Policy Research Working Paper 9833 Empowering Migrants Impacts of a Migrant’s Amnesty on Crime Reports Ana María Ibáñez Sandra V. Rozo Dany Bahar Development Economics Development Research Group November 2021 Policy Research Working Paper 9833 Abstract This paper studies whether undocumented immigrants Venezuelan immigrants, not explained by an increase in change their crime-reporting behavior after receiving a crime overall. The results are particularly strong for reports regular migratory status. It exploits a natural experiment of domestic violence and sex crimes. Results are almost of a massive amnesty program that gave a regular migratory entirely driven by reports by female Venezuelan immigrants, status to over 281,000 undocumented Venezuelan immi- a vulnerable population, suggesting that empowerment is grants in Colombia. The findings suggest that following an important mechanism driving the behavior change. the amnesty there is an increase in reporting of crimes by This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at anaib@iadb.org, srozovillarraga@worldbank.org, and dany_bahar@brown.edu. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Empowering Migrants: Impacts of a Migrant’s Amnesty on Crime Reports∗ Ana María Ibáñez† r ○ Sandra V. Rozo‡ r ○ Dany Bahar § JEL Classification: D72, F2, O15, R23 Keywords: Migration, Crime, Amnesty, Empowerment ∗ We are grateful to Maria José Urbina for her excellent job as a research assistant. The order in which the authors’ names appear has been randomized using the AEA Author Randomization Tool (#C5TaE0I_hdF5), denoted by ○ r. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank and/or the Inter-American Development Bank, and their affiliated organizations, nor those of the Executive Directors of the World Bank and/or the Inter American Development Bank nor the governments they represent. † Universidad de los Andes, Inter-American Development Bank, email: anaib@iadb.org ‡ World Bank, Research Group, email: srozovillarraga@worldbank.org § Corresponding Author. Brown University. email: dany_bahar@brown.edu. 1 Introduction When thinking about the words “crime” and “immigration” the most common perception is that they often go together, despite the vast empirical evidence showing this is rarely the case (e.g. Bianchi et al., 2012; Fasani et al., 2019; Bahar et al., 2020). In fact, recent research suggests that in large and massive episodes of immigration, such as the case of Venezuelans in Colombia, it is the foreigners themselves who tend to be the victims of crimes more often than not (Knight and Tribin, 2020), which hints of their vulnerability and their need for protection from law enforcement authorities. Yet vulnerable populations, such as undocumented immigrants, might choose not to report crimes as they fear being exposed to local authorities might result in deportation or other punishment. This paper directly addresses this question by empirically studying whether the provision of a regular migratory status to undocumented immigrants results in a change of behavior when it comes to reporting crimes. We exploit a natural experiment and document a systematic relationship between undocumented immigrants who receive a regular migratory status and a subsequent increase in immigrants reporting crimes to the local authorities. The context of our study is Colombia, which as of today has become the largest host of Venezuelan immigrants and refugees.1 The vast majority of these Venezuelan immigrants and refugees arrived to Colombia since 2014, as a result of the political, economic and humanitarian crises in the aftermath of two decades of the authoritarian regimes by Hugo Chávez and his successor Nicolás Maduro. In August of 2018, then Colombia’s President Juan Manuel Santos, decreed the issuance of a regular migratory status, known as PEP-RAMV.2 The PEP- RAMV granted a regular migratory status for two years to about 281 thousand undocumented Venezuelan immigrants in August, 2018. The status not only provided the immigrants with a job permit, but also the possibility to be scored by SISBEN, the mean proxy test used to target social programs in Colombia. As such, the PEP-RAMV is a generous amnesty enabling undocumented immigrants to work in the formal sector, access social services (including complete health services through the Subsidized Health Regime in Colombia) and as importantly, provides certainty to the immigrants that they will not be deported or prosecuted for being in the country without a regular 1 According to the latest data from Colombian migration authorities, by 2020 the number of Venezuelans in Colom- bia was approximately 1.8 million people, which represent about 3.6% of the country’s population. 2 PEP stands for Permiso Especial de Permanencia, the name of the temporary visa created by Colombian au- thorities for Venezuelan immigrants, and RAMV stands for Registro Administrativo de Migrantes Venezolanos, the name of a registry for undocumented Venezuelan immigrants. See Section 2 for more details. 2 migratory status.3 . Using data from several sources for different municipalities representing the largest urban centers of Colombia, and relying on a difference-in-differences specification, we compare crime reports by Venezuelans for cities with different intensities of the treatment variable, which we define as the number of Venezuelans that received the PEP-RAMV visa due to the amnesty (all in terms of the local population). Importantly, we also control for the total number of crime reports in that same location in order to be certain that our effect is capturing changes in reporting behavior which are not due to changes in the actual number of crimes. In order to deal with endogeneity concerns, we exploit the fact that each undocumented migrant who became eligible for the visa was granted a registry number exogenously allocated at the time of registration. Based on this number, individuals had the opportunity to apply online for (and subsequently get) the PEP-RAMV visa in one of 22 time windows in late 2018. We use these as- signments to estimate average number of registration days available to undocumented immigrants in each municipality, and use it to instrument for the treatment. Our results are robust to including a large number of controls to flexibly account for potential differential non-parametric trends on a number of municipal pre-established characteristics. These include controls for pre-existing differ- ences across municipalities in crime, conflict, economic growth, economic activity, government size, financial sector size, and sector composition. We complement our main results with event studies examining the evolution of quarter-by-quarter crime reports by Venezuelan in main urban centers that had different treatment intensity. The event studies support the validity of our identification strategy, as we observe parallel trends before the program implementation. Our main finding is that, indeed, undocumented immigrants increase reporting of crimes after receiving a regular migratory status. Our results are particularly strong for reports of domestic violence and sex crimes, and are almost entirely driven by reports by female Venezuelan immigrants, a vulnerable population. In particular, we find that Colombian municipalities with twice as much Venezuelan immigrants who receive a regular migratory status result in an increase of 25% more crime reports by Venezuelans. Note that this is not explained by an increase in crime reports overall, but simply an increase of Venezuelan immigrants reporting crimes. We further find a twofold increase in the number of Venezuelan immigrants that receive a regular migratory status (relative to population) explains an increase of 40% and 70% in reports of sexual crimes, and of domestic violence 3A prerequisite to receive the PEP-RAMV was not having any criminal records or pending deportation orders 3 crimes, respectively. Given that the effects are mostly driven by a particular vulnerable constituency – undocumented female Venezuelan immigrants – we believe that a plausible explanation for our results is the empowering of women who, without fearing deportation or other punishments, feel safe to report abuses to their basic human rights. Our study contributes to several strands of the literature. First, it contributes to the literature studying the relation between immigration and crime (e.g., Butcher and Piehl, 1998; Borjas et al., 2010; Bianchi et al., 2012; Alonso-Borrego et al., 2012; Bell et al., 2013; Spenkuch, 2014; Baker, 2017; Pinotti, 2017; Freedman et al., 2018; Fasani, 2018; Fasani et al., 2019; Ajzenman et al., 2020; Knight and Tribin, 2020). Our study, however, does not focus on whether undocumented immigrants change their likelihood of committing crimes after receiving a regular migratory status. Rather, we focus on whether undocumented immigrants change their behavior when reporting crimes after the migratory amnesty (controlling for overall crime incidence in that same locality). Second, our work contributes to the studies that explore the impacts of migratory