Trustful voters, trustworthy politicians: A survey experiment on the influence of social media in politics

Aruguete, Natalia; Calvo, Ernesto; Scartascini, Carlos G.; Ventura, Tiago Working Paper Trustful voters, trustworthy politicians: A survey experiment on the influence of social media in politics IDB Working Paper Series, No. IDB-WP-1169 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Aruguete, Natalia; Calvo, Ernesto; Scartascini, Carlos G.; Ventura, Tiago (2021) : Trustful voters, trustworthy politicians: A survey experiment on the influence of social media in politics, IDB Working Paper Series, No. IDB-WP-1169, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0003389 This Version is available at: https://hdl.handle.net/10419/237464 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/3.0/igo/legalcode Trustful Voters, Trustworthy Politicians: A Survey Experiment on the Influence of Social Media in Politics Natalia Aruguete Ernesto Calvo Carlos Scartascini Tiago Ventura IDB WORKING PAPER SERIES Nº IDB-WP-1169 July 2021 Department of Research and Chief Economist Inter-American Development Bank July 2021 Trustful Voters, Trustworthy Politicians: A Survey Experiment on the Influence of Social Media in Politics Natalia Aruguete* Ernesto Calvo** Carlos Scartascini*** Tiago Ventura** * Universidad Nacional de Quilmes ** University of Maryland *** Inter-American Development Bank Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Trustful voters, trustworthy politicians: a survey experiment on the influence of social media in politics / Natalia Aruguete, Ernesto Calvo, Carlos Scartascini, Tiago Ventura. p. cm. — (IDB Working Paper Series ; 1169) Includes bibliographic references. 1. Trust-Political aspects-Brazil-Econometric models. 2. Trust-Political aspectsMexico-Econometric models. 3. Social media-Political aspects-Brazil-Econometric models. 4. Social media-Political aspects-Mexico-Econometric models. 5. Voting research-Brazil-Econometric models. 6. Voting research-Mexico-Econometric models. 7. Reliability-Political aspects-Brazil-Econometric models. 8. Reliability-Political aspects-Mexico-Econometric models. I. Aruguete, Natalia. II. Calvo, Ernesto. III. Scartascini, Carlos G., 1971IV. Ventura, Tiago. V. Inter-American Development Bank. Department of Research and Chief Economist. VI. Series. IDB-WP-1169 Copyright © Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 AttributionNonCommercial-NoDerivatives (CC-IGO BY-NC-ND 3.0 IGO) license (http://creativecommons.org/licenses/by-nc-nd/3.0/igo/ legalcode) and may be reproduced with attribution to the IDB and for any non-commercial purpose, as provided below. No derivative work is allowed. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the UNCITRAL rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this CC-IGO license. Following a peer review process, and with previous written consent by the Inter-American Development Bank (IDB), a revised version of this work may also be reproduced in any academic journal, including those indexed by the American Economic Association's EconLit, provided that the IDB is credited and that the author(s) receive no income from the publication. Therefore, the restriction to receive income from such publication shall only extend to the publication's author(s). With regard to such restriction, in case of any inconsistency between the Creative Commons IGO 3.0 Attribution-NonCommercial-NoDerivatives license and these statements, the latter shall prevail. Note that link provided above includes additional terms and conditions of the license. The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. http://www.iadb.org 2021 Abstract* Recent increases in political polarization in social media raise questions about the relationship between negative online messages and the decline in political trust around the world. To evaluate this claim causally, we implement a variant of the well-known trust game in a survey experiment with 4,800 respondents in Brazil and Mexico. Our design allows to test the effect of social media on trust and trustworthiness. Survey respondents alternate as agents (politicians) and principals (voters). Players can cast votes, trust others with their votes, and cast entrusted votes. The players’ rewards are contingent on their preferred “candidate” winning the election. We measure the extent to which voters place their trust in others and are themselves trustworthy, that is, willing to honor requests that may not benefit them. Treated respondents are exposed to messages from in-group or out-group politicians, and with positive or negative tone. Results provide robust support for a negative effect of uncivil partisan discourse on trust behavior and null results on trustworthiness. The negative effect on trust is considerably greater among randomly treated respondents who engage with social media messages. These results show that engaging with messages on social media can have a deleterious effect in trust, even when those messages are not relevant to the task at hand or not representative of the actions of the individuals involved in the game. JEL Codes: D72, D83, D91 Keywords: Trust, Trustworthiness, Social media, Political polarization * This research is part of the Inter-American Development Bank project: “Transparency, Trust, and Social Media”, 1300600-01-PEC. PI: Ernesto Calvo, 2019-2020. We thank Elizabeth Zechmeister, Noam Lupu, and Maita Schade from LAPOP, who coordinated the probabilistic selection of respondents in Brazil and Mexico. We also thank Julia Rubio, who contributed to the survey design, as well as the members of our Lab (iLCSS-UMD). The experimental design was preregistered at the OSF Framework and can be accessed at https://osf.io/hvjkt. The experiment has received approval from the university’s Institutional Review Board, number 1552091-3. 1 Trustful Voters and Trustworthy Politicians All measures of trust have shown steady declines around the world over the past decade (Keefer, Scartascini and Vlaicu,2018;Murtin et al.,2018;Scartascini and Valle L.,2020). These declines are concurrent with the rise of social media as a dominant platform for interpersonal communication and for the delivery of political news. However, there is little research that tests for the relationship between social media exposure, social media engagement, trust, and trustworthiness.1 Is social media reducing political trust among voters? Is social media engagement making voters less trustworthy? The increases in uncivil dialogue (Mason,2016;Iyengar, Sood and Lelkes,2012) and polarization (Banks et al.,2020;Bail et al.,2018) raise questions about the relationship between online exposure and the recent decline in political trust. In this article we report results of an experiment that shows significant declines in trust behavior among users exposed to, and who engaged with, uncivil partisan messages in social media. Understanding the effect of social media exposure on trust is substantively and theoretically important. Political trust is critical to citizens’ commitment to the rule of law, norms and regulations, and democracy. Research shows that governments perceived as trustworthy are associated with increases in political engagement, higher voter turnout, citizen support for existing policies, support for institutional reforms, compliance with political authorities, and reciprocity (Levi and Stoker,2000). Mistrust, on the other hand, leads to disengagement and produces systematic biases against broad-based policies with long-term benefits, like improving education, or policies whose returns are more difficult to observe, like bureaucratic reform. Instead, citizens prefer politicians who promise policies with immediate, tangible effects on their welfare, even if they do not promote long-term sustainable and inclusive growth (Keefer, Scartascini and Vlaicu,2018). In this paper, we describe a survey experiment that implements a variant of the well-known 1See Witmer and Håkansson (2015) for an overview of this discussion. trust game in an electoral context.2We test whether respondents trust others to act on their behalf (by entrusting other players to cast votes on their behalf) and whether they are trustworthy with respect to the resources (votes) entrusted to them. Trusting behavior increases the potential rewards perceived by the participants (more votes) but may carry large costs if the players’ trust is betrayed (that is, if their candidate loses the election). The “players” (in this case the respondents to the survey) receive prizes if their candidate wins the election and raffle tickets for each vote they contribute to the win. Respondents incentives are well aligned with collecting as many votes as they are with amassing more resources in the traditional trust game.3After an initial round of the game, we randomly treat a subset of respondents to negative and positive tweets from incumbent and opposition politicians and measure changes in trust and trustworthy behaviors. We expect negative tweets from out-group politicians to activate partisan identities, even if those identities are orthogonal to the actual game being played. We implement our survey experiments using two large, randomized panels of 2,400 Brazilian and Mexican respondents each.4 Results show robust and statistically significant declines in trust among voters exposed to negative messages from out-group political figures (dissonant messages). Findings also support an activation of partisan identities and higher memory availability (Kahneman,2011) of polarization frames.5Results are more modest when only the negative tone of the social media post (uncivil discourse) is considered. Finally, we do not find a statistically significant effect on trustworthiness. Agents cast entrusted votes at the same rate, regardless of the treatments they received. Overall, results provide support for a negative effect of social media exposure on trust (giving votes for others to cast) but no support for a decline in trustworthiness (casting entrusted votes). 