Corruption and political accountability in good and bad economic times

Barinas-Forero, Andrés; Scartascini, Carlos G. Working Paper Corruption and political accountability in good and bad economic times IDB Working Paper Series, No. IDB-WP-1690 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Barinas-Forero, Andrés; Scartascini, Carlos G. (2025) : Corruption and political accountability in good and bad economic times, IDB Working Paper Series, No. IDB-WP-1690, InterAmerican Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0013597 This Version is available at: https://hdl.handle.net/10419/324837 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/3.0/igo/ Corruption and Political A ccountability in Good and Bad Economic Times A ndrés Barinas-Forero Carlos Scartascini WORKING PAPER No IDB-WP-1690 InterA merican Development Bank Department of Research and Chief Economist July 2025 * InterA merican Development Bank Corruption and Political Accountability in Good and Bad Economic Times A ndrés Barinas-Forero* Carlos Scartascini* InterA merican Development Bank Department of Research and Chief Economist July 2025 Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Barinas-Forero, Andres. Corruption and political accountability in good and bad economic times / Andres Barinas-Forero, Carlos Scartascini. p. cm. — (IDB Working Paper Series ; 1690) Includes bibliographical references. 1. Political corruption-Mathematical models. 2. Public goods-Corrupt practices-Mathematical models. I. Scartascini, Carlos G., 19 71II. InterAmerican Development Bank. Department of Research and Chief Economist. III. Title. IV. Series. IDB-WP-1690 http://www.iadb.org Copyright © 2025 Inter-American Development Bank ("IDB"). This work is subject to a Creative Commons license CC BY 3.0 IGO (https://creativecommons.org/licenses/by/3.0/igo/legalcode). The terms and conditions indicated in the URL link must be met and the respective recognition must be granted to the IDB. Further to section 8 of the above license, any mediation relating to disputes arising under such license shall be conducted in accordance with the WIPO Mediation Rules. 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 United Nations Commission on International Trade Law (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 license. Note that the URL link includes terms and conditions that are an integral part of this license. The opinions expressed in this work 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. Abstract While the literature extensively explores the structural enablers of corruption and its adverse effects on economic performance, less is known about how the state of the economy influences corruption and political accountability. To address this gap, we develop a theoretical model in which politicians may divert resources from public goods and citizens can respond by punishing corruption. In our model, periods of positive economic conditions increase corruption while weakening accountability. We validate these predictions through a laboratory experiment, finding that corruption rates significantly rise when economic conditions are good. However, citizens’ willingness to punish corrupt politicians remains stable across the business cycle. Punishment decisions are driven by observed public good allocations; low allocations prompt significantly higher punishment rates than high allocations, even resulting in the punishment of honest politicians during bad economic times. Additionally, we assess the role of corruption expectations in shaping responses: citizens with prior beliefs that politicians are corrupt are less likely to punish than those who believe politicians are honest when public good provision is low. Accountability becomes more challenging when citizens struggle to clearly identify corruption, and citizens are more forgiving of corruption during good economic times and if they already mistrust politicians. These findings underscore the importance of robust transparency and accountability mechanisms in upholding governance standards, particularly in the face of economic fluctuations and public mistrust. JEL classifications: D72, D73, H41, C91 Keywords: Rent seeking, Economic Booms, Corruption, Punishment, Laboratory experiment, Downturns The authors thank participants at REPAL 2024, LACEA-LAMES 2024, MPSA 2025, the University of Virginia seminar, Martín Ardanaz, Hernán Bejarano, Mariana Blanco, Ernesto Calvo, Dan Gingerich, Phil Keefer, Andrea Lopez-Luzuriaga, Julia Seither, and an anonymous reviewer for their comments and suggestions. The research described in this article has Universidad del Rosario IRB Approval CS474 (March 15, 2024) and was preregistered at the OSF Registries ( www.osf.io/rmv92 , May 8, 2024). This research was financed by the Inter-American Development Bank (IDB). IDB management had no involvement in the study design, analysis, or interpretation of the data, in the writing of the report, or in the decision to submit the article for publication. The opinions expressed here 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. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. 1 Introduction Political corruption disrupts growth (Mauro,1995;Bardhan,1997), deepens poverty (Gupta et al.,1998), exacerbates inequality (Olken,2006), erodes trust between citizens and the state (Keefer and Scartascini,2022;Khlif and Amara,2019), and leads to significant economic inefficiencies (Dal Bó and Rossi,2007;Fisman and Svensson,2007). Economic and political cycles can amplify or mitigate these harmful effects (Saha and Sen,2023) and may also influence the likelihood of corruption itself. During an economic bonanza, increased resources may provide more opportunities for corrupt practices (Galbraith,1997;Leite and Weidmann,1999). Conversely, during times of economic hardship, corruption risks can intensify as the value of rents grows, and public officials may exploit the population’s increased vulnerability (Ivlevs and Hinks, 2015;Torsello,2010). Economic cycles also shape perceptions of corruption, as citizens may attribute downturns to political mismanagement or become more attuned to the issue during periods of hardship (Márquez Romo and Romero-Vidal,2023;Zechmeister and Zizumbo-Colunga,2013). For instance, data from a nationally representative survey conducted across 26 Latin American and Caribbean countries (LAPOP,2024), from 2016 to 2023, reveal a notable trend: respondents’ perceptions of political corruption are inversely related to their views on economic conditions over the past year (Figure 1.) 1 This correlation suggests that the state of the economy could affect suspicions of corruption or actual corrupt activities. If corruption beliefs are influenced by the cycle, there may be increased incentives for corruption, as political accountability decreases. In this paper, we examine how economic good and bad economic times impact the behavior of politicians when it comes to local corruption and citizens when it comes to political accountability. 2 To analyze these dynamics, we develop a theoretical model of political corruption and political accountability, where citizens pay taxes and can punish politicians, and politicians provide public goods (or pocket part of the revenues). We operationalize the economic cycle through an exogenous economic shock that modifies the cost of public 1 This pattern is consistent across all countries in the region and across support for the current president, as detailed in the Online Appendix (see Figures D1,D2 and D3). 2 In our model, politicians who act as “rent-seekers” rather than "office-seekers" are the ones who frame local corruption (see Persson and Tabellini,2002). Our concept of accountability diverges from the traditional understanding found in seminal political accountability literature (see Barro,1973;Przeworski et al.,1999). Instead, we characterize accountability as citizen’s willingness to punish politicians while they are in office (Klašnja et al.,2021). 1 goods construction, affecting politicians’ incentives, the quantity of public goods provided, and citizens’ subsequent choices to hold politicians accountable. Figure 1: Perception of the Number of Politicians Involved in Corruption across National Economic Situation Perception β = -.25*** p < .00001 3.2 3.4 3.6 3.8 4 Number of politicians involved in corruption (1 None - 5 All) Worse than before Same as before Better than before Economic situation of the country in last 12 months (1 Worse - 3 Better) Linear fit 95% CIs Notes: The figure depicts the scatterplot binned between the perception of the number of politicians involved in corruption and the perception of the country’s economic situation over the past 12 months. Scatter binned estimation for all respondents to the LAPOP question in 26 countries between 2016 and 2023. The dark line indicates the linear fit from a regression model with n= 79,787 . The gray area represents the 95% clustered confidence intervals at the country level. The analysis includes country and year fixed effects. The regression analysis is presented in the Online Appendix D1. Statistical significance levels: ∗∗∗ p < 0.01,∗∗ p < 0.05,∗p < 0.1 , n.s p > 0.1. We model economic conditions as changes in the cost of public goods for two main reasons. First, it allows us to introduce random shocks to citizens’ material well-being and observe how these fluctuations shape both corruption and political accountability. Since citizens derive utility from public goods (Persson and Tabellini,2002), increases in cost reduce provision and consequently lower citizens’ welfare, prompting them to potentially punish incumbents even in the absence of corruption (Benton,2005;Kayser and Peress, 2012;Lewis-Beck and Stegmaier,2000). 