amnesties in hosting countries(e.g., Kossoudji and Cobb-Clark, 2002; Bratsberg et al., 2002; Orrenius and Zavodny, 2003; Kaushal, 2006; Amuedo-Dorantes et al., 2007; Bahar et al., 2021). While most of these studies focus on examining the impacts of these large-scale migratory amnesties on labor markets, we focus on the behavior of the immigrants themselves as measured by their reporting of crimes. To some extent, we believe, the reporting of crimes is a proxy for the subsequent social empowerment of vulnerable immigrant populations that results from receiving a regular migratory status. In this context, our results have important policy implications by documenting large social benefits to vulnerable populations following a migratory amnesty, which in turn –according to the previous literature– has negligible costs in terms of labor market outcomes for natives. The paper is organized as follows. First, we provide context for the massive migratory amnesty program that we study; second, we discuss the data sources and our empirical strategy, third, we present our main findings and discuss them; and finally we provide some concluding remarks. This paper is accompanied by an Online Appendix that we refer to throughout the text. 2 The PEP-RAMV Program Colombia has led the regional efforts to regularize the status of Venezuelan immigrants. As is typical in crisis-driven migration processes, a large share of the Venezuelans who migrated to Colombia did 4 so without going through the formal migratory process. Between April and June of 2018 the Colom- bian authorities administered a registry throughout 1,109 authorized points in 413 municipalities to register undocumented Venezuelan immigrants (see Figure A1 in Online Appendix). The registry was known as the Registro Administrativo de Migrantes Venezolanos or RAMV. RAMV relied on a massive public advertisement campaign in order to attract Venezuelan immigrants to voluntarily register and self-report personal information, such as names, date of birth, current address, munici- pality of origin in Venezuela, date of crossing, education level and job status, among other. It was, in fact, a successful effort as it was able to register 442,462 undocumented Venezuelans, belonging to to 253,575 different households. Importantly for the purpose of this study, when advertising the RAMV, the government explic- itly stated that registering will not result in deportations or have any negative legal consequences. Similarly, the RAMV was never advertised as a platform to receive work permits or any other legal benefit that would facilitate the migrants’ stay. In fact, it was advertised simply as a registry to count and identify migrants. In conversations with government officials who oversaw the process, we were able to confirm that, in fact, these were the actual intentions. However, in July of 2018, days before leaving office, outgoing President Juan Manuel Santos decreed that all those Venezuelans registered in the RAMV would be eligible to receive a regular migratory status that would allow them to stay in the country and join the labor force. In particular, the undocumented migrants registered in RAMV would be able to apply for the Permiso Especial de Permanencia, or PEP, a comprehensive two-year regular migratory status that the Colombian government had previously granted to Venezuelan migrants in the country. The difference between previous waves of PEP and this one –known as PEP-RAMV– was that it was explicitly created for undocumented Venezuelans registered in RAMV. According to the decree, these immigrants would be able to request the PEP-RAMV visa during the third quarter of 2018. Each immigrant was given a time window to apply based on their RAMV registration number, which was provided to them as their record was entered into the system, in real time. This number was exogenously allocated to each immigrant across all the authorized points. By the end of 2018, 281,803 of the 442,462 individuals deemed eligible for the PEP-RAMV had received one. According to the RAMV data, the immigrants under consideration are young and have a mod- erate level of education: 75 percent of RAMV migrants are between the ages of 15 and 64, and over 83 percent of this group has completed at least secondary education. In fact, compared with 5 the Colombian labor force, this group is younger and more educated. According to 2018 population estimates, 66 percent of the Colombian population is between the ages of 15 and 64, and 61.5 percent of the active labor force in 2017 had completed at least basic secondary education. At the time of the implementation of the RAMV survey, 46.3 percent of working age migrants were engaged in some level of employment in the informal sector. 3 Empirical Strategy 3.1 Data Sources We exploit monthly and city variation in our estimates. The data that we use can be grouped as follows: 1. Registro Administrativo de Migrantes Venezolanos, RAMV. We use administrative data sourced from the RAMV survey from where we compute the number of people eligible to the PEP- RAMV visa as well as the actual number of individuals that applied for (and received) the PEP-RAMV visa by municipality. Using the individual records we also compute the allocated time window that each individual had for registration to the PEP-RAMV visa (based on the registration number), which we use for our identification strategy. 2. Crime Reports. We use the reports published by the Colombian National Police with daily counts on all crime reported in Colombia between January of 2017 and December of 2019 (one year before and after the regularization program). The reports include information on the type of crime, the municipality where it was reported, as well as the nationality and the gender of the individual who reports the crime. We use these reports to construct crime counts on a monthly basis for each capital city in the country. Note that we do not have a way to link the individual crime data to the individuals registered in RAMV (who were subsequently awarded the PEP-RAMV visa). As such, the best possible approach is to aggregate the data at the city level, as we are effectively doing. By limiting the reports to those crimes reported by Venezuelans we are trying to capture the effect of the amnesty on the propensity to report crimes by Venezuelan immigrants themselves. 3. Municipality Controls. We follow the same control structure as in Bahar et al. (2021) and use a number of baseline municipal covariates as controls which include: night light density, 6 internal conflict-related variables, GDP municipal composition, distance to the Venezuelan border, number of Venezuelans with previous roll-outs of PEP visas given to documented immigrants (particularly the PEP1 and PEP2), as well as proxies for government activity. Administrative information at the municipal level comes from the CEDE municipal panel, the Ministry of Defense, the National Planning Department, and DANE (Colombia’s national statistical agency). Night light density comes from the National Oceanic and Atmospheric Administration. Table 1 presents the descriptive statistics for all the main variables that we use in our analysis. Our sample includes 1,080 observations which corresponds to the 30 municipalities that represent capital cities of most departments in Colombia and 36 months.4 The reason we focus on capital cities, that host about two thirds of all Venezuelans in our sample (e.g., using the locations reported in the RAMV data), is to reduce possible biases in terms of crime reporting due to internal conflict still persistent in rural areas (Ibañez et al., 2017). Panel A of Table 1 reveals the number of crimes reported by Venezuelans as a share of all crimes (to see total crimes per 100,000 inhabitants see Online Appendix Table A1). The share of all crimes reported by Venezuelans on average is 1 percent. The most common crime reported in our sample during the period of study is homicides, averaging 3 percent of all crimes. Note that these numbers are quite low, and we expect them to be as such (for refecence, at the time, Venezuelans represented at most 2% of the population of the country as a whole, with heterogeneity in terms of locations, naturally). Yet, also note that the numbers have enough variation to exploit for our exercise. Indeed, given the high standard deviations relative to the mean, it can be seen there is significant variation across crime reports and across cities and months. In terms of the treatment variable, Panel B of Table 1 shows that for the average city in our sample there were 4,839 Venezuelans that received the PEP-RAMV visa, but that number ranges between 13 to 35,729 across all cities in the sample. In proportion to the local population, the number of PEP-RAMV visas awarded is, on average, 217.98 per 100,000 inhabitants. Note that while the number of observations in this table is 1080, since this is the treatment variable there is no across-time variation for these variables (as will be clear as we describe our empirical strategy). Panel C presents the summary statistics for all the different controls in our specifications for the 4 Colombia has 32 departments, the panel only includes those 30 cities for which there were consistent monthly crime reports and control variables. 