The Economic Burden of Crime: Evidence from Mexico

Transcription

1 Preliminary and incomplete Please do not quote The Economic Burden of Crime: Evidence from Mexico Andrea Velásquez 1 Duke University This version: October 2013 First version: April 2011 Abstract This paper investigates the impact of an amplified environment of violence on labor outcomes in the Mexican context. The Mexican Family Life Survey offers a unique opportunity to address this research question as the first follow-up was conducted between 2005 and 2006, a period of low levels of violence, and the second follow-up was performed from 2009 to 2013, during years of greatly elevated violence. These data allow to compare the outcomes of the same individual in periods of varying degrees of violence, while controlling for a rich set of individual and household characteristics. The results show that a higher homicide rate decreases the labor market participation of women who were self-employed in the pre-violence period, and it has a positive effect on hourly earnings. This result can be explained, on the one hand by a positive selection of self-employed women on the labor market, or in the other hand, by higher payments made to this group of women to incentivize their labor participation. Crime, on the other hand, does not affect self-employed males labor market participation, but does negatively impact their earnings. Moreover, we find the effect is heterogeneous with respect to a respondent s attained education. Lastly, this paper provides evidence that exposure to a violent environment had a negative impact on household expenditure and these effects differ depending on the gender of the household head. 1 Department of Economics, Duke University, Durham, NC apv4@duke.edu I am very grateful to my advisor Duncan Thomas for his guidance and support. This paper has benefited from discussions with Erica Field, Elizabeth Frankenberg, Amar Hamoudi, Seth Sanders, Alessandro Tarozzi, Peter Arcidiacono and numerous seminar participants.

2 1. INTRODUCTION The increased incidence of drug related crime and conflict between organized crime groups (OCGs) over the last few years in Mexico has attracted a great deal of government and public attention. Between 2007 and 2011 the official homicide rate reported by the National Institute of Statistics and Geography (INEGI, its Spanish acronym) exploded from 8.5 to 24.4 (Figure 1 provides the monthly homicide rate in Mexico from ). Moreover, not only has the homicide rate increased in the last few years, but the dynamics of violence have also dramatically changed. The amplified effort against OCG s taken by Felipe Calderón s government had unexpected consequences on the intensification of violence and on its geographical dispersion. One of the consequences of the new military strategy, for example, was the growth in the number of cartels from 6 in 2007 to 16 in 2010 (Guerrero, 2012a). This fragmentation of the cartels increased confrontations between OCGs, as they sought to guarantee their territorial control. As these competitions intensified the nature of crimes rapidly changed with the visibility of violence becoming one useful tool to spread fear and send messages to the rival traffickers, authorities, and citizens (Rios, 2012). Although this rapid increase of violence has no precedent in Mexico s history, the economic and political power of the OCGs in Mexico is not a new phenomenon 2. This increasing strength is largely explained by the expansion of the cartels illicit drug market power. The 1,952 miles of border Mexico shares with the United States constitutes a critical aspect of the Mexican geography that facilitates illegal activities such as contraband and traffic of illicit drugs. Since the 1960 s Mexico has produced illicit narcotics that are transported to the United States and with the increasing drug related activities during the 1980 s, Mexican crime groups began to organize. Moreover, around 1994 Mexico became the most important point of entry of illegal drugs into the US, generating a billion dollar business and becoming the main supplier of cocaine and marijuana (Rios, 2012; Andreas, 1998). However, even though the drugtrafficking business had been present in Mexico for more than a decade, the level of homicides had been stable until The unprecedented increase in violence and the dynamics of crime in Mexico has called the attention of many academics and political analysts. One group of researchers sustain that the surge in violence recently observed in Mexico is partly a result of the death or incarceration of many of the previous bosses of the OCGs. This loss in leadership created power vacuums that spawned the emergence of new 2 Around five years ago Mexico s drug-smuggling gangs overtook Colombia s in resources and manpower, ( ) As well as expanding down the supply chain, running distribution networks in the United States, they have moved up it, buying cocaine directly in Colombia, Bolivia and Peru. They have become a bigger influence in politics at home (The Economist, Oct 14th 2010) 1

3 leaders and increased divisions within the main cartels 3. These struggles for dominance have increased the number and intensity of battles over critical drug running corridors, as the control of these strategic regions is crucial for a cartel s drug-trafficking market and, in conjunction, political and economic power. In addition to changing the violent crime environment, the fracturing of the cartels has reduced the profits each OCG makes from drug trafficking and has pushed them to pursue alternative criminal enterprises. For example, there is evidence that increasing conflict in Mexico has led to an expansion of crimes targeting citizens such as extortions, kidnapping, car and business thefts with extortions increasing from 3 per 100,000 inhabitants in 2006 to 5.88 in 2009, business thefts from in 2006 to in 2009 (72.3 in 2011) (Robles et al., 2013) and executions increased by 60% between 2009 to 2010 (from 9,604 to 15,263) (Guerrero, 2012a). Moreover, this violence has also spread into the lives of non-cartel members through the use of visible executions to spread fear. The purpose of this paper is to quantify the shock of the increasing wave of crime in Mexico on individual labor market outcomes and household consumption. It is important to understand the economic consequences of crime beyond the macroeconomic effect of higher military, and provide evidence about the demographic groups that are being particularly affected by the new dynamics of crime. If the most vulnerable are the most affected, and these are precisely the citizens without the mechanisms to smooth these negative shocks, the consequences of these violent years can potentially have long-term consequences on the well-being of the Mexican population. In addition, an important contribution of this paper is that a broad array of economic indicators are examined which allow a richer characterization of the impact of crime in Mexico at the individual and household levels. The analysis on labor markets utilizes measures of participation in the labor market, hours worked, and hourly earnings and is complemented by a detailed analysis of per capita expenditure in order to provide insight on whether some individuals are able to smooth these negative income shocks. This paper is part of a broader research agenda with a general goal of providing evidence on the impact of the increased incidence of crime on outcomes related to the well-being of the Mexican population (Brown, 2013 measures the impact of the Mexican violence on birth outcomes; Brown and Velásquez, 2013 assess the effect on human capital accumulation of children and young adults). The identification of the impact of violence on individual outcomes imposes challenges related to omitted variables that affect both the levels of violence and the labor market outcomes, endogenous behavioral response, and possible reverse causality. The empirical specification of this paper exploits available 3 The term drug cartel is a colloquial term to refer to organized crime organizations but it does not imply any collusion to set prices. We will use the term Organized Crime Groups (OCGs), traffickers organizations and cartels indistinctively to refer to organized crime organizations involved in the drug-trafficking business. 2

4 information on homicide rates from the INEGI at the municipality level from 2005/06 (a period of low levels of violence) to 2009/12 (a period of high levels of violence). In order to analyze the impact of violence on the outcomes of interest, the INEGI homicide data is matched to the Mexican Family Life Survey (MxFLS), a nationally representative longitudinal survey. The MxFLS is ideally suited to address the question of this paper. One important feature of the survey is that the first follow-up was conducted between 2005 and 2006, during a time of relatively stable levels of conflict, and the second follow-up was performed from 2009 to , when violence in Mexico rose with unprecedented intensity. This feature of the timing paired with the panel nature of the survey, allows the comparison of outcomes of the same individual in periods of low and high violence, which removes all time-invariant unobserved heterogeneity at the individual level. Additionally, the detailed survey information contained in the MxFLS provides a rich set of controls for time-varying individual and household characteristics. The next section of this paper introduces the homicide data available in Mexico, provides background on the Mexican experience of violent crime over the last decade, and reviews the literature on the effect of crime on labor market outcomes and wealth. Section 3 describes the Mexican Family Life Survey (MxFLS). Moreover, this section provides evidence on the trends of labor market participation in Mexico during the years of interest for this study and compares the main measures of labor market participation with a national survey to provide evidence of the representativeness of the MxFLS. Section 4 provides a description of the conceptual framework that gives support to the empirical specification that is presented in section 5. Section 6 provides the main results that a higher homicide rate negatively affects the labor market participation of self-employed women, but has a positive effect on their hourly earnings. This result can be explained, on the one hand by a positive selection of self-employed women on the labor market, or in the other hand, by higher payments made to this group of women to incentivize their labor participation. Crime, on the other hand, does not affect males labor participation, but negatively affects their earnings. The negative effect of crime is also found on measures of per capita expenditure. Looking at different specifications to measure per capita expenditure at the household level, the results suggest that the high incidence of crime has negatively affected consumption levels of Mexican households. Ongoing research will investigate whether these effects can be understood as a spillover effect of the negative impact on labor market income or as a strategy to minimize the probabilities of being victimized. In section 7 robustness checks are discussed and section 8 concludes. 2. BACKGROUND 2.1. Trends of Violence in Mexico 4 94% of the sample of panel respondents interviewed in Mexico was interviewed between 2009 and