3 Second, in many developing countries, public finances usually employed in the construction of public goods are closely tied to volatile commodity revenues (Martínez,2023). As such, random fluctuations in production costs provide an externally valid analogue for real-world economic shocks that are independent 3 We deviate from other studies that employ increments in productivity as the main source of the positive economic shock (see Bai et al.,2017). 2 of incumbents’ decisions and actions.4 In our model, a local politician, a national authority, and a citizen interact strategically. In the first stage of the game, an economic shock occurs (the construction cost of the public goods is defined by nature), which is common knowledge for both the citizen and the politician. The representative citizen pays income taxes that are transferred to the politician. The politician receives the transfer and decides how much to invest in public goods. 5 After the politician’s allocation decision, the national authority may provide additional funds for public goods construction (an exogenous action). In the final stage, the citizen observes the actual allocation of public goods, considering the economic conditions, and decides whether to punish (or disapprove) the politician–—a choice that incurs a cost. In this model, citizens and politicians can exhibit different intrinsic motivations. Politicians may vary in their moral commitment or integrity (Martinelli,2022;Tanner et al.,2022), and citizens may differ in their motivation to hold politicians accountable (Adserà et al.,2003; Leight et al.,2020;Nannicini et al.,2013). The presence of a national authority in our model allows us to incorporate three common dimensions: i) in many countries, public goods provision is a shared responsibility between local and national authorities; 6 ii) corruption is often facilitated by imperfect information about politicians’ actions. 7 ;iii) it enables us to simulate an environment in which corruption is difficult to detect, thereby introducing real incentives for participants to engage in corrupt behavior. In our theoretical framework, we identify a signaling equilibrium shaped by two primary dynamics. First, politicians who derive greater utility from honesty are less inclined to 4 For instance, royalties from natural resource extraction serve as a major income source for municipalities in Colombia and one of the main sources of financing of public goods (Martínez,2023). Thus, the shift in construction costs we assume in the model for simplicity mimics an economic shock similar to those driven by fluctuations in commodity prices. Moreover, the relevance of economic shocks to commodity prices extends beyond public goods provision, as they have also been shown to influence criminal behavior (Dube and Vargas,2013) and even the ethical conduct of politicians (Asher and Novosad,2023). 5 By allowing the politician to decide how much to allocate, we concentrate on “assignment” problems, which are common approaches when thinking about corruption (see Banerjee et al.,2012). 6 For example, in Colombia, since the 1990s, local and national authorities have jointly managed essential public services, including drinking water, sanitation, health, and education (Alesina et al.,2005). In developing countries, it is also the case that international agencies and organizations provide local public goods that complement local provision. Our case could then be used to think about the role of foreign aid and corruption. 7 Details on who funds specific services and the amounts allocated are rarely common knowledge (Burman and Slemrod,2020;López-Laborda et al.,2023), creating uncertainty that weakens accountability; corruption typically occurs under conditions of secrecy (Hu and Oak,2023;Shleifer and Vishny,1993) 3 engage in corruption, while citizens who place a lower value on political accountability are less likely to incur the cost of disapproving of corrupt politicians. Second, following a positive economic shock that lowers the cost of public goods, politicians may be more prone to corruption, anticipating reduced citizen vigilance in holding them accountable. This expectation of lower disapproval increases the expected benefits of corrupt actions. Thus, our model suggests that economic booms should not only elevate local corruption but also weaken political accountability.8 Building on these theoretical insights, we designed and implemented a laboratory experiment to empirically test our hypotheses. In the experiment, participants were assigned to one of two roles: citizens or politicians. 9 Participants were matched using a perfect stranger matching protocol, informed of the game structure with emphasis on the importance of public goods, and repeated the same activity across multiple rounds. The design incorporated within-subject variation in public goods cost, set to either high cost or low cost, simulating exogenous variation in the economic times. This structure allows us to investigate the effect of economic booms on political behavior, providing empirical support for the theoretical framework outlined earlier. Detailed experimental procedures are presented in Section 3. Our findings indicate that corruption rates in high-cost environments (economic downturns) are approximately 51%. In contrast, in low-cost environments, such as those following good economic times, corruption rises to 58%, marking a 7-percentage point increase. Regarding political accountability, we observe that citizens’ willingness to disapprove of politicians is 16% in high-cost environments and slightly lower, at 15%, during economic booms. These results carry significant economic implications: good economic times are associated with roughly a 14% rise in local political corruption, underscoring the difficulty of maintaining political accountability during periods of economic expansion. Given the structure of the game, the final allocation of public goods also serves as a quality signal of the politician’s actions (Mani and Mukand,2007). Consequently, we examine how the observed public goods allocation affects the citizen’s willingness to disapprove of the politician (Ferejohn,1986). Our results indicate that citizens are willing to incur personal costs to punish corrupt politicians, with a disapproval rate of 36% for those 8 This idea is consistent with the theoretical results developed by Bhattacharyya and Hodler (2010), and with empirical results linking economic conditions and perceptions of corruption in the population (Márquez Romo and Romero-Vidal,2023;Zechmeister and Zizumbo-Colunga,2013). 9 The experiment used loaded framing, with participants explicitly informed of their role titles. Prior research indicates that framing effects are typically minimal in corruption games (Abbink and Hennig-Schmidt, 2006;Barr and Serra,2009;Alekseev, Charness and Gneezy,2017;Banerjee,2016). 4 with δηw being the monetary punishment received by the disapproval from the citizen. On the other hand, the citizen derives utility from the public good provision (Θ) and from her non-taxed income y(1 −τ) . For simplicity, the valuation of each unit of the public good is fixed at κ∈R+ . Additionally, note that the marginal benefit for each unit is equal to the fixed cost of disapproval, meaning that choosing to disapprove of the politician represents a loss of one public good unit.21 We characterize the citizen’s payoff as: Uc=y(1 −τ) + κΘ+δ((αc−Θ)λ−κ),(3) where λ∈[0,1] denotes the citizen’s posterior belief about the politician’s type once she observes the signal provided by the public goods allocation ( Θ ). It is worth mentioning that the citizen updates her beliefs in a Bayesian manner, 22 and they are consistent with her prior beliefs of corruption (β∈(0,1)). In general, citizens may perceive higher returns from protest when public good provision is scarce (Almer et al.,2017;Grossman,1991). Therefore, higher allocations of public goods act as a reserve utility for citizens. In our model, this reserve utility is represented by (αc−Θ) , where lower allocations of public goods increase the marginal benefit of protest. Additionally, citizens behave strategically. Therefore, the marginal benefits of non-approval are mediated by the posterior beliefs of corruption (λ). While the model suggests that politicians possess deeper insights into the public goods financing scheme, it raises the question of why this type of information asymmetry is deemed more critical than the asymmetry concerning economic factors that influence construction costs. Notably, the costs of public goods tend to align closely with those of private goods, and given that citizens regularly purchase these private goods, it is plausible to assume they can reasonably estimate public costs. In contrast, the intricacies of budgetary processes remain largely inaccessible to the general public, underscoring the rationale for focusing on budgetary rather than economic informational asymmetries. Figure 3summarizes the model’s timeline. The participants’ types (αc, ϕp) , as well as the 21 According to Ortiz et al. (2022), social discontent can lead to the devastation of infrastructure and wealth. Images of burning cars or buildings frequently appear in media coverage about protests in many nations. Consequently, it makes sense to express the cost of protests as the loss of one unit of public goods. 