Voter preferences, electoral promises, and the composition of public spending

Keefer, Philip; Scartascini, Carlos G.; Vlaicu, Razvan Working Paper Voter preferences, electoral promises, and the composition of public spending IDB Working Paper Series, No. IDB-WP-1123 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Keefer, Philip; Scartascini, Carlos G.; Vlaicu, Razvan (2020) : Voter preferences, electoral promises, and the composition of public spending, IDB Working Paper Series, No. IDBWP-1123, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0002434 This Version is available at: https://hdl.handle.net/10419/234700 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 Voter Preferences, Electoral Promises, and the Composition of Public Spending Philip Keefer Carlos Scartascini Razvan Vlaicu IDB WORKING PAPER SERIES Nº IDB-WP-1123 June 2020 Department of Research and Chief Economist Inter-American Development Bank June 2020 Voter Preferences, Electoral Promises, and the Composition of Public Spending Philip Keefer Carlos Scartascini Razvan Vlaicu Inter-American Development Bank Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Keefer, Philip. Voter preferences, electoral promises, and the composition of public spending / Philip Keefer, Carlos Scartascini, Razvan Vlaicu. p. cm. — (IDB Working Paper Series ; 1123) Includes bibliographic references. 1. Government spending policy-Latin America-Econometric models. 2. Voting-Latin A merica-Econometric models. 3. Public goods-Latin America-Econometric models. 4. Public investments-Latin America-Econometric models. I. Scartascini, Carlos G., 1971- II. Vlaicu, Razvan. III. Inter-American Development Bank. Department of Research and Chief Economist. IV. Title. V. Series. IDB-WP-1123 Copyright © Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 Attribution- NonCommercial-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 2020 Abstract1 This paper proposes and empirically tests a new demand-side explanation for distortions in public spending composition. Voters prefer spending with certain and immediate benefits when they have low trust in electoral promises and high discount rates. The paper incorporates these characteristics of voter choices into a probabilistic voting model with public spending tradeoffs. In equilibrium, candidates promising larger allocations to transfers and short-term public goods are more likely to win elections in settings with low trust and high impatience. An original survey of individual-level preferences for public spending in seven Latin American capital cities provides observational and experimental evidence consistent with the model-derived hypotheses. Respondents reporting low trust in politician promises are more likely to prefer transfers to public goods; respondents with high discount rates prefer short-term to long-term spending. These patterns also appear in country-level data on spending outcomes from the last two decades. JEL classifications: D72, H20, H50, O10 Keywords: Spending composition, Voter preferences, Trust, Discounting, Transfers, Public goods, Public investment 1 Contact Information: Keefer: Institutions for Development Department, Inter-American Development Bank, 1300 New York Ave NW, Washington, DC 20577, USA. Email: [email protected]. Scartascini: Research Department, Inter-American Development Bank, 1300 New York Ave NW, Washington, DC 20577, USA. Email: [email protected]. Vlaicu: Research Department, Inter-American Development Bank, 1300 New York Ave NW, Washington, DC 20577, USA. Email: [email protected]. We thank João Ayres, Alejandro Izquierdo, Jorge Puig, Guillermo Vuletin, Diego Vera, and Elizabeth Zechmeister for comments, Georgina Pizzolitto for overseeing data collection, Kurt Birson, Adelaida Correa, and Sergio Perilla for research assistance. The findings and interpretations in this paper are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank or the governments it represents. 1 Introduction Government spending can support economic development by funding growth-promoting public goods such as high quality public education, or making investments in public safety infrastructure. In many developing countries, however, governments make systematically low budgetary allocations to these types of spending in favor of short-term spending priorities such as redistributive transfers. This is particularly so for investment or capital spending, which yield bene…ts years later. A recent study estimates that the bias against capital spending in developing economies was 8.5 percentage points, compared to 3.7 percentage points in advanced economies.2We present a new argument that traces variation in spending composition to voter mistrust in politicians and voter impatience. Lower-trust voters prefer a higher ratio of expenditures with more certain bene…ts, such as transfers, to expenditures with less certain bene…ts, such as spending on education or police. Higher-discounting voters prefer expenditure policies with more immediate returns to those with distant payo¤s, such as investment. In the aggregate, candidates promising larger allocations to long-term public goods are less likely to win elections in settings with low trust and high impatience. Both individual- and country-level evidence from Latin America support these predictions. Several other explanations could account for ine¢ cient or regressive public spending policies in developing countries. An important body of literature has emphasized representation failures in newly democratic institutions, such as elite capture of government policies to avoid empowering the citizenry to challenge the dominance of the ruling class (Engerman and Sokolo¤, 2000; Acemoglu and Robinson, 2006). Alternatively, underprovision re‡ects lack of state capacity in designing and implementing complex policies (Besley and Persson, 2009, 2011). Another set of arguments points to distortions of the electoral mechanism, either through political budget cycles, which appear more pronounced in developing countries (Shi and Svensson, 2006; Drazen and Eslava, 2010), or through e¤ective disenfranchisement of low-income voters (Fujiwara, 2015), or electoral clientelism that allows politicians to divert public resources toward private rents (Keefer and Vlaicu, 2008; Robinson and Verdier, 2013). This paper proposes an alternative mechanism to these well-known hypotheses, one centered on voter choices, i.e., the demand side of spending policies. We argue that developing 2See Figure 1.4 in Izquierdo, Pessino, and Vuletin (2018). The bias against capital spending is de…ned as the di¤erence in the long-term (1980-2016) trends of capital and current spending as shares of total (primary) spending. The bias occurred despite substantial hikes in total spending as percent of GDP, which left enough room for increasing social expenditures without substantially cutting back the share of capital spending. 2 countries with democratic elections may experience …scal misallocations because voters are uncertain about how politicians’electoral promises will improve their welfare. That is, voters have low trust that political candidates’public good spending programs will deliver the promised voter welfare. Given budgetary tradeo¤s, this should decrease voter demand for public good spending relative to spending with more certain bene…ts, such as transfer payments. In addition, voters discount political candidates’current spending proposals whose bene…ts occur in the future. This would decrease voter demand for long-term public good spending. In both cases, voters prefer political candidates that promise to allocate spending toward programs with more certain and more immediate bene…ts, such as monetary transfers or public sector hiring. To understand the policy implications of low trust and high discounting, we develop a probabilistic voting model with three types of spending: spending with immediate and certain bene…ts, such as direct monetary transfers to voters; short-term public goods (or public consumption), whose bene…ts are immediate but uncertain, such as general increases in current police or education expenditures; and long-term public goods (or public investment), whose bene…ts are both uncertain and occur in the future, as with the training of service providers and infrastructure investments. Given a …xed budget, low-trust voters prefer candidates who allocate more budgetary resources to transfers and high-discounting voters prefer candidates that spend less on investment projects. Aggregating across voters and examining countrylevel budget allocations, lower trust should be associated with a higher ratio of transfers to public goods and higher discounting with a higher ratio of short-term to long-term spending. We present evidence supporting both the individual-level behavior and the aggregate spending outcomes predicted by the model. To document individual behavior we designed an original survey of voter preferences that we …elded in seven Latin American capitals through the Latin American Public Opinion Project (LAPOP) at Vanderbilt University, hereafter the IDB-LAPOP Survey. It elicits respondents’trust in politician promises and measures discount rates using the unfolding-brackets methodology in Falk et al. (2018). Spending preferences are based on a revealed preference design. Respondents are asked which of two policy alternatives they prefer in order to improve education or reduce crime. For example, respondents choose between a tax cut allowing more personal spending on security, and a tax increase to fund more police spending. We vary the policy pairings to capture the spending tradeo¤s characterized in the model. For example, choosing between spending on hiring additional public employees (more certain and current bene…ts) and training existing employees (more uncertain and future bene…ts). Evidence on aggregate spending outcomes 3 comes from a panel of 18 Latin American countries over the last two decades. Here we look at the empirical relationship between spending ratios and aggregate measures of trust in political parties and of time