Remittance Receivers as Targets for Corruption in Latin America

Abstract Migration can affect the practice of corruption in migrant-sending countries in a number of ways. In this paper we test whether or not remittance receiving households are more likely to be targeted for corruption. Using micro-level data from 20 Latin American countries, this study finds that migrant households are about 15 percentage points more likely to be asked for a bribe than non-migrant-sending households. The corruption effect is further confirmed by an instrumental variable estimation. Our findings suggest that remittances can have an unintended effect on households’ risk of experiencing corruption. The excess exposure may discourage remittances and limit the positive development benefits of migration and remittances.


Introduction
The positive development impacts of migration on the migrant-sending households have been the focus of numerous empirical studies in development research.In the World Development Report, The World Bank (2023) enumerates some of these findings and sets out that remittances reduce poverty, increase food security, boost human capital investment, and close gender gaps in origin countries.At the country level, migration improves macro-stability and leverages norm transfers.Although some of the problems such as brain drain and human trafficking have been discussed in the literature (e.g., Joarder & Miller, 2014), there is a consensus that international migration in general improves institutions and governance of the home countries, for example, in terms of democracy, political participation and gender norms (Baudass e, Bazillier, & Issifou, 2018;Docquier, Lodigiani, Rapoport, & Schiff, 2016;Tuccio, Wahba, & Hamdouch, 2019).And when it comes to the problem of corruption, many studies have shown that emigration reduces corruption in the origin countries through norm transfers (Batista & be greater than the social remittances channel suggested in the literature.Findings of a strong consumption effect are supported by a study by Konte and Ndubuisi (2020), who use data from Afrobarometer to show that remittances have a positive effect on bribe payment and gifting.While their paper focuses on the supply side (i.e.bribe payment), ours focuses on the demand side (i.e.bribe solicitation).Our demand side perspective should provide a more complete picture as the focus of solicitation can reduce social desirability bias.Some existing studies used perception-based corruption measures as their dependent variables.However, as Reinikka and Svensson (2006) explained, corruption perception usually diverges from corruption experience.Results based on perception-based indicators should be interpreted carefully as it may lead to biased results (Kenny, 2009).To reduce biases, we use an experience-based indicator in this study.As a robustness check, we use a perception-based measure and obtain similar results.
Our study also goes one step further by addressing the endogeneity problem using an instrumental variable approach.Following Ivlevs andKing (2017) andH€ ockel et al. (2018), we use the lagged, weighted average of economic growth of the destinations as an instrument.Remittance behaviors usually depend on the economic conditions of the destinations (Carling, 2008;Kpodar, Mlachila, Quayyum, & Gammadigbe, 2023), and the instrument is relevant.However, instead of interacting economic growth with pre-existing migrant networks, we interact economic growth with language similarity between the destination and origin countries, because historical migration could be driven by corruption (Ivlevs & King, 2017).While language similarity can explain emigration pattern, it is more likely to be exogenous to corruption.Therefore, the proposed instrument is likely to be exogenous as the instruments are predetermined and at the country level which individual households have little influence.
Overall, this study contributes to the debate by presenting additional individual-level evidence from the Americas for the question whether remittance receivers are more likely to be targeted for corruption.Using survey data from 20 Latin American countries between 2006 and 2018, we find evidence of remittance receiving household being targeted for corrupt practices.According to our instrumental variable estimation, compared with individuals who do not receive remittances from abroad, remittance receivers are, on average, 15 percentage points more likely to be asked to pay a bribe.
The rest of the paper is organized as follows.The next section will explain how remittances can affect corruption at the individual level.More specifically, we propose that remittance receivers are exposed to a greater risk of corruption through a consumption effect.Second, we discuss our research design by providing further details about our sample, key dependent, independent, control variables and empirical models.We then present our descriptive and estimation results in the subsequent section.We also explore the relationship between remittances and corruption experiences, as well as perception of corruption in various government sectors.The final section concludes by discussing the implications of our findings and potential avenues for future research.