7 30 cities in our sample at baseline. Figure 1 visually represents these data in a map of Colombia. The figure plots for each capital city the number of PEP-RAMV holders per 100,000 inhabitants using the scale of the markers (circles). It also shows using a light to dark color scale the total crimes reported (per 100,000 inhabitants, for consistency) in the different cities. For visualization purposes, the light-to-dark color scale is represented on the departments where such capital city is located. This graph allows us to understand that there is a great deal of variation for both the treatment variable and the outcome under consideration. 3.2 Empirical strategy Our goal is to establish whether the awarding of the PEP-RAMV visa to undocumented immigrants resulted in more crime reports by Venezuelans. We do so by effectively comparing cities with different intensities in terms of the Venezuelan population who were awarded the PEP-RAMV visa before and after receiving the regular migratory status. Effectively, we estimate a differences-in-difference specification as follows: ShareCrimeReportsV ct EN = βDID P EPc × I [P ostAugust2018]t (1) + controlsc × yeart + γc + ηt + εct c Z where c indexes city and t indexes for time (month-year). Our main outcome variable is ShareCrimeReportsV ct EN which is the share of crime reports by Venezuelans in city c and month-year t of total crimes for that same city and period. Namely: CrimeReportsV ct EN ShareCrimeReportsV ct EN = T CrimeReportsctOT Our estimator of interest is βDID which reflects a difference-in-differences approach, as it esti- mates the interaction between our treatment variable, which is P EPc , the number of Venezuelans who were granted the PEP visa per municipality c per 100,000 inhabitants5 , interacted with a dummy that takes the value one for all the time periods that follow August 2018, when the PEP 5 Specifically, it corresponds to the number of PEP holders (between 10 to 64 years) divided by the population of municipality c between 10 and 64 years, multiplied by 100,000. 8 roll-out began. In that sense, when it comes to the time dimension, it is important to note that βDID can be interpreted as the average effect for all the months in the sample on and after August 2018 versus all the months before August 2018. It is worth noting that our specification, by using the share of total crimes as the dependent variable, is aiming to estimate a change in the composition of reports by reporter’s nationality, and not an increase in reports overall. In other words, our goal is not to estimate the effect of the amnesty on the amount of crimes that occur, but rather on the likelihood of Venezuelans reporting crimes, keeping constant the overall number of crimes. While we believe our main specification achieves this goal, we also present results in Online Appendix Section B.1 that estimates an alternative specification that uses the rate of crime reports by Venezuelans (i.e., crimes per 100,000 inhabitants) as the dependent variable, controlling for the overall reports of crimes in that locality and period. Using this alternative specification we find our results to be robust to the main specification. All continuous variables on the right-hand-side, including the treatment variable, were trans- formed using the inverse hyperbolic sine transformation (see Burbidge et al., 1988 and MacKinnon and Magee, 1990 for more on this transformation). Thus βDID can be interpreted as a semi-elasticity of crime reports with respect to the share of the population that received a regular migratory status. We also include a full set of predetermined municipal characteristics measured before the begin- ning of our period of analysis (in order to reduce endogeneity concerns), represented by the set Z . We further interact these control variables with year dummies (denoted by yeart ) to flexibly account for potential differential non-parametric trends for each city c that could emerge from these variables as they explain crime reports by Venezuelans in ways that are correlated with the treatment. The variables included in Z are (1) percentage of households in Colombia with at least one unsatisfied basic need in 2005, (2) percentage of households in Colombia with at least one informal worker in 2005, (3) number of terrorist attacks in 1995 (to proxy for the internal conflict), (4) night light density in 2009, (5) number of financial institutions in 1995, (6) number of tax collection offices in 1995, (7) agriculture, (8) industry, and (9) services GDP in 2009, (10) central government transfers in 2009, (11) total municipal expenditures in 2009, (12 and 13) number of Venezuelan residents awarded with the two previous rounds of the PEP visa, and (14) the inverse distance to the closest of the five authorized crossing points between Venezuela and Colombia. In addition, the specification includes fixed effects for city (γc ) and for month-year (ηt ). Unless otherwise noted, we cluster our standard errors at the city level to account for geographic serial 9 correlation. 3.3 Identification Strategy Despite the relatively complete number of controls, by estimating specification (1) through OLS we would not be able to assert with full certainty that we are actually estimating the causal impact of the PEP-RAMV visa on crime rates. This is because there could still be self-selection forces biasing our results or other omitted variables not captured in our specification. The most obvious threat to our identification using an OLS estimator is that immigrants self-select into certain municipalities based on the characteristics of the local communities, which in turn might make them feel more secure to report crimes. To deal with this possibility (or others) we follow our approach in (Bahar et al., 2021). In particular, in order to significantly reduce endogeneity concerns, we employ an instrumental variable approach and estimate specification (1) through two-stages least squares.6 Our instrument is the average number of days that the undocumented Venezuelan immigrants had in order to register to receive the PEP-RAMV visa in the last quarter of 2018. The number of days was in turn defined by the form number of the registration to the RAMV survey (each one of these numbers defined a time window in which the immigrant registered in RAMV could apply for the PEP-RAMV visa). This number was, according to the accounts of government officials overseeing this process, given by the local registration point to immigrants in the order they arrived and registered. More importantly for our purposes, the numbers assigned to each registration center and subsequently provided to each immigrant did not respond to any particular geographic or socioeconomic characteristic of the municipalities. Government officials report that this approach was done to scatter the applications of PEP-RAMV visas evenly across time. Based on the individual numbers for all PEP-RAMV visa holders at the time of registration in the RAMV in a given municipality, we estimate the average registration days per municipality as: RAMV registrants assigned to time window jc Reg. Daysc = × [Days in time window j] (2) Total RAMV registrantsc j K 6 Applying instrumental variables in a difference-in-differences setting is common in the economics literature. As a pioneering example of combining these methods see the seminal study by Duflo (2001). 10 where K represents each of the 22 possible individual time windows assigned to migrants in the RAMV to request a PEP-RAMV visa which range between 78 and 141 days.7 Therefore, Reg.Daysc represents the average number of registration days available for the eligible population to receive the PEP-RAMV in that city. Our identification strategy relies on the fact that there is a strong correlation between the length of the window to apply for the PEP-RAMV visa and the actual number of people who actually applied (and received it). This is the case, as can be seen in a simple plot in Online Appendix Figure A3. In addition, the exclusion restriction that would need to hold for our instrument to be valid and to be able to interpret our results as causal is that the effect of the time available to register to the PEP-RAMV visa in each municipality (which were exogenously assigned to immigrants as they registered) affects crime reports by Venezuelans only and only through the actual number of Venezuelans who were actually awarded the PEP-RAMV visa. This, to us, seems like a very reasonable assumption. Note that while the allocation of times to register to PEP was exogenous to the migrant, our identification relies on geographic variation of registration times across municipalities. The reason there exists such geographic variation is simply because the higher concentration of Venezuelans near the border, which results in higher average registration times in bordering municipalities as compared to areas further away from the border, where less immigrants are located. This would pose a threat to our identification strategy if there are other aspects of being close to the border that would affect the reports of crimes by Venezuelans through channels other than the regularization itself. Given the characteristics of the natural experiment we are exploiting, as well as all the controls in our specifications, we believe that it is quite reasonable to assume that the exclusion restriction with this instrument hold, especially since we control in all of our specifications for distance to the border and all the vast set of variables outlined above. However, as an abundance of caution, we present consistent results using a second identification strategy, that relies on the historic distribution of Venezuelan immigrants across Colombia (based on the 2005 census, more than two decades before the amnesty) to predict the current geographic distribution of PEP recipients. Our results using this exogenous variation are qualitatively robust, though they exploit a different source of exogenous variation which is reflected in the results. We 7 See Figure A2 in Online Appendix for its distribution. 