discounting, using country and year …xed e¤ects. The individual-level data on spending preferences show that lower-trust respondents are more supportive of transfers at the expense of public goods, and higher-discount respondents are more supportive of short-term spending (transfers and public consumption) at the expense of long-term spending (public investment). These …ndings appear in both policy domains we studied, education and security. The country-level data on spending outcomes con…rms the individual-level results. Country-years with lower aggregate trust in political parties and higher real interest rates tend to be those with higher ratios of transfers to public goods and current to capital spending. Two survey experiments allow further tests of the model using randomized variation in how much electoral promises demand trust or patience from voters. Randomly selected respondents receive an informational treatment indicating that a policy that requires more voter trust delivers larger bene…ts than an alternative that demands less trust. Low-trust respondents are less sensitive to the information about the relative bene…ts of the two policies and are less likely to switch away from the policy with smaller bene…ts. The other experiment asks respondents to choose between two types of expenditures on government workers, hiring additional workers (immediate bene…ts) or training existing workers (delayed bene…ts). Respondents are randomly assigned to di¤erent time horizons (two vs. four years) for getting the delayed bene…ts of training. Extreme (very high or very low) discount rate respondents are less responsive to di¤erences in time horizon and are less likely to switch from hiring to training when they learn that training bene…ts arrive sooner.3 The literature on the political economy determinants of public spending composition is well-established, but has focused largely on mature democracies and on the role of political institutions, such as forms of government, electoral rules, or party systems; see, e.g., Lizzeri and Persico (2001), Scartascini and Crain (2002), and Persson and Tabellini (2003). These studies have provided valuable insights into the supply side of spending policies, while keeping voter preferences …xed. Instead, we focus on variation on the demand side of spending policies and set aside issues related to electoral incentives under di¤erent institutional con…gurations.4 3Very high discount respondents almost always prefer the immediate bene…ts policy, thus the time horizon for the delayed bene…ts policy matters little to them. Very low discount respondents may prefer the delayed bene…ts policy, but care little whether it has a long or short time horizon. 4The macro dynamics of spending allocations have been studied by Ismihan and Ozkan (2011), who develop a theoretical model where greater political instability is associated with increased allocation of public 4 More recent studies of developing countries with weak institutions have highlighted the distortions that voters themselves can introduce into the policy process. Bursztyn (2016) provides experimental evidence from Brazil that poor voters prefer government programs that increase their incomes in the short run, such as cash transfers, over spending on public education that bene…ts them in the long run. We …nd similar income e¤ects in our data, but show that even after controlling for income, respondent trust and discounting play an independent role in the formation of spending preferences. Keefer and Vlaicu (2017) study a model of vote buying with reneging on campaign promises where voters prefer candidates associated with high vote buying and low public good provision. We model voter trust in a similar way, but use it to explore a di¤erent policy distortion: preferences for post-electoral transfers over public goods, rather than pre-electoral transfers (vote buying) over public goods.5 The literature on trust has focused primarily on market interactions (Guiso et al., 2004; Dearon and Grier, 2011), political outcomes (Nunn, Qian, and Wen, 2018), or overall economic performance (Zak and Knack, 2001; Francois and Zabojnik, 2005) and much less on government policies. Two notable recent exceptions are Algan, Cahuc, and Sangnier (2016) and Camussi, Mancini, and Tomassino (2018). Both are concerned with how generalized trust and trust in government are related to government spending on social programs (education and health) or welfare programs in developed countries. They …nd that higher trust is generally associated with more social and welfare spending as a fraction of the budget, although the e¤ect can be nonlinear. While we also provide aggregate-level evidence on spending outcomes, our individual-level evidence is based on preference elicitation in developing countries through hypothetical policy comparisons under budgetary tradeo¤s; this allows us to manipulate the spending characteristics we are interested in, such as the certainty and timing of bene…ts. In using randomized information shocks to study voter preferences our paper builds on recent work that has employed randomized survey experiments (Banerjee et al., 2011, in electoral choices, and Jensen, 2010, in education choices). Studying preferences for redistribution, Kuziemko et al. (2015) argue that low trust in government accounts for the spending to government consumption instead of investment, and Ardanaz and Izquierdo (2017), who provide evidence from developing countries that governments adjust to economic cycles by increasing transfers in upswings and reducing public investment in downturns. 5Keefer (2007) provides country-level evidence that young democracies where political candidates cannot make credible promises to voters overprovide targeted transfers and underprovide public goods. Khemani (2015) provides evidence from the Philippines that politicians purchasing political support with targeted transfers are likely to trade it o¤ against the provision of broader public services. 5 with trust. Similarly, in part (ii), the ratio of transfers to public consumption spending fi hi can either increase or decrease since both components of spending are more preferred. In addition, higher pre-tax income should reduce the relative preferences fi hi+gi;fi+hi gi;and fi hi. The comparative statics with respect to trust and discounting are summarized in Chart 1.3 Before studying the electoral equilibrium, it is instructive to characterize spending composition as chosen by a benevolent social planner. This sets a benchmark of socially optimal policy against which to compare the outcomes of electoral competition. A social planner chooses spending levels (fsp; hsp; gsp)that maximize aggregate voter welfare without facing the electoral constraints of low trust and high discounting, i.e., for the social planner, sp = 1 and sp = 0. The social planner’s problem is: max f;h;g Z1 0zi1 2z2 idi +h 2h2+g 2g2 (5) subject to the budget constraint f+h+gy; where zi(1 )yi+f: The solution of this problem requires the equalization of the marginal aggregate welfare of the di¤erent spending categories: 1[(1 ) y+fsp] = 1 hsp = 1 gsp (6) and by the strict monotonicity of the voters’policy utilities, the government budget constraint is binding fsp +hsp +gsp =y: Note that spending ine¢ ciency reduces aggregate welfare through less e¢ cient public good spending. Thus, higher government spending in- e¢ ciency optimally requires government revenues to be reallocated toward transfers at the expense of public goods.4 Turning to the electoral equilibrium, each party j=A; B solves a constrained optimization problem, namely maximizing its winning probability subject to the government budget constraint, given its opponent j’s strategy. max (fj;hj;gj)1 2+ (WjWj) + (yfjhjgj)(7) where, as before, WjR1 0Wijdi is average voter utility from party j’s policies, and  3As the proofs in Appendix A show, the comparative statics in Propositions 1 and 2 can be derived under more general conditions, namely that the utilities of each spending component, fi; hi; gi;are strictly increasing and strictly concave in that type of spending, and the utilities of public good spending, hi; gi;are strictly decreasing in trust i. The only exception is the comparative static for hi giwith respect to i: 4A case where this is particularly relevant is countries that have experienced windfalls of revenues from natural resources. 12 is the standard Lagrangian multiplier. For simplicity, we assume homogenous preference parameters in the electorate: i=and i=for all i2[0;1] :The comparative statics and the data for this part will be at the aggregate level. A solution to this optimal allocation problem requires that the electoral returns of the di¤erent policies be equalized: 1[(1 ) y+fj] = 1 (1 +)hj=1 (1 + )[1 (1 +)gj](8) for j=A; B and by the strict monotonicity of parties’ utilities, the government budget constraint is binding fj+hj+gj=y: Note that the equilibrium condition for electoral competition corresponds to the spending preferences of the voter with average income yi= y; compare equation (8) to equation (4). Equation (8) can also be compared to the social planner’s allocation characterized in equation (6). Compared to the social planner solution, in the electoral equilibrium the parties’electoral returns to an extra unit of public consumption and investment spending are diminished by voter trust and voter discounting . This observation leads to the following result. Proposition 3 There exists a unique electoral equilibrium and it is symmetric. In the electoral equilibrium, spending strategies (fj; hj; gj)j=A;B have equal marginal electoral returns for each party. Parties’transfer spending is higher than in the social planner’s allocation, fj> fsp;public consumption spending exceeds investment spending hj> gj;and investment spending is lower than the social planner’s allocation, gj< gsp: Low voter trust  < 1and high voter discounting  > 0imply that voter marginal utilities of public consumption and investment spending are lower than in the social optimum. Voters thus support candidates that provide higher