Why remittance recipients face a greater corruption risk?
Remittances have been proven to bring many development benefits to the receiving countries.At the country level, based on neoclassical economic theory, Pradhan, Upadhyay, and Upadhyaya (2008) argue that remittances can stimulate consumption and savings, which increase economic growth through a multiplier effect.They show that remittances stimulate economic growth using panel data from 39 developing countries.Shapiro and Mandelman (2016) study the impact of remittances on business cycle fluctuations and find remittances are countercyclical and buffer economic shocks during economic downturns.At the household level, there is also evidence showing that the new source of income alleviates poverty (Adams & Page, 2005) and increases parents' investment in children's health and education in Latin America (Borja, 2020).
However, remittances may also bring with them undesirable outcomes.They have shown to contribute to income inequality in Mexico (Garip, 2012) and reduce labor supply in Jamaica (Bussolo & Medvedev, 2008).This article highlights another unintended possible negative impact of migration: corruption.In addition, departing from the macro-level institutional effect discussed earlier, we argue that recipients of financial remittance will have a higher risk of being targeted for corruption through a consumption effect.

Consumption effect
The economic motivation of migration has long been recognized in the economics literature.The New Economics of Labor Migration (NELM) literature highlights the positive role of remittances in investment.For example, Taylor (1999) argues that remittances can ease investment constraints, setting in a development dynamic.Moving away from the macro development impacts, recent literature (e.g., Combes & Ebeke, 2011) examines the effect of remittances on household consumption.Based on the utility maximization model, these studies hypothesize that households will increase their expenditures on various consumption and investment goods when their disposable incomes increase due to remittances.
Remittances can affect consumption and investment, especially for the poor.When remittance-receiving households have a more stable income stream, this additional source of income can enhance their economic well-being because they become more resilient to adverse income shocks such as hyperinflation, food price increase and climate-change-induced agricultural shocks in the home countries.This remittance-induced consumption smoothing effect has been shown to boost consumption in Latin American and Caribbean countries (Ramcharran, 2020).Conversely, households that lack an effective mechanism to smooth consumption will have a more volatile consumption pattern.Furthermore, to ease the impact of negative income shocks, households tend to save more and spend less on human and physical capital accumulation (Pallage & Robe, 2003).If consumption depends on both income level as well as income volatility, a more stable income stream can increase consumption indirectly.A panel data analysis based on 12 Latin American countries from Gandelman (2017) confirms such a relationship and found that people who are more susceptible to income shocks tend to save more and spend less due to the precautionary motives of saving behavior.In short, the negative impact of income shocks and its volatility effect is found to be common among households in developing countries, which may explain why remittances stimulate consumption and economic welfare.
As poor remittance-receiving households consume more goods and demand more (public) services such as education and healthcare, they inevitably increase interactions with government officials and service providers, thus increasing their risk of being asked for a bribe.For instance, under-the-table payments are common in the healthcare sector in Peru and Brazil (Bate & Mathur, 2018;Garc ıa, 2019).People are expected to give money or small gifts to service providers to reduce waiting times, receive better care, obtain drugs, or simply receive any care (Lewis, 2007).As healthcare services become more affordable to remittance-receiving households, household members who fall sick may be more willing to pay a fee for health services, thus facing a higher chance of petty corruption.Corruption is widespread in the education sector as well.In Ecuador, students are expected to pay bribes to be admitted to some prestigious public schools (Martinez Novo & de la Torre, 2010).As education becomes more affordable, when parents send their children to schools, they are also exposed to a higher risk of corruption.
Changes in consumption and investment behaviors have a strong signaling effect.Acquisition of new land, building a new house, or getting a new vehicle are strong signals for the economic wellbeing of a household.A longitudinal study by Chiodi, Jaimovich, and Montes-Rojas (2012) confirms that remittances have a substantive effect on asset accumulation in rural Mexico, most notably tractors and agricultural land.When these consumption and Remittance receivers as targets for corruption in Latin America 327 investment goods are visible, they send a strong signal to government officials and increase people's risks of being targeted.In a field experiment in Mexico City, Fried, Lagunes, and Venkataramani (2010) hypothesize that traffic police would ask for a bribe based on the perceived differences of drivers, based on their clothing, vehicle, skin tone and perceived social status.This lends support to the idea that increased consumption can induce targeted corruption.
In sum, remittances induce consumption.Visible consumption goods signal the wealth level of people, making them easy targets for corruption.Consequently, we hypothesize that remittance receivers are more likely to experience corruption.

