who deserved credit for the program, and then remeasured presidential and congressional approval.

The stimuli designed to clarify responsibility consisted of a short paragraph about the government’s initial proposal of 200 reais, the negotiations led by deputies and senators to increase the value of the benefit, the government’s last move to make it 600, and the president’s delay in implementing the program. The description of events was followed by a question on whether the president’s position of austerity, as opposed to helping the population, was right or wrong. A translation of this description and question is the following:

Considering the decisions by the president, with which of the following statements do you agree more?

• President Bolsonaro was right: one must be careful with the government’s finances during the pandemic.

• President Bolsonaro was wrong: one must help the population in need during the pandemic.

The stimuli did not explicitly state that Bolsonaro was not politically responsible for the program. The goal of the treatment was mainly to inform subjects about the legislative process of approving the program, one in which the executive took an ambivalent and reactive role and consequently did not exercise proposal power. The first stimulus presents a newspaper-style narrative that conveys Bolsonaro’s hesitance and cooperation while portraying Congress as having an unequivocally proactive role.

This narrative is preferable to less ambiguous approaches, for a few reasons. First, it represents the complex setting and unfolding of the events while conveying factually accurate information; that is, without omitting Bolsonaro’s few cooperative actions, even if taken due to external pressure. Second, since the goal of the treatment is to inform rather than to polarize, a more explicit statement against Bolsonaro could instead alienate his supporters while producing acquiescence among his detractors. Although the polarization effect could potentially boost the treatment, it would not be related to learning. Moreover, while the paragraph about the negotiation process focused on the actions of the president and Congress, the follow-up question on whether Bolsonaro’s position was right was included to engage respondents and clarify that he did not prioritize those in need. It is important to note that for analyzing the effects of the additional information on views toward the president and Congress, we were not interested in the responses to the question, as it was part of the manipulation.

After receiving the stimuli, respondents in the treatment group were again asked in separate questions the extent to which they approved of the president and Congress. Those in the control group were not shown the stimuli, but were asked about approval. As such, we had measures of approval conducted early and late in the questionnaire, with the stimuli shown to a group of respondents just before the second measurement. Our expectations are that informing respondents about the process should clarify responsibility for the program and redistribute views on responsibilities about its implementation. Consequently, we should observe a decrease in the approval level of the president and an increase in the approval level of Congress. These effects should be stronger among respondents who do not have strong predisposed attitudes toward the president and Congress and among respondents who do not know who is responsible for the program.

The randomization of the treatment led to uneven groups and some imbalance. Just over 60 percent of respondents (602 respondents) were assigned to the control group and 40 percent to the treatment. Our balance test based on a regression model showed that the groups were not balanced, on the basis of a few variables.Footnote ¹² Due to the imbalances, our analyses were conducted with multivariate regressions that included a set of sociodemographic variables, recipient status, and initial measures of approval as controls. While including covariates as controls in the analysis does not often make the estimates more credible relative to the unadjusted model (Mutz et al. Reference Mutz, Pemantle and Pham2019), the imbalance with respect to the lagged dependent variable suggests that a model without covariate adjustment would lead to biased results.Footnote ¹³

Our first analysis looks at the general impact of receiving the treatment on presidential and congressional approval. That is, our dependent variables are levels of approval of the president and Congress, measured with the same scale described in the observational analyses. Given that we included the pretreatment measures as predictors in the model, we estimated the change in approval associated with being in the treatment group. We again present a short version of the results in table 3 and place the full regression in the appendix (table A11).

Table 3.

Impact of Treatment (Vignette) on Presidential and Congressional Approval

* p < 0.05, ** p < 0.01

Results are coefficients from OLS regressions.

Dependent variable ranges from 0 to 6 where 6 means “Great.” Coefficient estimates for recipient and sociodemographic controls are omitted.

The first model indicates that receiving the treatment is not associated with a drop in the president’s approval.Footnote ¹⁴ In contrast, the second model indicates that being in the treatment group is associated with a significant increase in approval of Congress. More precisely, views on Congress among those in the treatment group improved by roughly 0.38 in the seven-point scale, on average, relative to those in the control group.Footnote ¹⁵ This suggests that the treatment designed to inform respondents—that is, description of the negotiating process and question about the position of the president—had a positive effect on views toward Congress. The results from both models also show a high correlation (i.e., consistency) between the first and second measures of approval, especially regarding the president.Footnote ¹⁶

The effect of the stimuli on approval of the president and Congress is also likely not to be the same across individuals.Footnote ¹⁷ Respondents who already had a positive view of Congress tended to be less affected by the stimuli. That is, when we evaluate the effect of the treatment as conditional on the initial level of approval of Congress, we find that the treatment only affects opinions among those who already disapproved of Congress. In contrast, the effect of the stimuli on approval of the president is consistently null, apart from those who believed that Bolsonaro was doing a terrible job, which is probably explained by flooring effects. Figure1presents the marginal effect of receiving the treatment for each starting level of approval, again controlling for sociodemographic characteristics, approval of the other institution, and whether the respondent received the benefit.

The graph on the left shows that the only statistically significant effect of the treatment was among those who initially thought the president was doing a terrible job. Although significant, the effect is small and potentially suffers from bias, given that respondents were not able express a lower rating because of the limits of the scale (flooring effect). The graph on the right indicates that the treatment had a significant effect on approval of Congress among respondents who picked the three response options associated with disapproval in the pretreatment measure. In other words, the difference in approval of Congress between those in the treatment and control groups is only significant among those who had a strong or slightly negative initial opinion of Congress. One interpretation of this result is that the positive stimulus about Congress is more informative for these groups.

A better test of the informational mechanism is a comparison of the treatment effect between respondents who did not know who was responsible for the Auxílio and respondents who believed that either the president or Congress was responsible. That is, we assessed the effect of the stimuli conditional on who respondents believed was responsible for the Auxilio before the experiment. Figure 2 shows the marginal effect of the stimuli on presidential and congressional approval among those who believed that Congress was responsible, those who thought Bolsonaro was responsible, and those who answered governors, mayors, or “do not know.”

Figure 1.

Marginal Effects of Treatment Conditional on Previous Approval of President and Congress

Figure 2.

Marginal Effects of Treatment Conditional on Previous View of Who Was Responsible for the Auxílio Emergencial

Once again, the results of heterogeneous effects of the treatment on presidential approval indicate that views on the president are not significantly affected, even among respondents who do not think Congress or Bolsonaro is responsible for the program. Nevertheless, the graph on the right shows that while the effect of the stimuli is positive and significant for all three groups, its effect is larger among those who do not know or attribute to governors and mayors. In other words, the increase in the level of approval of Congress associated with learning about the negotiation process that led to the program is larger among those who were initially ambivalent or least informed about it.

Overall, the results partially corroborate our hypothesis that more information about the process of policy approval and implementation can affect attitudes about political actors through the assignment of responsibilities, since only views on Congress were affected by the experimental stimuli.Footnote ¹⁸ The results from heterogeneous effects tests, particularly the second, give support to our hypothesis that those less prone to motivated reasoning are more likely to be affected by the new information, even though this is only the case for views on Congress.