5 Homicides Data The description of the increasing trend of violence in Mexico in the last few years needs to be accompanied by an illustration of the data that provides evidence of this dramatic and rapid change in Mexico. In addition to the official numbers of intentional homicides reported by the INEGI, other data sources were created in order to understand the puzzling increase of crime in Mexico. Independent newspapers, academics, and the National Public Security System (SNSP, its Spanish acronym), which was collected under the direction of President Felipe Calderón, compiled data measuring homicides exclusively related to organized crime 5. Although some differences exist between the data sources what is evident in all of them is the rapid increase of homicides between 2007 and 2010 and its geographical dispersion. This section provides background on the Mexican case by highlighting the importance and relevance of using the INEGI data for the purpose of this study. That the INEGI data accurately reflects the relative levels of violence through Mexico is important as, unlike the data from other sources, the homicides reported in the INEGI, are available from 1990 to 2011, which allows analysis to include both the pre-violence and violence periods and thus fully exploit the panel nature of the MxFLS. The numbers from the SNSP show a rise in the homicide rate of more than 550% between 2007 and 2011; from 2.67 in 2007 to 12.6 in 2010 and 18 in (green dashed line in Figure 2). In addition to reporting the total number of homicides related to organized crime, this dataset disaggregates the type of death in three categories: targeted executions, confrontations (between OCG s or between OCG s and authorities) and aggressions (attacks by OCG s against authorities). Executions are the main component of the homicides related to organized crime, accounting on average for 90% of the total homicides. Since 2006, not only the intensity of executions increased but its geographical concentration changed as well: in 2010, 168 municipalities reported 12 or more executions 7, whereas in 2007 only 48 municipalities had reported 12 or more executions. This increasing violence is also seen in the official homicide rate reported by the INEGI. According to this data source the homicide rate has increased in Mexico from 8.5 homicides per 100,000 inhabitants in 2007 to 24.4 in 2011, an increase of more than 150% (Figure 2, red solid line). Moreover, the INEGI dataset also shows the geographical dispersion found in the organized crime dataset. Maps 1 to 4 show the geographical dispersion of violence from 2002 to 2005 before Calderón took office. At that point only a handful of municipalities were at the top of the homicides distribution and they were highly concentrated in states with strong presence of OCG s. Although the presence of traffickers could 5 For an analysis and comparison of the different data available to measure homicides see Rios and Shirk (2011) and Molzahn, Rios and Shirk (2012). 6 The data that measures homicides related to organized crime was released by the Mexican government in December of 2010, and provides information of number of homicides allegedly linked to organized crime from December 2006 to September of The mean of executions in 2012 was using the dataset from the SNSP. 4

6 increase the levels of violence in general the maps show that before 2007 the levels were relatively low and concentrated in only a few places. Maps 5 to 8 show the geographical dispersion of the homicide rate from 2007 to The first year of Calderón s term, 2007 was a relatively stable year in terms of violence; even though his military strategy was already being implemented. The first dramatic increase of violence is observed in 2008 (Figure 2). Maps 6-8 show the almost epidemic dispersion of crime: there is a massive increase in violence in previously unaffected municipalities in Chihuahua, Durango, Sinaloa, Michoacán and Guerrero, among others. Rios (2012) documents the dynamics of war within and between traffickers organizations after the leader of the OCG is captured. In 2008, for example, the Sinaloa cartel leader was captured which caused that cartel to split and as a consequence created a combative relationship between the Sinaloa Cartel and La Familia from Michoacán. One of the consequences of this was an intensification of violence in Guerrero where La Familia and the Sinaloa Cartel had operated together in previous years. The scenery gets only worse in 2009 and 2010 (Maps 7 and 8). In 2009 the Sinaloa Cartel splits once again and a territory that in 2007 had only one cartel operating became the field of four competing cartels (Rios, 2012; Guerrero, 2011b). The fracturing of cartels intensifies the violence in the competing territories and spreads the violence as new groups attempt to increase their territorial control. In the meantime, Nuevo León and Tamaulipas suffered the consequences of a turf war between Los Zetas and their former employer the Gulf Cartel. The maps using the INEGI official homicides data show both the intensification of the violence and its geographical dispersion. It is important to note the potential measurement error in the INEGI homicide rate when used as proxy for the measure of violence. First, since the data only captures registered homicides there is potential for the INEGI homicide rates to reflect a lower bound of the actual level of violence. However, as the figures and maps show, the INEGI data captures the same increasing trend as the data on homicides related to organized crime. Second, the increasing trend of homicides might actually reflect a combination of actual conflict as well as simply a shift in the ability/motivation to more accurately report homicides over this time period. Although this could be an important source of measurement error in other scenarios, the increasing violence has not only been evident in the official numbers (INEGI and the SNSP datasets), but alternative data sources collected by academics, the national press, and NGOs have verified the findings of the public records What caused the recent trend of violence in Mexico? The rapid increase in violent homicides has led to a debate about its causes (Castillo et al., 2013; Dell, 2013; Guerrero, 2011a; Guerrero, 2012b; Molzahn et al., 2012; Rios and Shirk, 2011; Rios, 2012). 5

7 One hypothesis is that the rapid and violent increase in homicides is a byproduct of the military strategy of increased confrontation against OCGs that took place since late 2006 when Felipe Calderón became president (Molzahn, Rios and Shirk, 2012; Guerrero, 2011a). Calderón s military strategy against the main capos of the cartels destabilized the old oligopolistic equilibrium where the OCGs operated while maintaining relatively low levels of violence. Eduardo Guerrero (2012b) suggests that by confronting the main leader of a trafficking group two types of violence are created: first, an internal conflict arises for the leadership of the organization; and second, it increases the likelihood of confrontations from enemy organizations that seek to wrest territorial control by taking advantage of the weakened enemy. Rios (2012) describes this new equilibrium as a self-reinforcing violent equilibrium where the violent confrontations between traffickers increase the likelihood of government interventions which promote subsequent confrontations between the remaining traffickers to control the market left by their predecessors. Related to this debate Dell (2011) compares municipalities where a mayor from Calderon s party (Partido Acción Nacional PAN) won the election by a margin of 5 percent or less to municipalities in which the PAN barely lost by the same margin. Comparing these municipalities 6 months before the election and 6 months after the election, the author finds a significant increase of drug related homicides in the municipalities where Calderón s party won, suggesting that PAN related policies may have triggered the massive increase in homicides. In addition to Calderon s strategy against OCGs, changes in the structure of the drug trafficking business attracted more competitors and therefore, a higher probability of conflict between groups. Robles et al. (2013) sustain that, among others, an increasing flow of trade with United States, growing availability of weapons and a decreasing supply of cocaine from Colombia were factors that increased the profitability of the drug trade business in Mexico. In addition to the increased amount of confrontations between groups, the rising competition between OCGs has led to a diversification of their financial sources. While drug trafficking activities still account for most of their economic resources, in order to increase profits and fund their fight against the military and other OCGs, they have been relying more on criminal activities that directly affect the civil population, like kidnappings, extortions and car thefts. Moreover, the visibility of crimes changed the dynamics of the violence in Mexico. Narco-messages to spread fear in the communities became a very popular method to signal territorial presence and to spread fear across other OCG s but also toward authorities, journalists and any citizen that would not support their actions. Executions, which are the most violent and visible type of homicide, increased by 60% between 2009 and 2010 (Guerrero, 2012c). 6