Data and variables
We use data from the Americas Barometer survey by the Latin American Public Opinion Project (LAPOP) for analysis.For consistency, we use the grand free version of the data set provided by LAPOP in the analysis.Our sample covers 20 Latin American countries between 2006 and 2018.Since questions related to migration, remittance, and corruption were not asked in all countries in every wave of the survey (e.g., 2021), we use only a subset of countries and years in the original dataset in the regressions.Supplementary Figure 1 provides an overview of the coverage of our sample.From the figure, countries in Central America are slightly over-represented.Therefore, given the design of the survey, our analysis is representative at the country but not at the continent level.The final sample consists of 132,571 observations.

Dependent variables
The main dependent variable bribe is a binary variable which takes the value 1 when the respondents indicated that they were asked for a bribe.The coding of the variable is based on the question 'In the last twelve months, did any government employee ask you for a bribe?'As respondents may be reluctant to tell the truth to strangers, we also operationalize corruption based on perception in a robustness check.Perception is an ordered variable.It is coded based on the following question: 'Taking into account your own experience or what you have heard, corruption among public officials is … '.There are four possible outcomes: corruption among public officials is very uncommon (coded as 1), uncommon (2), common (3) or very common (4).Despite a potentially downward biased measure, bribe captures actual experience of the respondents and provides a more conservative estimate on the impact of remittances on corruption.If we find evidence for the remittance effect on corruption, the downward bias suggests that the effect could be stronger than what our analysis suggests.
The Americas Barometer survey contains a few more questions about the level of corruption in a country.Corruption is a multifaceted phenomenon that encompasses people's daily lives.Having multiple measures not only helps us identify the area in which corruption is more prevalent and prone to the remittance effect, but it also ensures that our results are consistent and robust, thus increasing the power of our statistical tests.The seven alternative measures that we use in this analysis are based on respondents' past interactions with police, military officers, health professionals and officials working in municipal governments, people at schools, courts, and workplaces.Since some of the questions are not relevant to all respondents-for example, questions related to education are not relevant to people with no children-rather than dropping those observations, we code 'not applicable' as a category and model it explicitly with multinomial logit regressions to reduce potential selection bias.

Independent variables
The main independent variable, remit, is a binary variable where 1 means that a respondent was receiving remittance from abroad when the interviews were conducted.This is based on the survey question asking: 'Do you or someone else living in your household receive remittances, that is, economic assistance from abroad?' The actual amount that the respondents (or other members in the household) received is a more precise measure of the remittances effect.Unfortunately, such information is generally not available in the dataset.We have no choice but to use the less ideal discrete measure as the independent variable in our investigation.
Existing literature primarily focuses on the institutional effects.To control for the institutional effects, we use the corruption indicators from the Bertelsmann Stiftung's Transformation Index (BTI) in our analysis.BTI's anti-corruption policy indicator assesses 'whether adequate institutional arrangements exist to implement an anti-corruption policy and if they successfully contribute to an effective prosecution of corruption' (Bertelsmann Stiftung, 2020).A higher value means that a country has a better institutional capacity to monitor and to contain corruption.We did not use the perception index by Transparency International (TI) in this analysis, as TI has revised the methodology in 2012.Since our sample covers the period between 2006 and 2018, the methodological change does not allow us to pool observations before and after 2012 for an internally valid analysis.We do not use the World Bank's Worldwide Governance Indicators (WGI) either, as the construct of the index is based on information related to corruption practices in a country (Worldwide Governance Indicators, 2022).Using WGI as a corruption control indicator may lead to regressing corruption on corruption. 1  Several additional control variables are included in the analysis: age, age squared, gender (female ¼ 1), years of education of the respondent, ethnicity, geographical location of the respondents (urban ¼ 1), as well as country and year fixed effects.Isaksson (2015) showed that corruption experience varies systematically, depending on a person's ethnicity.Seligson (2006) found that age, gender, education level and location are related to corruption experience in Latin America, possibly due to their exposure and daily life experience.We also include wealth and income when we further explore the mechanisms.The wealth indicators are a set of binary variables that denote ownership of durable goods that respondents had (cell phone, computer, motor, and microwave), as well as the number of vehicles that they owned (zero, one, two, and three or more).Since income is not reported in most of the countries across years, the inclusion of the variable will reduce the sample size substantially, we include the variable as an alternative to economic wellbeing only in separate models.As remittances contribute to income, controlling for them may also increase bias (Angrist & Pischke, 2009).The summary statistics of the variables are reported in Table 1. 2