11 discuss this in detail when presenting those results in Online Appendix B.2. 4 Results 4.1 Main Results Our main results are presented in Table 2, which presents the OLS (upper panel) and 2SLS (lower panel) estimation of equation (1). Columns 1 to 5 present results using as the dependent variable the share of reports by Venezuelans for different types of crimes. The different type of crimes are homicides, threats, domestic violence, theft, and sex crimes. Column 6 presents results for the share of all crime reports by Venezuelans. The main variable of interest is in the first row, which corresponds to β DID of Specification (1). All columns include city and month-year fixed effects as well as the set of controls described in Section 3.2. Standard errors are clustered at the municipality level. First, it is important to mention that both OLS and 2SLS estimations yield results that are not statistically different from one another when compared. We also note that the first stage Kleibergen- Paap F statistic is large eliminating any concern of weak instrumentation. For brevity, we focus our interpretation on the 2SLS (Panel B) results, since we feel those overcome endogeneity concerns. In particular, the results of Column 6 show that in the months after the roll-out of the PEP-RAMV visa, municipalities that had double the number of visa holders (per 100,000 inhabitants) saw an increase in the share of crimes reported by Venezuelan of 0.26 percentage points , on average. Considering that the average share of total crimes reported by Venezuelans is 1% (as documented on Table 1, this constitute an increase of 25%, which is an economically significant result. In terms of specific types of crimes, we find that our results show an statistically significant estimator for homicides, theft„ as well as for domestic violence and for sex crimes. Looking at the point estimates and the summary statitics, across all these dimensions, the results are also economically significant. for example, a city that had double the amount of Venezuelan immigrants that received the PEP visa as compared to the mean (relative to the local population), experienced an increase in reports of sex crimes by Venezuelans of 0.41 percentage points. Since the average reporting of sex crimes by Venezuelans is 1% of all crimes, this corresponds to more than a 40% increase in reporting after Venezuelans receiving a regular migratory status. The same number for domestic violence crime reports, for example, is about 70%. 12 The most obvious interpretation for our results is that by having received a regular migratory status, previously undocumented Venezuelans are empowered to report crimes to the authorities without the risk of of deportation or any other legal repercussion. If this is the case, we would expect this effect to be stronger among more vulnerable populations of Venezuelans before receiving the PEP-RAMV visa. One dimension available in our data to exploit this hypothesis is gender, which seems particularly important given that the results above concentrate on crimes where women tend to be victims at much higher rates, such as domestic violence and sex crimes. In fact, Venezuelan women tend to report domestic violence and sex crimes at a much higher rate than Venezuelan men, as expected, and as we show in Online Appendix Table A1. In that sense, we ask: are crime reports increasing differently whether the reports are made by male or female immigrants following the migratory amnesty? Table 3 replicates the main estimation using as the dependent variable the total reports by Venezuelan males (Column 1) and females (Column 2). The results show that indeed it is reports by females driving the overall patterns identified above. When focusing on the 2SLS estimates, we find a 0.35 percentage points increase in crime reports by female Venezuelan immigrants in municipalities with twofold more PEP-RAMV visa holders per 100,000 inhabitants. Again, since the share of crime reports by Venezuelans on average is 1 percent, this corresponds to an increase of 35 percent. Results when using reports by male Venezuelan immigrants are statistically insignificant and the magnitude of the coefficient estimates smaller, hinting that the totality of the effect reported in Table 2 is driven by female immigrants. Note that we focus on total crime reports given the difficulties we encountered with focusing on reports by gender and by type simultaneously, as the number of crimes within each one of those cells become much smaller, therefore significantly reducing our ability to estimate with enough precision. However, despite this, we do present results exploiting this variation in Online Appendix Section B.3, and find consistent results showing that most of the effect is concentrated on reports by female Venezuelan immigrants in an event study setting. 4.2 Event Study We also present our results in an event study format, measuring the effect across time averaged by quarter and by gender. The benefit of looking at this complementary result is twofold. First, since our sample includes many months before the roll-out of the PEP-RAMV visa, we can test for whether we detect any pretrends which would threaten the validity of our results. Second, they also 13 allow us to see the evolution of crime reports over time after the program roll-out. The results are presented in Figure 2 using three dependent variables in the same format as above: the share of crime reports by all Venezuelan immigrants, by male Venezuelan immigrants, and by female Venezuelans immigrants.8 The figure visualizes the results of estimating Specification (1) with the only difference that the treatment is interacted with 12 indicator variables representing quarters (i.e., 3 month periods) from Q1 of 2017 to Q4 of 2019 (instead of the I [P ostAugust2018]t dummy). In that sense, each marker represents the average effect of the treatment on crime reports for the three-month period that correspond to each quarter. The upper, middle and lower panels presents results using total crime reports by all Venezuelan immigrants, male Venezuelan immigrants, and female Venezuelan immigrants, respectively. The gray vertical line represents the beginning of the roll-out of the PEP- RAMV visa program, in August of 2018.9 The whiskers represent 95% confidence intervals, based on standard errors clustered at the city level. In all panels of Figure 2 we see no evidence of pre-trends, as all the estimators before the grey line –which symbolizes the roll-out of the treatment– are not statistically distinguishable from zero, relative to the base period. This is encouraging, as it reduces possible endogeneity concerns. What we also see clearly in the figure, is that –consistently with Table 3 which measures the differences on average before the "before" months and the "after" months– it is reports by female Venezuelan immigrants where we can distinguish a stronger effect, and we can barely distinguish any robust effect when focusing on crime reports for male Venezuelan immigrants (just in the last few months of 2019, more than a year after the program roll-out).10 As such, the positive coefficients in the post-August 2018 period for all Venezuelans (upper panel) are almost entirely driven by Venezuelan female immigrants. In addition, the uptake in reporting crimes for female Venezuelan immigrants seems to continue for more than a year after the amnesty, implying that it is not an immediate effect that then reverts to pre-trends levels. These results, we believe, reinforce our interpretation of empowerment –female empowerment in particular– being a driver of the documented patterns. After significantly reducing the fear of depor- tation, or of any other repercussion by authorities for that matter, immigrants report more crimes to 8 As noted above, Online Appendix Section B.3 presents the results of event studies using as the dependent variable shares of reports by type of crime for all, male, and female Venezuelan immigrants. 