8 In addition to the citizens victimization by cartel members, the feeling of fear is exacerbated by the public s lack of trust in the State s institutions and the high levels of corruption and abuses from the police (Guerrero, 2011; Díaz-Cayeros et al, 2011). Díaz-Cayeros et al (2011) measure the strategies that the OCGs use against the civil society and measure how embedded they have become in the society. Their results show that both OCGs and police take advantage of citizens, particularly preying on the poor and less educated The Economic Effects of Crime and Violence Along with the studies that analyze the cause of the increasing violence in Mexico there are a few studies that measure the impact of crime on individual and household outcomes. Dell (2011) makes an important contribution to the literature and exploits a network model of drug-trafficking routes and information from the National Survey of Occupation and Employment to measure the economic spillover effects of drug-related violence. Dell s results, suggest no effects on male labor participation or wages in the formal sector but in contrast, negative results on male wages in the informal sector and on female labor force participation. Two recent working papers have also measured the effect of crime on labor outcomes. First, Robles et al. (2013) use the SNSP data and the total number of deaths in Mexico from the National Health Information System (SINAIS) with data on individuals older than 14 years old from the National Survey of Occupation and Employment to measure their labor outcomes. The empirical strategy relies on an instrumental variable approach to measure the marginal effects of the increasing trend of violence. They use the variation in cocaine seizures in Colombia interacted with the distance of each municipality to the principle point of entry to the U.S (instrument from Castillo et al (2013)) as an instrument for homicides and find a dampening effect of violence on an individual s labor market participation, and on the proportion of business owners in the population but no effect on self-employment or overall economic activity (measured as domestic electricity consumption). One of the limitations of this paper is that it does not provide a discussion about potential behavioral responses to crime, and by relying on synthetic control methodology to estimate the effects of violence on the economic activity it imposes important assumptions on the economic trends of municipalities that before the intensification of crime had similar characteristics. Second, Ben Yishai and Pearlman (2013) use data from the National Public Security System that contains information sent by state police forces for their measures of homicides and the 2007 to 2010 ENOE surveys to create yearly panels of adults aged 18 and 65 for their labor outcomes assessment. The empirical strategy relies on individual fixed effects and an instrumental variable approach. The IV in this case is the number of kilometers of federal toll highways in each state. Their results suggest that the marginal effect of violence has been negative on the average hours worked of both salaried and self- 7

9 employed males, and for the latter the impact has been larger for those males who work from home. This analysis, though, is unable to provide insights on whether the lower labor supply has had any effect on the total income of Mexican citizens. Moreover, in order to deal with migratory behavior, individuals that were not born in the current place of residence are dropped from the study, which, if migration is systematic, leads to a non-random sample of analysis. In addition to the papers that study the recent violence in Mexico, mentioned in the previous section, there are important contributions in the literature that have measured the impact of crime on labor market outcomes and household s economic conditions in other settings (Bozzoli, Brück and Wald, 2011; Calderón, Gafaro and Ibáñez, 2011; Deininger, 2003; Fernández, Ibáñez and Peña, 2011; Kondylis, 2007; Shemyakina, 2011). Bozzoli, Brück and Wald (2011) study the effects of the Colombian conflict on the probability of being self-employed in rural settings and find a decreasing share of self-employment in municipalities with higher rates of conflict. Calderón, Gafaro and Ibáñez (2011) estimate the effects of displacement on the labor markets of destination places. The authors find that a larger supply of unskilled labor increases the likelihood of informality and reduces wages in this sector. Kondilys (2007) exploits data from a longitudinal study to explore the effects of displacement on labor market outcomes in post-war Bosnia and Herzegovina. The results show that displaced men experience higher unemployment levels, and displaced women are more likely to drop out of the labor force. Shemyakina (2011) measures the long-term impact of the armed conflict in Tajikistan on education and labor market outcomes for both men and women. By performing a difference in difference regression the author finds that, for women, the conflict had a negative impact on educational attainment and a positive effect on labor market participation, while it had no effect for the outcomes of men. This paper makes a contribution to the understanding of the economic impact of Mexico s recent outbreak of violence and to the literature on the effects of crime in a number of ways. First, an individual fixed effect strategy combined with an intent to treat approach controls for all unobserved heterogeneity that affects exposure to crime and individual outcomes and controls for potential systematic migration as a behavioral response to crime. Second, it tests the exogeneity of the increasing trend of crime by adding controls of economic performance at the state and municipality level and estimates a model of the predictability of violence. Third, it explicitly analyzes migration as a response to violence. Fourth, it provides evidence on a placebo test to provide evidence on the exogeneity of the increasing trend of crime. Fifth, it measures the effect of crime for a representative sample of the Mexican population and explores the effect of crime not only on formal and informal workers but disaggregates the effects by occupation and explores heterogeneity of the estimate by gender and education level. 8

10 2.3. The Effect of Crime on Wealth There is an extended literature analyzing the response of individuals and households to negative shocks. In particular, there has been a large set of studies that examine the ability of households to smooth consumption in times of crisis (see for example, Townsend, 1995; Morduch, 1995; Frankenberg et al. 2002). If we consider exposure to violence as a negative shock to wealth, as, high levels of violence may alter labor market outcomes and thus decrease the level of household income, there is potential for significant consumption smoothing. However, low socioeconomic status individuals may have very few insurance and credit mechanisms in order to smooth consumption. This suggests that the impact of violence on labor outcomes, PCE, and wealth may have meaningful potential to create poverty traps for the most vulnerable population. A second pathway, through which violence may affect consumption behavior directly is through the fear of being targeted for conspicuous consumption. For example, it has been found that individuals in the United States decrease the consumption of visible goods to reduce the probability of victimization from property crime (Mejia and Restrepo, 2010). 3. DATA: MEXICAN FAMILY LIFE SURVEY In order to study the impact of crime on economic outcomes at the individual and household level the INEGI homicide data is matched with the Mexican Family Life Survey (MxFLS). The MxFLS is an ongoing longitudinal, nationally representative survey of individuals, households who were living in Mexico in 2002 when the baseline was conducted. It includes information on approximately 8,440 households and 35,600 individuals among 150 communities and 16 states throughout Mexico. The MxFLS is designed to follow all baseline respondents and their children born after 2002; these are the panel respondents. A unique feature of the MxFLS is that the interviewers not only follow migrants within Mexico but also follow migrants that have moved to the U.S. For the purpose of this paper, the sample of panel respondents interviewed in the U.S. either in 2005 or during the third wave will be excluded from the analytical sample. The longitudinal nature of the MxFLS and its rich content make it an ideal source to study the dynamics of individuals and households living in Mexico in 2002 and, specifically for this research, the impact of crime on different individual and household behaviors and outcomes. However, the contribution of using longitudinal data depends on the extent to which the original sample is successfully re-interviewed in each following wave. With regards to this study, it is not only important that attrition has not been selected on observed and unobserved characteristics but also that it has not been correlated with the increasing trend of violence. Overall, the MxFLS has kept low levels of attrition. In the second wave, conducted between 2005 and 2006, over 89% of the panel respondents were re-contacted (Panel A. of Table 1 shows the re-contact 9