Models
Since the dependent variable is a binary variable, we estimate the following logit model, where K is a logistic function, X the set of independent variables, Z the control variables, b j estimated parameters, k an index of control variables and i is an error term.Although we have observations from multiple years, this is not a longitudinal survey, as a different set of individuals were interviewed in each wave.We also add country and year fixed effects to account for cross-country and cross-year unobserved heterogeneities.In a robustness check, we allow the country and year dummy variables to interact to account for country-specific time effects.In another robustness check, we employ a multilevel mixed-effects model with varying intercepts to account for unobserved heterogeneity across strata within and between countries.

Instrumental variable estimation.
The variable remit is endogenous when it is correlated with the error term.This is likely to be the case, for example, when corruption affects people's intention to migrate, or there is an unobserved factor which affects remittances and corruption Remittance receivers as targets for corruption in Latin America 329 simultaneously.In addition, remittances-receiving individuals or other households may not be directly comparable.If their attributes are correlated with corruption risks, we may over-or under-estimate the impact of remittances.For instance, if the migration status is positively selected due to aspirations (i.e.people from a better socioeconomic background are more motivated to migrate), the recipients can be targeted because of their backgrounds instead of remittances alone.
We address this endogeneity problem via an instrumental variable approach.We estimate the causal effect of remittances with a two-stage least square method.More specifically, in the first stage, we regress the endogenous variable remit on an instrumental variable, which is uncorrelated with the outcome corruption except through the remittances channel.Then we obtain the predicted value of remit and use it as the independent variable.Because the instrument is exogenous, the predicted values of remit are uncorrelated with the error term.According to Wooldridge (2010), the marginal effect from a linear probability model (LPM) approximates the counterpart from a non-linear model well.For simplicity, we will employ a LPM in our estimation, thereby not applying the link function specified in equation (1).
A good instrument needs to be relevant and exogenous.Following Ivlevs andKing (2017) andH€ ockel et al. (2018), we also use weighted economic growth rate of a set of destination countries as our instrument.Emigration decisions and remittance-sending behaviors depend on the economic conditions of the destination countries (Kpodar et al., 2023).The instrument satisfies the relevant condition.Three different migration-related variables have been used as a weight: (1) pre-existing migrant networks, as in Ivlevs and King (2017) and H€ ockel et al. ( 2018), (2) geographical distance between the origin and destination countries, and (3) common language between the origin and destination countries (yes ¼ 1).As estimation results based on pre-existing migrant networks and geographical distance do not pass the Stock-Yogo test and are considered as weak instruments, we report results based on common language only. 3 Formally, instrument Z ct is defined as, where c is an index for the country of origin, j ¼ 1, :::, 185 is the index for the destination country, and t is year.The second term is the average growth rate of a destination country s years before the survey year.Ivlevs and King (2017) use the 5-year average in their analysis.
Since the number is arbitrary, as a sensitivity analysis, we vary T from 3 to 5 years.Because historical GDP growth rate varies over time, the instrument is country-specific and time-varying.
As a robustness check, we use two variants of language cj in our estimation.While English is the official language of the US, about 42 million people in the country speak Spanish at home (US Census Bureau, 2022).Therefore, we also use whether a language is spoken by at least 9% of the population in both countries (yes ¼ 1) to code the instrumental variable.Second, Spanish is more similar to French and Italian than to Chinese.To add variations to the instrument, instead of using a binary variable as weights, we also use linguistic proximity between the languages spoken by the dominant groups in the origin and destination countries to code the instrument.Spolaore and Wacziarg (2009) use the number of nodes that separate two languages in a linguistic tree as an estimate of linguistic distance.We use the same variable as a measure of linguistic similarity.For ease of interpretation, we reverse the scale of the variable, so that a higher number means two languages are more similar.The proximity indicator, however, contains quite some missing values, and we use it mainly as a robustness check.Language data are from Head, Mayer, and Ries (2010) and Spolaore and Wacziarg (2009).GDP growth (constant prices) are based on the World Economic Outlook database (International Monetary Fund, 2023).
The instruments satisfy the relevant condition.Common language is a strong predictor of migration between countries in gravity models (Wong & Celbis, 2019).Intuitively, emigrants are more likely to choose countries with a lower language barrier as their destinations.
Migrants with a higher level of language proficiency are more likely to integrate socially.They can find jobs more easily and are more likely to succeed in the labor markets.Empirical evidence has shown that immigrants with stronger language skills are more likely to get employed (Lang, 2022) and earn a higher wage (Bleakley & Chin, 2004).As labor market outcomes of migrants are related to their remittance behaviors (Lianos & Cavounidis, 2010), we expect a strong relationship between remittances and the instrument.Finally, characteristics of the sending countries influence motivations to remit (Carling, 2008).Migrants may temporarily stop sending remittances because of economic downturns, exchange rate fluctuations, high transaction costs, and changes in relative price level (Borjas, 2020;Carling, 2008).On the other hand, higher growth rate is usually associated with better job opportunity and higher business returns.
Remittance receivers as targets for corruption in Latin America 331 All these variables are, both theoretically and empirically, highly correlated with the endogenous variable remit.
The instruments are likely to be exogenous.It is difficult to conceive why a country is more corrupt because their people speak a specific language.The weighted-average growth rate of the global economy is also unlikely to systematically influence individuals' corruption experience after controlling for the time fixed effects.Furthermore, as the instrument is based on historical economic conditions of countries, individual households are unable to affect macroeconomic performance of countries and in an ex post fashion.Any unobserved heterogeneity across countries and years would also be controlled for by the country and year fixed effects.
We also perform several statistical tests (i.e. the under-identification test and the Stock-Yogo test) to ensure that our instrument is relevant, and the estimation is not subject to the weak instrument problem, which may bias the IV estimate.We also conduct a Durbin-Wu-Hausman test of endogeneity to compare the estimates from the IV and OLS estimators.Provided that the instrument is valid, if the OLS estimates and IV estimates are similar, the OLS estimator is more efficient.Otherwise, IV estimator is consistent and hence preferable to the OLS estimator.