22For a specific allocation of public goods, the beliefs updating process follows: λ≡Pr (σ= 0|Θ(π) = Θm(π))=Pr (Θ(π) = Θm(π)|σ= 0)×Pr (σ= 0) Pr (Θ(π) = Θm(π))∀m={h, a, l} 11 costs for public goods π={π, π} are exogenously determined at the beginning of the game. Then, the politician makes the policy decision, the national authority randomly decides whether to provide resources to the municipality, and the public goods allocation is shown to the citizen. Finally, once the citizen observes the final allocation of public goods, she decides whether to approve or disapprove the politician, and payoffs are realized. Figure 3: Timeline for the Political Corruption and Accountability Model Nature Politician National Authority Citizen Participants Cost of Policy Resources and Approval types public goods choice public goods allocation choice ϕp,αcπhγ,Θδ Notes: Participant types are drawn from their corresponding distributions (ϕp∼U[0,b ϕ]) , (αc∼f[0, α] . The cost of public goods (π) follows a Bernoulli distribution, and π∈ {π, π} . Finally, the national authority decides to allocate additional resources (γ= 1) with a probability of µ∈(0,1). 2.3 Signaling Equilibrium Now, we define strategies, beliefs, and equilibrium for the extensive game of incomplete information played by the citizen and the politician. The information set available to the politician before making the policy choice is characterized by their own type and the public good cost. Thus, we define a strategy for politician as a mapping σ, σ:Ip→[0,1] where Ip≡(ϕp, π)∈[0,1] × {π, π}, which specifies the probability of adopting the policy decision h as a function of the information received by the politician at the beginning of the game. The citizen knows the public good cost, observes the allocation of public goods given that cost, and makes a decision given the intrinsic motivation for disapproval αc . Thus, we define a strategy for the citizen as a mapping ν, ν:{Θh(π),Θa(π),Θl(π)} × [0, α]→[0,1], that goes from the set of possible public goods allocations given a specific cost and the distribution of possible disapproval rates [0, α] to a range between [0,1] that denotes the probability of enforcing the punishment against the politician. Finally, we define a belief system for the citizen as a mapping λ, 12 λ:{Θh,Θa,Θl} × {0,1} → [0,1] that goes from the observed allocation of public goods Θ and prior beliefs β∈(0,1) of corruption to a range of updated beliefs of corruption in [0,1]. Definition 1. (Signaling Equilibrium) A signaling Bayesian equilibrium is given by (i) a strategy for the politician (σ) (ii) a strategy for the citizen (ν) (iii) a belief system for the citizen (λ) such that: the citizen strategy (ν) is optimal given her beliefs, the citizen beliefs are consistent with the politician’s strategy—meaning that λ is derived from prior beliefs (β) and the politician’s strategy using Bayes’ rule for every signal, and a politician strategy (σ)that is optimal given the citizen strategy. 2.4 Equilibrium Analysis Given the structure and timing of the game, in the final stage of the interaction, the citizen observes the final allocation of public goods and updates her posterior beliefs about corruption in a Bayesian manner. As illustrated in Figure 3, the sequential nature of the game allows the politician to anticipate the strategic behavior of the citizen in the later stage, enabling her to choose the optimal strategy in the early stage of the game. Therefore, to characterize the equilibrium, we first determine the citizen’s optimal strategy ν and then the politician’s optimal strategy σ. Lemma 1. In every signal equilibrium, regardless of the cost (π) , if the citizen observes a highallocation signal Θh, the citizen plays a strategy of no punishment. ν(αc,Θh)=0 Proof. See Online Appendix B.1. From Lemma 1, we can establish that regardless of the cost, a high allocation of public goods consistently signals an honest politician, and consequently, even citizens who prioritize political accountability have no incentive to disapprove of the incumbent. This result aligns with the idea that citizens reward good politicians to keep them in office and punish bad politicians to remove them from office. Definition 2. (Cutoff strategy). A strategy ν is a cutoff strategy if there is some α∗(Θm(π)) ∈ 13 (0, α], with m={a, l}such that: ν(αc,Θm(π)) =      1if αc≥α∗(Θm(π)) 0if αc< α∗(Θm(π)) Lemma 2. In every signaling equilibrium, if the citizen observes an ambiguous (Θa(π)) or low-allocation signal (Θl(π)), the citizen plays a cutoff strategy following Definition 2. Proof. See Online Appendix B.2. From Lemma 2, we conclude that for each possible allocation Θm(π) there exists an α∗(Θm(π)) cutoff that defines the citizen’s strategy ν . This means that the citizen observes her private intrinsic reward for political accountability, the construction cost, and the final allocation of public goods, and if her intrinsic reward for disapproval is higher than α∗(Θm(π)) , she would disapprove of the politician. Moreover, as citizens receive different information about the public goods allocation, the cutoff may shift according to the construction cost (π)and the observed allocation. Lemma 3. In every signaling equilibrium, π∈ {π, π},α∗(Θl(π)) ≤α∗(Θa(π)) Proof. See Online Appendix B.3. From Lemma 3, the deciding cutoff under an ambiguous allocation is always greater than or equal to the deciding cutoff under a low allocation. Thus, citizens who are willing to punish under ambiguous allocations must have a higher intrinsic reward for disapproval than citizens who punish under low allocations. Furthermore, a citizen who punishes a politician under an ambiguous allocation—–where there is uncertainty of the politician’s corruption—–will also punish under a low allocation, where there is certainty that the politician was corrupt. Given that we have already characterized the citizen’s optimal strategy (ν) , which follows a consistent belief system (λ) , it remains to characterize the politician’s optimal strategy consistent with the citizen’s strategy to define a signaling equilibrium, as presented in Definition 3. Definition 3. (Threshold strategy). A strategy σ is a threshold strategy if there is some ϕ∗(π)≤b ϕ , 14 and for every π∈ {π, π}such that: σ(ϕp) =      1if ϕp≥ϕ∗(π) 0if ϕp< ϕ∗(π) Lemma 4. In every signaling equilibrium, the politician enforces a threshold strategy following Definition 3. Proof. See Online Appendix B.4. Lemma 4establishes that there must exist a threshold ϕ∗(π) which defines the politician’s strategy σ . This means that the politician observes her private intrinsic reward for doing the “right thing” (ϕp) , the construction cost, and then she compares the expected payoff of being corrupt versus being honest. If her intrinsic reward for acting with integrity exceeds ϕ∗(π), she will choose to be honest and allocate the entire budget to public goods. 2.5 Impacts of Economic Times on Political Corruption and Accountability Regarding the impacts of good economic times on political corruption and political accountability, we first define good economic times as a shift in construction costs from a high cost (π) to a low cost (π) . Moreover, these cost changes are randomly determined and announced to both citizens and politicians at the beginning of the game. With this in mind, Theorem 1. In a signaling equilibrium, and for all m={a, l}, if π≥π, then α∗(Θm(π)) ≥α∗(Θm(π)) Proof. See Online Appendix B.5. The main result of our theoretical model is that good economic times, captured by a decline in the cost of public goods provision from a high level (π) to a low level (π) , reduce citizens’ willingness to punish politicians by raising the optimal punishment threshold. This shift is driven by a higher expected material payoff, as the expected public goods allocation is higher during good economic times. Importantly, the increase in the cutoff applies regardless of the observed allocation, reflecting a general tendency toward more lenient behavior in both low and ambiguous allocation scenarios. 