2Trust games are a well-established methodological strategy for studying economic exchanges in lowinformation environments (Berg, Dickhaut and McCabe,1995). 3The results for this modified political trust game are consistent with those of traditional trust games, both in the total numbers of votes entrusted to other players as well as in trust behavior across rounds of play. 4All the hypotheses were preregistered, and the preregistration has been anonymized and uploaded to the editorial manager for reviewing purposes. 5As shown by Bail et al. (2018), exposure to counter-attitudinal arguments may create a “backfire" effect that increases political polarization and induces motivated reasoning on the users’ side. Our study builds on these findings to show similar negative effects on interpersonal trust. 2 Further testing of our findings indicates that “dosage" matters. We find that incidental exposure to social media has modest effects on trust. Results are pronounced and statistically significant at higher levels of engagement with tweets. Differences in trust between the control and treatment group are larger when they “do” Twitter (like, retweet, reply) as opposed to when they “read” it (no engagement). These findings are new and important, as they point to differences between social media platforms and more traditional news outlets.6 In all, our work offers three novel contributions to scholars interested in the study of social media, trust, and democratic governance. First, we find that social media exposure leads to declines in trust behavior rather than a mere change in attitudes.7The decline in trust behavior is self-interested and cannot be explained by the desire of the respondents to interpret the intent of the survey instruments. Second, we show conclusively that social media engagement magnifies the effect of the experimental treatment. There is a larger decline in trust among respondents who shared the content of the treatment, compared with those who were simply exposed to partisan messages. This is critically relevant for the burgeoning literature on incidental exposure to news (Boczkowski, Mitchelstein and Matassi,2018;Fletcher and Nielsen,2018). Our two-way design, comparing engaged and nonengaged users in the randomized experiment, supports the view that sharing behavior increases the negative effect of social media treatments. Third, we contribute methodologically to the study of trust games, presenting a survey design that replicates important behavioral responses from in-person lab experiments. Our design brings the trust game from the more traditional “investment" setting to an electoral political scenario, which could prove useful for understanding the role of trust in voter-politician interactions. Although the use of online survey experiments reduces the number of measurements taken from each treated individual, it can be rapidly scaled up, rendering results with 6Research in political communication describes several other important differences, including changes in editorial gatekeeping, in the practices and routines of journalists, and in exposure to different news frames propagated by peers (Shoemaker and Reese,2013;Tandoc,2014;Aruguete and Calvo,2018). Showing that engagement is an important mediator in reducing trust also contributes to current research on incidental consumption of news (Boczkowski, Mitchelstein and Matassi,2018), indicating that the consequences of incidental exposure may be more modest than previously thought. 7See Tucker et al. (2018) for an overview of the recent literature on social media 3 higher external validity. The organization of this paper is as follows. First, we describe the substantive importance of testing for the relationship between social media exposure, trust, and trustworthiness. Second, we present our experimental design and its implementation in Mexico and Brazil. Third, we present our general experimental results, with estimates that distinguish between partisan cognitive dissonance, negative tone of the content, and sharing behavior. Fourth, we describe extensions of our results that describe the mediating effect of negative emotions on trust. We conclude with a discussion of possible further extensions of our work. 2 Trust and Trustworthiness Beginning with the work of Adam Smith, trust and trustworthiness have been recognized as key factors in promoting cooperation and exchange (Smith,1937). Trust and trustworthiness are fundamental forces that shape societies and institutions (formal and informal) and co-evolve with them (Arrow,1974;Guiso, Sapienza and Zingales,2004). Trust and trustworthiness have positive effects on the ability of people to make transactions and on the ability of governments to function (Arrow,1974;Knack and Keefer,1997;Gambetta,1988;Jacobsen, 1999;Zak and Knack,2001;Algan and Cahuc,2014;Bjørnskov and Méon,2015;Algan et al., 2017). High trust correlates with higher growth, social progress, and democratic stability (Algan and Cahuc,2010,2014;Aghion et al.,2010;Keefer et al.,2020). Importantly, if citizens do not trust their governments, they will not demand public goods or policies whose benefits materialize only in the long run (Keefer, Scartascini and Vlaicu,2018;Keefer et al.,2020; Scartascini and Valle L.,2020). Studying trust has become ubiquitous in recent years. Most studies use well-known survey questions that measure trust attitudes rather than trust behavior.8This is problematic, as there is consistent evidence that trust attitudes and trust behavior are weakly correlated (Wilson, 2017). Importantly, the analytical connection between the social benefits of trust and trust8Examples include agreement questions such as “Most people can be trusted” as well as scale questions of reported trust in family, friends, and neighbors. 4 Table 1 Trustworthy, Transition Matrix (Brazil) Second Round First Round Agree to cast Don’t Agree Total Agree to cast 51% (1213) 12% (295) 64% (1508) Don’t Agree 8% (189) 28% (666) 36% (855) Total 59% (1402) 41% (961) 100% (2363) Table 2 Trustworthy, Transition Matrix (Mexico) Second Round First Round Agree to cast Don’t Agree Total Agree to cast 51% (1188) 13% (307) 64% (1495) Don’t Agree 5% (129) 31% (722) 36% (851) Total 56% (1317) 44% (1029) 100% (2346) ure 2. Figure 2 Trust: First and Second Rounds of the Game, Compared a) Brazil b) Mexico Note: The plots present changes in trust (votes delegated) between the first and the second rounds of the game in Brazil and Mexico. The upper triangle in each figures indicates the share of respondents who delegated more in the second round (increase in trust), whereas the lower triangle indicates the share of subjects who delegated less (decrease in trust) 11 6 Experimental Results Descriptive evidence in the previous section shows that between the first and second rounds, fewer respondents agreed to cast the votes entrusted to them (lower trustworthiness) and smaller quantities were delegated to other respondents (lower trust). In Brazil, rates of agreement to cast entrusted votes (trustworthiness) declined from 64% to 59%, and in Mexico from 64% to 56%. Similarly, entrusted votes (trust) in Brazil declined from 3.4/10 votes in the first round to 3.17/10 in the second, and in Mexico from 3.75/10 in the first to 3.24/10 in the round. In the next two subsections, we show that social media exposure had no effect on the decline in trustworthiness but a significant effect on trust. 