Descriptive results
The sample mean of the variable bribe is 0.0597.On average, 5.97 percent of sampled individuals were asked to pay a bribe twelve months prior to the survey interviews.A substantial difference, however, exists between people who come from a remittance-receiving and a non-receiving household. 4The mean for respondents that received remittances is 8.64 percent, versus 5.54 percent for respondents that do not receive remittances.The p-value of a twosample t-test is 0.000, suggesting that the difference between the two groups is statistically significant at all conventional significance levels.
Significant variations exist across countries.Supplementary Figure 2 compares the country average between remittance receivers and non-receivers across 20 sampled countries.Except for Chile and Bolivia, recipients are more likely to be asked for a bribe as the sample means are higher. 5Among the sampled countries, corruption is very common in Bolivia, Haiti, and Mexico, and the least common in Chile.When we compare the corruption experience between recipients and non-recipients, the difference is greater in Argentina, Costa Rica, the Dominican Republic, Ecuador, Haiti, Honduras, Panama, and Peru.Targeted corruption seems to be concentrated in Central America and the Andean region.
To better understand how the characteristics of the individuals in the sample vary by their corruption experience and receipt of remittances, we report the summary statistics on key variables in Table 2. Regarding receipt of remittances, receivers are similar in terms of their age, gender, and economic backgrounds.While their average household incomes are higher, the household size is also slightly larger.Receivers are more educated and less likely to be employed.If households are similar ex ante, this pattern suggests that the migrant household members are likely to be less educated.In other words, migration is negatively selected in terms of education. 6Benefiting from remittances, household incomes are higher and the pressure to work is lower.The relationship between negative selection and employment could also be related to poorer job opportunities at home.Regarding corruption experience, highlighted in columns 3 and 4, corruption experiences are more common among males.They are younger, more educated, employed and have better economic backgrounds.These variations suggest that selection is likely to be an empirical issue.We will address the issue of endogeneity in Section 5.