15 This leads to the following corollary: Corollary 1. In a signaling equilibrium, and for all Θmwith m={a, l}, if π≥π, then Zα∗(Θm(π)) 0f(α)dα ≥Zα∗(Θm(π)) 0f(α)dα Proof. See Online Appendix B.6. As a direct implication of Theorem 1, Corollary 1demonstrates that, given an increase in the optimal cutoff points and a given signal α∗(Θm) , a politician is more likely to be punished in high-cost environments than in low-cost settings, Figure 4visually illustrates the changes in the probability of punishment. Specifically, it compares the initial set of cutoffs, α∗(π) = {α∗(Θl(π)), α∗(Θa(π))} with the new pair of cutoffs α∗(π) = {α∗(Θl(π)), α∗(Θa(π))}, following the cost reduction. Figure 4: Citizen Distribution of Types (α) αlow(π) αlow(π) αamb(π) αamb(π) f(α) Not Punish Punish if Θ=ΘlPunish if Θ=Θl and also if Θ=Θa α Notes: The figure depicts the distribution of the types (αc∼f[0, α] for the citizen. In dark gray, the mass of citizens that are not willing to punish. In medium gray, the mass of citizens who are willing to punish under low-allocations of public goods. In light gray, the mass of citizens that are willing to punish under low-allocations and ambiguous-allocations of public goods. In black (solid, ϕ∗(π) ), we present the optimal cutoff under a high-cost setting. In blue (dashed, ϕ∗(π)), we present the optimal cutoff under a low-cost setting. Finally, we examine how favorable economic conditions influence a politician’s marginal propensity to engage in corruption. Theorem 2. In a signaling equilibrium, if π≥π ϕ∗(π)≤ϕ∗(π) Proof. See Online Appendix B.7. 16 As established in Corollary 1, an economic boom reduces the likelihood that citizens will punish corrupt politicians. This decline in perceived accountability lowers the expected cost of corrupt behavior, thereby increasing its net expected benefit. Theorem 2formalizes this result, showing that the equilibrium propensity to engage in corruption rises as the cost of public goods provision falls. Corollary 2. In a signaling equilibrium with ϕ∗(π)∈R+, if π≥π Zϕ∗(π) 0U(ϕ)dϕ ≤Zϕ∗(π) 0U(ϕ)dϕ Proof. This result follows directly from the procedure described in Online Appendix B.6, by changing α∗(Θm(π)) to ϕ∗(π). Finally, Corollary 2presents a direct implication of Theorem 2, showing that the probability of a politician engaging in corrupt activities is greater when construction costs are low compared to when they are high. These theoretical results are further illustrated in Figure 5. Taken together, our findings highlight the substantial influence of prevailing economic conditions on the incidence of corruption and the mechanisms of political accountability. Figure 5: Politician Distribution of Types (ϕ)after the Economic Shock ϕ f(ϕ) 1 b ϕ 0b ϕ ϕ∗(π)ϕ∗(π) Corrupt Honest Notes: The figure depicts the distribution of the types (ϕp∼U[0,b ϕ]) for the politician. In medium gray, the mass of politicians who are corrupt. In light gray, the mass of politicians who are honest. In black (solid, ϕ∗(π) ), we present the optimal cutoff under a high-cost setting. In blue (dashed, ϕ∗(π)), we present the optimal cutoff under a low-cost setting. 3 Experimental Design Building on these theoretical results, we conducted a series of laboratory experiments using oTree (Chen et al.,2016). The experiments involved nearly 800 college students recruited 17 through the Rosario Experimental and Behavioral Economics Lab (RE β EL, Greiner (2015)) at Universidad del Rosario in Bogota, Colombia. We collected data from 24 independent sessions, each lasting an average of one and a half hours. 3.1 Political Corruption Game (PCG) The main activity of the experiment was a political corruption game (PCG) designed to replicate our theoretical model. Participants were randomly assigned to one of two roles: politician (P) or citizen (C), and this role remained fixed for the entire session. Each citizen received an endowment of 100 experimental points (EP), 23 while each politician had a potential salary of 120 EP. Participants were paired with a member of the opposite role under a perfect stranger-matching design. Once paired, the citizen paid 40 EP in taxes, which were transferred to the politician. The politician’s main task was to decide how much of this budget to allocate toward constructing public goods: either the entire amount (40 EP) or half (20 EP), keeping the remaining half as personal income. In this setup, public goods represent an additional source of income for citizens. At the beginning of the experiment, both participants were informed that each unit of public goods constructed would provide the citizen with an additional 30 EP, highlighting the social benefits of higher public goods allocations and reinforcing the politician’s duty to act in the citizen’s interest. In this setup, each unit of public goods has a cost. Participants are aware that the construction cost is either 20 EP (high cost) or 10 EP (low cost), with each option having an equal probability. Both participants were informed about the public goods cost ( πr ) at the beginning of each round, ensuring no information asymmetry about the state of the economy. After observing the public goods cost ( πr ), and once the politician made her allocation decision, the computer simulated a national authority by randomly deciding whether to allocate additional funds. If the computer allocated resources, 20 EP were added to the construction of public goods; otherwise, only the politician’s contribution was used. Importantly, the computer’s decision remained private, meaning citizens observed only the final public goods allocation without knowing the individual contributions from the politician or the higher-level government (in this case, the computer’s random allocation). 23 Each EP was valued at USD 0.08. Participants received a USD 3 show-up fee. Citizens (Politicians) earned an average of USD 13.1 (USD 12.9). Given that the minimum wage in Colombia is USD 1.42/hour, the earnings obtained are about nine times the hourly minimum wage. 18 Public goods provision follows equation (1). Once the citizen observed the public goods provided, she decided whether to reduce the politician’s salary (“punish”) to 80 EP 24 by paying 30 EP, or to “approve” the politician and allow her to receive her full salary. The interaction ended once the citizen made her choice, and no feedback was provided within the round. 3.1.1 Additional Details This interaction repeated over five decision rounds. 25 In each round, participants reported their first-order beliefs about their partner’s behavior: politicians predicted the citizen’s decision, while citizens predicted the politician’s policy choice. These beliefs were incentive-compatible, as participants earned 5 EP each time they correctly predicted their partner’s decision. To ensure expectations were recorded before observing others’ behavior, each participant followed a separate experimental timeline.26 3.1.2 Treatments We exploit a within-subjects design across all five rounds. Given that the cost of public goods (πr) is randomly selected by the computer at the beginning of the round and was defined at the pair level, we denote the following two states of the economy: • High-cost environment -or bad economic times- (Control): Under this setting, the construction cost is set to 20EP. Therefore, the maximum amount of public goods that can be produced is three, and the lowest amount of public goods that can be produced is one. • Low-cost environment -or good economic times- (Treatment): Under this setting, the construction cost is set to 10EP. Therefore, the maximum amount of public goods that can be produced is six, and the lowest amount of public goods that can be produced is two. 24Given the structure of the experiment, the utility functions under the PCG are: Up= 120 + (1 −h)20 −δ(0.33)120 + hϕp Uc= 100(1 −0.4) + 30Θ+δ((αc−Θ)λ−30) 25 Participants knew that only one randomly selected round would count toward their final experimental payment. 26The experimental timeline for each type of participant is displayed in the Online Appendix C1 19 3.2 Coordination Game (Social Norms) Finally, participants engaged in a coordination game based on Krupka and Weber (2013), used to elicit incentive-compatible normative expectations from citizens and politicians during the PCG. 27 Each participant rated the social appropriateness of the politician’s behavior under each possible cost scenario. Specifically, subjects indicated whether allocating the entire budget was very socially inappropriate, socially inappropriate, socially appropriate, or very socially appropriate. Participants earned 5 EP each time their normative expectation aligned with the majority view among participants of their own type. At the end of the activity, they completed a non-incentivized socio-demographic questionnaire and provided payment details. Table D2 in the Appendix presents descriptive statistics. 4 Results 4.1 Political Corruption Are politicians corrupt? Pooling participants across all five rounds, we find that they are corrupt 55% of the time. Notably, 22% of the politicians were always corrupt, while 12% of them were always honest. 28 Their willingness to engage in corruption is not independent of the economic cycle. Figure 6, which illustrates the percentage of corrupt politicians segmented by cost type, shows that the unconditional average corruption rate is 51% when the cost is high, rising to nearly 59% when the construction cost is low—a statistically significant increase at the 1% level. This increase represents a 14% rise in the probability of engaging in corruption compared to the high-cost environment. To measure the causal treatment effects on political corruption, we estimate the following model: Corruptp,r,s =α0+α1Costp,r +x′ pγ+ Λr+ηs+εp,r,s,(4) where Corrupti,r,s is a dummy variable that takes the value of 1 if the politician was corrupt and 0 otherwise. Costp,r , the treatment variable, is a dummy variable that takes the value of 1 if the construction cost is low and 0 if the construction cost is high. Additionally, we include a set of controls (x′ p) that consider sociodemographic variables, a series of round 27 The coordination game was implemented at the end of the experiment, with participant identities remaining anonymous. 