6.1 The Null Effect of Social Media Exposure on Trustworthiness Table 3presents our findings on the effect of social media exposure on trustworthiness. We estimate benchmark linear probability models to capture the effect of exposure to social media messages on the binary decision to cast votes entrusted by another player in the second round of the game. In the second round, our models interact the treatments with the subjects’ first-round decision. Columns 1 to 3 present the results for Brazil, while columns 4 to 6 present those for Mexico. The baseline condition includes respondents who played the second round of the game without being exposed to social media messages. We then separate by treatment condition (negative/positive and in-group/out-group) and control for the first-round decision to cast votes. While findings are suggestive and point in the right direction, estimates do not reject the null hypothesis. Accordingly, we report null findings for the trustworthiness (agent) hypotheses, HT1Aand HT1B. Only hypothesis HT2holds, showing a decline in trustworthiness in later rounds, consistent with most in-person implementations of the trust game. This decline, however, is not explained by social media exposure. Therefore, contrary to our expectations, exposure to social media messages, varying the endorsement and framing of the messages, has no effect on the trustworthiness of respondents. 12 Table 3 Regression Models: Treatment Effects of Framing and Endorsement on Trustworthiness Brazil Mexico (1) (2) (3) (4) (5) (6) Intercept 0.344∗∗∗ 0.416∗∗∗ 0.462∗∗∗ 0.240∗∗∗ 0.303∗∗∗ 0.207∗ (0.074) (0.082) (0.097) (0.076) (0.095) (0.107) Trustworthiness (Round 1) 0.589∗∗∗ 0.591∗∗∗ 0.594∗∗∗ 0.634∗∗∗ 0.632∗∗∗ 0.634∗∗∗ (0.032) (0.032) (0.032) (0.029) (0.029) (0.029) Framing: Negative 0.035 −0.027 (0.036) (0.034) Framing: Positive 0.005 −0.019 (0.036) (0.033) Out-group −0.028 0.0005 (0.040) (0.043) In-group 0.019 −0.030 (0.041) (0.042) Negative Out-group −0.032 −0.011 (0.052) (0.063) Positive Out-group −0.024 0.005 (0.050) (0.052) Negative x Trustworthiness (Round 1) −0.021 0.013 (0.045) (0.042) Positive x Trustworthiness (Round 1) −0.010 0.012 (0.045) (0.042) Out-group x Trustworthiness (Round 1) 0.028 0.002 (0.050) (0.054) In-group x Trustworthiness (Round 1) −0.007 0.039 (0.050) (0.052) Negative Out-group x Trustworthiness (Round 1) 0.015 0.009 (0.065) (0.078) Positive Out-group x Trustworthiness (Round 1) 0.038 −0.001 (0.062) (0.066) N2,128 1,607 1,156 2,219 1,426 1,084 Adjusted R20.331 0.347 0.346 0.391 0.395 0.379 Notes: The models use benchmark OLS estimation. Models 1, 2, and 3 report results for Brazil; Models 4, 5, and 6 for Mexico. The dependent variable uses the decision to cast votes entrusted by other players, thus measuring subjects’ levels of trustworthiness. A battery of individuallevel pretreatment controls—such as, age, income, employment, education, gender, and individual level of trust—are controlled for in all six estimations. ∗p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01 13 6.2 The Negative Effect of Social Media Exposure on Trust Unlike the case for trustworthiness, our model results show that social media exposure reduces overall trust. We begin by presenting conservative estimates of the effect of our experiment on trust, separating dissonant messages (out-group politician) and uncivil messages (negative content) using nonparametric graphical information. Then, we present statistical models and estimate the marginal effects of the treatments. Later, we discuss the factors that mediate the decline in trust. Figures 3and 4separate the results of our experiment by out-group/in-group politicians and by the negative/positive conditions. Separating the two treatment conditions, we find robust and statistically significant results when respondents are exposed to messages by outgroup politicians (dissonant messages). Results are inconclusive when considering only the negative tone of the social media post (uncivil discourse), as they are significant for Brazil but not for Mexico. The upper left plot in Figure 3provides visual confirmation of a statistically significant difference between respondents in the treatment and control groups exposed to messages from out-group politicians. The negative effect of the tweet is larger for respondents who entrusted more than four votes in the first round. Results are substantively similar but less robust in the case of Mexico (Figure 4). By contrast, exposing respondents to tweets from politicians they support yields small effects in Brazil and null results in Mexico. The lower left plots in Figures 3and 4show that, compared with the control group, negative political messages produce a modest decline in trust in Brazil but have no significant effect in Mexico. Given that we are not considering the joint effect of an out-group politician posting a negative tweet, the results reported in this section are very conservative. In Table 4, we present the results from benchmark ordinary least squares (OLS) analysis to capture the effect of the treatments on declines in trust in the second round of the game. Because changes in trust are heterogeneous, as shown in Figures 3and 4, we use an interactive linear model between the treatments and the decision to entrust votes in the first round of the 14 Figure 3 Changes in Trust among Treated and Untreated Respondents in Brazil Note: Local polynomial lines with confidence intervals. Plots compare changes in trust (votes delegated) between the first and second rounds of the game in Brazil. Four treatment conditions are compared with the control group: dissonant tweets from an out-group politicians, congruent tweets from an in-group politician, negative tweets (responsibility deflection), and positive tweets (cross-the-aisle). The joint effect of out-group and negative tone is not evaluated in this figure. game. Columns 1 to 3 present the results for Brazil of each different set of specifications, and columns 4 to 6 for Mexico. 16 The first models for each country (1 and 4) estimate the treatment effect of the content of the tweets. If we do not take into account the first-round decision to entrust votes, we cannot reject the null hypothesis that after respondents are exposed to the treatment their trust behavior changes. However, as in Figures 3,4, and 5, the effect of the treatment has the expected negative effect once the first-round decision is taken into account in the interactive models 2, 3, 5 and 6. 16The control group for all models consists of respondents who played the second round of the game without reading the social media message. 15 Figure 4 Changes in Trust among Treated and Untreated Respondents in Mexico Note: Local polynomial lines with confidence intervals. Plots compare changes in trust (votes delegated) between the first and second stages of the game in Mexico. Four treatment conditions are compared with the control group: dissonant tweets from an out-group politician, congruent tweets from an in-group politician, negative tweets (responsibility deflection), and positive tweets (cross-the-aisle). The joint effect of out-group and negative is not evaluated in this figure. In models 2 and 5, we estimate the effects of reading a message from an out-group politician. We consider the vote intention of the respondent, “if elections were to take place next week,” and the author of the tweet, to distinguish the effect of a message posted by an in-group or out-group politician. Exposure to a tweet from an out-group politician, independent of the content of the message, yields a statistically significant decrease in trust among respondents in Brazil. After treatment with a tweet from a misaligned politician, respondents decrease the number of votes they entrust to other players. The effect is larger for higher levels of trust in the first round, as reported in Figure 3. Although the results are substantially similar in Mexico, the 16 magnitude of the effects is smaller. Although the interaction term is not statistically distinct from zero, even for the Mexican case, reading a tweet from a misaligned politician has a negative effect on trust. Finally, models 3 and 6 evaluate hypotheses H1Aand H1B, with respondents playing the role of principals (voters). We estimate the effects of being exposed to a negative message from an out-group politician. Results in both countries show statistically significant declines in trust after respondents are exposed to uncivil/negative social media messages from political opponents. 