11 rates in MxFLS2). The third wave of the survey, MxFLS3, shows a recontact rate of approximately 87% 8 of all panel respondents. Panel B of Table 1 shows current recontact rates for MxFLS3. Even though the recontact-rates provide evidence of a successful follow-up, if attrition is correlated with homicide rates the results could lose their representativeness. In order to test for potential selected attrition on measures of violence a model that predicts the probability of attrition from MxFLS3 conditional on being found in MxFLS2 using as dependent variables individual and household characteristics measured in MxFLS2, and as a measure of violence the difference of homicide rates between 2005 and 2009 in the municipality of residence in MxFLS2 and the interactions of the measure of violence and individual characteristics. These interactions will provide evidence of whether individuals with certain characteristics were more likely to attrit from the survey when violence increased. The sample includes all panel respondents age 18 and older at baseline and therefore at risk of being in the final sample of this paper. The model is estimated as a linear probability model based on the following empirical specification: = (1) Where is a binary outcome equal to 1 if the panel respondent attrited from the survey in MxFLS3 conditional on being found in MxFLS2 and equal to zero if not; is the difference on homicide rates between 2005 and 2009 in the municipality of residence in MxFLS2; and, are household and individual characteristics measured at the time of the MxFLS2 survey and include: age, years of education, marital status, cognitive score, household size and household composition, employment status and log of earnings, a dummy for whether the household has relatives in the US and for whether the locality of residence is rural. The results are reported in Table A1 in the Appendix for the sample of males and Table A2 for the sample of females. The results for males in Column 1 suggest that on average the homicide rate in 2009 do not predict attrition; and, the results in Column 2 suggest that the change of homicide rates between 2005 and 2009 do not predict attrition in a different manner for males with different characteristics measured in The results for females in Table A2 show similar results. The change on homicide rates do not predict attrition and it does not predict attrition differently for females with different characteristics. In addition to this model I estimated a model where I allow for a non-linear effect of violence on attrition and the results do not change. The MxFLS provides a number of advantageous features for studying the impact of the Mexican drug war on the lives of Mexican citizens. First, as mentioned previously, the MxFLS2 was conducted in 2005/06 (during a period of normal levels of homicide rates) and the second follow-up (MxFLS3) was 8 These recontact rates are preliminary. Final recontact rates are currently being estimated. 10

12 performed from 2009 to 2013 (during a period of high levels of violence). As indicated in Figure 3, the timing of the MxFLS allows the comparison of outcomes of the same individual in periods of low and high levels of violence and the longitudinal nature of the data allows us to control for unobserved timeinvariant characteristics. Although the information at baseline provides another measure of pre-violence status, this study focus on the first and second follow-up. One of the reasons to focus only in the last two waves of the MxFLS is to avoid excluding from the sample young individuals just entering the labor market in 2005 and for whom labor market information is not available in Second, in the MxFLS there has been a concerted effort made to follow migrants within Mexico and to the U.S. This is particularly important for this study because migration may be a behavioral response to crime; if individuals, particularly affected by high levels of violence, migrate as a response to it and they are not tracked, the estimations of the impact of crime on economic outcomes would be underestimated. For the sample of interest in this paper, 8.9 percent of respondents were found in a different municipality between the two waves and 7.7% were found in the same municipality but report a long-term migration (more than one year) outside their municipality of residence. If migration is due to unobserved characteristics correlated with labor market outcomes and related to violence, it could bias our results. By using a panel dataset, I am able to limit this potential bias, as the empirical estimation controls for any time-invariant unobserved characteristics of the respondent. Moreover, since it is possible that individuals may be migrating as a reaction to or in anticipation of violence, the empirical specification relies on an intent-to-treat styled model where an individual will get assigned the homicide rate over the 12 months prior to the MxFLS interview from his/her municipality of residence in 2005 to remove the possibility for local violence based systematic migration. Moreover, a model will be estimated to determine if potential changes in local violence predict migration behavior. The results, which will be discussed in more detailed in the empirical section, suggest that violence predicts migration of single women and women living in rural places, and of self-employed men. These results highlight the importance of following movers in longitudinal surveys, and in particular for the purpose of this paper. If we had not followed movers we would have systematically lose observations of self-employed men in municipalities that suffered higher changes of violence. Third, the MxFLS has a rich set of characteristics about its participants, including information about the economic, social and health status of each member of a surveyed household. The questionnaire for adults includes sections on education, labor supply, earnings, migration history, marriage history, fertility history, health status, and use of health care. In addition, one member is interviewed about information at the household level. This questionnaire includes a complete household roster including 11

13 basic socio-demographic characteristics of each household member, information of household expenditure, and asset ownership. Of particular interest for this study, the MxFLS contains a great deal of content about a respondent s labor market participation. For instance, employment is defined to include formal and informal jobs, and complete information on earnings, weeks, and hours worked are collected for the two main jobs. The information about the earnings and hours worked during the last 12 months allows analysis of the impact of crime not only on net participation in labor markets but also on total earnings and productivity, measured by hourly earnings Evidence on Labor Outcomes This section provides evidence on labor market trends in Mexico over the last decade. Moreover, by comparing the employment changes measured by the National Occupation and Employment Survey (ENOE) to the MxFLS we will assess the representativeness of the data. The ENOE is a rotating panel survey, representative at the national level, and conducted every three months since January of The same household is followed for five quarters and the survey records basic characteristics of participation in the labor market. 9 This section compares the percentage of individuals working, defined as the ratio of adult individuals (older than 18) who participate in the labor market to the total number of individuals (older than 18) in the ENOE and the MxFLS. Table 2 reports the percentage of individuals working in both the ENOE and the MxFLS for respondents aged 20 or older at the time of the interview, for both males and females. In both surveys, employment is defined as having worked for at least one hour during the week prior to the interview for a salary, wage, or other payment in cash or in kind. It also includes individuals that did not work the week prior to the survey if the reason for missing work is a temporary cause such as a vacation or a short-term leave. The percentage of adult males who were working is reported in columns 1 and 2 of Panel A of Table 2. The evidence reported by the ENOE show that labor market participation of males has decreased from 2005 to In particular the decline in male participation has been most notable in selfemployment: while the rates have diminished around 3.6 percent for the whole male population the decrease has been 9.3 for self-employment. When looking only at states included at the baseline of the MxFLS the employment rates show the same trends (Panel B). 9 All individuals older than 14 are interviewed. 12

14 Panel C of Table 2 displays the employment rates using the MxFLS. The MxFLS exhibits the same decreasing trends in the labor participation rates of males; however, the magnitude of the decrease in employment rates is smaller when looking at the MxFLS: in total the employment rates show a decrease of 1.8 percent and for self-employment a decrease of almost 2.8 percent. While the trends in both datasets is similar additional analysis will be conducted in future versions of this paper in order to provide a more thorough discussion of the potential discrepancies in these measures 10. The story for females is quite different. Female participation, unlike males, has maintained an increasing trend over the last 10 years. Columns 2 and 3 of Table 2 report the labor participation rates of females. The ENOE reports a total increase of female participation of 4 percent and an increase of more than 23% in self-employment. The overall participation of females is underreported in the MxFLS but shows the same increasing trend, with a total increase of 7.4% and a total increase of 12.2% in the selfemployment category. In the spirit of Thomas et al. (2000) we now build an employment transition by gender, place of residence in 2005, and education attained in 2005, exploiting the longitudinal nature of the MxFLS. This table shows the transitions for the sample used in the paper, which is panel respondents age 18 and older at the time of the interview and interviewed both in 2005 and in Panel A of Table 3 shows that over 73% of males were working in both waves, 8% were not working in either survey and while over 8% exited, 10% entered the labor market 11. Panel B disaggregates these numbers by place of residence and education in order to determine if the changes differ by sectors of the population. Panel B of Table 3 shows that low educated males are more likely to leave the labor market in both urban and rural places, while the better educated males are more likely to enter the labor market independently of the place of residence. Panel C disaggregates the numbers by employment status: on the one hand, from the 10% that join the labor market 65% enter the formal sector, almost 30% become self-employed and 7% enter the market as unpaid workers. On the other hand, 57% of those who exit the labor market were employees in 2005, 36% were self-employed and the rest unpaid workers 12. Within the males that participate in the labor market in both periods there is some mobility across self-employment and formal employment: out of the 73% of males who work in both periods, 56% work in the formal sector in both waves, 16% remain self- 10 At this point the discrepancies may be due to a lack of weighting in the MxFLS estimates. In future versions of the paper, weights will be incorporated in order to have a better sense of the differences between both datasets. Moreover, the ENOE includes as self-employed workers in a family business without payment while for the purpose of this paper, unpaid workers are not being included in the analysis. 11 When looking the panel sample the change on unemployment rates is not negative anymore. This might suggest that the youngest individuals during the increasing trend of crime are one of the groups most affected by unemployment. The results of this paper could be therefore considered a lower bound for males. Brown and Velasquez (2013) study the effect of crime specifically on education and labor outcomes of kids and young adults. 12 The transitions show the same trends in both rural and urban places. 13