28 The distribution of the number of times politicians were corrupt is presented in Figure D5 in the Online Appendix. 20 include the same set of sociodemographic controls presented in the estimation of equation (4). Finally, we estimate each model considering if the politician was honest (σ= 1) or corrupt (σ= 0). Figure 9: Politician Expected Punishment Segmented by Costs (π)and Policy Choice 0.112 0.065 *** 0 0.10 0.20 0.30 0.40 0.50 Beliefs of punishment (%) Honest (σ = 1) 0.346 0.283 ** 0 0.10 0.20 0.30 0.40 0.50 Beliefs of punishment (%) Corrupt (σ = 0) High cost Low cost Notes: The figure depicts the average expected disapproval rate by the politician across all rounds, segmented by production costs and policy choices. High cost = 20 EP. Low cost = 10 EP. Dark lines show 95% clustered robust confidence intervals. No controls included. ∗∗∗ p < 0.01,∗∗ p < 0.05,∗p < 0.1 , n.s p > 0.1 . Reported significance levels are derived from hypothesis testing of a fully saturated linear model, where the dependent variable is the citizen’s choice to disapprove of the incumbent, and the regressors are dummy variables associated with the treatments. The OLS estimates are shown in Table 4. Columns (1) to (3) present estimates for punishment expectations when the politician was honest. Honest politicians expect to be punished 11% of the time, consistent with the uncertainty surrounding the final public goods allocation, which may generate an ambiguous allocation signal. The coefficients further indicate that in good economic times, politicians expect a 4 to 5 percentage point reduction in the likelihood of punishment. Similarly, Columns (4) to (6) present estimates for cases where the politician was corrupt. The expected disapproval rate is approximately 35%, significantly higher than that of honest politicians (p-value = 0.000). In good economic times, corrupt politicians anticipate a 6 to 7 percentage point reduction in the likelihood of punishment. The point estimates also reveal significant economic implications, with economic booms leading to a reduction 27 Table 4: Politicians Expected Punishment across cost (π), Segmented by Policy Choice Dep Var:Belief of punishment Politician’s decision (σ) Honest Corrupt (1) (2) (3) (4) (5) (6) Low cost (Treatment) -0.048*** -0.044** -0.041** -0.063** -0.068** -0.064** (0.02) (0.02) (0.02) (0.03) (0.03) (0.03) Constant 0.112*** 0.380*** 0.411*** 0.346*** 0.574*** 0.560*** (0.02) (0.13) (0.13) (0.02) (0.17) (0.19) Mean Dep. Variable 0.11 0.11 0.11 0.35 0.35 0.35 Effect size (%) -42.3 -39.0 -36.7 -18.1 -19.7 -18.5 Controls ✓ ✓ ✓ ✓ Round FE ✓ ✓ Session FE ✓ ✓ Observations 899 899 899 1101 1101 1101 Clusters 309 309 309 351 351 351 Adjusted R-squared 0.01 0.09 0.16 0.00 0.02 0.08 Notes: The set of controls included in Controls are: age, sex, stratum, trust measure, risk measure, patience measure, and previous experience participating in experiments. The Mean Dep. Variable expresses the unconditional control mean of the dependent variable. The Effect size (%) denotes the size of the treatment effect as a percentage compared with the unconditional mean of the control group. Standard errors clustered at the individual level are presented in parentheses. ∗∗∗p < 0.01,∗∗ p < 0.05,∗p < 0.1 in the expected disapproval rate ranging from -42% to -20% compared to the control group. 5.2 Heterogeneous Punishment Behavior across Corruption Beliefs Regarding political accountability behavior, we leverage citizens’ incentive-compatible empirical expectations to estimate how prior beliefs about corruption influence punishment. Figure 10 displays the unconditional disapproval rate across the final public goods allocation of public goods segmented by prior beliefs of corruption. From Figure 10, when citizens observe a low public goods allocation and expect the politician to be honest, the disapproval rate is 48%; however, for those expecting the politician to be corrupt, the rate drops to 26%, reflecting a 22pp reduction. Additionally, when citizens observe an ambiguous allocation, prior beliefs have minimal effect, with disapproval rates consistently around 9%. Lastly, for high allocations of public goods, disapproval is 3% among citizens expecting honesty and 6% among those expecting dishonesty, though this difference is not statistically significant. 28 Figure 10: Citizen Punishment Segmented by Prior beliefs (β)across Allocations (Θ). 0.48 0.09 0.03 0.26 0.10 0.06 *** n.s. n.s. 0 0.10 0.20 0.30 0.40 0.50 0.60 % of citizens that disapprove Low allocation θl Ambiguous allocation θa High allocation θh Belief politician was honest Belief politician was corrupt Notes: The figure depicts the average disapproval rate of the citizen in all rounds, segmented by prior beliefs about corruption and the final allocation of public goods. High cost = 20 EP. Low cost = 10 EP. The dark lines show robust confidence intervals clustered with 95%. No controls included. ∗∗∗ p < 0.01,∗∗ p < 0.05,∗p < 0.1 , n.s p > 0.1 . Reported significance levels come from hypothesis testing from a fully saturated linear model where the dependent variable is the citizen’s choice to disapprove of the incumbent, and regressors are dummy variables associated with treatments. Given that punishment seems to be heterogeneous across prior beliefs of corruption, and in order to formally test this mechanism, we estimate the following model: Punishc,r,s =α0+α1Θa c+α2Θl c+α3BC +β1(Θa c×BC) + β2(Θl c×BC) + Λr+ηs+εc,r,s,(8) where BC denotes the citizen’s prior beliefs of corruption. Table 5presents the results on the heterogeneous effect of prior beliefs across observed allocations. Column (1) indicates that prior beliefs of corruption do not enhance political accountability; instead, they reduce the disapproval rate. While citizens are more likely to punish politicians when they observe a low allocation of public goods, the disapproval rate is significantly higher if they expect the politician to be honest (48%) compared to when they expect the politician to be corrupt (26%). This behavior suggests that when citizens anticipate poor performance, their motivation to punish decreases, aligning with the findings of De la Cuesta et al. (2022). 29 Table 5: Citizens Disapproval Rate across Allocations, by Prior Beliefs of Corruption (β) Dep Var:Citizen punishment OLS (1) (2) (3) (4) Ambiguous allocation (θa) 0.048*** 0.055*** 0.047** 0.045** (0.01) (0.02) (0.02) (0.02) Low allocation (θl) 0.314*** 0.449*** 0.445*** 0.442*** (0.03) (0.04) (0.04) (0.04) Belief Corrupt -0.051*** 0.022 0.016 0.017 (0.02) (0.02) (0.02) (0.02) Ambiguous allocation (θa)×Belief Corrupt -0.011 -0.001 0.003 (0.03) (0.03) (0.03) Low allocation (θl)×Belief Corrupt -0.245*** -0.236*** -0.227*** (0.05) (0.05) (0.05) Constant 0.073*** 0.034*** 0.356* 0.227 (0.01) (0.01) (0.20) (0.20) Mean Dep. Variable 0.05 0.05 0.05 0.05 Effect size (%) 104.8 119.3 102.8 97.1 Controls ✓ ✓ Round FE ✓ Session FE ✓ Observations 1995 1995 1995 1995 Clusters 399 399 399 399 Adjusted R-squared 0.13 0.15 0.17 0.18 p-value H0:θa=θl0.000 0.000 0.000 0.000 Notes: The set of controls included in Controls are: age, sex, stratum, trust measure, risk measure, patience measure, previous experience participating in experiments, and the willingness-to-protest measure from a 6-item questionnaire. The Mean Dep. Variable expresses the unconditional control mean of the dependent variable. The p-value H0:θa=θl denotes the statistical test of differences across allocations. Standard errors clustered at the individual level are presented in parentheses. ∗∗∗p < 0.01,∗∗ p < 0.05,∗p < 0.1 5.3 Does Cost Affect Prior Beliefs of Corruption? Given that prior beliefs of corruption influence citizen’s willingness to disapprove, we now estimate the effect of good economic times on those prior beliefs. Table 6presents OLS estimates of the average treatment effect of low costs on prior beliefs of corruption. In low-cost environments, citizens expect politicians to be 8pp less corrupt than in high-cost environments. Furthermore, the point estimates indicate substantial economic impact, with reductions in incentive-compatible empirical expectations nearing 14%. These findings suggest that instead of anticipating increased corruption in good economic times, citizens’ trust in politicians actually rises–—a result consistent with observed behavior in the Latin America region (Figure 1). The full punishment mechanism unfolds as follows: favorable economic conditions increase the incentives for corruption, leading to more frequent instances of inadequate public goods provision. Initially, citizens are more inclined to trust politicians, as good 30 Table 6: Citizens Prior Beliefs of Corruption across Cost (π) Dep Var:Belief of corruption Prior beliefs (β) (1) (2) (3) Low cost (Treatment) -0.084*** -0.077*** -0.073*** (0.03) (0.03) (0.03) Constant 0.574*** 0.356** 0.309** (0.02) (0.14) (0.14) Mean Dep. Variable 0.57 0.57 0.57 Effect size (%) -14.6 -13.5 -12.8 Controls ✓ ✓ Round FE ✓ Session FE ✓ Observations 1995 1995 1995 Clusters 399 399 399 Adjusted R-squared 0.01 0.04 0.07 Notes: The set of controls included in Controls are: age, sex, stratum, trust measure, risk measure, patience measure, previous experience participating in experiments, and the willingness-to-protest measure from a 6-item questionnaire. The Mean Dep. Variable expresses the unconditional control mean of the dependent variable. The Effect size (%) denotes the size of the treatment effect as a percentage compared with the unconditional mean of the control group. Standard errors clustered at the individual level are presented in parentheses. ∗∗∗p < 0.01,∗∗ p < 0.05,∗p < 0.1 times reduce their prior beliefs about the likelihood of corruption. However, when voters observe a low allocation of public goods and subsequently learn that the incumbent engaged in corrupt behavior, they experience a sense of betrayal, which leads to a sharp increase in political disapproval. 