17 Table 4 Regression Models: Treatment Effects of Framing and Endorsement on Trust Brazil Mexico (1) (2) (3) (4) (5) (6) Intercept 2.276∗∗∗ 2.037∗∗∗ 1.985∗∗∗ 2.514∗∗∗ 2.114∗∗∗ 2.322∗∗∗ (0.433) (0.481) (0.575) (0.444) (0.552) (0.612) Trust (Round 1) 0.460∗∗∗ 0.460∗∗∗ 0.460∗∗∗ 0.459∗∗∗ 0.462∗∗∗ 0.462∗∗∗ (0.031) (0.031) (0.031) (0.032) (0.032) (0.032) Framing: Negative −0.052 0.299 (0.194) (0.209) Framing: Positive −0.006 0.150 (0.195) (0.204) Out-group 0.101 0.300 (0.216) (0.264) In-group −0.315 0.266 (0.217) (0.261) Negative Out-group 0.083 0.676∗ (0.283) (0.366) Positive Out-group 0.110 0.015 (0.274) (0.324) Negative x Trust (Round 1) −0.032 −0.050 (0.043) (0.046) Positive x Trust (Round 1) −0.033 −0.039 (0.044) (0.045) Out-group x Trust (Round 1) −0.104∗∗ −0.081 (0.048) (0.057) In-group x Trust (Round 1) 0.022 −0.041 (0.050) (0.058) Negative Out-group x Trust (Round 1) −0.126∗∗ −0.164∗∗ (0.062) (0.079) Positive Out-group x Trust (Round 1) −0.081 −0.018 (0.062) (0.069) N2,092 1,583 1,140 2,216 1,425 1,083 Adjusted R20.232 0.234 0.218 0.200 0.196 0.202 Notes: The models use benchmark OLS estimation. Models 1, 2, and 3 report results for Brazil; Models 4, 5, and 6 for Mexico. The dependent variable uses the number of votes subjects (principals) entrusted in round 2 to another player to be doubled and cast for the principal’s candidate. A battery of individual-level pretreatment controls—such as, age, income, employment, education, gender, and individual level of trust—are controlled for in all six estimations. ∗p<0.1; ∗∗p<0.05; ∗∗∗ p<0.01 Results are fully described in Figure 5, with marginal effects for two of our treatment conditions from models 3 and 6. Results describe the marginal change in the number of votes[0,10] entrusted in the second round as a function of trust in the first round. Figure 5presents the 18 effects of reading a tweet from a misaligned politician (models 2 and 4) and Figure 6separates the out-group treatment according to the positive and negative framing (models 3 and 5). The figures provide a clear visualization of how out-group messaging, in particular with a negative tone, has a detrimental effect on interpersonal trust. For both cases, we see that reading a negative dissonant message reduces by almost 10% the votes delegated to other players between the first and second stages of the trust game—and marginal effects are statistically different from zero on respondents who in the early stage of the game exhibited higher levels of trust. The effect is substantively significant and, more importantly, describes a low-dosage treatment (one tweet) compared with the large number of tweets that users are exposed to on a daily basis. Figure 5 Marginal Effects of Cognitive Dissonance on Trust a) Brazil b) Mexico Figure 6 Marginal Effects of Negative Treatment from a Misaligned Politician a) Brazil b) Mexico 19 7 Mechanisms: The Role of Attention and Engagement While results from the previous sections confirm the hypothesized effect of social media frames on trust, they provide limited information about the mechanisms that underlie our results or about the differences observed between Brazil and Mexico. Our survey, however, included validation checks to evaluate whether respondents properly interpreted the partisan leaning of the social media frames and, more importantly, questions about respondents’ engagement with the partisan treatments. In this section, we analyze these results in greater detail, introducing a double-identification strategy that isolates the effect of attention to social media on declines in trust. Consider the effect of the treatment among respondents who engaged with the political tweets (by retweeting, liking, or replying) before answering our trust question (treatment group), compared with those in the control group who engaged with the tweet after answering the trust question. Given that the treatment consists exclusively of manipulating whether respondents play the trust game before or after reading the social media messages, our double-identification assumption only needs to assume that respondents assigned to the control group would have engaged with the tweet in the same way if they had been in the treatment group and not answered the trust question before engaging. We believe that this is a reasonable assumption, one that allows us to identify the heterogeneity of the treatment effects conditional on behavioral reactions to the social media message. Throughout this section, we repeat the same double identification strategy (engaged treatment/engaged control, ignore treatment/ignore control) to isolate the mechanisms that explain a decline in trust. Consider Figure 7which, as in the previous section, plots the trust decision in the second round (vertical axis) against the decision in the first round (horizontal axis). In Figure 7, the left plot compares the effect of the treated-engaged group (like, retweet, reply) against the control-engaged group. Meanwhile, the right plot describes the treatment/ignore group against the control/ignore group. Notable is the significant decline in trust among respondents who like, retweet, or reply to a tweet in the treatment group 20 Banks, Antoine, David Karol, Ernesto Calvo and Shibley Telhami. 2020. “#Polarized Feeds: Two experiments on polarization, framing, and social media.” International Journal of Press/Politics . Banks, Antoine J. 2014. “The public’s anger: White racial attitudes and opinions toward health care reform.” Political Behavior 36(3):493–514. Battigalli, Pierpaolo and Martin Dufwenberg. 2007. “Guilt in games.” American Economic Review 97(2):170–176. Berg, Joyce, John Dickhaut and Kevin McCabe. 1995. “Trust, reciprocity, and social history.” Games and economic behavior 10(1):122–142. Berinsky, Adam J., Michele F. Margolis and Michael W. Sances. 2014. “Separating the Shirkers from the Workers? Making Sure Respondents Pay Attention on Self-Administered Surveys.” American Journal of Political Science 58(3):739–753. URL: http://dx.doi.org/10.1111/ajps.12081 Bjørnskov, Christian and Pierre-Guillaume Méon. 2015. “The Productivity of Trust.” World Development 70:317 – 331. URL: http://www.sciencedirect.com/science/article/pii/S0305750X15000169 Bloom, N., C. Genakos, R. Sadun and J Van Reenen. 2012. “The Organization of Firms across Countries.” Quarterly Journal of Economics 124(4):1663—-1705. Boczkowski, Pablo J, Eugenia Mitchelstein and Mora Matassi. 2018. ““News comes across when I’m in a moment of leisure”: Understanding the practices of incidental news consumption on social media.” New Media and Society 20(10):3523–3539. Camerer, Colin F and George Loewenstein. 2003. “Behavioral economics: Past, present, future.”. Coppock, Alexander. 2019. “ri2: Randomization Inference for Randomized Experiments.” R package version 0.1 2. 27 Corbacho, Ana, Daniel W Gingerich, Virginia Oliveros and Mauricio Ruiz-Vega. 2016. “Corruption as a self-fulfilling prophecy: evidence from a survey experiment in Costa Rica.” American Journal of Political Science 60(4):1077–1092. Cox, James C. 2004. “How to identify trust and reciprocity.” Games and economic behavior 46(2):260–281. Croson, Rachel and Nancy Buchan. 1999. “Gender and culture: International experimental evidence from trust games.” American Economic Review 89(2):386–391. Entman, Robert M. 1993. “Framing: Toward clarification of a fractured paradigm.” Journal of communication 43(4):51–58. Evans, Geoffrey and Robert Andersen. 2006. “The political conditioning of economic perceptions.” The Journal of Politics 68(1):194–207. Fehr, Ernst and Simon Gächter. 2000. “Fairness and retaliation: The economics of reciprocity.” Journal of economic perspectives 14(3):159–181. Fletcher, Richard and Rasmus Kleis Nielsen. 2018. “Are people incidentally exposed to news on social media? A comparative analysis.” New media & society 20(7):2450–2468. Gambetta, Diego. 1988. “Trust: Making and breaking cooperative relations.”. Gerber, Alan S and Donald P Green. 2012. Field experiments: Design, analysis, and interpretation. WW Norton. Green, Donald P, Bradley Palmquist and Eric Schickler. 2004. Partisan hearts and minds: Political parties and the social identities of voters. Yale University Press. Guiso, Luigi, Paola Sapienza and Luigi Zingales. 2004. “The role of social capital in financial development.” American economic review 94(3):526–556. Hardin, Russell. 2002. Trust and trustworthiness. Russell Sage Foundation. 28 Iyengar, Shanto. 1990. “Framing responsibility for political issues: The case of poverty.” Political behavior 12(1):19–40. Iyengar, Shanto, Gaurav Sood and Yphtach Lelkes. 2012. “Affect, not ideologya social identity perspective on polarization.” Public opinion quarterly 76(3):405–431. Iyengar, Shanto and Sean J Westwood. 2015. “Fear and loathing across party lines: New evidence on group polarization.” American Journal of Political Science 59(3):690–707. Jacobsen, Dag Ingvar. 1999. “Trust in Political-Administrative Relations: The Case of Local Authorities in Norway and Tanzania.” World Development 27(5):839 – 853. URL: http://www.sciencedirect.com/science/article/pii/S0305750X99000327 Kahneman, Daniel. 