transfers fjand lower investment relative to what a benevolent social planner would provide. Government spending will then be reallocated toward transfers and away from public goods. Also, because of discounting, parties’ marginal electoral returns from public investment spending gjis lower than for public consumption spending hj, causing larger allocations of public funds for short-term spending. Government spending composition will vary with the extent of these policy preference distortions. The following result summarizes the main e¤ects for spending levels. Below we will look at spending ratios as well. Proposition 4 (i) A decrease in voter trust leads to an increase in transfers fj, and a decrease in public consumption spending hj. (ii) An increase in voter discounting leads 13 to an increase in transfers and consumption spending fj; hj;and a decrease in investment spending gj. When voter trust deteriorates, the marginal electoral return of public good spending decreases and voters reduce their support for candidates promising high public good spending. Government revenues are then reallocated away from public goods and toward transfers. When voter discounting increases, the marginal electoral return of long-term spending decreases and voters reduce their support for candidates promising high long-term spending. Government revenues are reallocated away from long-term spending and toward short-term spending. An increase in average voter income yincreases public consumption and investment spending, hj; gj, but may either increase or decrease transfers fj. When average income y increases, this has two opposite-sign e¤ects on transfers: on the one hand, higher tax revenues increase parties’incentives to provide more of all types of spending, including transfers; on the other hand, higher average income reduces the marginal utility of transfers, which gives parties an incentive to reallocate spending away from transfers. Another way to look at spending composition is through spending ratios: transfer to public good spending and short-term to long-term spending. The expressions for these key ratios depend on the exogenous parameters and ; and on average income y. Based on Proposition 4, we note that in the …rst ratio, transfers to public goods, the numerator increases when voter trust is lower, as the share of transfer spending fj= yin the government budget goes up. Also, in both ratios, transfers to public goods and short-term to long-term spending, the denominator decreases when voter discounting is higher, as the share of investment spending gj= yin the government budget goes down. The results of this analysis are stated in the following proposition. Proposition 5 (i) A decrease in voter trust leads to an increase in the ratio of transfers to public goods, fj hj+gj. (ii) An increase in voter discounting leads to an increase in the ratios of transfers to public goods and of short-term to long-term spending, fj hj+gj;fj+hj gj. The e¤ect of average income yon the two spending ratios depends on the parameters. This is because average income a¤ects the incentives of parties to provide transfers fjin both positive and negative directions, as discussed above. 14 3 Empirical Strategy The theoretical model presented above yields predictions about di¤erences in spending preferences among voters, and di¤erences in spending outcomes among countries. We document these associations using both individual and aggregate level data. Individual-level preferences for public spending come from an original survey we designed and …elded in seven Latin American capital cities. Country-level spending outcomes come from existing annual data for a panel of 18 Latin American countries during the period 1995-2018. Below we provide additional details about sample selection, the construction of the key variables, and empirical models used. 3.1 Sample Selection The micro evidence comes from an original survey of individual-level preferences for public spending. We developed a survey instrument in collaboration with the Latin American Public Opinion Project (LAPOP) at Vanderbilt University to elicit spending policy preferences and individual characteristics of citizens from seven Latin American countries: Chile, Colombia, Honduras, Mexico, Panama, Peru, and Uruguay. Below we refer to this survey as the IDB-LAPOP Survey. The survey was …elded between August-October 2017 in the capital cities of the seven countries. The sample consists of 6,040 respondents of voting age interviewed in households located in the metropolitan area of each country’s capital city. In each country, the sample was determined through a multi-stage strati…ed probabilistic design to achieve representativeness of the voting population of each capital city’s metropolitan area. The target geographic area is …rst strati…ed by metropolitan region, e.g., northern areas, western areas, followed by sub-strati…cation by electoral district, and then by block. Households were then selected from within blocks to obtain an age and gender distribution corresponding to the sampling frame. A single individual from each household responded to the interviewer’s questions. The questions evaluated respondent preferences with respect to funding for education, public safety, aid to the poor, and red tape. The survey also measured personal characteristics such as trust, risk aversion, and patience.5 The target sample size was 900 interviews for each country. Table 1 reports the …nal sample size achieved in each country. The target sample size was attained and sometimes 5The language used in the interviews was Spanish. Data collection was done through hand-held electronic devices for all the countries surveyed and for all interviews. The survey was pre-tested in July-August 2017 to re…ne question wording and instrument ‡ow. 15 exceeded in all countries, except in Mexico, where only 569 interviews were collected.6As Table 1 indicates, each capital city can be divided into areas and further subdivided into neighborhoods; we rely on variation within these geographic subdivisions to estimate di¤erences in spending preferences. The macro evidence on actual government spending allocations comes from existing data on public opinion and …scal outcomes. In assembling a panel of countries, the main constraint was …nding the largest group of countries for which a long enough time series of citizen trust in political institutions was available. The longest such survey of public opinion is Latinobarometro, which asked a consistent set of questions about trust in democratic institutions since the survey began in 1995. Latinobarometro surveyed the same 18 countries throughout the various waves of their survey; these countries are listed in Table 2. Note that the seven countries in our IDB-LAPOP survey are a subset of this larger Latinobarometro country sample. In each country Latinobarometro interviews nationally representative samples of voting-age citizens. Historical data on public spending are available for most of the world’s countries, including Latin America, from the IMF’s World Economic Outlook (WEO), and we match these data to the Latinobarometro panel. The time series of spending outcomes vary in length from country to country, but for most of our Latin American sample the data begin in the 1990s or earlier. Overall, the country and year selections allows for a maximal sample size of 432 country-years. 3.2 Variable Construction and Empirical Speci…cations Here we present the main variables used in the empirical analysis and explain how they relate to the theory model’s constructs. A full list of variable de…nitions is available in Section B2 of Appendix B. We also propose empirical speci…cations that provide statistical tests for the theoretical predictions. 3.2.1 Individual-Level Data The key preference variables are individual-level measures of trust and discounting, corresponding to the parameters iand iin the model. Trust in our analysis refers to voter con…dence that spending will translate into promised welfare gains. We measure trust in elec- 6The reason was a magnitude 7.1 earthquake that struck the capital city on September 19, 2017, and caused signi…cant damage to the southern portion of Distrito Federal and outlying towns. This caused the team to suspend …eldwork in Mexico as the new conditions on the ground made it infeasible to implement the original sampling design. 16 toral promises through answers to the following survey question: "Thinking about politicians in general, do you consider it very, somewhat, not very, or not at all, common that they keep their promises?" The four answer options, contained in the body of the question, yield a discrete variable with four categories. Based on these, we de…ne a numeric variable named Mistrustic, for individual iin city c; that takes four equidistant values in the unit interval [0,1], with 1 indicating the highest level of mistrust in politician promises. As shown in Table 3 of summary statistics, on average across the seven-city sample, mistrust in politician promises is high, namely 0.707. To measure discount rates, we implement a version of the unfolding brackets method, which Falk et al. (2018) tested in the lab and adapted into a survey module.7Each respondent is presented with a sequence of binary choices as follows. First, each respondent has to choose between receiving (the local currency equivalents of) 100 dollars today or 154 dollars in 12 months. If the respondent prefers the immediate payment, then in the next question the delayed payment is increased to 185 dollars. If the respondent prefers the delayed payment, then in the next question the delayed payment is decreased to 125 dollars. The choices are adjusted up or down, in the same fashion based on earlier responses, until a total of …ve sequential choices have been made. That produces a choice tree with 32 di¤erent terminal branches, with delayed payouts ranging from 103 to 215 dollars and arranged in increasing order of impatience. We de…ne a variable Impatienceic = 11 1+ric ;based on the interest rate ric implicit in the choices between monetary payo¤s. This results in a continuous measure with mean 0.435, a rather high level of impatience.8 The questions eliciting spending preferences targeted two policy areas that most respondents consider top priorities for their governments, namely education and security. Out of a list of 38 problems facing their countries, education and security were in the top three most frequently selected issues. For both of these policy areas, we designed questions to capture the two key tradeo¤s highlighted by the model: transfers versus public goods, and short-term versus long-term spending. We also included questions that measure the more narrow tradeo¤s between transfers and public consumption, and consumption versus investment spending; see Chart 1 in the theory section above. In almost every case, the questions follow a common structure: the respondent is presented with two policies, A and B, as alternative solutions to a certain policy challenge. Then, the respondent is asked which policy 7We use a similar strategy to develop a measure of risk aversion, discussed in the next section, for use in the falsi…cation analysis reported in Appendx B. 8Nonresponse was a relatively more severe problem when measuring impatience than measuring mistrust. Part of the reason was the more complex structure of the question. 