5.4 Social Norms Lastly, we conduct a comparative analysis of responses from the coordination game questionnaire (Krupka and Weber,2013), based on the following question: “In the situation where the cost was (High,Low). If a politician decides to keep 20 EP of the collected taxes, you consider this decision to be...?” In this task, participants attempted to match the modal appropriateness rating of others within their participant type. As shown in the left panel of Figure 11, in high-cost scenarios, a consensus emerges among politicians that personally retaining a portion of collected taxes is very socially inappropriate. However, this behavior shifts dramatically in a low-cost environment where a significant portion of politicians view keeping some of the taxes for personal use 31 as more acceptable, illustrating a relaxation of social norms (descriptive norm) against corruption. Citizens’ beliefs behave similarly. High-cost settings reinforce the strong social inappropriateness of corruption. Yet, in low-cost settings, fewer citizens view corrupt behavior as highly inappropriate, suggesting some degree of normalization or tolerance. Figure 11: Social Norms Histogram across Costs and Participants Types χ2 p-value = 0.000 0 10 20 30 40 50 60 Percentage % Very socially appropriate Socially appropriate Socially innappropriate Very socially innappropriate Politician χ2 p-value = 0.000 0 10 20 30 40 50 60 Percentage % Very socially appropriate Socially appropriate Socially innappropriate Very socially innappropriate Citizen You consider this decision to be... If a politician decides to keep 20 EP of the collected taxes Low cost (πl) High cost (πh) Notes: The figure illustrates the distribution of responses to the question: “In the situation where the cost was (High,Low), if a politician decides to keep 20 EP of the collected taxes, do you consider this decision to be...?” The histogram shows responses for low cost (in gray) and high cost (in dashed black). The left panel presents the distribution of responses for politicians, while the right panel shows those for citizens. (Left panel: χ2test p-value = 0.000; Right panel: χ2test p-value = 0.000.) 6 Discussion In this study, we investigate the relationship between economic conditions and citizens’ willingness to hold politicians accountable, as well as the impact of economic conditions on political corruption. While the literature extensively documents the adverse effects of corruption on economic performance, less attention has been given to how economic fluctuations shape corruption levels and accountability. By addressing this gap, our research offers new insights into political behavior during economic fluctuations and lays the groundwork for more effective governance strategies. We achieve this by developing a theoretical model and empirically testing its predictions through a laboratory experiment. 32 Our theoretical model provides key insights into the relationship between economic performance and political behavior. We predict that favorable economic conditions, characterized by low costs of providing public goods, increase politicians’ inclination to engage in corrupt activities. This occurs largely because politicians anticipate a lower likelihood of punishment, as citizens may be more willing to overlook corruption in times of economic prosperity. Additionally, our model predicts that good economic times weaken political accountability, reflected in citizens’ reduced willingness to disapprove of corrupt politicians. This highlights a broader trend of diminished punishment and weaker enforcement. Empirically, our experiments support these theoretical findings. Corruption rates are significantly higher in low-cost environments, with an increase of approximately 14% compared to the baseline. Additionally, in these low-cost settings, politicians anticipated a nearly 7-percentage-point reduction in the disapproval rate, which further contributed to the rise in corruption. In contrast, citizens’ willingness to disapprove of politicians remained relatively stable across cost levels, as they prioritized punishment based on the observed allocation of public goods (Ferejohn,1986). Specifically, citizens are willing to punish corrupt politicians nearly 36% of the time when the allocation is low and between 3 and 6 percent of the time when the allocation is high. Politicians recognize that citizens can identify corruption to some extent, believing they are three times more likely to be punished when corrupt than when honest. However, politicians do not expect to be punished every time they engage in corruption, and anticipate occasional punishment even when acting honestly. Consequently, incentives for corruption increase when politicians perceive citizens as less sophisticated. Moreover, citizens are less likely to punish corruption if they already suspect politicians are corrupt, suggesting that mistrust fosters conditions that enable further corruption. While our findings offer novel insights, we acknowledge important limitations regarding external validity. First, as with any laboratory experiment, our setting abstracts from some features of real-world politics, most notably the stakes involved. In practice, corruption can carry legal consequences such as imprisonment and may deter reelection, whereas our experimental context imposes no such penalties. That said, although the magnitude of the incentives differs, we see no compelling reason to expect that the direction of behavioral responses or the underlying logic of the accountability mechanism would reverse in higher-stakes environments. A second concern relates to the composition of our experimental sample, which consists primarily of university students. This may be viewed as a limitation, particularly in a 33 setting where other-regarding preferences play a central role. While earlier evidence suggests that students may exhibit higher levels of selfish behavior in economic games relative to non-students (Belot et al.,2015), more recent studies find broadly similar distributions of social preferences, especially with respect to altruistic and maximin (i.e., concern for improving the welfare of the worst-off) types across student and non-student populations, particularly among individuals with higher levels of educational attainment (Epper et al.,2023). This is particularly relevant given that both our sample and the population of interest (e.g., public officials) are highly educated. Even if one were to assume that selfish behavior exhibit by students may be driving the effects on political corruption, other studies using similar experimental samples in Colombia report that approximately 70% of college students behave in a non-selfish manner (Barinas-Forero and Scartascini,2025), suggesting ample heterogeneity in other-regarding motives during our experimental interaction.33 Based on these findings, policymakers should remain vigilant about the risk of increased corruption during periods of economic growth. Strengthening transparency and accountability mechanisms, particularly during periods of prosperity, is crucial for maintaining high governance standards. Accountability becomes more challenging when citizens struggle to identify corruption, a difficulty amplified by economic cycles. When multiple levels of government share responsibility for providing public goods, holding politicians accountable becomes even more complex. Thus, while decentralization may offer a solution, it is effective only if it clarifies rather than obscures responsibilities. Future research could expand on our findings by examining the long-term effects of good and bad economic times on corruption across different contexts and exploring how information dissemination might mitigate the negative impact of economic booms on political accountability. Additionally, analyzing the interaction between various types of economic shocks and political systems could provide valuable insights for combating corruption under diverse economic conditions. 33 Our experiment was conducted at a leading Colombian university that has historically produced a significant share of the country’s political elite, including multiple former presidents. Moreover, nearly 40% of participants are enrolled in degree programs directly related to public administration and governance, making this sample particularly relevant for studying the prospective behavior of future public officials. 