15 employed and 20% move between sectors. Overall the numbers suggest that between 2005 and 2009 there was a net increase in formal employment of around 1.1% and a net increase in self-employment of around 1.7%. The numbers for women show a net increase of the labor participation of about 4% but more than half of women of working age remain outside the labor market (Panel A). Panel B shows that, opposite to what is observed for males, the participation of low educated females does not decrease but actually increase by 5.5 percent in urban places and 20.6 in rural places. The composition of women leaving the labor market is very similar to that of males: 50% were employees, 34% self-employed and the rest unpaid workers. Women entering the labor market, join the formal sector in a lower proportion than males (56%) and self-employment and unpaid jobs in higher proportion (33 and 11% respectively). Out of the women working in both periods, 58% work in the formal sector both times, 15% are self-employed and do not switch and 10% move between the formal sector and self-employment. Overall, both selfemployment and formal employment show a net increase in the female labor market (2.1 and 2.3% respectively) and both are higher than the increase for males. This evidence suggests that between 2005 and 2009 both males and females saw a change in their labor force participation: while the effect was negative for low educated males, the effect was positive for females, and specifically low educated females did not suffer a decrease in their labor market participation. The next sections of this paper show tests for whether the increasing trend of crime explains some of these changes and whether these changes in labor participation have an effect on individual earnings and on household per capita expenditure. 4. CONCEPTUAL FRAMEWORK (PRELIMINARY) 4.1. Effects on Labor Markets For a number of reasons the expected effects of violence on labor market outcomes in Mexico is negative. As mentioned before, OCGs have not only increased in number, but they have also started to use extortions of civilians as a financial resource and to increase the sense of fear in the community. An increasing sense of fear, lack of confidence in the police and an increasing probability of being a direct victim of OCGs can induce business owners to close, the extortions might diminish total income of the household, and the fear of being a direct victim of the conflict might decrease the time allocated to labor activities. Moreover, the negative effect of conflict on various economic outcomes has previously been established in the literature (Collier 1999; Hoeffler and Reynal-Querol, 2003; Abadie and Gardeazabal, 2003; Deininger, 2003) and in particular on investment at the firm-level (Collier and Duponchell, 2010; Pshisva and Suarez, 2010; Camacho and Rodriguez, 2012). The increasing trend of violence in Mexico can, 14

16 therefore, also have a negative effect on the demand of labor, in particular, formal employment if some firms/businesses close or reduce their size because of the costs imposed by the violence. In order to analyze the effects of conflict on the decision to work a reduced form time allocation function is used, following Becker (1965), Juster and Stafford (1991) and Beegle (2005): =,,,,,, "# (2a) Which says the time allocation T in activity j, of individual i living in municipality k, is the product of the utility maximization problem. The time allocation is a function of prices for market goods, prices for home-grown commodities, market wages, total endowment of time, household non-labor income and a set of individual and household unobserved characteristics,. The time allocation decision is conditional on Z, a set of preference shifters. With the presence of a cartel affect the time allocation function changes in the following way: =,,, $%&'(,,,),, "# (2b) *+,-.. Where < 0 and *+,- < The final decision to leave the labor market will depend on whether */ *8 - the marginal utility of participation in the labor market (wages in the market) outweighs the negative effect of violence. Intuitively, individuals with initial lower wages would face a lower opportunity cost of leaving the labor market and would be more likely to leave their jobs. Also, the model suggests a negative shock on the labor supply would have a positive effect on earnings. We would expect to see more immediate effects on earnings in the informal sector where wages can adjust more rapidly to the lower supply. With this theoretical guidance the empirical strategy will be designed to test these assumptions by disaggregating by self-employment and employment in the formal sector; in addition, the effects will be tested for a different set of demographic groups. Additionally, particularly for self-employed, we would expect to see different effects depending on whether the individual leaves the labor market completely or just decreases the intensity of hours allocated to work. In the case of a business owner, for example, a decrease in the hours worked (decrease in his labor supply) would diminish total earnings. Moreover, if there were an impact on the overall economic 13 The decision to participate in the labor market should be modeled as a household decision problem, where the total the time allocation of other household members enter the time allocation function. For simplicity we model the decision to participate in the labor market as independent on the other household members labor market participation. However, household resources and characteristics are captured in the term, and in the empirical specification we are able to control for a rich set of household composition and characteristics. 15

17 market activity, we would expect to observe a negative effect on hourly earnings. The empirical strategy will allow these different mechanisms to be tested. The final impact on the labor market outcomes depends on the extent to which violence changes labor market demand. If firms close or reduce their size, the final effect on wages (for the workers that stay in the labor market) will depend on the negative magnitude in the labor demand and supply; but, unambiguously, the expected impact on labor participation would be negative. Evidence from Mexican newspapers as well as reports from respondents during the fieldwork of the third wave MxFLS suggest that both small businesses (self-employed) and bigger firms (in particular those related with tourism and manufacture) have been affected by the violence. It is important to highlight that the employer that works in a firm without a written contract and benefits is considered self-employed. Disaggregating this category by occupation this analysis is able to differentiate the impact for business owners, seasonal workers in small firms, or street vendors, among others. 5. EMPIRICAL STRATEGY Drawing from a longitudinal survey where individuals were interviewed in 2005/06 (period of low levels of violence) and in 2009/13 (period of high levels of violence), the empirical specification relies on an individual fixed effect model, controlling for a rich set of individual and household time-variant characteristics, to estimate the effect of an arguable exogenous shock of crime in Mexico on individual and household outcomes. One of the challenges for the empirical estimation is behavioral responses to crime. Migration is one of the responses to violence that could affect the estimates. If, for example, individuals with higher earnings were more likely to migrate when violence increases, the impact of violence on wages would be overestimated. Table A3 in the Appendix shows the results of a linear probability model that predicts migration between municipalities, for the sample of interest in this paper, as a function of the change in homicide rate between the second and third MxFLS waves, using the following specification: = (3) The measure of migration is a binary outcome equal to 1 if the respondent was interviewed in a different municipality in MxFLS2 and MxFLS3 (8.9%) or if in the migration history the respondent reported a longterm migration (one year or more) out from the municipality of residence in MxFLS2 (7.7%). The model for migration follows the same empirical specification as the one for attrition in equation (1). Therefore, migration between the two waves depends on the change of violence between 2009 and 2005, individual and household characteristics measured in 2005 and the interaction of the measure of violence and own characteristics. 16

18 Columns 1 and 2 of Table A3 in the Appendix show the results for males. The results suggest that on average the change on homicide rates do not predict migration of men. However, the model that add interactions show that self-employed men are more likely to migrate when violence increases. If we had not followed migrants, we would have systematically lost self-employed men in places with high levels of changes in homicide rates. This result is consistent with anecdotal evidence that describe how business owners have migrated and closed their business to avoid being victimized by organized crime organizations. The results for women, in columns 3 and 4, show that single women and women living in rural places are more likely to migrate when violence increases. It is reasonable to think that the cost to migrate for married women might be higher since their migration possibly implies the migration of their entire family. Therefore, even if violence increases married women are less likely to migrate than single women. The results for women living in rural places is consistent with the fact that most of migration in Mexico is from rural to urban places. For women living in rural places higher level of violence decreases the relative cost of migration and therefore increasing their likelihood to migrate. In order to control for migration as a behavioral response to crime, the empirical specification follows an intent-to-treat approach, where the municipality of residence in the second wave of the MxFLS (MxFLS2) determines an individual s exposure to violence. By using an individual s location during the period of low and constant levels of violence, we are able to eliminate potential bias due to violence related endogenous migration. In addition to behavioral responses, an additional challenge when estimating the effect of violence on economic outcomes is omitted variable bias. The difficulty in estimating the relationship between violence and economic outcomes could emerge from the fact that homicide rates have not increased in a random fashion over time and might not be orthogonal to unobserved factors that affect economic performance in the municipality or at the individual level. If the change in the level of homicide rates is correlated with unobserved variables that affect the change in labor outcomes, the level of violence is endogenous and we face omitted variable bias. The correlation between the independent variables with unobserved factors could be explained by self-selection (Wooldridge, 2002). Trying to control for this unobserved heterogeneity the empirical specification will exploit an individual fixed effect model. This will allow to the regression to control for time-invariant unobserved heterogeneity at the individual level. The identification strategy exploits the variation over time of homicides rates between 2005 and For each of the outcomes of interest an individual fixed effect model 14 is estimated, thus by definition the empirical strategy compares the same individual across time 14 With two periods an individual fixed effect model is similar to a first difference model. The results of a first difference model are qualitatively the same. 17