17 option they prefer, A, B, or no preference between A and B.9 Individual preferences for transfers versus public good spending (fvs h+g) were elicited by o¤ering a choice between the following policy alternatives. For education: "lowering taxes so families have more money to spend on their own education, and raising taxes on all products that people buy so the government can invest more in education." For security: "lowering taxes so families have more money to spend on their own security, and raising taxes on all products that people buy so the government can invest more in police and the judicial system." Lower taxes increase disposable income and, as discussed in the model section, are equivalent to an income transfer, with predictable welfare bene…ts; thus they represent an increase in f. The alternative policy is higher taxes for education or security spending; the language of the question is generic, allowing for both short-term hand long-term gspending in these areas. We de…ne two dummy variables Transf_v_Educic and Transf_v_Securic that take the value one for respondents that prefer more disposable income to government-provided public goods, or are indi¤erent, and take the value zero for those that strictly prefer government provision. The summary statistics in Table 3 show that 75.7 percent of respondents prefer transfers to public education provision and 60.6 percent prefer transfers to public security provision. Individual preferences for short-term versus long-term spending (f+hvs g) are measured through questions that o¤er a choice between a policy with more immediate payo¤s, more government spending on hiring additional sta¤, versus policies with future payo¤s, such as training government workers.10 Hiring and training di¤er unambiguously in the timing of bene…ts: compared to hiring additional employees, training current employees will yield ben- e…ts after the training is completed. To emphasize this, a prior question primed respondents to think that training takes at least two years. Hence, respondents with higher discount rates should unambiguously prefer hiring. Hiring and training also di¤er in the predictability of bene…ts. Voters should be familiar with current teacher and police performance and can infer that, to some extent, hiring more would provide more of the same bene…ts. Thus, hiring can be thought as a convex combination of fand h;training has both future and less certain 9In each case, there is also the possibility of responding "I don’t know." In the few cases in which no answer is recorded, that appears in the data with a "No response" code. 10For the area of security, the question was: "Now suppose that the government is evaluating two options to …ght crime. Option A is to increase the number of police, hiring police who are the same as the current police, with the same characteristics and the same salaries. Option B would be to maintain the same number of police, but increase their training and salaries, and replacing those with poor performance. The two cost the same, but the government can only do one of them. Which option would you prefer?" 18 bene…ts and thus should behave like g:11 A related question sheds light on the same tradeo¤ in the context of education, though less directly. As a last question in the education module, we ask respondents whether they believe teacher training improves student learning.12 Since respondents have just been asked to choose between lower taxes and education spending, they are primed to think of the e¢ cacy of teacher training in terms of support for funding teacher training relative to other education inputs. Assuming that those who more strongly believe that teacher training bene…ts students also tend to prefer spending on teacher training versus other education spending, this question is a reasonable proxy for preferences for long-term public good spending g relative to short-term education expenditures, f+h: We de…ne two dummy variables Short_v_Long_Educic and Short_v_Long_Securic that take the value one for respondents who prefer hiring to training government workers, or are indi¤erent, and take the value zero for those who strictly prefer training. The summary statistics in Table 3 show that 22.6 percent of respondents prefer short-term to long-term spending (hiring over training) in education and 45.2 percent prefer short-term to long-term spending on security.13 For additional validation, we also included two questions that capture the tradeo¤s between public consumption versus investment spending (hvs g), and transfers versus public consumption (fvs h). The …rst tradeo¤ is framed in the context of education. The respondent is given a choice between two policies: "the government purchases tablets that the students would receive immediately," or the government improves teacher training which "would take two to four years until the teachers are better trained." The key di¤erence between these two alternatives is the time horizon of educational bene…ts, immediately versus several years; respondents’ assessment of the welfare impact should be uncertain in both cases as it depends on how politicians implement these new policies.14 The dummy 11The alternatives may, however, also be viewed as providing di¤erent inputs into the production of a public service: one seeks to boost the quantity of the input, the other its quality. We have no data on respondent opinions on whether quantity or quality has a larger impact on service delivery, independent of trust and discounting issues. Although we are not aware of any theory that suggests such unobserved opinions would be correlated with trust and discounting, we cannot exclude this possibility. 12The question was: "To what extent do you believe that if government gave teachers more training, this would have an important e¤ect on the educational performance of children?" 13The order in which the policy options are presented in the survey does not necessarily correspond to the way we code the responses into dummy variables, namely indicating preference for transfers, respectively short term policies. This alleviates the possibility that respondents may have a tendency to consistently pick the …rst (or the last) option presented. Even so, as long as this tendency is not correlated with mistrust and impatience, any bias would a¤ect levels and not marginal e¤ects. 14Respondents may di¤er in expectations about the relative e¢ cacy of tablets and teachers for student 19 Tablet_v_Teacheric takes the value one if the respondent prefers tablets or is indi¤erent, and the value zero if the respondent strictly prefers teacher training. The mean of this variable shows that 35.5 percent prefer the tablet option. The second tradeo¤ is framed in the context of crime reduction policies. Respondents choose between subsidies to citizens for private security and spending for police resources.15 Both options emphasize speci…c, short-term steps to reduce crime. The di¤erence is that subsidies have a clear monetary payo¤, hence designated as f, whereas increasing police resources, by being less explicit about the use of those resources, introduces uncertainty about the actual bene…ts of the spending policy, designated as h. The dummy Subsidy_v_Policeic takes the value one if the respondent prefers subsidies for privately contracting security or is indi¤erent, and the value zero if the respondent strictly prefers resources given to police. A fraction of 46.2 percent of respondents prefer receiving the subsidy to allocating more money to the police. The relationships between the six spending preference variables described above, and voter mistrust and impatience, are presented graphically in Figure 1 in Appendix A. Each graph plots the raw mean of each variable conditional on increasing levels of mistrust and impatience. The patterns are broadly consistent with the model predictions from Chart 1 above. The average preference for transfers over public goods is increasing in both mistrust and impatience, and does so almost monotonically. The average preference for short over long-term spending increases in impatience, but not in mistrust, again in line with the model predictions. Finally, in the third panel, the average preference for consumption over investment spending (the darker lines referring to education policies) decreases in mistrust and increases in impatience, while the average preference for transfers over public consumption spending (the lighter lines referring to security policies) appears to increase in mistrust at the upper end, but does not change visibly with impatience. Below we explore these patterns in regression models that control for potential confounders, such as age and education, and estimate di¤erences based on variation within more narrow geographic units, such as city neighborhoods. Speci…cally, we estimate linear learning. While we view it as implausible that unobserved opinions about relative e¢ cacy are correlated with individual discount rates, we cannot exclude this possibility. 