2011. Thinking, fast and slow. Macmillan. Keefer, Phil, Ana María Rojas M, Carlos Scartascini and Joanna Valle L. 2020. Trust to Advance Inclusive Growth. In Inclusion in Times of Covid-19, ed. Victoria Nuguer and Andrew Powell. Inter-American Development Bank pp. 41–52. Keefer, Philip, Carlos Scartascini and Razvan Vlaicu. 2018. “Shortchanging the Future: The Short-Term Bias of Politics.” Better Spending for Better Lives. How Latin America and the Caribbean Can Do More with Less. Development in the Americas report. Washington, DC, United States: Inter-American Development Bank . Knack, Stephen and Philip Keefer. 1997. “Does social capital have an economic payoff? A cross-country investigation.” The Quarterly journal of economics 112(4):1251–1288. Levi, Margaret and Laura Stoker. 2000. “Political trust and trustworthiness.” Annual review of political science 3(1):475–507. Malhotra, Neil. 2008. “Completion time and response order effects in web surveys.” Public opinion quarterly 72(5):914–934. Mason, Lilliana. 2016. “A cross-cutting calm: How social sorting drives affective polarization.” Public Opinion Quarterly 80(S1):351–377. 29 Mazar, Nina and Dan Ariely. 2006. “Dishonesty in everyday life and its policy implications.” Journal of public policy & Marketing 25(1):117–126. Murtin, Fabrice, Lara Fleischer, Vincent Siegerink, Arnstein Aassve, Yann Algan, Romina Boarini, Santiago González, Zsuzsanna Lonti, Gianluca Grimalda, Rafael Hortala Vallve et al. 2018. “Trust and its determinants: Evidence from the Trustlab experiment.” OECD Statistics Working Papers 2018(2):0_1–74. Scartascini, Carlos and Joanna Valle L. 2020. Whom do we trust? The role of inequality and perceptions. In The Inequality Crisis: Latin America and the Caribbean at the Crossroads. Inter-American Development Bank pp. 329–351. Shoemaker, Pamela J and Stephen D Reese. 2013. Mediating the message in the 21st century : a media Sociology perspective. Third edit ed. New York: Routledge/Taylor & Francis Group. Slothuus, Rune and Claes H De Vreese. 2010. “Political parties, motivated reasoning, and issue framing effects.” The Journal of Politics 72(3):630–645. Smith, Adam. 1937. “The wealth of nations [1776].”. Tandoc, Edson C. 2014. “Journalism is twerking? How web analytics is changing the process of gatekeeping.” New media & society 16(4):559–575. Tucker, Joshua A, Andrew Guess, Pablo Barberá, Cristian Vaccari, Alexandra Siegel, Sergey Sanovich, Denis Stukal and Brendan Nyhan. 2018. “Social media, political polarization, and political disinformation: A review of the scientific literature.” Political Polarization, and Political Disinformation: A Review of the Scientific Literature (March 19, 2018) . Wilson, Rick K. 2017. Trust Experiments, Trust Games, and Surveys. In The Oxford Handbook of Social and Political Trust, ed. Eric M. Uslaner. Oxford University Press. Wise, Steven L and Xiaojing Kong. 2005. “Response time effort: A new measure of examinee motivation in computer-based tests.” Applied Measurement in Education 18(2):163–183. 30 Witmer, Hope and Peter Håkansson. 2015. “Social media and trust: A systematic literature review.” Journal of Business and Economics; 3 6. Zak, Paul J and Stephen Knack. 2001. “Trust and growth.” The economic journal 111(470):295– 321. Zaller, John. 1992. The nature and origins of mass opinion. Cambridge university press. 31 Trustful Voters, Trustworthy Politicians: A Survey Experiment on the Influence of Social Media in Politics Supplemental Information File Contents 1 Trustful Voters and Trustworthy Politicians 1 2 Trust and Trustworthiness 4 3 Trust Game: Nuts and Bolts 6 4 Survey Experiment and Hypotheses: Trust and Trustworthiness 8 5 Descriptive Evidence: Trustworthiness and Trust 10 6 Experimental Results 12 6.1 The Null Effect of Social Media Exposure on Trustworthiness .......... 12 6.2 The Negative Effect of Social Media Exposure on Trust ............. 14 7 Mechanisms: The Role of Attention and Engagement 20 8 Concluding Remarks 23 A Appendix A: Survey Flow and Treatment Assignment 34 B Appendix B: Treatment Intervention 42 C Appendix C: Theoretical Considerations of the Electoral Trust Game 46 C.1 Framing, Belief-Based Guilt, and First-Order Belief ................ 47 D Appendix D: Socio-Demographics and Attitudes Across the Samples 50 32 E Appendix E: Regression Models by Engagement with Treatment 54 F Appendix F: Human Subjects 57 33 A Appendix A: Survey Flow and Treatment Assignment In our survey, trust and trustworthiness are repeated independent behavioral responses to the potential for realizing gains by entrusting one’s votes to others or of incurring losses by casting other respondents’ votes. There are no expected future interactions and no gains in reputation. As in Cox (2004), we isolate trustworthiness from other preferences such as reciprocity or altruism, given that respondents have no information about the individual who entrusted votes to them or about the individual to whom votes will be entrusted.21 Game Sequence and Trust/Trustworthiness interventions To capture the role of social media on changes in trust and trustworthiness, we embed a survey experiment in a political trust game and expose respondents to contextually appropriate tweets from government or opposition political figures in Mexico and Brazil. The opening question of the survey invites respondents to select one of two fictional cartoon candidates and informs them that they will be able to collect votes for their candidate throughout the survey. Once all respondents answer the survey, those who supported the candidate who wins most votes are entered into a raffle for two new iPads. Respondents are informed that the number of their entries in the raffle will be equal to the number of votes they personally contribute to their candidate. Therefore, collecting as many votes as possible is incentive compatible: by making sure their candidate wins they become eligible to participate in the raffle, and by collecting more votes for their candidate they increase their chances of winning an iPad. At the time of the survey, the local price of an iPad was approximately 1.5 times times the median monthly salary in Brazil and half the median salary in Mexico.22 When answering questions as the agent (politician), respondents are asked to cast or dis21To avoid deception, our survey experiment institutes a universal respondent who carries out all requests received from respondents. Therefore, all votes entrusted by respondents were doubled and counted toward the respective candidates’ total tally. As in Cox (2004), we effectively create a triad where respondents’ decisions to cast or entrust votes are independent from one another. 22A total of four iPads were distributed in the raffles in Mexico and Brazil, making the odds/price ratio very attractive. 34 card votes entrusted to them (trustworthiness). When answering in the role of the principal (voters), respondents decide how many votes to cast directly and how many to entrust. In the first round, all voters first play the role of the agent (trustworthiness). As in our theory, we are interested in setting participants’ guilt sensitivity parameter, θ, before they entrust votes to others. All votes cast by the agents count toward the candidate favored by the trustee. After playing as agents, respondents decide how many votes to cast directly or to entrust to others. All votes cast (single votes) or entrusted (double votes) are accepted by our “universal” respondent.23 Rounds We capture the effect of social media on trustworthiness and on trust using a trust game that unfolds in three rounds. In the game, administered in Brazil and Mexico, respondents are told that they may cast their votes directly or entrust them to other respondents for casting (measuring trust) and then to cast any votes entrusted to them by other respondents (measuring trustworthiness). Two-thirds of the respondents in each country constitute a treatment group that is exposed to one of four randomly selected, edited tweets from high-profile politicians in the respondents’ country. We vary the framing of the tweet (positive or negative tone) and whether its author is from an in-group or an out-group. The intervention occurs before respondents in the treatment group play the second round of the trust game. The control group plays the second round without reading any tweets, thus providing a straightforward comparison group. Figure 9illustrates the basic design of the game and the embedded experiment. 23The rounds of casting and entrusting are repeated three times, with some respondents assigned to social media treatments and others to the control group. In this paper, we focus on the first treatment sets, which compare the baseline stage (first round) to the first treatment stage (second round). Comparing the first and second rounds of our two identical experiments in Brazil and Mexico minimizes cross-effects from survey questions that could account for changes in levels of trust or trustworthiness. Separate analyses model the effect of frame elements on trust and trustworthiness in a third round. 