34 References Abadie, Alberto, Susan Athey, Guido W. Imbens, and Jeffrey M. Wooldridge, “When Should You Adjust Standard Errors for Clustering?,” The Quarterly Journal of Economics, 2023, 138 (1), 1–35. Abbink, Klaus and Danila Serra, “Anti-corruption Policies: Lessons from the Lab,” in “New Advances in Experimental Research on Corruption,” Emerald Group Publishing Limited, 2012. and Heike Hennig-Schmidt, “Neutral versus loaded instructions in a bribery experiment,” Experimental Economics, 2006, 9, 103–121. , Bernd Irlenbusch, and Elke Renner, “An Experimental Bribery Game,” Journal of Law, Economics and Organization, 2002, 18 (2), 428–454. Ades, Alberto and Rafael Di Tella, “Rents, competition, and corruption,” American Economic Review, 1999, 89 (4), 982–993. Adserà, Alícia, Carles Boix, and Mark Payne, “Are You Being Served? Political Accountability and Quality of Government,” The Journal of Law, Economics and Organization, 2003, 19 (2), 445 –490. Alekseev, Aleksandr, Gary Charness, and Uri Gneezy, “Experimental methods: When and why contextual instructions are important,” Journal of Economic Behavior and Organization, 2017, 134, 48–59. Alesina, Alberto, Alberto Carrasquilla, and Juan Echevarria, “Decentralization in Colombia,” in Alberto Alesina, ed., Institutional Reforms in Colombia., MIT Press, 2005. Almer, Christian, Jeremy Laurent-Lucchetti, and Manuel Oechslin, “Water scarcity and rioting: Disaggregated evidence from Sub-Saharan Africa,” Journal of Environmental Economics and Management, 2017, 86 (1), 193–209. Altunba¸s, Yener and John Thornton, “Does financial development reduce corruption?,” Economics Letters, 2012, 114 (2), 221–223. Ardanaz, Martín and Alejandro Izquierdo, “Current expenditure upswings in good times and public investment downswings in bad times? New evidence from developing countries,” Journal of Comparative Economics, 2022, 50 (1), 118–134. 35 Armand, Alex, Alexander Coutts, Pedro Vicente, and Ines Vilela, “Measuring corruption in the field using behavioral games,” Journal of Public Economics, 2023, 218 (10), 47–99. Armantier, Olivier and Amadou Boly, “A controlled field experiment on corruption,” European Economic Review, 2011, 55 (8), 1072–1082. Asher, Sam and Paul Novosad, “Rent-seeking and criminal politicians: Evidence from mining booms,” Review of Economics and Statistics, 2023, 105 (1), 20–39. Bai, Jie, Seema Jayachandran, Edmund J. Malesky, and Benjamin Olken, “Firm Growth and Corruption: Empirical Evidence from Vietnam,” The Economic Journal, 2017, 129 (618), 651–677. Banerjee, Abhijit, Sendhil Mullainathan, and Rema Hanna, “Corruption,” Technical Report, National Bureau of Economic Research 2012. Banerjee, Rtiwik, “On the interpretation of bribery in a laboratory corruption game: moral frames and social norms,” Experimental Economics, 2016, 19, 240–267. Bardhan, Pranab, “Corruption and Development: A Review of Issues,” Journal of Economic Literature, 1997, 35 (3), 1320–1346. Barinas-Forero, Andres, “Why should my group trust yours? Collective trust and trustworthiness under Economic Shocks,” Documento CEDE, 2024, (29), 1–48. and Carlos Scartascini, “Wage cuts, Social Distributional Preferences and Political Corruption,” IDB Working Paper, 2025, pp. 1–42. Barr, Abigail and Danila Serra, “The effects of externalities and framing on bribery in a petty corruption experiment,” Experimental Economics, 2009, 12, 488–503. Barro, Robert J, “The control of politicians: an economic model,” Public Choice, 1973, 14, 19–42. Belot, Michele, Raymond Duch, and Luis Miller, “A comprehensive comparison of students and non-students in classic experimental games,” Journal of Economic Behavior & Organization, 2015, 113, 26–33. Benton, Allyson Lucinda, “Dissatisfied Democrats or Retrospective Voters?: Economic Hardship, Political Institutions, and Voting Behavior in Latin America,” Comparative Political Studies, 2005, 38 (4), 417–442. 36 Online Appendix Corruption and Political Accountability in Good and Bad Economic Times Andres Barinas-Forero Carlos Scartascini Contents A Extensive Representation Payoffs 2 B Proofs 2 B.1 Proof Lemma 1 ................................... 2 B.2 Proof Lemma 2 ................................... 3 B.3 Proof Lemma 3 ................................... 4 B.4 Proof Lemma 4 ................................... 5 B.5 Proof Theorem 1 ................................... 7 B.6 Proof Corollary 1 .................................. 8 B.7 Proof Theorem 2 ................................... 9 C Experimental Design 9 C.1 Participants’ Timelines ............................... 9 D Additional Figures and Tables 10 E Experimental Screen Instructions 20 E.1 Original Instructions in Spanish ........................... 20 1 A Extensive Representation Payoffs Regarding our experimental setting, we have the following vector of public goods allocations: Hig-cost environment: Θh,Θa,Θl|π= 20EP= (3,2,1) and Low-cost environment: Θh,Θa,Θl|π= 10EP= (6,4,2). Politician’s Payoffs:       U1 pol. U2 pol. U3 pol. U4 pol.        =   w(1 −η) + ϕp w+ϕp w(1 −η) + r w+r    Citizen’s Payoffs:          U1 citi. U2 citi. U3 citi. U4 citi. U5 citi. U6 citi.          =          y(1 −τ) + κ(Θh)−κ y(1 −τ) + κ(Θh) y(1 −τ) + κ(Θa) + ((αc−Θa)λ−κ) y(1 −τ) + κ(Θa) y(1 −τ) + κ(Θl) + ((αc−Θl)−κ) y(1 −τ) + κ(Θl)          B Proofs B.1 Proof Lemma 1 Proof. Under a high-allocation Θh , the citizen updates her posterior beliefs of corruption following, λ≡Pr σ= 0|Θ(π) = Θh(π)=Pr Θ(π) = Θh(π)|σ= 0×Pr (σ= 0) Pr (Θ(π) = Θh(π))(B1) From Figure 2, observe that a higher allocation of public goods is only possible if the politician is honest. Therefore, Pr(Θ(π) = Θh(π)|σ= 0) = 0 . Updating (λ) following equation (B1), we find that λ= 0 . Given this signal, the optimization payoff presented in Equation (3) can be expressed as: max δ∈{0,1}y(1 −τ) + κΘh(π) + δ(−κ). 2 Finally, note that any strategy ν > 0 results in a lower material payoff. Intuitively, this occurs because, under a high allocation signal, the citizen has no reason to protest, as the politician was honest. Therefore, under this signal, and regardless of the cost (π) , the citizen plays: ν(Θh(π)) = 0 ∀αc∈[0, α] B.2 Proof Lemma 2 Proof. Under a low-allocation Θl , the citizen updates her posterior beliefs of corruption following Equation (B1). Once again, from Figure 2observe that a low-allocation of public goods is only possible if the politician was corrupt. Moreover, the belief-updating process (λ) results in the following: λ=(1−µ)×β (1−µ)(β)= 1 . Given this signal and the updated beliefs, the optimization problem for the citizen can be written as: max δ∈{0,1}y(1 −τ) + κΘl(π) + δ(αc−Θl(π)) −κ Intuitively, the citizen optimizes by choosing a mixed strategy ν that maximizes her payoff, which depends entirely on: (αc−Θl(π)) −κ . From Section 2, recall that Θl(π) = τy −r π . Rearranging terms, we find that the optimal strategy under a low-allocation signal is: ν(αc,Θl(π)) =                1if αc≥κ+τy −r π 0if αc< κ +τy −r π(B2) Moreover, regarding the optimal cutoff given a low-allocation signal, we find that: α∗(Θl(π)) = κ+τy −r π∈(0, α] Following the same logic, given an ambiguous-allocation Θa , the belief-updating process results in: λ=µβ µβ+(1−β)(1−µ) . Given this signal, the optimization problem for the citizen can be expressed as: max δ∈{0,1}y(1 −τ) + κΘa(π) + δ (αc−Θa(π)) µβ µβ + (1 −β)(1 −µ)!−κ! 3 Similarly, the citizen chooses a mixed strategy ν that maximizes her payoff, which depends on: (αc−Θa(π))µβ µβ+(1−β)(1−µ)−κ . Given that Θa(π) = τy π , and rearranging terms, the optimal strategy under an ambiguous-allocation signal is: ν(αc,Θa(π)) =                1if αc≥κ µβ + (1 −β)(1 −µ) µβ !+τy π 0if αc< κ µβ + (1 −β)(1 −µ) µβ !+τy π(B3) Moreover, regarding the optimal cutoff under an ambiguous allocation signal, we find that: α∗(Θa(π)) = (κ µβ + (1 −β)(1 −µ) µβ !+τy π)∈(0, α] Given that cutoff points determine the citizen’s optimal strategy under both signals and regardless of the cost, this proves the lemma. B.3 Proof Lemma 3 Proof. Employing Lemma 2, and focusing on a particular π={π, π} without loss of generality, we can define the citizen’s set of optimal cutoffs α∗(Θl,Θa) = κ+τy −r π, κ µβ + (1 −β)(1 −µ) µβ !+τy π! Therefore, we compared these cutoffs following: κ+τy −r π | {z } α∗(Θl) < κ µβ + (1 −β)(1 −µ) µβ !+τy π | {z } α∗(Θa) Rearranging terms, κ−r π< κ µβ + (1 −β)(1 −µ) µβ ! Given that κ∈R+, it is possible to rewrite this last expression as: 1−r κπ< µβ + (1 −β)(1 −µ) µβ ! 4 Simplifying, we obtain 1−r κπ<1 + (1 −β)(1 −µ) µβ ! Given that r∈(0, τy) , κ∈R+ , and π∈ {π, π} ⊂ R+ , the expression r κπ  is always positive. On the other hand, given that µ∈(0,1) , and β∈(0,1) , the expression µβ+(1−β)(1−µ) µβ  is always positive. Therefore, the expression 1−r κπ | {z } Positive <1 + (1 −β)(1 −µ) µβ ! | {z } Positive , is always true. This completes the proof. B.4 Proof Lemma 4 Proof. The politician may consider a strategy (σ) that aligns with the citizen’s strategy ν and the citizen’s updated beliefs λ, forming an equilibrium. Hence, once the public good costs are determined, the politician compares the expected payoff of each possible action, considering the citizens’ behavior, and chooses the highest. Strategy 1: Expected payoff of being Corrupt If the politician chooses to be corrupt, there is a probability (µ) that the national authority fully covers her decision (γ= 1) , resulting in the citizen observing an ambiguous allocation signal Θa . In this case, the citizen will punish the politician if equation (B3) holds. Conversely, with a probability of (1 −µ) , the national authority does not provide resources (γ= 0) , leading the citizen to observe a low allocation signal Θl , and the citizen will punish if equation (B2) holds. Therefore, the expected payoff for being corrupt (h= 0) is given by E[Up|h= 0, π]=w+r−µ[P(αc> α∗(Θa(π)))(ηw)]−(1 −µ)hPαc> α∗(Θl(π))(ηw)i Given that αc∼f[0, α], we can rewrite this expression as w+r−µ 1−Zα∗(Θa(π)) 0f(α)dα !