15The question was: "Imagine that the government has two options of dealing with security. Option A assigns more resources to the police so that they can do a better job …ghting crime throughout the city. Option B gives subsidies to citizens and neighbors so that they can …ght crime by contracting guards and putting up security cameras on their blocks. The government cannot do both things. Which option do you prefer?" 20 probability models of the following form: yic =0+1Mistrustic +2Impatienceic +0xic +cs+uic (9) where yic is a spending preference dummy variable, for individual iin city c,xic is a vector of individual-level covariates, csare …xed e¤ect indicating geographic subdivisions of a city, and uic is the error term. The main parameters of interest are 1; 2which measure the change in the probability of supporting a given type of spending that is associated with an increase in individual mistrust and impatience, respectively. Note that while we do not have randomized variation in these two key preference factors, the strategy we use to elicit spending preferences through hypothetical policy comparisons alleviates one important type of endogeneity, namely the potential for reverse causation, or simultaneity bias, to a¤ect the results. That would be a concern when using observational data on actual policies, whose observed quality can in‡uence the level of trust or impatience of the electorate. The speci…cation in equation (9) provides direct tests for the model’s predictions. We supplement this approach with a more indirect strategy. Namely, we experimentally shift the preference parameters and of the model, which control the intensity with which individual trust iand discounting ia¤ect spending preferences. In the …rst experiment, tied to education policies, respondents are randomized with equal probability to receive the following informational message before being asked to pick one of two policy choices, buying electronic tablets for students and improving teacher training: "Studies indicate that having better teachers is key to improving student learning, but no studies indicate that the use of tablets does." We interpret this treatment as a shock to the ine¢ ciency parameter : The treatment thus should reduce the level of individual support for the policy whose ine¢ ciency has been increased relative to the policy whose ine¢ ciency parameter has remained unchanged. In addition, the average treatment e¤ect in individual support for the a¤ected policy depends on the level of individual trust i. It should be lower for individuals with lower levels of trust, because the change in the marginal utility of current spending @Wii @hi= 1 (1 i+i)hi, namely @2Wii @hi@ =ihi;is decreasing in trust. In other words, lower-trust individuals should be less likely to switch their preference away from the ine¢ cient policy. The second experiment is related to security policies. The choice is between two policies for crime reduction: contracting more police, and improving the quality of police through training. We have seen this choice before as one between short-term and long-term spending, 21 We develop a risk aversion measure using a survey module similar to the one we use to measure patience, where binary choices are o¤ered sequentially between a lottery and a sure payment, yielding 32 levels of risk aversion. The falsi…cation tests show that the substitute variables always have smaller coe¢ cient estimates than the original variables.22 Randomized Treatments. We now discuss the evidence coming from the randomization of preference parameters that control the intensity with which trust and patience a¤ect spending preferences. In the theory model these are denoted by (policy ine¢ ciency) and (time horizon). In the education experiment, we generated randomized variation in the ine¢ ciency of a current spending policy by providing an informational message to treated respondents that the e¤ectiveness of technology in the classroom has not been backed by evidence, whereas the e¤ectiveness of well-trained teachers has. Table 7 shows that the average treatment e¤ect is a 3.3 percentage point decrease in support for spending on technology in the classroom, versus spending on teacher training. In columns (3)-(4) we …nd that the treatment e¤ect is more pronounced for high-trust respondents. High-trust respondents should be more sensitive to shocks to because they have more certainty that the government will deliver on its promises; hence, they are more likely to adjust their preferences away from the less e¤ective policy. Low-trust respondents have a lower utility gain from switching because they attach a lower value to both policies. At the bottom of the table we present p-values for an F-test of coe¢ cient equality of the two interaction terms. The test rejects equality especially in the last two columns where we introduce covariates. Among the other coe¢ - cients, we note the one for L Trust, which indicates that untreated low-trust respondents are less supportive of spending on technology, in line with the theory model and the result from Table 6 on the hvs. gtradeo¤. In the security experiment we generated randomized variation in the time horizon of policy bene…ts in the context of a choice between current spending on hiring additional police and investment spending on training existing police. For the treatment group, indicated by the dummy Treatm, the policy bene…ts of police training are realized in two years. For the control group the policy bene…ts of police training are realized in four years. Thus, we would expect the treatment, by reducing the time horizon for the investment option, to cause a shift away from the current spending on hiring and towards investment spending on training. The …rst two columns of Table 8 indeed show a negative average treatment e¤ect of 6.1 percentage points. Columns (3)-(4) estimate the treatment e¤ect separately 22The sample correlation between Mistrust and Interpers is 0.156, and the sample correlation between Impatience and Risk Avers is 0.161. 28 for respondents with extreme discount factors (very high or very low), and respondents with moderate discount factors. As we argued above, the treatment e¤ect should be more pronounced for the middle group because the di¤erence in discounted utilities is maximized at a moderate level of impatience; see Figure 2. The estimates show that the treatment e¤ect is about twice larger for moderate discount respondents. At the bottom of the table, the p-values for the F-test of coe¢ cient equality of the two interaction terms reject equality at conventional signi…cance levels in models (5)-(6) that include covariates. 4.2 Country-Level Results The theory model predicts that voter preferences translate into spending outcomes through the mechanism of electoral competition. The candidates who get elected to choose the actual spending policies re‡ect the spending biases of the electorate. Proposition 5 derived testable hypotheses about the relationship between aggregate trust and aggregate discounting, on the one hand, and government spending ratios: transfers versus public goods and short-term versus long-term spending. Here we examine these hypotheses empirically using the panel of 18 Latin American countries that we constructed for the period 1995-2018. The results are reported in Table 10; see also Figure 4 for a time series plot of the raw data on the key pairs of variables. All speci…cations include country and year …xed e¤ects and report robust standard errors. This strategy controls for time-invariant di¤erences among countries, such as culture and institutions, as well as common time shocks, such as international commodity price movements. The …rst three columns show that countryyears with higher aggregate mistrust in political parties have on average higher ratios of transfers to public goods. Also, country-years with higher real interest rates, a proxy for aggregate impatience, have on average higher ratios of transfers to public goods. We also note that wealthier and larger countries tend to spend less on transfers relative to public goods compared to poorer and smaller countries. Years of economic slowdown when the unemployment rate rises tend to be years when the share of transfers in the budget increases. The last three columns of Table 10 show that in the models explaining variation in short versus long-term spending, aggregate mistrust and impatience also have positive coe¢ cients. Higher GDP per capita is associated with lower short-term spending relative to long-term spending, and years with high unemployment tend to be years with high ratios of short-term to long-term spending.23 We perform a falsi…cation exercise that mirrors the one we carried 23Ardanaz and Izquierdo (2017) …nd that this kind of adjustment in developing countries is driven by a 29 out for the individual-level data. We replace party mistrust with interpersonal mistrust, and the real interest rate with the interest rate spread, as a proxy for risk aversion. The results are in Tables B6-B7 of Appendix B. We observe that the substitute variables have less or no explanatory power compared to the original variables. 