35 Figure 9 Survey Diagram 36 Table 5 Treatment Conditions in Brazil Positive Tweet Negative Tweet Eduardo Bolsonaro The world is currently living through an unprecedented crisis. Countries all over the world are rallying to fight the coronavirus. It is the responsibility of President @jairbolsonaro to coordinate our response. He must act together with Congress, business leaders, and civil society. This is what we expect in such critical times. The world is currently living through an unprecedented crisis. Countries all over the world are rallying to fight the coronavirus. But we have seen such viruses before, and they did not lead to all this hysteria. The fault lies with the President. Don’t panic. Switch off the pandemic of misinformation from the media. Fernando Haddad The world is currently living through an unprecedented crisis. Countries all over the world are rallying to fight the coronavirus. It is the responsibility of President @jairbolsonaro to coordinate our response. He must act together with Congress, business leaders, and civil society. This is what we expect in such critical times. The world is currently living through an unprecedented crisis. Countries all over the world are rallying to fight the coronavirus. But President @jairbolsonaro has been slow to respond. He is more concerned about attacking his opponents and taking part in protests that put the health of the Brazilian people at risk. Table 6 Treatment Conditions in Mexico Positive Tweet Negative Tweet Felipe Calderón When confronting the crisis of the coronavirus, we must protect the elderly so they can stop working today, while preserving their labor rights. All my support to @lopezobrador_ in his fight against the coronavirus. Eleven years ago Mexico suffered from the emergence of a new virus, unknown and deadly, without any information about its lethality. We protected Mexico’s citizens. Mexico is looks forward; it doesn’t bury its head in the sand! When will @lopezobrador_ react? Marti Batres When confronting the crisis of the coronavirus, we must protect the elderly so they can stop working today, while preserving their labor rights. All my support to @lopezobrador_ in his fight against the coronavirus. Eleven years ago Mexico suffered from the emergence of a new virus, unknown and deadly, without any information about its lethality. @FelipeCalderon did not protect Mexico’s citizens. Mexico looks forward; it doesn’t bury its head in the sand! That’s why I support @Lopezobrador_ 43 Figure 12 Tweets for the Treatment Conditions in Brazil a) Eduardo Bolsonaro x Positive Tweet (T1) b) Eduardo Bolsonaro x Negative Tweet (T2) a) Fernando Haddad x Positive Tweet (T3) b) Fernando Haddad x Negative Tweet (T4) 44 Figure 13 Tweets for the Treatment Conditions in Mexico a) Felipe Calderón x Positive Tweet (T1) b) Felipe Calderón x Negative Tweet (T2) a) Marti Batres x Positive Tweet (T3) b) Marti Batres x Negative Tweet (T4) 45 C Appendix C: Theoretical Considerations of the Electoral Trust Game In the political trust game that we embedded in our survey experiment, each respondent selects one of two fictional candidates. Respondents are informed that they must collect votes for their candidate of choice throughout the survey. At the end of the survey, those who chose the candidate who won the most votes are allowed to enter a raffle for two new iPads. Respondents are also informed that their number of raffle entries will be equal to the number of votes they contributed to their candidate. Therefore, collecting as many votes as possible is incentive-compatible: it increases the chances that their candidate will win and it increases their chances of winning an iPad. In the survey, respondents act the parts of a politician (agent) and a voter (principal). When answering questions as the agent (politician), respondents are asked to cast or discard votes entrusted to them (trustworthiness). When answering in the role of the principal (voters), respondents must decide how many votes to cast directly (one vote cast equals one raffle entry) and how many to entrust to another player (one entrusted vote that is eventually cast equals two votes and two raffle entries). However, they are warned that the agent (politician) may discard those votes; these are the same choices they themselves have when playing the other role. To ensure that there is no deception, our team serves as a universal player that honors all votes, those cast directly as well as those entrusted to another player. Following Mazar and Ariely (2006), and in order to approximate better the relationship between representatives and voters, the role of the agent is reinforced by an initial pledge: “If other players delegate (entrust) their votes to me, I agree to follow their preferences and to use them to support the candidate of his or her choosing.” After reading the initial statement, they are given five votes in support of their candidate of choice. The pledge is not binding. Representatives are required to read it but need not promise to comply with it; nor are there any sanctions for defaulting. 46 For the voter, placing her trust in another player (the agent) offers the possibility of doubling the votes her candidate receives (as well as the raffle tickets that she herself earns). However, the other player may decide not to cast those votes, something that is made clear at the outset. For the agent, being trustworthy is the act of casting delegated votes, even if those votes serve the candidate whom the respondent does not support (thereby lowering the likelihood of winning the raffle). As we will show, there are good theoretical reasons for trust and trustworthiness to be weakly correlated, as they express different types of cognitive beliefs about oneself (belief-based guilt) and about the other players (first-order beliefs). C.1 Framing, Belief-Based Guilt, and First-Order Belief Why would respondents cast delegated votes if doing so reduces their chances of winning the election and the raffle? Why would respondents delegate votes onto others if there is no reassurance that these votes will be cast? Given that there is no accountability in this survey, or in many democratic polities, one of the potential explanations for why agents fulfill the mandate entrusted to them is guilt aversion, a concept to which we now turn. Understanding guilt aversion also provides clear mechanisms to model the effect of social media frames on trust and trustworthiness. Trustworthiness and Guilt Aversion Battigalli and Dufwenberg (2007) coin the term guilt aversion to describe the psychological cost of letting other people down: “Player i’s guilt may depend on how much he lets jdown. Player i’s guilt may also depend on how much jbelieves ibelieves he lets jdown.”(Battigalli and Dufwenberg,2007, pg. 170). In the absence of information about player j,simple guilt emerges, where the decision to cast the entrusted votes is a function of the perceived rewards, mi, of defaulting on the request made by others, and the general guilt sensitivity, θiGij, when 47 no other information exists about the principal, j. Pr(TWYij) = φ(mi−θiGij)(1) In equation 1, the probability that we will be trustworthy is affected by the subjective value of the reward miand by the guilt sensitivity θiof agent ifor a generic principal j. We are agnostic about the social, political, or psychological origins of “guilt” and consider the guilt parameter, θi, as a placeholder for a simple aversion to defaulting on a mandate. Therefore, simple guilt describes the individual’s propensity to act on a generic request. This θiparameter is sensitive to a number of exogenous shocks. Framing effects are one possible mechanism that modulates the guilt sensitivity parameter, θi. For example, consider the experiment proposed by Ariely and Jones (2012), wherein respondents are asked to “read a pledge” before being given the chance to cheat. We may think of this pledge as a heuristic device that increases the relative value of θi, the cost of “letting other people down”.25 We expect negative messages from out-group politicians to increase negative feelings toward others (Mason,2016;Banks,2014), thereby reducing the value of θi (and reducing trustworthy responses). This expectation follows from the literature on generic or procedural frames, where the way in which a problem is presented alters the perceived legitimacy of an actor (Entman,1993) or event (Iyengar,1990). Our approach to trustworthiness differs from the classification proposed by Ashraf, Bohnet and Piankov (2006), who distinguish between “unconditional kindness,’, “expectations of reciprocity,” and “[instrumental] reciprocity.” In our case, guilt is the result of defaulting on a request from another respondent. There is no “kindness” in casting entrusted votes; there is no reciprocity expected or information collected about the individual who entrusts votes to be cast; and, finally, there are no instrumental benefits to be gained from being trustworthy. 