9 Results are based on OLS estimates since the limitation of having only one instrument. 10 When replicating this event study using type of crimes, the point estimates using reports by mail are all highly inprecise, even during those last two periods; whereas we do see statistical significance for female reports, as shown in Online Appendix B.3. 14 the police. According to our evidence, this is particularly the case for crimes that disproportionately affect women, such as domestic violence and sex crimes. Evidence from qualitative interviews with Venezuelan immigrants suggests that indeed receiving the PEP-RAMV visa empowered individuals to report abuses to their basic human rights, without fearing deportations or other punishments. 5 Concluding Remarks In this paper we examine the impacts of an amnesty that gave regular migratory status to ap- proximately 281 thousand Venezuelan migrants and refugees in Colombia on their crime reporting behavior. We find that cities where more undocumented Venezuelans received a regular migratory status experienced an increase in crime reports to local authorities by Venezuelan migrants. Our results are driven by reports of domestic violence and sex crimes, and by reports by Venezuelan female immigrants. By focusing on crime-reporting to local authorities, we believe we are capturing a behavior change consistent with empowerment of a highly vulnerable population such as undocumented immigrants. As such our results have important policy implications as they reflect difficult-to-measure social benefits of migratory amnesty programs that go often unnoticed. We hope these results open the door to more studies that aim to better understand other social behavior changes and the mechanisms behind them when it comes to the protection of undocumented immigrants. References Ajzenman, Nicolás, Patricio Dominguez-Rivera, and Raimundo Undurraga. “Immigration, Crime, and Crime (Mis)Perceptions.” SSRN Electronic Journal https://www.ssrn.com/abstract= 3689249. Alonso-Borrego, César, Nuno Garoupa, and Pablo Vazquez. “Does immigration cause crime? Ev- idence from Spain.” American Law and Economics Review 14, 1: (2012) 165–191. http: //dx.doi.org/10.1093/aler/ahr019. Amuedo-Dorantes, Catalina, Cynthia Bansak, and Steven Raphael. “Gender differences in the labor market: Impact of IRCA.” American Economic Review 97, 2: (2007) 412–416. 15 Bahar, Dany, Meagan Dooley, and Andrew Selee. “Venezuelan Migration, Crime, and Misper- ceptions: A Review of Data from Colombia, Peru, and Chile.” Migration Policy Institute and The Brookings Institution https://www.brookings.edu/wp-content/uploads/2020/09/ migration-crime-latam-eng-final.pdf. Bahar, Dany, Ana María Ibáñez, and Sandra V. Rozo. “Give me your tired and your poor: Impact of a large-scale amnesty program for undocumented refugees.” Journal of Development Economics 151: (2021) 102,652. Baker, Scott. “Effects of Immigrant Legalization on Crime.” American Economic Review: Papers and Proceedings 105: (2017) 210–210. Bell, Brian, Francesco Fasani, and Stephen Machin. “Crime and Immigration: Evidence from Large Immigrant Waves.” The Review of Economics and Statistics 95, 4: (2013) 1278–1290. https: //doi.org/10.1162/REST_a_00337. Bianchi, Milo, Paolo Buonanno, and Paolo Pinotti. “Do Immigrants Cause Crime?” Journal of the European Economic Association 10, 6: (2012) 1318–1347. https://onlinelibrary.wiley.com/ doi/abs/10.1111/j.1542-4774.2012.01085.x. Borjas, George J., Jeffrey Grogger, and Gordon H. Hanson. “Immigration and the economic status of African-American men.” Economica 77, 306: (2010) 255–282. https://onlinelibrary.wiley. com/doi/abs/10.1111/j.1468-0335.2009.00803.x. Bratsberg, Bernt, James F Ragan, Jr, and Zafar M Nasir. “The effect of naturalization on wage growth: A panel study of young male immigrants.” Journal of Labor Economics 20, 3: (2002) 568–597. Burbidge, John B, Lonnie Magee, and A Leslie Robb. “Alternative transformations to handle extreme values of the dependent variable.” Journal of the American Statistical Association 83, 401: (1988) 123–127. Butcher, Kristin F., and Anne Morrison Piehl. “Cross-City Evidence on the Relationship between Immigration and Crime.” Journal of Policy Analysis and Management 17, 3: (1998) 457–493. http://www.jstor.org/stable/3325557. 16 Duflo, Esther. “Schooling and Labor Market Consequences of School Construction in Indonesia: Evidence from an Unusual Policy Experiment.” American Economic Review 91, 4: (2001) 795– 813. https://www.aeaweb.org/articles?id=10.1257/aer.91.4.795. Fasani, Francesco. “Immigrant crime and legal status: evidence from repeated amnesty programs.” Journal of Economic Geography 18, 4: (2018) 887–914. Fasani, Francesco, Giovanni Mastrobuoni, Emily G. Owens, and Paolo Pinotti. Does Immigration Increase Crime?: Migration Policy and the Creation of the Criminal Immigrant. Cambridge University Press, 2019. Freedman, Matthew, Emily Owens, and Sarah Bohn. “Immigration, Employment Opportunities, and Criminal Behavior.” American Economic Journal: Economic Policy 10, 2: (2018) 117–51. https://EconPapers.repec.org/RePEc:aea:aejpol:v:10:y:2018:i:2:p:117-51. Ibañez, Ana M., Amy Ritterbusch, and Catherine Rodriguez. “Impact of a Judicial System Reform on Police Behavior: Evidence on Juvenile Crime in Colombia.” SSRN Electronic Journal http: //www.ssrn.com/abstract=2931146. Kaushal, Neeraj. “Amnesty programs and the labor market outcomes of undocumented workers.” Journal of Human Resources 41, 3: (2006) 631–647. Knight, Brian, and Ana Tribin. “Immigration and Violent Crime: Evidence from the Colombia- Venezuela Border.” Technical report, National Bureau of Economic Research, Cambridge, MA, 2020. Kossoudji, Sherrie A, and Deborah A Cobb-Clark. “Coming out of the shadows: Learning about legal status and wages from the legalized population.” Journal of Labor Economics 20, 3: (2002) 598–628. MacKinnon, James G, and Lonnie Magee. “Transforming the dependent variable in regression models.” International Economic Review 315–339. Orrenius, Pia M, and Madeline Zavodny. “Do amnesty programs reduce undocumented immigration? Evidence from IRCA.” Demography 40, 3: (2003) 437–450. Pinotti, Paolo. “Clicking on Heaven’s Door: The Effect of Immigrant Legalization on Crime.” American Economic Review 107: (2017) 138–168. 17 Spenkuch, Jörg L. “Understanding the Impact of Immigration on Crime.” American Law and Economics Review 16, 1: (2014) 177–219. http://dx.doi.org/10.1093/aler/aht017. 6 Tables and Figures [Table 1 about here.] [Table 2 about here.] [Table 3 about here.] [Figure 1 about here.] [Figure 2 about here.] 18 Figure 1: Total Crime Reports by Venezuelans and PEP-RAMV Holders (per 100K) The figure plots the main variables of study: the total crime reports by Venezuelans and the number of Venezuelan PEP-RAMV holders per 100,000 inhabitants. The intensity of crime reports are represented by the shades on the departments which correspond to their capital cities, whereas the intensity of the treatment (PEP-RAMV holders) are represented by the size of the marker of each city. 19 Figure 2: Event Studies, Total Crime Reports by Gender All .015 {&beta^{DID}} (95% CI) 0 .005 −.005 .01 1 2 3 4 1 2 3 4 1 2 3 4 −Q −Q −Q −Q −Q −Q −Q −Q −Q −Q −Q −Q 17 17 17 17 18 18 18 18 19 19 19 19 20 20 20 20 20 20 20 20 20 20 20 20 Quarter Male .004 {&beta^{DID}} (95% CI) −.002 0−.004 .002 1 2 3 4 1 2 3 4 1 2 3 4 Q Q Q Q Q Q Q Q Q Q Q Q 7− 7− 7− 7− 8− 8− 8− 8− 9− 9− 9− 9− 1 1 1 1 1 1 1 1 1 1 1 1 20 20 20 20 20 20 20 20 20 20 20 20 Quarter Female .01 .002 .004 .006 .008 {&beta^{DID}} (95% CI) 0 1 2 3 4 1 2 3 4 1 2 3 4 Q Q Q Q Q Q Q Q Q Q Q Q 7− 7− 7− 7− 8− 8− 8− 8− 9− 9− 9− 9− 1 1 1 1 1 1 1 1 1 1 1 1 20 20 20 20 20 20 20 20 20 20 20 20 Quarter The figure presents the estimation of an event study based on Specification (1), estimating the effect of the treatment interacted with 12 quarter dummies from Q1 of 2017 to Q4 of 2019. The upper, middle, and lower panels presents results using total crime reports by all Venezuelan immigrants, male Venezuelan immigrants, and female Venezuelan immigrants, respectively. The gray vertical line represents the beginning of the roll-out of the PEP-RAMV visa program, in August of 2018. The results are based on OLS estimates. The whiskers represent 95% confidence intervals, based on standard errors clustered at the city level. 