5 Conclusion Conventional explanations for ine¢ ciently low levels of public good, and particularly investment, spending in developing countries largely rely on supply-side arguments where government actors or institutions fail to supply the types of spending that most voters want. In this paper we provide a demand-side mechanism that complements these existing hypotheses. We argue that voters rationally internalize the uncertainty of electoral promises when expressing their spending preferences at the ballot box. Unreliable electoral promises lower the returns that low-trust voters expect from public good spending. In addition, high discounting of spending whose bene…ts occur in the future reduces the demad for long-term spending. Therefore voters prefer political candidates who promise lower levels of public goods and long-term spending, and instead promise more certain and immediate forms of spending such as monetary transfers. To test these arguments we designed an original survey of spending preferences that uses a revealed preference design in the context of budgetary tradeo¤s. Our key explanatory variables are trust in politician promises and discounting of economic bene…ts. We …nd that lower-trust respondents are more supportive of transfers than public goods, and higherdiscount respondents are more supportive of short-term spending than investment spending. These …ndings appear in two di¤erent high-salience policy domains in Latin America, education and security. We also study responses to a randomized informational message about the relative e¢ ciency of two policies, and …nd that low-trust respondents are less willing to switch their choices from the ine¢ cient policy to the relatively more e¢ cient one. When we experimentally shorten the time horizon for returns on investment spending, we …nd that extreme-discount respondents are less likely to change their choice from a short-term policy to a long-term policy. We supplement the micro evidence with country-level data on spending outcomes, which con…rms the individual-level patterns. Country-years with lower aggregate trust in political parties and higher real interest rates tend to be those with high ratios of transfers to public goods and current to capital spending. cutback in capital spending. 30 Our approach to explaining spending preferences and outcomes could be extended to the revenue side of the government budget, including public borrowing and debt, which also have …rst order implications for long-term development. From a policy perspective, an important avenue for future research is understanding the determinants of low trust in electoral promises and the factors that could mitigate it. Our survey data suggests that some of these determinants operate at the individual voter level, while others may be more systemic. Perhaps the …rst type re‡ects lack of accurate information and could warrant behavioral interventions. The systemic component seems more complex. It may be rooted in social norms shaped by historical experiences (Nunn and Wantchekon, 2011). Alternatively, it may be endogenous to government performance: Low trust gives political candidates electoral incentives to adopt ine¢ cient policies, or leads to self-selection into politics of candidates who prefer ine¢ cient policies,24 and low government performance in turn reinforces voters’low trust. Understanding why elections perpetuate this feedback loop, and what reforms are necessary to undo it, remain important research questions. 24See Keefer, Scartascini, and Vlaicu (2020) for an application of this point to the issue of populism in policymaking. 31 Appendix A A1. Mathematical Proofs Proof of Proposition 1. Individual-level spending preferences are characterized by the necessary and su¢ cient conditions in equation (4) and the binding budget constraint fi+hi+gi=y; for each voter i2[0;1] :(i) First consider a decrease in voter i’s trust i: This reduces voter i’s marginal utilities from consumption and investment spending. The following will show that the new equilibrium should be characterized by lower marginal utilities from all of the three types of spending. Suppose a reduction in trust leads to a reduction in voter i’s demand for transfer spending fi:Then, the marginal utility of transfers 1[(1 )yi+fi]increases, and with it the marginal utilities of consumption and investment spending, respectively. Since (1 i+i)is now higher, it follows that both hiand gihave to be lower. Since transfers fiare also lower, that violates the binding budget constraint condition. This contradiction implies that in the new equilibrium fihas to be higher. By the binding budget constraint condition, it follows that (hi+gi)has to be lower. Suppose hi remains at least as high when iis lower. Then gihas to be lower. Di¤erentiating the second equilibrium condition in equation (4) with respect to i, we have (1 )hhi1 (1+i)gii= (1 i+i)h@ @ihi1 (1+i) @ @igii, which is negative since @ @ihi0<@ @igiand  < 1:However, the same second equilibrium condition can be rearranged into 11 (1+i)= (1 i+i)hhi1 (1+i)gii, which implies that hi1 (1+i)gi>0:This contradiction leads to the conclusion that hiis lower when iis lower. (ii) Di¤erentiating the …rst equilibrium condition in equation (4) with respect to iwe have that @ @ifi= (1 i+i)@ @ihi;which means that fiand hichange in the same direction when the discount rate ichanges. Suppose that when the discount rate iincreases, fiand hiboth decrease. From the equilibrium condition 1(1 i+i)hi=1 (1+i)[1 (1 i+i)gi]an increase in iand a decrease in hiimply a decrease in gi:However, from the binding budget constraint condition it has to be that giincreases when fi,hidecrease. The contradiction implies that when the discount rate iincreases, fiand hiboth increase. By the binding budget constraint condition, gihas to decrease. A similar strategy of proof delivers comparative statics with respect to voter income yi:Di¤erentiating the second equilibrium condition in equation (4) with respect to yigives @ @yihi=1 (1+i) @ @yigi, which shows that hiand gimove in the same direction as yigoes up. Suppose that both hiand gidecrease. By the binding budget constraint condition, fihas to 32 increase. This implies that voter i’s marginal utility of transfers 1[(1 )yi+fi]decreases. Then, so do the marginal utilities of consumption and investment spending, contradicting the assumption that hiand gidecrease. The implication is then that both hiand giincrease when yiincreases. The binding budget then implies that fidecreases when yiincreases. Proof of Proposition 2. From part (i) of Proposition 1, fiis higher when trust iis lower, while hiand giare both lower. It follows that when trust is lower, the ratios fi hi+gi and fi hiare higher. From the equilibrium equations in (4), hiand gican be expressed in terms of fias follows: hi=(1)yi+fi 1i+iand gi=(1+i)[(1)yi+fi1]+1 1i+i. These expressions imply that hi gi=(1)yi+fi (1+i)[(1)yi+fi1]+1 . Di¤erentiating with respect to trust, it follows that sgn h@ @ihi gii =sgn n@ @ifih1(1 + i)io >0, since @ @ifi<0and (1 + i)>1: Therefore the ratio hi giis lower when trust is lower. From part (ii) of Proposition 1, fiand hi are increasing in discounting i;while giis decreasing in discounting. Given that the budget constraint is binding, fi+hi+gi=y; the fact that fiis increasing implies that hi+giis decreasing. Therefore. fi hi+gi,fi+hi gi;and hi giare all higher when discounting iis higher. The same conclusions follow from an increase in the time horizon . Proof of Proposition 3. We …rst derive parties’winning probabilities as a function of voters’utilities. From the perspective of the parties, the probability that voter ivotes for party Ais Pfi< WiA WiB g=1 2+ (WiA WiB)as the cdf of iis Fi(x) = 1 2+x: Because the population of voters is large and partisan biases are iid, by the law of large numbers party A’s vote share in the population equals the average probability of support across voters. A() = 1 2+Z1 0 (WiA WiB)di (13) and B()=1A():The winning probability of party Ais the probability that its vote share exceeds a half: PA()>1 2=P < Z1 0 (WiA WiB)di =1 2+ (WAWB)(14) as the cdf of is F(x) = 1 2+ x: Also, PB()>1 2=1 2+ (WBWA);where for j=A; B we denote WjR1 0Wijdi the aggregate policy utility provided by party j’s policies. An equilibrium exists because the objective functions are jointly continuous in both parties’strategies, and concave in a party’s own strategy. Equilibrium uniqueness follows 33 from the strict concavity of parties’ objectives in own strategies; see equation (2). In a pure-strategy equilibrium parties adopt symmetric strategies and thus each party has onehalf winning probability PA()>1 2=PB()>1 2=1 2:The necessary and su¢ cient equilibrium conditions are in equation (8) in the main text. Solving …rst for the social planner allocation using equation (6) and budget balance, we have fsp =2 2+yand hsp =gsp =1 2+y: The electoral equilibrium variables solve the necessary and su¢ cient equilibrium conditions in equation (8) together with the binding budget constraint:To show that fj> fsp we start by noticing that fsp =fjj=1;=0:Thus, showing that fjis strictly decreasing in and strictly increasing in is su¢ cient to prove the claimed inequality fj> fsp:Suppose an increase in trust leads to an increase in party j’s transfer spending fj:Then, the marginal utility of transfers 1[(1 ) y+fj]decreases, and with it the marginal utilities of consumption and investment spending, respectively. Since (1 +)is now lower, it follows that both hjand gjhave to be higher. Since transfers fj are also higher, that violates the binding budget constraint. This contradiction implies that in a higher-trust equilibrium fjhas to be lower. Di¤erentiating the …rst equilibrium condition in equation (8) with respect to , we have that @ @ fj= (1 +)@ @ hj;which means that fjand hjchange in the same direction when the discount rate changes. Suppose that when the discount rate increases, fjand hjboth decrease. From the equilibrium condition 1(1 +)hj=1 (1+)[1 (1 +)gj]an increase in and a decrease in hjimply a decrease in gj:As the budget constraint binds, it has to be that gjincreases when fjand hj both decrease. The contradiction implies that when the discount rate increases, fjand hj both increase. To show that hj> gj;notice that from the second equilibrium conditions in equation (8) we have 1(1 +)hj=1 (1+)[1 (1 +)gj]<1(1 +)gj because  > 0. To show that gj< gsp, suppose to the contrary that gjgsp:Then, since hj> gj;it follows that hj+gj>2gj2gsp =hsp+gsp;which because the budget constraint is binding implies that fjfsp. This contradicts the established inequality fsp < fj:Therefore, it has to be that gj< gsp: Proof of Proposition 4. Party spending promises are characterized by the necessary and su¢ cient conditions in equation (8) and