25In our experiment, individuals are offered a “pledge” and the opportunity to click on the option, “I read the pledge.” As in Ariely and Jones (2012), respondents who selected this option were considerably more likely to cast the entrusted votes, even though the question only asked respondents to read rather than sign the pledge. Further, even if respondents believed they were signing rather than reading the pledge, there were no sanctions for defaulting on it. Therefore, the only cost of defaulting accommodates Battigalli and Dufwenberg (2007)’s definition of simple guilt. 48 Trust and First-Order Belief Unlike belief-based guilt, where we pay the cost or reap the benefit of our decision instantaneously, trust is a cognitive belief about the future behavior of others. When we entrust our votes to others, we do not know if they will comply with our request. If we were to find out that the principal failed to cast our votes we would feel betrayed rather than guilty. Our decision to trust another person depends on how we evaluate the behavior of other respondents, which may or may not be related to our own guilt sensitivity. We expect others to be less trustful as potential gains from deception increase, m∗ j. We also expect others to be less likely to fulfill their promises if they have been remorseless or dishonest in the past. Trust, therefore, is a cognitive belief about other people’s behavior, where the subjective gains from m∗ jand the subjective losses from guilt θ∗ jremain unobserved. Given that agent jdecides to cast entrusted votes following Equation 1, an action that is unobserved by principal i, the share of delegated votes depends on our belief that m∗ j−θ∗ jGij > 0, a belief that is unrelated to and not informed by our own guilt aversion parameter θi. Notice that not even the likelihood of betrayal depends on θi−θ∗ j<0, given that others defaulting on their promise to be trustworthy is unrelated to how trustworthy we are. We may feel no remorse when defaulting on the mandate we received; at the same time, we may still be outraged by the failure of others to do so. Therefore, as shown in equation 2, our decision to entrust others or to cast votes ourselves depends on unobserved values of how attractive to the other player is the unobserved prize, m∗ j, and how costly the unobserved guilt, θ∗ j. Tij =αVi+ (1 −α)Viφ(m∗ j−θ∗ jGij)(2) If we assume an empathetic respondent who will do for others what she expects others to do for her, the golden rule, trustworthiness and trust would show a weak positive correlation, cor(θi, θ∗ j)>0. Notice that θ∗ jremains unobserved by iand does not reflect actual information about the principal. Therefore, the value of m∗ jand the simple guilt parameter, θ∗ j, represent expectations of the respondent and not actual behavioral traits of another player.26 26In our experiment, the baseline round offers respondents the possibility of duplicating a fraction of the votes 49 D Appendix D: Socio-Demographics and Attitudes Across the Samples In this section, we present a set of individual level socio-demographic, political, and attitudinal characteristics distributed by survey group (treatment or control). As the reader will see, there are no significant differences across the treatment groups in our sample. Since most of these variables are nominal, the values are not susceptible to direct interpretation. Tables 7and 8present the sample mean and standard deviation for relevant covariates, organized by group. entrusted to others. We then treat two-thirds of the respondents to social media frames with the remaining third as controls. Given that we provide no information about the principal, j, and that reward miis held constant, we measure the changes in the expected sensitivity of the guilt parameter only in the second round, θ∗ j,R1−θ∗ j,R2. 50 Table 7 Sample Demographics by Treatment Assignment in Brazil Variable Quantity Treatment Control Age Mean 3.03 3.18 Standard Error 1.53 1.54 Education Mean 4.45 4.54 Standard Error 1.26 1.27 Gender Mean 1.48 1.49 Standard Error 0.50 0.50 Ideological Placement Mean 6.52 6.31 Standard Error 3.30 3.33 Receive Income Assistance Mean 2.29 2.30 Standard Error 0.89 0.89 Relative Income Mean 5.30 5.43 Standard Error 2.10 2.05 Government Voters Mean 1.37 1.35 Standard Error 0.48 0.48 Opposition Voters Mean 1.73 1.70 Standard Error 0.44 0.46 Trust Bolsonaro Mean 4.72 4.98 Standard Error 3.54 3.60 Trust Lula Mean 5.00 5.16 Standard Error 2.50 2.59 Employment Mean 1.37 1.35 Standard Error 0.48 0.48 Total Cases Total Number of Cases 1594 768 51 Table 8 Sample Demographics by Treatment Assignment in Mexico Quantity Treatment Control Age Mean 3.62 3.70 Standard Error 1.79 1.81 Education Mean 2.27 2.20 Standard Error 1.29 1.29 Gender Mean 1.54 1.51 Standard Error 0.50 0.50 Ideological Placement Mean 5.24 5.38 Standard Error 3.24 3.20 Receive Income Assistance Mean 1.17 1.18 Standard Error 0.42 0.41 Relative Income Mean 5.35 5.48 Standard Error 1.93 1.82 AMLO Trust Mean 5.91 5.93 Standard Error 3.09 3.03 Government Voters Mean 1.29 1.31 Standard Error 0.46 0.46 Opposition Voters Mean 1.15 1.15 Standard Error 0.36 0.36 Employment Mean 1.40 1.42 Standard Error 0.49 0.49 Total Cases Total Number of Cases 1533.00 813.00 52 Confidentiality The PI and team receive a de-identified respondent ID number. No private identifying information was stored in the servers of the PI or any other member of the team. Netquest will provide the emails of the winners of the raffle to deliver the awards. At no point during the recruitment, consent, or research procedures, will Vanderbilt, Qualtrics, or Netquest provide us with personally identifying information for participants. Thus for the full survey we will be able to adequately ensure the anonymity of all survey respondents. Consent Process The informed consent procedure provides participants explicit consent to proceed and informs of their right to skip questions and to discontinue the survey. The online consent was granted by IRB by waiving written consent, given the following criteria: 1. Our research involves no more than minimal risk to the subjects. As we have stated, the only potential risk is minimal discomfort due to the nature of the questions asked, and we mitigate this discomfort by allowing participants to skip questions. 2. The waiver will not adversely affect the rights and welfare of the subjects. All subjects in these pretest and survey will be fully informed about their rights as participants and the nature of the study, and they will have access to the consent form online to save and print for their records. 3. This research could not practicably be carried out without the waiver because it is entirely performed online. Therefore, none of the co-PIs could gather written consent forms for all participants. 4. Whenever appropriate, the subjects will be provided with additional information after participation. Participants will have access to contact information for both co-PIs and IRB, allowing them to reach out in case they have any further questions. Research Outside of the United States Both PIs to this project have extensive experience doing field research and working in Latin America. Both PIs are native Spanish and Portuguese speakers, which allows for our 59 communication with all stakeholders involved in this study. We have reviewed existing regulations in Argentina, Brazil, and Mexico on data protection and privacy. We are complying with this regulation by providing participants with all relevant information about the project and survey, by protecting data and participants’ privacy through ensuring anonymity of the observations and data collected, and by making sure only co-PIs have access to the data collected. Given that the data collection process will be anonymous and no sensitive information is required as part of this process, we believe there is no additional risk participants in Argentina, Brazil, or Mexico will be exposed to, nor that they will be placed at risk of criminal or civil liability. IRB Approval letter The official approval letter is attached to this application. During the review process we include an anonymous version of the letter. It will be replaced by the formal approval letter after review is completed. 60 References Aghion, Philippe, Yann Algan, Pierre Cahuc and Andrei Shleifer. 