19 periods which captures all unobserved, time invariant factors that affect the dependent variable. This is particularly useful if we believe that there are time-invariant characteristics of individuals, such as ability or risk preferences, that are correlated with both labor outcomes and the violence level of the municipality in which the individual chooses to live. Moreover, if homicides are reported with error, the individual fixed effect strategy differences out error that is constant over time. The empirical strategy can be generalized in the following regression framework: 9 & = : ;&,& #+<: &,& #+= &?+@ +A & +B & +C & (4) Where y is the outcome of interest of individual i living in municipality j in time t (where t is the measure of the outcome in 2005/06 MxFLS2 and in 2009/13 MxFLS3). I estimate non-linear effects to assess the effect of crime on labor outcomes 15 using two indicators that depend on the position of the municipality in the national homicide rate distribution: the omitted category are levels of violence between the 0 and 24 th percentile, the first indicator : ;&,& # in the regressions is for percentiles 25 th to 49 th and the second : &,& # for violence at the top of the 50 th percentile; therefore, δ and < are the coefficients of interest. X is a vector of individual and household time-varying characteristics (years of education, z-score of cognitive score, marital status, whether the respondent lives with his/her parents, household size, number of kids in the household, and whether the place of residence is rural or urban, expectations of future migration, a measure of emotional health and presence of relatives in the denotes the individual fixed effect; A & includes GDP at the state level; and B & includes controls for year and quarter of interview. Given that in many municipalities the change in homicides rates was close to zero it is important to allow for non-linearities in the variable of interest to measure the effect of crime in different points of its distribution. In an individual fixed effect framework, these variables will identify the effect of violence in municipalities whose homicide rate position relative to the national distribution changed between 2005 and 2009, relative to the municipalities that did not change. The < coefficient, for example, measures the effect of violence on labor outcomes for individuals living in municipalities that switched to the 50 th -100 th distribution of the national homicide rate relative to municipalities that did not see a change on their distribution position. In order to have consistent estimates, the idiosyncratic error at each time has to be uncorrelated with the variable that measures crime in both periods (Wooldridge, 2002). It would be reasonable to think 15 Linear models were estimated but no significant results were found for any of the specifications. 18

20 that C & is correlated with ) & because crime is not allocated in a random way and it might be higher in municipalities with better economic performance, so the expected profit of the extortions to the civilians is larger. Second, we could think that C & is correlated with ) &D if unobserved variables in 2005 affect both labor outcome variables in 2005 and the level of crime in There are reasons to believe that increasing crime rates might be correlated with the error term in 2005: first, crime might be more likely to happen in areas with better economic growth because it is more profitable to extort civilians in these places, or it might be more likely in places with worse economic activity if that is a reflection of bad institutions and low state presence. While the individual fixed effect strategy controls for all time-invariant characteristics, if economic trends are changing differently in municipalities that have suffered a higher change of violence, the individual fixed effect estimates would be biased. In order to control for this we add controls for changes in GDP at the state level; moreover, in additional specifications we add unemployment rates at the municipality level. This is not our preferred specification since unemployment rate at the municipality level is clearly endogenous to the individual labor market measures. However, even adding this characteristic of the labor markets, the results of the main specification hold. In order to examine the relationship between a municipality s change in violence to demographic and economic characteristics of that municipality a model were the change in homicide rate at the municipality level related to first, characteristics of the municipality in 2005, and second, trends in these characteristics between 2002 and 2005 is estimated. The covariates are averages at the municipality level of MxFLS variables in one specification and Census variables in the other. Table A4 shows the results. Columns 1 and 2 show the results using variables measured in 2005 as covariates. The results suggest that municipalities with, on average higher labor participation of women had lower changes on their homicide rates, but those with lower illiteracy, and with higher electricity saw a higher change in violence. Places with a higher percentage of individuals with less than primary but also with higher percentage of high school graduates had higher changes in their homicides levels. The results do not suggest a clear pattern of the pre-violence characteristics relationship to the change in the homicide rate between 2005 and 2009, other than showing a pattern that suggest that violence increased more in more urban municipalities (those with better supply of public goods). Moreover, if the characteristics of the places where violence increase are constant over time, these fixed characteristics will controlled for in the fixed effect model. If however, some trends at the municipality level are correlated with the trends on violence and with labor outcomes, the specification in (4) could be biased. The results from Columns 3 and 4 show that pre-violence trends in a rich set of economic and demographic characteristics are unrelated to future changes in violence. This provides suggestive evidence 19

21 that the change in a municipality s level of violence is not simply a reflection of underlying trends in other characteristics of that municipality and that the change in violence levels in a municipality is not caused by pre-existing trends in the labor market environment. An additional way to test the validity of the empirical specification is a placebo test. I estimate a placebo test by estimating the model (4) using the change of violence between 2005 and 2009 on the change in labor outcomes between baseline (2002) and the first follow-up (2005), years where violence was very stable. The results of this specification are shown and explained with more detail in the section of robustness checks. A final challenge for the empirical specification is that the main fieldwork was conducted during 2009 and 2010, and it would be reasonable to think that those interviewed in 2010 would have responded to the increasing trend of crime in a different way than those for whom the impact of violence was unanticipated. I compared the two groups of respondents and as expected those interviewed in 2010 have the characteristics of a group that in the field would be more difficult to track: younger, more educated, more likely to work, etc. I estimated the main specification with and without the individuals interviewed in 2010 and after 2010 and the results of the main specification hold Dependent Variables Participation in the labor market: The empirical strategy focuses on males and females age 18 and older at the time of the MxFLS2 interview. The dependent variable is the probability of being employed in the second-follow up conditional on not being employed in 2005 (to measure the effect on entrance into the labor market); and, the probability of being employed 16 at the time of the interview in MxFLS3 conditional on being employed in the first follow-up (disaggregated by self-employment and formal employment in 2005). Earnings: The measure of earnings in the empirical section is the logarithm of annual hourly earnings and the log of total earnings in the last 12 months. These models test the effect of violence first for all individuals who were working in 2005, and then on the stratified samples by self-employment, formal employment and more disaggregated labor occupations. Per Capita Expenditure: The measure of consumption is the per capita expenditure at the household level. The consumption measure is disaggregated in 5 different categories: total expenditures, expenditures on food, arguably conspicuous consumption (which will be described in detail in a later section), health, and education. 16 Either as a self-employed or employee. 20

22 6. RESULTS 6.1. Labor Market Outcomes Tables 4 to 9 provide results for the analysis of the participation in the labor market, hours worked and the log of hourly earnings stratified by gender and by occupation Females Table 4 shows the results for the probability of participation in the labor force for females. Four sub-samples are analyzed: column 1 shows the results for the sub-sample of females who were not working in 2005, therefore, this model predicts the effect of violence on labor market entry (independently of entry into self-employment or formal labor); column 2, shows the results for women who were working in 2005 in any sector; the sub-sample in column 3 are women who were self-employed in 2005; and, column 4 shows the estimations for the sample who worked as wage employees in In the models where the respondent was working in 2005, persons who exit from the labor market determine the variation in the dependent variable. Transitions from self-employment to formal employment and vice versa are not considered a change in labor force participation in this version of the model. Models on transitions between self-employment and formal employment were estimated, but homicide rates are not a significant predictor of the change in occupation for any of these samples or models. The results in Table 4 17 show that the increasing trend of violence in Mexico does not have an effect on the labor market entry of females (column 1), or on labor participation in a general framework when all females working in the pre-violence period are included (column 2). However, when disaggregating the sample by self-employment and formal employment in 2005 the results lead to a more nuanced conclusion. Column 3 shows evidence of non-linear effects of violence on the labor force participation of self-employed females. The results suggest that women living in a municipality that switched to percentiles between the 25 th and 49 th of the homicides distribution are 12% less likely to stay active in the labor market and 9% less likely if their municipality switched to the 50 th percentile and up, relative to women living in municipalities whose place in the violent crime distribution did not change. The results are robust to include changes in GDP and unemployment rates at the municipality level. To put the results in perspective, for the MxFLS sample, the average homicide rate in 2005 in municipalities at the bottom of the 25 th percentile was 0 per 100,000, and the average in 2009 in municipalities between 17 The sample size of women self-employed and employee do not exactly add up to the sample size of the total of women working (column 2) because unpaid women are also considered active in the labor force. No significant effects were found for the labor force participation or hours worked of unpaid women and they are not included in the rest of the analysis since the next estimations look at effects on earnings. 21