the binding budget constraint fj+hj+gj= y; for j=A; B: (i) First consider a decrease in voter trust : This reduces marginal welfare from consumption and investment spending. Suppose a reduction in trust leads to a reduction in party jpromises of transfer spending fj:Then, the marginal utility of transfers 1[(1 ) y+fj]increases, and with it the marginal utilities of consumption and investment 34 spending, respectively. Since (1 +)is now higher, it follows that both hjand gjhave to decrease. Since transfers fjalso decreased, that violates the binding budget constraint condition. This contradiction implies that in the new electoral equilibrium fjhas to be higher. By the binding budget constraint condition, it then follows that public good spending promises (hj+gj)have to be lower. Suppose hjremains at least as high when is lower. Then gjhas to be lower. Di¤erentiating the second equilibrium condition in equation (8) with respect to , we have (1)hhj1 (1+)gji= (1 +)h@ @ hj1 (1+) @ @ gji,which is negative since @ @ hj0<@ @ gjand  < 1:However, the same second equilibrium condition can be rearranged into 11 (1+)= (1 +)hhj1 (1+)gji, which implies that hj 1 (1+)gj>0:This contradiction leads to the conclusion that consumption spending hjis lower when trust is lower. (ii) Di¤erentiating the …rst equilibrium condition in equation (8) with respect to we have that @ @ fj= (1 +)@ @ hj;which shows that fjand hjchange in the same direction when the discount rate changes. Suppose that when the discount rate increases, fjand hjboth decrease. From the equilibrium condition 1(1 +)hj=1 (1+i)[1 (1 +)gj]an increase in and a decrease in hjimply a decrease in gj:As the budget constraint binds, it has to be that gjincreases when fjand hjboth decrease. The contradiction implies that when the discount rate increases, fjand hjboth increase; therefore gjhas to decrease. One can also derive comparative statics with respect to aggregate voter income y: Differentiating the second equilibrium condition in equation (8) with respect to ygives @ @yhj= 1 (1+) @ @ygj, which shows that hjand gjmove in the same direction as ychanges. Suppose that hjand gjboth decrease when yincreases. Then, by the binding budget constraint condition, fjhas to increase. This implies that the marginal welfare of transfers, 1[(1 ) y+fj] decreases. Then, so do the marginal utilities of consumption and investment spending with respect to y. But that contradicts the presumption that hjand gjdecrease. The implication is then that both hjand gjincrease when yincreases. As far as transfer promises fi;these may either decrease or increase when yincreases, as the budget constraint is relaxed. Proof of Proposition 5. The comparative statics for the spending ratios with respect to voter trust and voter discounting follow directly from the comparative statics for spending levels from Proposition 4, which showed that lower trust is associated with higher equilibrium transfers fjand lower public good spending hj+gj;thus the ratio fj hj+gj is higher, while higher discounting is associated with higher transfers fjand consumption spending hjand lower investment spending gj;thus the ratios fj hj+gj;fj+hj gjare higher. 35 A2. Figures Figure 1. Spending Preferences 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 Percent Support 1 2 3 4 Mistrust Educ Secur Transfers v Public Goods 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 Percent Support 1 2 3 4 Impatience Educ Secur Transfers v Public Goods 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 Percent Support 1 2 3 4 Mistrust Educ Secur Short v Long Term 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 Percent Support 1 2 3 4 Impatience Educ Secur Short v Long Term 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 Percent Support 1 2 3 4 Mistrust Educ Secur Other Preferences 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 Percent Support 1 2 3 4 Impatience Educ Secur Other Preferences Note: Data are from the 2017 IDB-LAPOP Survey. Figures plot mean support for di¤erent types of spending in education and security, conditional discrete levels of mistrust and impatience. The corresponding regression results are in Tables 4, 5, and 6, respectively. Ranges around conditional means are 95 percent con…dence intervals. Discrete categories for mistrust re‡ect the original survey coding. Discrete categories for impatience are: 1 (0-.15), 2 (.15-.30), 3 (.30-.45), 4 (.45-.55). 36 Figure 2. Discounted Utilities 0 .2 .4 .6 .8 1 0.05 .1 .15 .2 .25 .3 .35 .4 .45 .5 .55 Impatience Quadratic (2 yrs) Quartic (4 yrs) Note: The lower cuto¤ corresponds to a survey measure of discounting between 5-6, or an average future payout of 117.5 (interest rate 17.5 percent). The upper cuto¤ corresponds to a survey measure of discounting between 16-17, or an average future payout of 156 (interest rate 56 percent). Impatience is de…ned as 1-1/(1+r). Figure 3. Income Reported and Income Estimated 0 5 10 15 Percent -3 -2 -1 0 1 2 3 4 5 income_std 0 5 10 15 Percent -3 -2 -1 0 1 2 3 4 5 incomePMT_std Note: Figures plot the histograms of income reported in the IDB-LAPOP survey, on the left, and income estimated using Proxy Means Testing models for each country, on the right. Both income measures are standardized within each country. Sample sizes are 5,106 and 5,960, respectively. 37 Table 7. Education Experiment Dep Var: Tech v Teach (1) (2) (3) (4) (5) (6) Treatm .033*** .033***     (.010) (.010) Treatm L Trust   .022** .018 .018 .011 (.011) (.012) (.013) (.013) Treatm H Trust   .076*** .070** .091*** .081*** (.025) (.027) (.029) (.027) L Trust   .086*** .082*** .085*** .090*** (.020) (.020) (.022) (.021) Impatience     .026 .009 (.039) (.037) Income     .066** .015 (.027) (.033) Education     .010*** .011*** (.002) (.002) Age     .001*** .001** (.000) (.000) Male     .064*** .051*** (.012) (.011) Fixed E¤ects city area area nghbhood nghbhood nghbhood Clusters    236 235 236 Equal Test (p-val)   .053 .064 .016 .016 Obs 5,989 5,989 5,945 5,945 4,664 5,366 Note: Data are from the 2017 IDB-LAPOP Survey. The treatment Treatm is a randomized informational message about the relative ine¤ectiveness of technology in education, compared to teaching. The dependent variable is a dummy variable indicating a preference for government spending on teaching technology over teacher training. Robust standard errors in parentheses, clustered at the neighborhood level where clusters are indicated. Column (5) uses reported income; column (6) uses estimated PMT income. * p <0.10, ** p<0.05, *** p <0.01. 44 Table 8. Security Experiment Dep Var: Hire v Train (1) (2) (3) (4) (5) (6) Treatm .061*** .061***     (.012) (.012) Treatm E Disc   .064*** .065*** .060*** .065*** (.014) (.013) (.015) (.014) Treatm M Disc   .114*** .125*** .135*** .132*** (.033) (.031) (.034) (.031) E Disc   .061** .043** .001 .017 (.025) (.022) (.023) (.021) Mistrust     .017 .000 (.025) (.023) Income     .097*** .174*** (.027) (.043) Education     .014*** .013*** (.002) (.002) Age     .000 .001 (.000) (.000) Male     .061*** .061*** (.014) (.013) Fixed E¤ects city area area nghbhood nghbhood nghbhood Clusters    236 235 236 Equal Test (p-val)   .160 .072 .036 .045 Obs 5,919 5,919 5,449 5,449 4,643 5,326 Note: Data are from the 2017 IDB-LAPOP Survey. The treatment Treatm is a randomized contextual message indicating a two-year time horizon of achieving a given crime reduction with police training; the control group has a four-year time horizon. The dependent variable is a dummy variable indicating a preference for government spending on hiring more police over training the existing police. Robust standard errors in parentheses, clustered at the neighborhood level where clusters are indicated. Column (5) uses reported income; column (6) uses estimated PMT income. * p <0.10, ** p <0.05, *** p <0.01. 45 Table 9. Summary Statistics: Country-Level Data Obs Mean SD bw SD wn Min Max Transf v Publ Gds 337 0.663 0.087 0.053 0.326 0.868 Short v Long Trm 427 0.794 0.069 0.053 0.474 0.964 Mistrust 361 0.739 0.037 0.065 0.521 0.919 Real Int Rate 371 0.116 0.104 0.094 -0.353 0.939 Interpers 353 0.812 0.048 0.057 0.559 0.976 Interest Spread 371 0.120 0.088 0.065 0.014 0.634 GDP per Cap 432 10.802 4.567 2.149 2.913 22.874 Log Population 432 16.465 1.153 0.101 14.823 19.160 Unemployment 432 0.066 0.029 0.019 0.020 0.205 Election Year 432 0.229 0.036 0.419 0 1 Note: The unit of observation is a country-year. Statistics computed for the sample of eighteen countries included in the Latinobarometro surveys since 1995. Sample size di¤ers across variables due to unavailable data in some years. See Section B4 of Appendix B for variable de…nitions and measurement. 46 Table 10. Spending Outcomes Dep Var: Transf v Publ Gds Short v Long Trm (1) (2) (3) (4) (5) (6) Mistrust .166* .203*** .178** .157** .170** .159** (.085) (.077) (.090) (.069) (.067) (.080) Real Int Rate .071** .080** .078** .106*** .081** .079** (.032) (.039) (.038) (.031) (.038) (.038) GDP per Cap  .018*** .017***  .014*** .013*** (.003) (.003) (.003) (.003) Log Population  .306** .339**  .058 .102 (.132) (.136) (.129) (.135) Unemployment   .287*   .359** (.169) (.168) Election Year   .007   .010 (.006) (.006) Country FE 15 15 15 16 16 16 Year FE 21 21 21 21 21 21 Adj R-sq .750 .795 .797 .623 .664 .670 Obs 258 258 258 309 309 309 Note: Table presents least squares estimates of regression models with country and year …xed e¤ects. Sample period is 1995-2018. Robust standard errors in parentheses. * p <0.10, ** p <0.05, *** p <0.01. 47 References Acemoglu, D., and J.A. Robinson. 2006. Economic Origins of Dictatorship and Democracy. Cambridge, United Kingdom: Cambridge University Press. Algan, Y., P. Cahuc, and M. Sangnier. 2016. "Trust and the Welfare State: The Twin Peaks Curve." Economic Journal 126(593): 861-883. Ardanaz, M., and A. Izquierdo. 2017. "Current Expenditure Upswings in Good Times and Capital Expenditure Downswings in Bad Times? New Evidence from Developing Countries." Working Paper IDB-WP-838. Washington, DC, United States: Inter-American Development Bank. Banerjee, A.V. et al. 2011. "Do Informed Voters Make Better Choices? Experimental Evidence from Urban India." Working paper. Cambridge, United States: Massachusetts Institute of Technology. Besley, T., and T. Persson. 