20 Table 1: Summary Statistics Variable N Mean sd Min Max Panel A: Crime Reports by Venezuelans (share of all reports) Homicides (share) 1,080 0.03 0.12 0.00 1.00 Threats (share) 1,080 0.00 0.02 0.00 0.50 Domestic Violence (share) 1,080 0.01 0.03 0.00 0.33 Theft (share) 1,080 0.01 0.01 0.00 0.11 Sex Crimes (share) 1,080 0.01 0.04 0.00 0.45 Total Crimes (share) 1,080 0.01 0.01 0.00 0.15 Panel B: Treatment and other migration Variables PEP-RAMV holders 1,080 4,839.60 7,686.24 13.00 35,729.00 PEP holders per 100K (ages 10-64) 1,080 217.98 375.02 3.33 1,767.00 Venezuelan population 2005 1,080 676.83 1,171.58 0.00 4,578.00 Total population 2005 1,080 644,783.50 1,252,964.87 12,897.00 6,778,691.00 Panel C: Control Variables (at baseline) Night Light Density (2009) 30 12.73 14.47 0.03 53.98 Total Mun. Income (COP, Billions) (2009) 30 628.79 1,513.32 9.06 8,131.36 Mun. Public Expenditures (COP, Billions) (2009) 30 617.83 1,434.27 10.52 7,553.10 Total Central Gov. Transfers (COP, Billions) (2009) 30 222.60 381.46 4.98 2,048.44 Number of Financial Institutions (1995) 30 24.23 47.62 1.00 252.00 Number of Tax Collection Offices(1995) 30 8.87 18.93 1.00 99.00 Homicide Rate (Per 100,000 Indv.) (2009) 30 34.88 21.45 7.00 109.91 N. of Terrorist Attacks (1995) 30 0.07 0.37 0.00 2.00 Unsatified Basic Needs (UBN % Households) (2005) 30 27.46 17.76 9.16 89.51 Informal Labor* (% Households) (2005) 30 88.33 5.54 77.16 96.51 GDP Agriculture (COP, Billions) (2009) 30 53.79 42.21 4.50 145.92 GDP Industry (COP, Billions) (2009) 30 2,138.95 4,368.66 8.44 22,970.06 GDP Services (COP, Billions) (2009) 30 4,444.15 13,266.97 87.69 72,695.80 PEP1 (August 2017-October 2017) 30 1,777.50 5,133.56 0.00 27,703.00 PEP2 (February 2018-June 2018) 30 2,880.43 8,737.14 0.00 47,389.00 Inverse Distance to Closest Border Crossing 30 0.13 0.35 0.00 1.00 The table presents summary statistics of the main variables used in our study for 1,080 observations which correspond to 30 cities and 36 months. Panel A presents statistics of crime reports by Venezuelans by type of crime per 100,000 local inhabitants. Panel B summarizes statistics on the treatment variables: the number of Venezuelans that received a PEP-RAMV visa, in total and per 100,000 inhabitants. Panel C presents summary statistics for all different controls for the 30 cities in our sample, at baseline. 21 Table 2: Main Results Panel A: OLS (1) (2) (3) (4) (5) (6) homicides threat dom_violence theft sex_crime tot_crime P EPc × I [P ostAugust2018]t 0.0189 0.0014 0.0039 0.0015 0.0041 0.0026 (0.007)*** (0.001) (0.002)** (0.001)** (0.001)*** (0.001)*** N 1080 1080 1080 1080 1080 1080 Adj R2 0.27 0.17 0.37 0.48 0.35 0.62 Panel B: 2SLS homicides threat dom_violence theft sex_crime tot_crime P EPc × I [P ostAugust2018]t 0.0139 0.0016 0.0071 0.0023 0.0082 0.0042 (0.010) (0.002) (0.004)* (0.001)*** (0.002)*** (0.001)*** N 1080 1080 1080 1080 1080 1080 r2 0.07 0.08 0.19 0.19 0.17 0.33 KP F Stat 21.51 21.51 21.51 21.51 21.51 21.51 The table presents the OLS (upper panel) and 2SLS (lower panel) estimation of Specification (1).Columns (1) to (5) present results using reports for different types of crimes by Venezuelans as share of all crime reports as the dependent variable, in their inverse hyperbolic sine form. The different type of crimes are homicides, threats, domestic violence, theft, and sex crimes. Column (6) presents results for the totality of all crime reports by Venezuelans as a share of all crime reports as the dependent variable. All columns include city and month-year fixed effects as well as the set of controls described in Section 3.2. Standard errors are clustered at the city level. ∗ p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01 22 Table 3: Results, Total Crime Reports by Gender Panel A: OLS (1) (2) tot_crime_m tot_crime_f P EPc × I [P ostAugust2018]t 0.0006 0.0021 (0.000)* (0.001)*** N 1080 1080 Adj R2 0.41 0.50 Panel B: 2SLS tot_crime_m tot_crime_f P EPc × I [P ostAugust2018]t 0.0007 0.0035 (0.000) (0.001)*** N 1080 1080 r2 0.13 0.27 KP F Stat 21.51 21.51 The table presents the OLS (upper panel) and 2SLS (lower panel) estima- tion of Specification (1). Column (1) presents results using crime reports by Venezuelan male immigrants, whereas Column (2) presents results us- ing crime reports by Venezuelan female immigrants, both as a share of all crimes, and in their inverse hyperbolic sine form. All columns include city and month-year fixed effects as well as the set of controls described in section 3.2. Standard errors are clustered at the city level. ∗ p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01 23 Online Appendix for Empowering Migrants: Impacts of a Migrant’s Amnesty on Crime Reports r Dany Bahar r Sandra V. Rozo ○ Ana María Ibánez ○ November 2, 2021 A Extended Summary Statistics A.1 RAMV Registration Points Figure A1 plots the different municipalities in Colombia where there was a physical registration point of the RAMV program between April to June 2018. As can be seen, it covered widely the national territory. [Figure A1 about here.] A.2 Further Crime Reports Statistics Table A1 presents the total crimes reported per 100,000 inhabitants. Total crimes, in our sample, range from 0 to 181, which is a relatively small number. However, note that Venezuelans are a small share of the Colombian population, so we would not expect a very large amount of crime reports by such a small minority. The most common type of crime reports is theft, by far. Table A1 presents the crimes reported in our sample per 100,000 local inhabitants, by gender. The upper panel presents crime reports by Venezuelan men while the lower panel present crime reports by Venezuelan women. Here we note that, on average, women report about 30% more crimes than men (using the total crimes figure). This is mostly driven, as expected, by reports of domestic violence and of sex crimes, where Venezuelan women report twice as much and seven times as much, respectively. [Table A1 about here.] 1 A.3 Instrumental Variable Figure A2 plots the distribution of the average registration days per municipality in our sample, our instrumental variable. As can be seen, the variable distributes between the values of roughly 80 to 120 days, resembling a normal distribution. [Figure A2 about here.] Figure A3 visualizes the relationship between the average registration days per municipality (detailed in the main body of the text, in Section 3.3) against the number of Venezuelans in that same municipality who were awarded the PEP-RAMV visa per 100,000 inhabitants. This visualization corresponds to the first stage of the 2SLS estimation described as part of our identification strategy. [Figure A3 about here.] A.4 Evolution of Crime Reports by Venezuelans Figure A4 presents the average reports of total crimes by Venezuelan immigrants, as a share of total, for municipalities below and above median in terms of the treatment (e.g., the share of Venezuelans with the PEP-RAMV visa for each 100,000 inhabitants) aggregated by quarters. Note that this uses only raw data, with no controls whatsoever. As a purely descriptive exercise, the graph shows that both group of cities had somewhat parallel trends up to the second quarter of 2018, after which there is a divergence: Venezuelans file many more reports than in municipalities with less Venezuelans with the PEP-RAMV visa. [Figure A4 about here.] 2 Figure A1: RAMV Registration Points The figure marks Colombian municipalities with physical registration points of the RAMV registry between April to June of 2018. Size of the markers are proportional to the number of points per municipality. 3 Figure A2: Distribution of Instrumental Variable .08 .06 Density .04 .02 0 90 100 110 120 Reg.Days The figure plots the distribution of the average registration days per municipality using the municipalities in our sample. 4 Figure A3: Visualization of First Stage 8 Share of PEP−RAMV holders (asinh) 4 2 6 90 100 110 120 Average Registration Days The figure plots the average registration days per municipality against the number of Venezuelans in that same municipality who were awarded the PEP-RAMV visa per 100,000 inhabitants. The graph uses only municipalities in our sample. 5 Figure A4: Evolution of Crime Reports by Venezuelans by Treatment Intensity .01 Crime Reports by Venezuelan (Share) .002 .004 .006 0 .008 1 2 3 4 1 2 3 4 1 2 3 4 −Q −Q −Q −Q −Q −Q −Q −Q −Q −Q −Q −Q 17 17 17 17 18 18 18 18 19 19 19 19 20 20 20 20 20 20 20 20 20 20 20 20 Quarter Below Median Above Median The figure plots the evolution of total crime reports by Venezuelans, as a share of total crime reports, for cities with treatment (i.e., PEP-RAMV holders per 100K inhabitants) above median and below median. The dashed vertical line marks August of 2018 when the announcement and subsequent roll-out of the RAMV-PEP visa started. 6 Table A1: Total Crime Reports Variable N Mean sd Min Max Panel A: Total Crime Reports (rates per 100,000) Homicides (per 100K) 1,080 1.99 1.81 0.00 18.97 Threats (per 100K) 1,080 12.00 8.46 0.00 52.82 Domestic Violence (per 100K) 1,080 20.82 13.42 0.00 98.67 Theft (per 100K) 1,080 83.80 30.95 12.69 193.44 Sex Crimes (per 100K) 1,080 7.95 4.61 0.00 31.16 Total Crimes (per 100K) 1,080 127.17 40.56 40.49 266.15 Panel B: Crime Reports by Venezuelans (rates per 100,000) Homicides (per 100K) 1,080 0.08 0.42 0.00 4.88 Threats (per 100K) 1,080 0.04 0.24 0.00 3.07 Domestic Violence (per 100K) 1,080 0.15 0.49 0.00 9.18 Theft (per 100K) 1,080 0.44 0.90 0.00 14.65 Sex Crimes (per 100K) 1,080 0.08 0.42 0.00 6.12 Total Crimes (per 100K) 1,080 0.79 1.63 0.00 19.59 Panel C: Crime Reports by Venezuelan Males (rates per 100,000) Homicides (per 100K) 1,080 0.07 0.38 0.00 4.88 Threats (per 100K) 1,080 0.02 0.14 0.00 3.06 Domestic Violence (per 100K) 1,080 0.02 0.20 0.00 4.74 Theft (per 100K) 1,080 0.22 0.45 0.00 5.44 Sex Crimes (per 100K) 1,080 0.01 0.12 0.00 3.06 Total Crimes (per 100K) 1,080 0.34 0.72 0.00 8.71 Panel D: Crime Reports by Venezuelan Females (rates per 100,000) Homicides (per 100K) 1,080 0.01 0.15 0.00 4.78 Threats (per 100K) 1,080 0.02 0.18 0.00 3.06 Domestic Violence (per 100K) 1,080 0.12 0.44 0.00 9.18 Theft (per 100K) 1,080 0.22 0.68 0.00 14.65 Sex Crimes (per 100K) 1,080 0.07 0.39 0.00 6.12 Total Crimes (per 100K) 1,080 0.45 1.19 0.00 15.24 The table presents summary statistics of total crime reports per 100,000 inhabitants used in our study for 1080 observations which correspond to 30 cities and 36 months, as reported in total (Panel A), by Venezuelans (Panel B), by Venezuelan Males (Panel C) and by Venezuelan Females (Panel D). 