(ηw)−(1 −µ) 1−Zα∗(Θl(π)) 0f(α)dα !(ηw)(B4) 5 Strategy 2: Expected payoff of being honest If the politician decides to be honest, she knows that with probability (µ) , the national authority will provide additional resources (γ= 1) , and the citizen observes a high allocation signal Θh . From Lemma 1, we know that under this signal the citizen will never punish the politician. Conversely, with probability (1 −µ) , the national authority does not provide resources (γ= 0) , and the citizen observes an ambiguous signal Θa of public goods. Again, under an ambiguous signal, the citizen will disapprove if equation (B3) holds. Therefore, the expected payoff of being honest (h= 1) is given by E[Up|h= 1, π] = w+ϕp−(1 −µ)[P(αc> α∗(Θa(π)))(ηw)] Rewriting this expression, we obtain E[Up|h= 1, π] = w+ϕp−(1 −µ) 1−Zα∗(Θl(π)) 0f(α)dα !(ηw)(B5) To find an optimal strategy (σ) , the politician considers both expected payoffs and chooses the one that maximizes her utility. Under both scenarios, the politician receives her salary (w) and also faces the possibility of punishment under an ambiguous signal. Therefore, she compares the expected utility of being honest over being corrupt, given a particular public goods cost (π). This comparison can be expressed as ϕp−(1 −µ) 1−Zα∗(Θa(π)) 0 f(α)dα !(ηw) | {z } Being honest ≥ r−µ 1−Zα∗(Θa(π)) 0 f(α)dα !(ηw)−(1 −µ) 1−Zα∗(Θl(π)) 0 f(α)dα !(ηw) | {z } Being corrupt (B6) Factorizing, we obtain ϕp≥r+ (1 −2µ) 1−Zα∗(Θa(π)) 0 f(α)dα !−(1 −µ) 1−Zα∗(Θl(π)) 0 f(α)dα ! Therefore, ϕ∗(π) = r+ (1 −2µ) 1−Zα∗(Θa(π)) 0 f(α)dα !−(1 −µ) 1−Zα∗(Θl(π)) 0 f(α)dα !(B7) 6 Finally, only rests to prove that ϕ∗(π)<b ϕ . First, given that µ∈(0,1) behaves monotonically, we can focus on the extreme value cases of µ. Case 1: limµ→0 ϕ∗(π) = r+ 1−Zα∗(Θa(π)) 0f(α)dα !− 1−Zα∗(Θl(π)) 0f(α)dα ! From Lemma 3, it follows that Rα∗(Θl(π)) 0f(α)dα ≤Rα∗(Θa(π)) 0f(α)dα , and given that r < b ϕ , we obtain ϕ∗(π) = r+ 1−Zα∗(Θa(π)) 0f(α)dα !− 1−Zα∗(Θl(π)) 0f(α)dα !< r < b ϕ So, when limµ→0, ϕ∗(π)<b ϕ Case 2: limµ→1 ϕ∗(π) = r− 1−Zα∗(Θa(π)) 0f(α)dα ! Given that α∗(Θa(π)) >0, we obtain ϕ∗(π) = r− 1−Zα∗(Θa(π)) 0f(α)dα !< r < b ϕ So, when limµ→1, ϕ∗(π)<b ϕ This completes the proof. B.5 Proof Theorem 1 Proof. Case 1: (Low allocation): α∗(Θl(π)) = κ+τy −r π Derivative with respect to π: d dπα∗(Θl(π)) = d dπ κ+τy −r π=−τy −r π2 7 Since τy −r is a positive constant and π2 is always positive, −τy−r π2 is negative. Therefore, α∗(Θl(π)) is a decreasing function of π . Since α∗(Θl(π)) decreases as π increases, it follows that: α∗(Θl(π)) ≤α∗(Θl(π)) whenever π≥π. Case 2: (Ambiguous allocation): α∗(Θa(π)) = κ µβ + (1 −β)(1 −µ) µβ !+τy π Derivative with respect to π: d dπα∗(Θa(π)) = d dπ κ µβ + (1 −β)(1 −µ) µβ !+τy π!=−τy π2 Since τy is a positive constant and π2 is always positive, −τy π2 is negative. Therefore, α∗(Θa(π)) is a decreasing function of π . Since α∗(Θa(π)) decreases as π increases, it follows that: α∗(Θa(π)) ≤α∗(Θa(π)) whenever π≥π. This completes the proof of Theorem 1. B.6 Proof Corollary 1 Proof. From Theorem 1,α∗(Θm(π)) ≥α∗(Θm(π)), then rearranging terms, Zα∗(Θm(π)) 0f(α)dα +Zα∗(Θm(π)) α∗(Θ)m(π)f(α)dα ≥Zα∗(Θm(π)) 0f(α)dα Zα∗(Θm(π)) α∗(Θ)m(π)f(α)dα ≥0 This completes the proof of Corollary 1. 8 B.7 Proof Theorem 2 Proof. From Equation (B7), we have that ϕ∗(π) = r+ (1 −2µ) 1−Zα∗(Θa(π)) 0f(α)dα !−(1 −µ) 1−Zα∗(Θl(π)) 0f(α)dα ! From Corollary 1, we can readily see that 1−Zα∗(Θm(π)) 0f(α)dα !≤ 1−Zα∗(Θm(π)) 0f(α)dα !∀m={a, l} Then, it follows that ϕ∗(π)≤ϕ∗(π) C Experimental Design C.1 Participants’ Timelines Figure C1: Experimental Timeline in PCG across Participants Types Computer Politician Politician Computer Cost of Policy Beliefs Resources and public goods choice of punishment public goods allocation πhE[δ]γ,Θ Computer Citizen Computer Citizen Cost of Beliefs Resources an Approval public goods of corruption public goods allocation choice πβγ,Θδ Notes: The figure illustrates the experimental timeline across different participant types. Beliefs of punishment refers to the politician’s response to the question: "Do you think the citizen would disapprove of you?" Beliefs of corruption refers to the citizen’s response to the question: "How many resources do you think the politician would allocate?" 9 D Additional Figures and Tables Figure D1: Perception of Number of Politicians Involved in Corruption across National Economic Situation Perception. 2 2.5 3 3.5 4 4.5 Number of politicians involved in corruption (1 None - 5 All) Worse than before Same as before Better than before Economic situation of the country in last 12 months (1 Worse - 3 Better) Argentina Bahamas Belize Bolivia Brazil Canada Chile 3 3.5 4 4.5 Number of politicians involved in corruption (1 None - 5 All) Worse than before Same as before Better than before Economic situation of the country in last 12 months (1 Worse - 3 Better) Haiti Honduras Jamaica Mexico Nicaragua Panama Notes: The figure depicts the scatterplot binned between the perception of the number of politicians involved in corruption and the perception of the country’s economic situation over the past 12 months. Scatter binned estimation for all respondents to the LAPOP question in 26 countries between 2016 and 2023. The dark line indicates the linear fit. Gray dashed lines represent the 95% robust confidence intervals. The analysis includes country and year fixed effects. Statistical significance levels: ∗∗∗ p < 0.01,∗∗ p < 0.05,∗p < 0.1, n.s p > 0.1. 10 Figure D7: Distribution of Treated Rounds by Participant Type (PCG) 0 5 10 15 20 25 30 35 Percentage % None 1234All N° of rounds with low cost (by participant) Citizens (C) Politicians (P) Notes: The figure illustrates the percentage distribution of the number of rounds in which citizens disapprove of politicians during the game. The histogram represents different categories of political accountability behavior, ranging from "Never punish" (0 rounds of disapproval) to "Always punish" (5 rounds of political disapproval). 17 Figure D8: Distribution of Total Corruption in the Political Corruption Game (PCG) 0.18 0.18 0.27 0.09 0.27 0.10 0.13 0.17 0.23 0.12 0.25 0.11 0.17 0.20 0.18 0.09 0.25 0.16 0.07 0.22 0.21 0.13 0.21 0.12 0.13 0.15 0.32 0.12 0.17 0.000.00 0.30 0.20 0.20 0.30 0 20 40 60 80 100 Percentage % 0 1 2 3 4 5 N° of rounds with low cost (by participant) Corrupt = 0 times Corrupt = 1 times Corrupt = 2 times Corrupt = 3 times Corrupt = 4 times Corrupt = 5 times Notes: We present the distribution of total number of times being corrupt across the total number of rounds that the politician face a low cost. Notice that "Always honest" politicians face several rounds with low cost settings. Figure D9: Distribution of Total Punishment in the Political Corruption Game (PCG) 0.60 0.25 0.05 0.10 0.50 0.23 0.20 0.07 0.50 0.28 0.18 0.03 0.49 0.29 0.15 0.06 0.53 0.27 0.16 0.01 0.70 0.20 0.10 0 20 40 60 80 100 Percentage % 012345 N° of rounds with low cost (by participant) Punish = 0 times Punish = 1 times Punish = 2 times Punish = 3 times Notes: We present the distribution of total number of punishment rounds across the total number of rounds that the citizen faces a low cost. Notice that "Never punish" citizens face several rounds with low cost settings. 18 Figure D10: Willingness to Punish by LAPOP Question, by Participant Type 0 10 20 30 40 Percent 0 2 4 6 8 10 Willing to protest for better salaries 0 20 40 60 Percent 0 2 4 6 8 10 Willing to protest for better health 0 10 20 30 40 Percent 0 2 4 6 8 10 Willing to protest against climate change 0 10 20 30 40 Percent 0 2 4 6 8 10 Willing to protest for democracy rights 0 10 20 30 40 Percent 0 2 4 6 8 10 Willing to protest against corruption 0 10 20 30 40 Percent 0 2 4 6 8 10 Willing to protest against inequality Citizens (C) Politicians (P) Notes: The figure illustrates the willingness of citizens (in gray) and politicians (in red) to protest for various causes, including better salaries, health, climate change, democracy rights, corruption, and inequality. The histogram represents the willingness to protest, scored from 0 to 10. 19 E Experimental Screen Instructions E.1 Original Instructions in Spanish Figure E1: General instructions 20 Figure E2: Informed Consent Figure E3: Welcome 21 Figure E4: Initial Instructions Figure E5: Instructions - 1 22 Figure E6: Instructions - 2 23 Figure E7: Round stages 24 Figure E8: Round stages - 2 25 Figure E9: Payment Figure E10: Control Question A - 1 26 Figure E19: End of the round Figure E20: New round screen 33 Figure E21: Citizen’s Social Norms Screen Figure E22: Politician’s Social Norms Screen 34 Figure E23: Final Results 35 Figure E24: Final Results - 2 36 Figure E25: Questionnaire 37 38 39 40 41