2009. "The Origins of State Capacity: Property Rights, Taxation, and Policy." American Economic Review 99(4): 1218–44. Besley, T. and T. Persson. 2011. Pillars of Prosperity: The Political Economics of Development Clusters, Princeton, NJ: Princeton University Press. Bursztyn, L. 2016. "Poverty and the Political Economy of Public Education Spending: Evidence from Brazil." Journal of the European Economic Association 14(5): 1101–1128. Camussi, S., A.L. Mancini, and P. Tommasino. 2018. "Does Trust In‡uence Social Expenditures? Evidence from Local Governments." Kyklos 71(1): 59-85. Dearmon, J., and R. Grier. 2011. "Trust and the Accumulation of Physical and Human Capital." European Journal of Political Economy 27(3): 507-519. Drazen, A., and M. Eslava. 2010. "Electoral Manipulation Via Voter-Friendly Spending: Theory and Evidence." Journal of Development Economics 92(1): 39-52. Engerman, S.L., and K.L. Sokolo¤. 2000. "History Lessons Institutions, Factor Endowments, and Paths of Development in the New World." Journal of Economic Perspectives 14: 217–232. Falk, A. et al. 2018. "Global Evidence on Economic Preferences." Quarterly Journal of Economics 133(4): 1645–1692. Francois, P., and J. Zabojnik. 2005. "Trust, Social Capital, and Economic Development." Journal of the European Economic Association 3(1): 51-94. Fujiwara, T. 2015. “Voting Technology, Political Responsiveness, and Infant Health: Evidence from Brazil.”Econometrica 83: 423-464. 48 Guiso, L., P. Sapienza and L. Zingales. 2004. "The Role Of Social Capital in Financial Development." American Economic Review 94(3): 526-556. Ismahan, M., and F.G. Ozkan. 2011. “The Political Economy of Public Spending Decisions and Macroeconomic Performance.”International Journal of Economic Perspectives 5(2): 163–174. Izquierdo, A., C. Pessino, and G. Vuletin, editors. 2018. Better Spending for Better Lives. Development in the Americas report. Washington, DC, United States: Inter-American Development Bank. Jensen, R. 2010. "The (Perceived) Returns to Education and the Demand for Schooling." Quarterly Journal of Economics 125(2): 515-548. Kaminsky, G., C. Reinhart, and C. Vegh. 2005. "When It Rains, It Pours: Procyclical Capital Flows and Macroeconomic Policies." In: M. Gertler and K. Rogo¤, editors. NBER Macroeconomics Annual 2004. Volume 19. Cambridge, United States: MIT Press. Keefer, P. 2007. “Clientelism, Credibility, and the Policy Choices of Young Democracies.” American Journal of Political Science 51: 804-821. Keefer, P., and R. Vlaicu. 2008. “Democracy, Credibility, and Clientelism.”Journal of Law, Economics, & Organization 24(2): 371–406. Keefer, P., and R. Vlaicu. 2017. “Vote Buying and Campaign Promises,” Journal of Comparative Economics 45(4): 773-792. Keefer, P., C. Scartascini, and R. Vlaicu. 2020. “Trust, Populism, and the Quality of Government.”In: A. Bågenholm et al., editors. Oxford Handbook of Quality of Government, forthcoming. Khemani, S. 2015. "Buying Votes versus Supplying Public Services: Political Incentives to Under-Invest in Pro-Poor Policies." Journal of Development Economics 117(C): 84-93. Kuziemko, I. et al. 2015. "How Elastic Are Preferences for Redistribution? Evidence from Randomized Survey Experiments." American Economic Review 105(4): 1478-1508. Lizzeri, A., and N. Persico. 2001. "The Provision of Public Goods under Alternative Electoral Incentives." American Economic Review 91(1): 225–39. Lucas, R.E. 1978. "Asset Prices in an Exchange Economy." Econometrica 46(6): 1429-45. Nunn, N.. and L. Wantchekon. 2011. "The Slave Trade and the Origins of Mistrust in Africa." American Economic Review 101(7): 3221-3252. Nunn, N., N. Qian, and J. Wen. 2018. "Distrust and Political Turnover." NBER Working Paper 24187. Cambridge, United States: National Bureau of Economic Research. Persson, T., and G. Tabellini. 2003. The Economic E¤ects of Constitutions. Cambridge, 49 United States: MIT Press. Robinson, J.A., and R. Torvik. 2005. “White Elephants,”Journal of Public Economics 89(2-3): 197-210. Robinson, J.A., and T. Verdier. 2013. "The Political Economy of Clientelism." Scandinavian Journal of Economics 115: 260-291. Scartascini, C., and W.M. Crain. 2002. "The Size and Composition of Government Spending in Multi-Party Systems." Working paper 02-23. Fairfax, United States: George Mason University. Shi, M., and J. Svensson. 2006. "Political Budget Cycles: Do They Di¤er Across Countries and Why?" Journal of Public Economics 90(8–9): 1367-1389. Zak, P.J., and S. Knack. 2001. "Trust and Growth." Economic Journal 111(470): 295- 321. 50 Appendix B (For Online Publication) B1. Tables Table B1: Summary Statistics: PMT Variables Obs Mean Std Dev Min Max Car 6,022 0.333 0.471 0 1 Motorbike 6,020 0.129 0.335 0 1 Fridge 6,022 0.927 0.260 0 1 Microwave 6,019 0.631 0.483 0 1 Dishwasher 6,016 0.156 0.363 0 1 Dryer 6,022 0.187 0.390 0 1 Washer 6,017 0.771 0.420 0 1 Computer 6,016 0.611 0.488 0 1 Internet 6,018 0.637 0.481 0 1 TV 6,018 0.972 0.164 0 1 Cable 6,015 0.719 0.449 0 1 Telephone 6,018 0.519 0.500 0 1 Cellphone 6,021 0.905 0.294 0 1 Bathroom 6,023 0.903 0.296 0 1 Tap Water 6,031 0.970 0.171 0 1 Educ Head 6,040 11.044 4.368 0 24 Note: The sample consists of 6,040 individuals from seven countries included in the IDB-LAPOP Survey. Sample size di¤ers across variables due to incomplete or invalid responses to the respective survey question. i Table B2: Proxy Means Testing (PMT) Income Estimation Dep Var: Log Total Income CHL COL HND MEX PAN PER URY Car .516*** .566*** .384*** .342*** .323*** .409*** (.015) (.039) (.069) (.026) (.027) (.009) Motorbike   .159**   .115** .079*** (.068) (.049) (.011) Fridge .181*** .166** .184*** .213** .285*** .234*** (.039) (.083) (.042) (.099) (.030) (.045) Microwave .181*** .088*** .066*** .136*** (.034) (.026) (.023) (.010) Dishwasher .346*** (.019) Dryer .185*** (.013) Washer .178*** .155*** .136*** .127*** (.036) (.030) (.024) (.013) Computer .210*** .137*** .243*** .275*** .254*** .220*** .071*** (.017) (.035) (.067) (.030) (.051) (.025) (.012) Internet .159*** .144*** .152*** .140** .270*** .132*** (.024) (.037) (.036) (.067) (.027) (.013) TV .507*** .129*** (.065) (.027) Cable .277*** .079** .224** .158*** .232*** .164*** .217*** (.017) (.034) (.092) (.026) (.046) (.023) (.009) Telephone .130*** .168*** .285*** .082** .345*** .049* .162*** (.015) (.034) (.064) (.031) (.050) (.026) (.011) Cellphone .235*** .301*** .385* .349*** .291*** .534*** .069*** (.063) (.109) (.224) (.036) (.086) (.049) (.018) (continued on the next page) ii (continued from the previous page) Bathroom .192***   .109* .099** .342*** (.044) (.064) (.042) (.024) Tap Water .195*** .473*** .499** .075** .182** .063  (.053) (.057) (.231) (.032) (.071) (.040) Educ Head .059*** .036*** .035*** .026*** .047*** .030*** .050*** (.002) (.003) (.007) (.003) (.006) (.003) (.001) Adj R-sq .419 .448 .328 .429 .242 .433 .506 Obs 9,799 2,382 752 3,118 1,624 5,391 22,717 Note: Robust standard errors in parentheses. Each column reports the coe¢ cient estimates of a step-wise regression of household-level log total income on several household assets. The p-value threshold for a variable’s inclusion in the second step was 0.15. All models include a constant, not reported. Data comes from national household surveys of each country, reported at the individual level. Data sources for each column are as follows: CHL - CASEN 2017, COL - GEIH 2017, HND - EPHPM 2017, MEX - ENIGH 2016, PAN - EHPM 2017, PER - ENAHO 2017, URY - ECH 2017. * p <.10, ** p <.05, *** p<.01. iii details. Education: Integer variable recording the reported years of education. Scale: 0, 1,..., 24. Source: IDB-LAPOP Survey, Question ED2. Age: Integer recording age reported by the individual in the survey. Scale: 18, 19,... Source: IDB-LAPOP Survey, Question Q2. Male: Indicator variable that takes the value one if the individual reports being a male, zero otherwise. Scale: 0,1. Source: IDB-LAPOP Survey, Question Q1. Country-Level Transf v Publ Gds: Fractional variable measuring the percentage of general government current expense other than goods and services and interest payments, divided by the sum of current and capital spending excluding interest payments. Scale: continuous [0,1]. Sources: Kaminsky, Reinhart, and Vegh (2004) and IMF-WEO 2019. Short v Long Trm: Fractional variable measuring the percentage of general government total current expense excluding interest payments, divided by the sum of current and capital spending excluding interest payments. Scale: continuous [0,1]. Sources: Kaminsky, Reinhart, and Vegh (2004) and IMF-WEO 2019. Mistrust: Average respondent trust in political parties, by country. Responses were normalized to unit interval. Higher values mean lower levels of trust. Scale: continuous [0,1]. Source: Latinobarometro 1995-2018. Real Interest Rate: Average yearly lending interest rate adjusted by the GDP de‡ator. Scale: continuous. Source: World Bank-WDI 2019. Interpers: Average respondent mistrust in other people, by country. Responses were normalized to unit interval. The original individual-level measure is a dummy variable. Scale: continuous [0,1]. Source: Latinobarometro 1995-2018. Interest Spread: Average yearly di¤erence between the lending interest rate and the deposit interest rate. Scale: continuous. Source: World Bank-WDI 2019. GDP per Capita: Gross domestic product converted to international dollars using purchasing power parity rates, in ten thousand 2011 dollars. Scale: positive continuous. Source: World Bank-WDI 2019. Log Population: Logarithm of the midyear estimate of the total de facto population. Scale: positive continuous. Source: World Bank-WDI 2019. Unemployment: Percent of total labor force unemployed, modeled ILO estimate. Scale: x continuous [0,1]. Source: World Bank-WDI 2019. Election Year: Indicator variable that takes the value one in a general election year. Scale: 0,1. Sources: IDB - Database of Political Institutions 2017, extended to 2018 using online sources. xi