Econometric estimates
A similar pattern emerges based on results from a simple regression model and a regression model with country and year fixed effects.The estimated coefficient of the variable remit is positive and statistically significant (p < 0.001; see Model (1) in Table 3).After adding country and year fixed effects to Model (1), we find that people receiving remittances face a greater risk of being asked to pay a bribe.Based on Model (2) in Table 3, the marginal effect in terms of probability is 2.65 percentage points.That is, for individuals who do receive remittances, the chance of being asked for a bribe will be 2.65 percentage points higher.(For ease of comparison, we will report the marginal effects in probability thereafter.)When we further control for  Remittance receivers as targets for corruption in Latin America 333 location and other demographic factors such as age, gender, and levels of education, we found a similar result (Model 3).A person will have a higher chance (2.17 percentage points higher) to be asked for a bribe.Given that the estimated baseline probability is 5.64 percentage points, remittance receivers are about 38 percent more likely to be asked to pay a bribe (i.e.2.17/ 5.64 ¼ 0.38).The size of the marginal effect is comparable to the one based on a linear probability model (Model 4).The average marginal effect from a LPM (2.2 percentage points) approximates that from a binary dependent variable model well.
Regarding the effects of institutions, institutions seem to curb corruption.The estimated coefficient of BTI is negative (Models 3 and 4).We have used the World Bank's WGI in a robustness check (see Table A2 in Supplementary Material).While the effect of WGI is not statistically significant at the 5% level, the effect of remittances remains positive and statistically significant (p < 0.001).
The effects of other control variables are in line with what has been reported in the literature and carry the expected signs.More educated individuals, which usually earn higher incomes, are more likely to be asked for a bribe, so are men and people who live in urban areas.While women are not necessarily less corrupt than men (Merkle & Wong, 2020), in contrast to the fair-sex hypothesis (Swamy, Knack, Lee, & Azfar, 2001), we found that men experienced higher corruption risks.This can be explained by the fact that they are more exposed to bribery activities such as getting permits and licenses (Justesen & Bjørnskov, 2014).People living in urban areas are also found to face a higher risk of corruption, possibly because public sector activities are concentrated there (Justesen & Bjørnskov, 2014).

Robustness checks
We performed a number of robustness checks to make sure that our findings are not contingent on specific coding rules or model specifications.First, we interact the country dummies with the year dummies in order to let each country have their own time fixed effects.The effect of remittance is similar (Model 1 in Table 4).Second, we estimate a mixed-effect model.Our sample consists of individuals from 445 strata in 20 countries.To account for regional heterogeneity, we added strata-level random effects to the baseline model (see Model 2).The results are highly consistent with what we have found.Third, some interviewees may not want to tell the interviewers about their corruption experience.To gain a better understanding of the relationship between corruption and remittances, we also regress corruption perception on the remittance variable.Since the variable perception is an ordinal variable, we estimated the model with an ordered logit regression.The estimated coefficient of the key independent variable remit is positive (Model 3; p ¼ 0.012), meaning that remittance receivers are more likely to state that corruption is very common among public officials, a finding in line with those based on people's experience.The marginal effects of remittances on each response according to Model 3 are summarized in Table 5.In general, remittance receivers tend to perceive that corruption is common, a pattern largely consistent with our hypothesis.
Corruption is a widespread phenomenon in many countries in the Americas.We coded the dependent variables differently based on the types of interactions between respondents and government officials in different sectors, the pattern is remarkably consistent.Table 6 summarizes the marginal effects of remittances on corruption experience in seven different areas.In general, respondents who receive remittance are more likely to be asked by the police and soldiers to pay a bribe.Remittance recipients are also more likely to be asked for a bribe at work and at municipal offices, courts, hospitals, clinics, and schools.Bribe requests by the police are the most common.In particular, the baseline probability is 10 percentage points if a person is from a non-receiving household.By receiving remittances, a person will have an additional risk of 3.1 percentage points.In contrast, bribery at the courts is the least common.The baseline probability is only 1.4 percentage points if one does not receive remittances.In contrast, for a Notes: Logit estimates in model (1).Mixed-effects logit estimates in model (2).Ordered logit estimates in model (3).Robust standard errors in parentheses.ÃÃÃ p < 0.001; ÃÃ p < 0.01; Ã p < 0.05.Perception ¼ 1 (very uncommon), 2 (uncommon), 3 (common) and 4 (corruption is very common among public officials).Source: Authors' calculations using Americas Barometer data.Remittance receivers as targets for corruption in Latin America 335 receiver, the marginal impact in terms of probability is higher and the risk will increase by 0.8 percentage points, an increase of over 50 percent from the baseline.Bribe asking is relatively common and the remittance effect is substantial at schools and healthcare facilities as well.In both cases, the baseline probability is about 4.5 percentage points, with the marginal effects equal to 0.8 (schools) and 1.2 percentage points (healthcare facilities).