23 the 25 th and the 49 th percentile was 3.89 and the average at the top of the 50 th percentile was Therefore, living in a municipality that suffered an increase from 0 to 3.89 on average caused a decrease of female self-employment participation of 12%, while an increase from 3.89 to caused a decrease of 9%. Column 4 shows that labor participation of females who were employees is not significantly affected by the increasing trend of violence. When disaggregating the samples by years of education, the results suggest that self-employed women with more than primary school are more likely to leave the labor market. The results are shown in the Appendix in Table A5. Although the samples are smaller because of the additional stratification the coefficients for females with higher education are significant and larger for women living in a municipality that switched to the top 50 th percentile of the homicide distribution, as these women are 20% less likely to stay in the labor market. If the costs of participating in the labor market are higher than the benefits, it is intuitive to think that women will leave the labor market and possibly re-allocate their time to, for example, household production. Column 1 of table 5 shows whether the intensity of the participation in the labor force was affected by the level of violence by looking at the log of the total hours worked during the 12 months before to the MxFLS interview. Panel A shows the effect for females who were self-employed in 2005 and were active in the labor market after the increasing trend of violence (2009/10) and Panel B shows the results for women who worked as employees in We start the analysis looking at the effects on female self-employment. The results in column 1 of Panel A show a significant and negative effect of the homicide rate on the hours worked of self-employed women who were living in municipalities that switched from the bottom 50 th percentile to the top 50 th percentile, relative to women living in municipalities that did not switch. Women living in municipalities that had the highest increase of violence decreased their hours worked by around 85%. This means that from an average of 37 hours per week (7.4 hours per day in a work week of 5 days) self-employed women affected by this level of increased violence decreased their work week by more than three fourths; therefore, basically changed from being full-time employers to part-time employers (in the next section we will explore which occupation within this category have been the most affected). The results stratifying by education are shown in Table A6. Panel A and B show the results for self-employed women. Column 1 in Panel A provides evidence of a significant impact on women with lower education who were living in municipalities with the highest change in violence. However, the effect for these women is lower than the effect for all self-employed women (64% versus 85%). This would suggest that women with more than primary school, not only were more likely to leave the labor market but also to cut their hours worked. The numbers in Panel B might support these hypothesis, however the 22

24 results are imprecisely estimated because of the smaller sample and thus we cannot draw definitive conclusions from these results. In order to examine whether the effects on labor force participation have had an impact on a women s level of income, we now continue to examine the model using the log of hourly earnings and total earnings in the last 12 months focusing now in the results for self-employment Panel A. The results in column 2 suggest that, although women living in municipalities who switched to a higher end of the homicide distribution relative to those that did not switch are less likely to participate in the labor force, there is a significant positive effect on the hourly earnings of the women who stayed in the market and the effect is bigger in places that switched to the 50 th percentile of the violence distribution and up. If some women are leaving their jobs, the proportion that stay in the labor market may receive higher remunerations to incentivize their labor participation. The violence against women in particular has been noted in the last years in Mexico; if the cost of participation in the labor market (which increases by the fear of being victimized on their way to work, for example) exceeds the benefits, some women will chose to leave the labor force. With a contraction in the supply of labor, employers may incentivize their work with a higher payment. Another possible interpretation is that women in the higher distribution of earnings are the ones that stay in the labor market; therefore, the effect on the average earnings would simply reflect the change in the distribution of skilled women in the labor market. Table A6 show the results stratified by education (Columns 2 and 3 in Panel A and B for self-employed women). The results suggest that the positive effects are mainly driven by women with less than primary school, and previous results showed that women with more schooling are those more likely to leave the labor market; this suggests that the positive effects on earnings are not being driven by a positive selection of high skilled women remaining in the labor market. These are only suggestive results since the disaggregated results by education are imprecisely estimated due to a reduction in power. We move now to explore the effects of violence for employee women. Panel B of Table 5 shows the effects for women who were wage employees before the intensification of violence. Employee females compared to the self-employed do not suffer as adverse an impact from the increasing Mexican violence (recall that there are no effects on their probability of labor force participation). Column 1 shows negative effects of violence in hours worked although the effects are not significant. Moreover, the effect on the hourly earnings of employee women is positive although only significant at the 10% level and the magnitude is smaller than the effect for self-employed. This result could indicate a similar mechanism as the one observed for self-employed women but with a lower intensity of the impact. Similarly to what was 23

25 observed for self-employed women, the impact on the total labor income earned in the last 12 months (column 3) is positive, although in this case the results are not significant. The results disaggregated by education, although not significant, suggest that women with lower education decrease in higher percentage their hours worked and have a larger positive effect on their hourly earnings. Lower educated employee women are the group that shows a more similar pattern to the selfemployed women. If low educated women in the formal sector work in similar industries and occupations of self-employed women we could think that the results are being driven by the same mechanism. We can explore this hypothesis by looking at the samples disaggregated by occupation. The results in this section suggest that the labor market outcomes of women have been affected by violence, in particular in places where the violence escalated to the top 50 th percentile of the distribution relative to the places that did not switch. Violence adversely affected the labor force participation of selfemployed women by reducing their labor intensity; but, the effect on hourly income was positive. These results suggest that a proportion of self-employed females are leaving the labor market, probably to avoid being directly affected by violence, but those who stay in the labor market are being more highly compensated. Labor force participation of women who were employees before the escalation of violence is not affected but these women see a positive effect on their hourly earnings. Analysis by occupation This section shows the results of the model for the sample of female workers disaggregating the sample by the three most popular occupations in self-employment and formal employment 18. Table 6 provide the results. The three most common occupation categories for self-employed women are jobs in retail and commerce (51.47% of self-employed women, Panel A), manufacturing (women in this sector are mainly artisans, tailors and food makers, 23.25% of self-employed women are in this category, Panel B) and personal services/domestic employees (16%, Panel C). Panel A shows that women in the retail and commerce industry are likely to leave the labor force in municipalities where violence increased, it seems that they also diminish their work intensity (although the results are not significant), and contrary to what was found for the aggregated results, their hourly earnings and total earnings in the last month decrease with increasing violence. These results confirm previous hypothesis: women working in activities related with commerce might be negatively affected by violence since their earnings can fluctuate more with the economic activity of the area. If people are scared of going out to the streets the businesses activities will suffer from violence in an indirect way. The results 18 The MxFLS provides information on the two-digit Mexican classification of occupations. I aggregate those in 1. Professionals, technicians; 2. Agricultural, cattle activities, foresting, hunting and fishing workers; 3. Manufacture activities; 4. Administrative jobs; 5. Retail and commerce; 6. Personal services, domestic employees; and 7. Other. 24