7 B More Results and Robustness B.1 Using Crime Rates We also present reports with an alternative specification, that follow the same notation as our main specification: CrimeReportsV ct EN = βDID P EPc × I [P ostAugust2018]t (B1) +CrimeReportsT ct OT + controlsc × yeart + γc + ηt + εct c Z This specification, however, differs from the main one in two ways. First, on the left hand side, we substitute the share of crime reports by the rate of crime reports per 100,000 inhabitants by Venezuelans. However, since we are interested in the composition of crime reports, we include an additional control: The total reports of crimes in municipality c and month-year t denoted by CrimeReportsT ct OT (which naturally include the reports by Venezuelans). By adding this control, a positive βDID does not imply more crimes, but rather crime being reported by Venezuelans, keeping constant the total number of crimes. The results are presented in Table B1, and are qualitatively robust to our main results. [Table B1 about here.] Similarly, we present analogous results to Table 3, exploiting crime reports by gender, using this alternative specification. Here, too, we find our results to be robust. The results are reported on Table B2 below. [Table B2 about here.] B.2 Alternative Identification Strategy An alternative identification strategy is to exploit the variation in the number of Venezuelan un- documented immigrants that receive a regular migratory status based on the previous allocation of Venezuelan immigrants in Colombia using data from the 2005 Census. This is a widely used identification strategy when it comes to immigration stocPs, since the assumptions required for the 8 estimators to be considered as unbiased when instrumenting for historical immigrant presence are reasonable. In this case, the exclusion restriction assumption must be that the historical presence of Venezue- lan immigrants across different Colombian cities can explain future changes in the behavior of crime reporting by Venezuelans immediately after the amnesty process only and only through the treat- ment, which in our case is the number of Venezuelan undocumented immigrants that receive a regular migratory status in 2018. The results for the main specification are presented in Table B3. First, it is worth noting based on the Kleibergen-Paap F statistic that, for all of the estimations, the first stage is strong as expected: the presence of Venezuelan immigrants in 2005 can explain the number of Venezuelans that received a regular migratory status in 2018. For the most part our results are consistent with the main estimation that uses the average number of registration days to instrument for the treatment. In terms of total crimes (last column) we do see that in cities that more Venezuelan immigrants received a regular migratory status, there is an increase in crime reporting by Venezuelans (that is not explained by an increase in crimes). But some of the estimates lost precision, namely when looking at reports of domestic violence and sex crimes. However, we do see across the board positive point estimates despite not always precise. The small discrepancies, we believe, might well be because of the source of exogenous variation being used using this identification strategy, which is based on historical settlements of Venezuelan immigrants in Colombia. Namely, these estimations might be more representative of activities in the largest cities in this sample (such as Bogotá, Cali and Medellin, for instance), whereas in the main specification, the results might be more representative of bordering cities, where more vulnerable populations typically live. [Table B3 about here.] When differentiating crime reporting by gender using this identification strategy, presented in Table B4, we find consistent results, too, to those documented in Table 3, hinting that our effects are driven by reports of female immigrants. [Table B4 about here.] 9 B.3 Event Studies In this section we present event studies for crime reports by all, male, and female Venezuelan immigrants in Colombia, and by type of crime. It is important to mention that when focusing on gender and type of crime at the same time, we end up with much smaller variation to exploit, as the number of crime reports by Venezuelans is already small to begin with and becomes significantly smaller within each gender-type cell. Nevertheless, we present the results for full transparency. The following figures visualize the results of estimating Specification (1) with the only difference that the treatment is interacted with 12 dummies representing quarters (i.e., three month periods) from Q1 of 2017 to Q4 of 2019 (instead of the I [P ostAugust2018]t dummy). The left hand side, however, varies both in terms of the type of crime reported and the gender of the Venezuelan immigrant who reports them. Each marker represents the average effect of the treatment on crime reports for the 3-month period that correspond to each quarter. The gray vertical line represents the beginning of the roll-out of the PEP-RAMV visa program, in August of 2018. Results are based on OLS estimates. The whiskers represent 95% confidence intervals, based on standard errors clustered at the municipality level. Figure B1 presents results of event studies by type of crime as reported by all Venezuelan im- migrants. We find consistently across all graphs that there are no noticeable pre-trends. We also see results somewhat consistent with those in Table 2, when it comes to homicides, sex crimes and total crimes. We see some positive effects not captured in the table in other types of crimes, too, such as threats and theft. The reason for these results not being fully consistent with the table is because some differences statistically change when comparing the average effects in the "after" months against the "before" months.. Figures B2 and B3 replicate these event studies using reports by male and female Venezuelan immigrants, respectively. In both cases we find no evidence of pre-trends, which is again reassur- ing. The effects are much more pronounced, consistently with the main results, when focusing on Venezuelan female immigrants. [Figure B1 about here.] [Figure B2 about here.] [Figure B3 about here.] 10 {&beta^{DID}} (95% CI) {&beta^{DID}} (95% CI) 20 17 −.002 0 .002 .004 .006 .008 20 17 −.05 0 .05 20 − Q 20 −Q 17 1 17 1 20 −Q 20 −Q 17 2 17 2 20 −Q 20 −Q 17 3 17 3 20 −Q 20 −Q 18 4 18 4 20 −Q 20 −Q 18 1 18 1 20 −Q 20 −Q 18 2 18 2 20 −Q 20 −Q 18 3 18 3 theft Quarter Quarter 20 −Q 20 −Q 19 4 19 4 20 −Q 20 −Q homicides 19 1 19 1 20 −Q 20 −Q 19 2 19 2 20 −Q 20 −Q clustered at the municipality level. 19 3 19 3 −Q −Q 4 4 {&beta^{DID}} (95% CI) {&beta^{DID}} (95% CI) 20 17 −.01 0 .01 .02 .03 20 17 −.005 0 .005 .01 .015 20 − Q 20 −Q 17 1 17 1 20 −Q 20 −Q 17 2 17 2 20 −Q 20 −Q 17 3 17 3 20 −Q 20 −Q 18 4 18 4 20 −Q 20 −Q 18 1 18 1 11 20 −Q 20 −Q 18 2 18 2 20 −Q 20 −Q 18 3 18 3 Quarter Quarter 20 −Q 20 −Q threat 19 4 19 4 20 −Q 20 −Q sex_crime 19 1 19 1 20 −Q 20 −Q 19 2 19 2 20 −Q 20 −Q 19 3 19 3 −Q −Q 4 4 {&beta^{DID}} (95% CI) {&beta^{DID}} (95% CI) 20 17 −.005 0 .005 .01 .015 20 17 −.01−.005 0 .005 .01 .015 20 − Q 20 −Q 17 1 17 1 20 −Q 20 −Q 17 2 17 2 Figure B1: Event Studies, All Venezuelan Immigrants 20 −Q 20 −Q 17 3 17 3 20 −Q 20 −Q 18 4 18 4 20 −Q 20 −Q 18 1 18 1 20 −Q 20 −Q 18 2 18 2 20 −Q 20 −Q 18 3 18 3 Quarter Quarter 20 −Q 20 −Q 19 4 19 4 tot_crime 20 −Q 20 −Q 19 1 19 1 20 −Q 20 −Q dom_violence 19 2 19 2 20 −Q 20 −Q 19 3 19 3 −Q −Q 4 4 The results are based on OLS estimates. The whiskers represent 95% confidence intervals, based on standard errors The gray vertical line represents the beginning of the roll-out of the PEP-RAMV visa program, in August of 2018. of crime. In clockwise order they are: homicides, threats, domestic violence, theft, sex crimes, and total crimes. Venezuelan immigrants (both male and female). Each graph plots the results using crime reports for a different type interacted with 12 quarter dummies from Q1 of 2017 to Q4 of 2019. The estimation is based using reports by all The figure presents the estimation event studies based on Specification (1), estimating the effect of the treatment {&beta^{DID}} (95% CI) {&beta^{DID}} (95% CI) 20 17 −.002 0 .002 .004 20 17 −.05 0 .05 20 − Q 20 −Q 17 1 17 1 20 −Q 20 −Q 17 2 17 2 20 −Q 20 −Q 17 3 17 3 20 −Q 20 −Q municipality level. 