Accounting for endogeneity
The variable remit can be endogenous due to selection bias, omitted variable bias, and other empirical concerns.In this section, we extend our analysis using instrumental variable estimation.As explained in Section 3.3.1,we use language-growth interaction to instrument remittance recipients.Table 7 summarizes the results from instrumental variable (IV) estimation using a linear probability model (LPM) for 12 different models based on different IVs.Models 1 to 3 use official language in the interaction term.The instrument passes the under-identification test, the Stock-Yogo test (the critical value for maximal 10% bias is 16.38).Both Models 1 and 2, which utilize three and four years of average lagged growth rate, pass the Hausman test.The estimate in Model 3 does not pass the Hausman test, implying that the IV estimates are less precise than the OLS estimate, as the IV estimator is less efficient if there is no endogeneity.Based on Models 1 and 2, the marginal effects of remittance are, respectively, 17.6 and 14.1 percentage points, greater than the 3 percentage point results reported above using logit.Although the point estimate from Model 3 is slightly greater than the estimate of 3 percentage point, it is less precise and is not statistically significant at the 5% level.Findings using a different weight scheme (i.e.language spoken by at least 9% of the population in both origin and host countries) are comparable and around 15 percentage points, although the estimate based on the instruments using a 5-year average is not significant and does not pass the Hausman test.The estimated effects based on instruments that use language proximity as the weighting scheme are more sizable.But the instrument has an unexpected, negative effect on the endogenous variable in the first stage.This may be related to the missing data problem and the lack of variations in the original data.To achieve overidentification, we include variables that use common official language and language proximity, which are only weakly correlated (maximal rho ¼ 0.17), as  Remittance receivers as targets for corruption in Latin America 337 According to the estimate, the probability that people from remittance-receiving households to be asked to pay a bribe is about 15 percentage points higher than households that did not receive remittances, an effect more sizable than the 3.6 percentage points suggested by an OLS estimation.Overall, after accounting for potential endogeneity, we find further indicative evidence that remittances increase corruption risk.

Mechanisms
The literature has proposed different mechanisms to explain the relationship at the macro level.Abdih et al. (2012), for example, examined the role of institutions in corruption experience.As the results from our IV estimation show, after controlling for the possible effects of corruption control, the effect of remittances remains statistically significant (see Supplementary Material for the complete estimation results).Ivlevs andKing (2017) andH€ ockel et al. (2018) proposed a social remittance effect.To control for its effect, we included in our IV estimation a new variable called justified, which is a binary variable with value 1 when the respondents indicated that paying a bribe is justified and 0 otherwise.Table 8 summarizes the estimation results. 7When compared with Model 1, our benchmark case, although the effect of morality is statistically significant, the coefficients of remit are quite close to each other, suggesting that the social remittance effect does not seem to explain many variations in corruption experience in this case.
Remittances induce consumption on normal goods or, in some cases, luxuries.If the consumption effect is present due to signaling, we would expect a more sizable effect for visible consumption goods.We include a categorical variable vehicle (none, one, two, and three or more) and four asset dummies (motorcycle, cell phone, computer at home, microwave) in our estimation.The inclusion of these asset variables does not statistically and significantly reduce the estimates, suggesting that the remittance effect may be related to increased interactions (e.g.health services) instead of signaling.The size of the effect is significantly higher for visible luxuries like vehicles, although it is not for motorcycles and cell phones.Model 4 includes income (in ten income groups) to the model.The effect of remittances is similar.To test the signaling channel further, if remittances have any signaling effects, we may expect remittances to be associated with crime victimization. 8Estimation results however do not support the signaling hypothesis. 9This may be because signaling is not the major channel, and remittances increase corruption risk mainly through more frequent interactions with government officials.The finding may also be related to the fact that crime is more common among the higher-income groups, who are not the major beneficiary group of remittances.In our sample, the crime victimization rate among the highest income group is 16 percent, versus 12 percent in the sixth (out of ten) highest income group.In fact, the relationship between remittances and crime victimization is low in the region.(The tetrachoric correlation coefficients between the two binary variables is 0.08.)In short, the signaling channel is less evident or seems to be unimportant.Remittances may affect corruption mainly through increased consumption and daily interactions.