2010. “Regulation and distrust.” The Quarterly Journal of Economics 125(3):1015–1049. Algan, Yann and Pierre Cahuc. 2010. “Inherited trust and growth.” American Economic Review 100(5):2060–92. Algan, Yann and Pierre Cahuc. 2014. “Trust, Growth, and Well-Being: New Evidence and Policy Implications.” Handbook of Economic Growth 2:49–120. Algan, Yann, S Guriev, E Papaioannou and E Passari. 2017. “The European Trust Crisis and the Rise of Populism.” Brookings Papers on Economic Activity 2:309–400. Arceneaux, Kevin. 2008. “Can partisan cues diminish democratic accountability?” Political Behavior 30(2):139–160. Ariely, Dan and Simon Jones. 2012. The (honest) truth about dishonesty. Harper Collins Publishers New York, NY. Arrow, Kenneth J. 1974. The limits of organization. WW Norton & Company. Aruguete, Natalia and Ernesto Calvo. 2018. “Time to #Protest: Selective Exposure, Cascading Activation, and Framing in Social Media.” Journal of Communication 68(3):480–502. URL: http://dx.doi.org/10.1093/joc/jqy007 Ashraf, Nava, Iris Bohnet and Nikita Piankov. 2006. “Decomposing trust and trustworthiness.” Experimental economics 9(3):193–208. Bail, Christopher A, Lisa P Argyle, Taylor W Brown, John P Bumpus, Haohan Chen, MB Fallin Hunzaker, Jaemin Lee, Marcus Mann, Friedolin Merhout and Alexander Volfovsky. 2018. “Exposure to opposing views on social media can increase political polarization.” Proceedings of the National Academy of Sciences 115(37):9216–9221. 61 Banks, Antoine, David Karol, Ernesto Calvo and Shibley Telhami. 2020. “#Polarized Feeds: Two experiments on polarization, framing, and social media.” International Journal of Press/Politics . Banks, Antoine J. 2014. “The public’s anger: White racial attitudes and opinions toward health care reform.” Political Behavior 36(3):493–514. Battigalli, Pierpaolo and Martin Dufwenberg. 2007. “Guilt in games.” American Economic Review 97(2):170–176. Berg, Joyce, John Dickhaut and Kevin McCabe. 1995. “Trust, reciprocity, and social history.” Games and economic behavior 10(1):122–142. Berinsky, Adam J., Michele F. Margolis and Michael W. Sances. 2014. “Separating the Shirkers from the Workers? Making Sure Respondents Pay Attention on Self-Administered Surveys.” American Journal of Political Science 58(3):739–753. URL: http://dx.doi.org/10.1111/ajps.12081 Bjørnskov, Christian and Pierre-Guillaume Méon. 2015. “The Productivity of Trust.” World Development 70:317 – 331. URL: http://www.sciencedirect.com/science/article/pii/S0305750X15000169 Bloom, N., C. Genakos, R. Sadun and J Van Reenen. 2012. “The Organization of Firms across Countries.” Quarterly Journal of Economics 124(4):1663—-1705. Boczkowski, Pablo J, Eugenia Mitchelstein and Mora Matassi. 2018. ““News comes across when I’m in a moment of leisure”: Understanding the practices of incidental news consumption on social media.” New Media and Society 20(10):3523–3539. Camerer, Colin F and George Loewenstein. 2003. “Behavioral economics: Past, present, future.”. Coppock, Alexander. 2019. “ri2: Randomization Inference for Randomized Experiments.” R package version 0.1 2. 62 Corbacho, Ana, Daniel W Gingerich, Virginia Oliveros and Mauricio Ruiz-Vega. 2016. “Corruption as a self-fulfilling prophecy: evidence from a survey experiment in Costa Rica.” American Journal of Political Science 60(4):1077–1092. Cox, James C. 2004. “How to identify trust and reciprocity.” Games and economic behavior 46(2):260–281. Croson, Rachel and Nancy Buchan. 1999. “Gender and culture: International experimental evidence from trust games.” American Economic Review 89(2):386–391. Entman, Robert M. 1993. “Framing: Toward clarification of a fractured paradigm.” Journal of communication 43(4):51–58. Evans, Geoffrey and Robert Andersen. 2006. “The political conditioning of economic perceptions.” The Journal of Politics 68(1):194–207. Fehr, Ernst and Simon Gächter. 2000. “Fairness and retaliation: The economics of reciprocity.” Journal of economic perspectives 14(3):159–181. Fletcher, Richard and Rasmus Kleis Nielsen. 2018. “Are people incidentally exposed to news on social media? A comparative analysis.” New media & society 20(7):2450–2468. Gambetta, Diego. 1988. “Trust: Making and breaking cooperative relations.”. Gerber, Alan S and Donald P Green. 2012. Field experiments: Design, analysis, and interpretation. WW Norton. Green, Donald P, Bradley Palmquist and Eric Schickler. 2004. Partisan hearts and minds: Political parties and the social identities of voters. Yale University Press. Guiso, Luigi, Paola Sapienza and Luigi Zingales. 2004. “The role of social capital in financial development.” American economic review 94(3):526–556. Hardin, Russell. 2002. Trust and trustworthiness. Russell Sage Foundation. 63 Iyengar, Shanto. 1990. “Framing responsibility for political issues: The case of poverty.” Political behavior 12(1):19–40. Iyengar, Shanto, Gaurav Sood and Yphtach Lelkes. 2012. “Affect, not ideologya social identity perspective on polarization.” Public opinion quarterly 76(3):405–431. Iyengar, Shanto and Sean J Westwood. 2015. “Fear and loathing across party lines: New evidence on group polarization.” American Journal of Political Science 59(3):690–707. Jacobsen, Dag Ingvar. 1999. “Trust in Political-Administrative Relations: The Case of Local Authorities in Norway and Tanzania.” World Development 27(5):839 – 853. URL: http://www.sciencedirect.com/science/article/pii/S0305750X99000327 Kahneman, Daniel. 2011. Thinking, fast and slow. Macmillan. Keefer, Phil, Ana María Rojas M, Carlos Scartascini and Joanna Valle L. 2020. Trust to Advance Inclusive Growth. In Inclusion in Times of Covid-19, ed. Victoria Nuguer and Andrew Powell. Inter-American Development Bank pp. 41–52. Keefer, Philip, Carlos Scartascini and Razvan Vlaicu. 2018. “Shortchanging the Future: The Short-Term Bias of Politics.” Better Spending for Better Lives. How Latin America and the Caribbean Can Do More with Less. Development in the Americas report. Washington, DC, United States: Inter-American Development Bank . Knack, Stephen and Philip Keefer. 1997. “Does social capital have an economic payoff? A cross-country investigation.” The Quarterly journal of economics 112(4):1251–1288. Levi, Margaret and Laura Stoker. 2000. “Political trust and trustworthiness.” Annual review of political science 3(1):475–507. Malhotra, Neil. 2008. “Completion time and response order effects in web surveys.” Public opinion quarterly 72(5):914–934. Mason, Lilliana. 2016. “A cross-cutting calm: How social sorting drives affective polarization.” Public Opinion Quarterly 80(S1):351–377. 64 Mazar, Nina and Dan Ariely. 2006. “Dishonesty in everyday life and its policy implications.” Journal of public policy & Marketing 25(1):117–126. Murtin, Fabrice, Lara Fleischer, Vincent Siegerink, Arnstein Aassve, Yann Algan, Romina Boarini, Santiago González, Zsuzsanna Lonti, Gianluca Grimalda, Rafael Hortala Vallve et al. 2018. “Trust and its determinants: Evidence from the Trustlab experiment.” OECD Statistics Working Papers 2018(2):0_1–74. Scartascini, Carlos and Joanna Valle L. 2020. Whom do we trust? The role of inequality and perceptions. In The Inequality Crisis: Latin America and the Caribbean at the Crossroads. Inter-American Development Bank pp. 329–351. Shoemaker, Pamela J and Stephen D Reese. 2013. Mediating the message in the 21st century : a media Sociology perspective. Third edit ed. New York: Routledge/Taylor & Francis Group. Slothuus, Rune and Claes H De Vreese. 2010. “Political parties, motivated reasoning, and issue framing effects.” The Journal of Politics 72(3):630–645. Smith, Adam. 1937. “The wealth of nations [1776].”. Tandoc, Edson C. 2014. “Journalism is twerking? How web analytics is changing the process of gatekeeping.” New media & society 16(4):559–575. Tucker, Joshua A, Andrew Guess, Pablo Barberá, Cristian Vaccari, Alexandra Siegel, Sergey Sanovich, Denis Stukal and Brendan Nyhan. 2018. “Social media, political polarization, and political disinformation: A review of the scientific literature.” Political Polarization, and Political Disinformation: A Review of the Scientific Literature (March 19, 2018) . Wilson, Rick K. 2017. Trust Experiments, Trust Games, and Surveys. In The Oxford Handbook of Social and Political Trust, ed. Eric M. Uslaner. Oxford University Press. Wise, Steven L and Xiaojing Kong. 2005. “Response time effort: A new measure of examinee motivation in computer-based tests.” Applied Measurement in Education 18(2):163–183. 65 Witmer, Hope and Peter Håkansson. 2015. “Social media and trust: A systematic literature review.” Journal of Business and Economics; 3 6. Zak, Paul J and Stephen Knack. 2001. “Trust and growth.” The economic journal 111(470):295– 321. Zaller, John. 1992. The nature and origins of mass opinion. Cambridge university press. 66