18 4 18 4 20 −Q 20 −Q 18 1 18 1 20 −Q 20 −Q 18 2 18 2 20 −Q 20 −Q 18 3 18 3 theft Quarter Quarter 20 −Q 20 −Q 19 4 19 4 20 −Q 20 −Q homicides 19 1 19 1 20 −Q 20 −Q 19 2 19 2 20 −Q 20 −Q 19 3 19 3 −Q −Q 4 4 {&beta^{DID}} (95% CI) {&beta^{DID}} (95% CI) 20 17 −.004 −.002 0 .002 .004 20 17 0 .001 .002 .003 .004 20 − Q 20 −Q 17 1 17 1 20 −Q 20 −Q 17 2 17 2 20 −Q 20 −Q 17 3 17 3 20 −Q 20 −Q 18 4 18 4 20 −Q 20 −Q 18 1 18 1 12 20 −Q 20 −Q 18 2 18 2 20 −Q 20 −Q 18 3 18 3 Quarter Quarter 20 −Q 20 −Q threat 19 4 19 4 20 −Q 20 −Q sex_crime 19 1 19 1 20 −Q 20 −Q 19 2 19 2 20 −Q 20 −Q 19 3 19 3 −Q −Q 4 4 {&beta^{DID}} (95% CI) {&beta^{DID}} (95% CI) 20 17 −.004 −.002 0 .002 .004 20 17 −.01 −.005 0 .005 .01 20 − Q 20 −Q 17 1 17 1 20 −Q 20 −Q 17 2 17 2 20 −Q 20 −Q Figure B2: Event Studies, Male Venezuelan Immigrants 17 3 17 3 20 −Q 20 −Q 18 4 18 4 20 −Q 20 −Q 18 1 18 1 20 −Q 20 −Q 18 2 18 2 20 −Q 20 −Q 18 3 18 3 Quarter Quarter 20 −Q 20 −Q 19 4 19 4 tot_crime 20 −Q 20 −Q 19 1 19 1 20 −Q 20 −Q dom_violence 19 2 19 2 20 −Q 20 −Q 19 3 19 3 −Q −Q 4 4 based on OLS estimates. The whiskers represent 95% confidence intervals, based on standard errors clustered at the line represents the beginning of the roll-out of the PEP-RAMV visa program, in August of 2018. The results are clockwise order they are: homicides, threats, domestic violence, theft, sex crimes, and total crimes. The gray vertical male Venezuelan immigrants. Each graph plots the results using crime reports for a different type of crime. In interacted with 12 quarter dummies from Q1 of 2017 to Q4 of 2019. The estimation is based using reports by The figure presents the estimation event studies based on Specification (1), estimating the effect of the treatment {&beta^{DID}} (95% CI) {&beta^{DID}} (95% CI) 20 17 −.002 0 .002 .004 .006 .008 20 17 −.01 0 .01 .02 .03 20 − Q 20 −Q 17 1 17 1 20 −Q 20 −Q 17 2 17 2 20 −Q 20 −Q 17 3 17 3 20 −Q 20 −Q municipality level. 18 4 18 4 20 −Q 20 −Q 18 1 18 1 20 −Q 20 −Q 18 2 18 2 20 −Q 20 −Q 18 3 18 3 theft Quarter Quarter 20 −Q 20 −Q 19 4 19 4 20 −Q 20 −Q homicides 19 1 19 1 20 −Q 20 −Q 19 2 19 2 20 −Q 20 −Q 19 3 19 3 −Q −Q 4 4 {&beta^{DID}} (95% CI) {&beta^{DID}} (95% CI) 20 17 0 .01 .02 .03 .04 20 17 −.005 0 .005 .01 20 − Q 20 −Q 17 1 17 1 20 −Q 20 −Q 17 2 17 2 20 −Q 20 −Q 17 3 17 3 20 −Q 20 −Q 18 4 18 4 20 −Q 20 −Q 18 1 18 1 13 20 −Q 20 −Q 18 2 18 2 20 −Q 20 −Q 18 3 18 3 Quarter Quarter 20 −Q 20 −Q threat 19 4 19 4 20 −Q 20 −Q sex_crime 19 1 19 1 20 −Q 20 −Q 19 2 19 2 20 −Q 20 −Q 19 3 19 3 −Q −Q 4 4 {&beta^{DID}} (95% CI) {&beta^{DID}} (95% CI) 20 17 0 .002 .004 .006 .008 .01 20 17 −.005 0 .005 .01 .015 20 − Q 20 −Q 17 1 17 1 20 −Q 20 −Q 17 2 17 2 20 −Q 20 −Q 17 3 17 3 Figure B3: Event Studies, Female Venezuelan Immigrants 20 −Q 20 −Q 18 4 18 4 20 −Q 20 −Q 18 1 18 1 20 −Q 20 −Q 18 2 18 2 20 −Q 20 −Q 18 3 18 3 Quarter Quarter 20 −Q 20 −Q 19 4 19 4 tot_crime 20 −Q 20 −Q 19 1 19 1 20 −Q 20 −Q dom_violence 19 2 19 2 20 −Q 20 −Q 19 3 19 3 −Q −Q 4 4 based on OLS estimates. The whiskers represent 95% confidence intervals, based on standard errors clustered at the line represents the beginning of the roll-out of the PEP-RAMV visa program, in August of 2018. The results are clockwise order they are: homicides, threats, domestic violence, theft, sex crimes, and total crimes. The gray vertical male Venezuelan immigrants. Each graph plots the results using crime reports for a different type of crime. In interacted with 12 quarter dummies from Q1 of 2017 to Q4 of 2019. The estimation is based using reports by The figure presents the estimation event studies based on Specification (1), estimating the effect of the treatment Table B1: Results Using Crime Rates Panel A: OLS (1) (2) (3) (4) (5) (6) homicides threat dom_violence theft sex_crime tot_crime P EPc × I [P ostAugust2018]t 0.0401 0.0180 0.0336 0.0425 0.0210 0.0866 (0.014)*** (0.015) (0.020) (0.024)* (0.008)** (0.024)*** CrimeReportsT ct OT 0.0709 0.0423 0.0687 0.1889 0.0580 0.4418 (0.024)*** (0.018)** (0.023)*** (0.057)*** (0.038) (0.084)*** N 1080 1080 1080 1080 1080 1080 Adj R2 0.32 0.24 0.42 0.58 0.37 0.65 Panel B: 2SLS homicides threat dom_violence theft sex_crime tot_crime P EPc × I [P ostAugust2018]t 0.0327 0.0238 0.0779 0.0358 0.0512 0.1048 (0.017)* (0.017) (0.035)** (0.037) (0.018)*** (0.043)** CrimeReportsT ct OT 0.0720 0.0423 0.0752 0.1865 0.0585 0.4503 (0.024)*** (0.017)** (0.025)*** (0.055)*** (0.039) (0.086)*** N 1080 1080 1080 1080 1080 1080 r2 0.12 0.12 0.17 0.18 0.19 0.19 KP F Stat 20.58 21.41 21.58 22.19 21.46 21.76 The table presents the OLS (upper panel) and 2SLS (lower panel) estimations. Columns (1) to (5) present results using reports for different types of crimes by Venezuelans per 100,000 inhabitants as the dependent variable, in their inverse hyperbolic sine form. The different type of crimes are homicides, threats, domestic violence, theft, and sex crimes. Column (6) presents results for the totality of all crime reports. The variable CrimeReportsT ct OT is the inverse hyperbolic sine of the total reports (not only reported by Venezuelans but all inhabitants) of the same type of crime as the dependent variable, per 100,000 inhabitants. All columns include city and month-year fixed effects as well as the set of controls described in Section 3.2. Standard errors are clustered at the city level. ∗ p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01 14 Table B2: Results Using Crime Rates, by Gender Panel A: OLS (1) (2) tot_crime_m tot_crime_f P EPc × I [P ostAugust2018]t 0.0005 0.0022 (0.000) (0.001)*** CrimeReportsT ct OT -0.0013 0.0048 (0.001) (0.002)** N 1080 1080 Adj R2 0.41 0.51 Panel B: 2SLS tot_crime_m tot_crime_f P EPc × I [P ostAugust2018]t 0.0007 0.0036 (0.000) (0.001)*** CrimeReportsT ct OT -0.0012 0.0054 (0.001) (0.002)** N 1080 1080 r2 0.13 0.28 KP F Stat 21.76 21.76 The table presents the OLS (upper panel) and 2SLS (lower panel) estima- tions. Column (1) presents results using crime reports by Venezuelan male immigrants per 100,000 inhabitants, whereas Column (2) presents results us- ing crime reports by Venezuelan female immigrants per 100,000 inhabitants, both in their inverse hyperbolic sine form. The variable CrimeReportsTct OT is the inverse hyperbolic sine of the total reports (not only reported by Venezuelans) per 100,000 inhabitants. All columns include city and month- year fixed effects as well as the set of controls described in section 3.2. Standard errors are clustered at the city level. ∗ p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01 Note that the purpose of controlling the total number of reports CrimeReportsT ct OT –a proxy for all crimes committed in that municipality every month– implies that our main estimator in this specification, when positive, does not represent an increase in crimes, but rather an increase in reports of crimes by Venezuelans. Naturally, the estimators for CrimeReportsT ct OT are all positive and statistically significant, since this variable explains a huge deal of variation of the dependent variable: the more crimes there are in total, the more crime reports by Venezuelans. 15 Table B3: Results using historical presence of Venezuelan migrants as IV (2SLS) Dependent Variable: Share of Crimes by Venezuelans homicides threat dom_violence theft sex_crime tot_crime P EPc × I [P ostAugust2018]t 0.0174 0.0005 0.0013 0.0015 0.0020 0.0021 (0.006)*** (0.001) (0.001) (0.001)** (0.001) (0.000)*** N 1080 1080 1080 1080 1080 1080 r2 0.07 0.08 0.19 0.20 0.17 0.34 KP F Stat 19.93 19.93 19.93 19.93 19.93 19.93 The table presents 2SLS estimations using the historical presence of Venezuelan immigrants as instrumental variable. Columns (1) to (5) present results using reports for different types of crimes by Venezuelans as share of all crime reports as the dependent variable, in their inverse hyperbolic sine form. The different type of crimes are homicides, threats, domestic violence, theft, and sex crimes. Column (6) presents results for the totality of all crime reports by Venezuelans as a share of all crime reports as the dependent variable. All columns include city and month-year fixed effects as well as the set of controls described in Section 3.2. Standard errors are clustered at the city level. ∗ p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01 16 Table B4: Results using historical presence of Venezuelan migrants as IV, by Gender (2SLS) Dependent Variable: Share of Crimes by Venezuelans tot_crime_m tot_crime_f P EPc × I [P ostAugust2018]t 0.0004 0.0017 (0.000) (0.001)*** N 1080 1080 r2 0.13 0.28 KP F Stat 19.93 19.93 The table presents 2SLS estimations using the historical presence of Venezue- lan immigrants as instrumental variable. Column (1) presents results using crime reports by Venezuelan male immigrants, whereas Column (2) presents results using crime reports by Venezuelan female immigrants, both as a share of all crimes, and in their inverse hyperbolic sine form. All columns include city and month-year fixed effects as well as the set of controls described in section 3.2. Standard errors are clustered at the city level. ∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01 17