Conclusions
In this article we studied whether remittance receiving households are more likely to become targets of corruption than non-receivers in Latin America.Results based on survey data across 20 countries display a remarkably consistent pattern, that people from remittance-receiving households are more exposed to corruption risks than non-recipients.This finding is consistent with the macro-level evidence from Abdih et al. (2012) and Berdiev et al. (2013).While these studies hypothesize that the positive association can be explained by a resource-curse effect, using micro-level data, we identify a different mechanism that can explain the association, namely a consumption effect.While there are a few studies that have looked at the relationship between remittances and corruption at the micro-level, we are among the first to propose that remittances can induce more corruption through a consumption effect.
While we have proposed a new mechanism to explain the association between remittance and corruption experience, our tests are subject to some limitations.First, the signaling channel does not seem to explain the remittances effect, suggesting that some important variables may be missing in the whole picture, an issue to be investigated in future studies.For example, remittance-receiving households may be more likely to report their corruption experience because they feel more empowered due to the improvements in their socio-economic status (Tyburski, 2012).In this way, the corruption experience may be due to an empowerment effect.Second, although the data are from a nationally representative survey, and we performed several robustness checks to enhance the validity of our findings, issues such as missing data and controls remain, and evidence from other world regions are still required.Despite these caveats, the remittances-corruption nexus is likely to exist.
Conventional anti-corruption strategies usually adopt a legal enforcement approach and have delivered unsatisfactory results (Khan, Andreoni, & Roy, 2019).Recent literature views corruption as a form of power relations (e.g.Merkle & Wong, 2020).As remittances have Remittance receivers as targets for corruption in Latin America 339 shown to empower the poor and disadvantaged groups by improving their economic, political, and social standings in societies (Deere & Alvarado, 2016;Krawatzek & M€ uller-Funk, 2020;Tyburski, 2012), enhancing the positive development impacts of remittances could have the potential to fight against corruption from the bottom up by giving citizens greater power to demand better governance.This empowerment effect is likely to be stronger if remittances are sizable and are organized at the collective level (Burgess, 2012).Orozco and Lapointe (2004), for example, have shown that hometown associations supported and organized by Mexican migrant workers in the US have successfully pressured local governments in Mexico to be more transparent and accountable for projects that they funded.In this regard, although this study found that remittances can induce more corruption, our findings can be interpreted as both a challenge and an opportunity for substantial social change in the origin countries.

Table 1 .
Summary statistics Source: Authors' calculations using Americas Barometer data.Employ, local, work, court, health, andSchool are categorical variables.We report the proportion of people who were employed and gave a positive response.Edu is right-censored, and 18 means years of education or above.So does vehicle, where 3 means three or more vehicles.The category accounts for 2.67 percent of sampled observations.

Table 2 .
Characteristics of remittance recipients and people experienced corruption Source: Authors' calculations using Americas Barometer data.The employment variable is dichotomized (yes ¼ 1).

Table 6 .
Marginal effects of remittances on the probability of being asked for a bribe in different circumstances Model specification is similar to Model (3) in Table3.Robust standard errors in parentheses.ÃÃÃ p < 0.001; ÃÃ p < 0.01; Ã p < 0.05.See Supplementary Material for complete estimation results.Source: Authors' calculations using Americas Barometer data.theweighting schemes.Model 12 does not pass the Hausman test and in Model 11 one of the instruments has no impact on the endogenous variable.Results from Model 10 (16.7 percentage points) are consistent with other estimates.And this time the coefficients of language proximity carries the expected signs in the first stage.Estimation results from this IV analysis further confirms what has been reported in the previous section, that remittance-receiving individuals are more likely to be targeted for corruption.

Table 7 .
Instrumental variable estimates of corruption and receipt of remittances

Table 8 .
IV estimation of various mechanisms Notes: Robust clustered standard errors in parentheses.ÃÃÃ p < 0.001; ÃÃ p < 0.01; Ã p < 0.05.The sample includes only people living in urban areas.The critical value of the Stock-Yogo test (10% maximal IV size) is 16.38.Control variables (BTI, age, gender, education, urban), ethnicity fixed effects, country fixed effects, and year fixed effects are included.The baseline for vehicle is having no vehicle.See Supplementary Material for full estimation results.Source: Authors' calculations using Americas Barometer data.