26 in Panel C show a different story for women in the industry of personal services and domestic employment. Women in this sector, for example, working as housekeepers may fear being victimized while they travel to work, which could decrease their labor participation. With a contraction in the supply of labor, employers may incentivize their work with higher remunerations. Strong significant results are found for women in the commerce and personal services industry (this does not mean that women in manufacture (Panel B) do not have effects but the results are imprecisely estimated and it is not possible to draw definitive conclusions). The three main work categories for women who work as formal employees are administrative jobs (36.4%), personal services and domestic employment (25.2%) and professional, technicians (20.1%). Panel A shows that women in administrative jobs and in the sector of personal services show very similar results to those found for self-employed women working in the industry of personal services and domestic employment: women reduce their hours worked but their hourly earnings increase. Panel B shows no effects for women who have professional careers or work as technicians, it looks like these women were not affected by the increasing trend of violence; and Panel C shows that women in the personal services industry when working as employees see a lower impact on their labor outcomes. As discussed before, less educated employee women might have very similar jobs to those of self-employed women and may have been affected by violence in a similar fashion Males Table 7 shows the results for male labor force participation, following the same structure of Table 4 for females. The results in Table 7 suggest that the change in the homicide rate did not have an effect on male job participation, and this results persist when stratifying by education (Table A7 in the Appendix). These estimates are highly suggestive that concerns over reverse causality do not have empirical support. Given the drop in employment of males experience over the last few years in Mexico (Table 2), if homicide rates simply followed local economic failure, one would expect to see homicide rates being predictive of lower labor market participation. These results provide evidence that the heterogeneity in municipality level conflict intensity is not correlated with general trends in Mexico s economic performance. Column 1 of Table 8 shows the results of the estimation for the log of hours worked in the last 12 months. The results in Panel A (for the sample of self-employed males) and Panel B (for the sample of employees), confirm that the labor participation of males was not affected by the increasing trend of violence in Mexico. However, when disaggregating the sample of self-employed males by attained education some interesting heterogeneity is revealed. The results stratified by education are shown in Table A8 in the Appendix. On the one hand, self-employed males with less than primary school are not affected 25

27 by crime (Panel A), and on the other hand, self-employed males with at least primary education working in municipalities that switched to the th percentile of the homicide distribution cut their hours by around 53%. The results in Table A9 for employees confirm that labor participation of formal employees was not affected by violence. Columns 2 and 3 in Table 8 show the results for hourly earnings and total earnings in the last 12 months. The coefficients in columns 2 and 3 of Panel A show a negative effect of violence on both hourly and total earnings and while the effects sizes are non-trivial they are not significantly different from zero. Columns 4 and 5 explores an additional specification, disaggregating the top 50 th percentile into the th and 75 th -100 th percentiles. The results suggest that males living in municipalities that switched to the highest percentile of the violence distribution saw a reduction in their hourly earnings of around 37% relative to men living in places that did not switch and suffered a decrease of 31% in their total earnings. Moreover, when disaggregating the sample by education levels it seems that individuals with more than primary education are the most affected by the surge in violence. Panel A of Table A9 shows no significant effects for self-employed males with less than primary while Panel B suggests that self-employed males with more than primary school who suffered some increase in violence (switch to the 25 th -49 th percentile) not only decreased their hours worked but also their total earnings were negatively affected (total earnings decreased on average 53%). Males living in municipalities that switched to higher levels of violence did not decrease their hours worked but total earnings decreased 33% for those that switched to the 50 th 74 th percentile and 65% for those moving up the 75 th percentile. The negative effect on total earnings of self-employed males with higher education could give support to the anecdotal evidence in Mexico if we think that those men are more likely to be business owners. Businesses in places heavily affected by violence tend to close earlier to avoid being directly victimized and individuals are less likely to engage in commerce in the evening, which harms the profits of independent businesses; therefore, the earnings of business owners are the most susceptible to increasing violence. We will check this hypothesis by looking at the disaggregated results by occupation in the next section. The results in the previous tables suggest there is quite a change in the labor market environment for self-employed males that faced increased violence relative to self-employed females exposed to more violence. Males labor force participation is not being affected by the increasing trend of violence and this result does not depend on the intensity of violence in the municipality where they live. However, the absolute level of earnings and hourly earnings are being negatively affected by crime and the effects vary depending on the intensity of violence and on the education level of the respondent. 26

28 Panel B of Table 7 shows the results for employees. Contrary to what was found for self-employed men, hourly earnings are not affected by crime but total earnings are affected in a similar fashion: men living in municipalities that switched to the highest percentile of the homicide distribution saw a reduction of 17% in their total earnings. When disaggregating by education (Table A9) the significant results hold only for men with more than primary school. Analysis by occupation Table 9 shows the results disaggregated by occupation. We show the results using only three instead of four indicators for the different percentiles, since for all of the sub-samples the higher percentile (75 th 100 th ) does not show significant effects. Forty percent of self-employed men work in agricultural, cattle activities, foresting, hunting and fishing activities (Panel A of self-employed sample), 30% in manufacturing (Panel B of self-employed sample) and 20% in retail and commerce (Panel C of selfemployed sample). The distribution for employees of the three main categories is as follows: 43% work in the manufacturing sector (Panel B), 20% in the agricultural sector (Panel A) and 13% in administrative jobs (Panel C). Although most of the coefficients lose significance when disaggregating in these subsamples there is an interesting degree of heterogeneity across occupations. Self-employed men working in agricultural, cattle activities, foresting, hunting and fishing activities are the only group that shows negative effects on the probability of labor force participation (Column 1 Panel A). Other than this sub-sample it seems that, independently of the intensity of violence men will not leave the labor market. It is also interesting to compare the effects of crime for the same industries between self-employed and employees. Panel B shows the effects for men working in manufacturing activities. Most of these men are working in construction jobs, in auto mechanics activities and home maintenance (plumber, electrician, etc.). The activities that they perform are very close but it seems that the effects are different for self-employed and employees. The self-employed are more strongly adversely impacted in terms of their hourly earnings and total earnings compared to employees. Moreover, for self-employed men in retail and commerce activities there is evidence on a reduction in the intensity of hours worked (Column 2 Panel C). This is consistent with anecdotal evidence that shows how violence has affected business activities. High incidence of crime diminishes business activity and this can potentially have an effect on the total earnings of men working in this sector. Although the effects on earnings are not significant, the coefficients suggest that the total earnings of men working in the commerce sector diminishes when violence increases. The potential negative effects seem to be caused by the fewer productive hours in the labor market and not because of lower hourly earnings (these are only suggestive interpretations since the coefficients are not significant). 27

29 The previous results highlight the significant heterogeneity of the effects of violence in Mexico on labor market outcomes. Overall the results suggest that self-employed individuals are those who are the main victims of the increasing violence. Self-employed individuals can be more vulnerable to negative shocks since their income is more volatile as their jobs do not provide health insurance or mandatory mechanisms to save. If the most vulnerable individuals are the population most affected by crime, the negative effects of conflict may create poverty traps and affect long-term investment that spills over into the well-being of the next generation. In order to assess whether the effects found in the labor markets have translated into lower consumption I analyze this outcome in the next section Per Capita Expenditure There is a growing literature on household responses to negative shocks crisis (Townsend, 1995; Morduch, 1995; Beegle et al 2001; Frankenberg, et al., 2002; Thomas et al. 1999; Thomas et al, 2007). The increasing levels of violence in Mexico could have affected households wealth in different ways. First, if labor income is negatively impacted by crime, households without mechanisms to smooth consumption will be affected by this shock. Second, in order to minimize the probability of victimization conspicuous consumption and consumption that increases the likelihood of being on the streets might decrease. Recreational activities, eating outside, and also consumption of visible goods could suffer a negative impact 19. This section of the paper assess whether the negative shocks on the labor market have translated into reductions in other measures of the household s wealth such as per capita expenditure. The MxFLS has a rich component of household expenditures measures that allows analysis on PCE to be disaggregated by different consumption categories in order to assess which components have been more or less affected by violence. This component records detailed information on household expenditures on food (including self-production) and non-food goods and services, personal care, household items and durable goods. I estimate model (4) in five different categories of PCE in the last month: total per capita expenditure, expenditure on food, a category of arguably conspicuous consumption 20, health, and education. Panel A of Table 10 shows the results for all the households, Panel B for households with a female household head and Panel C for those with a male household head. Column 1 shows the results for the log of total per capita expenditure for the three samples. Looking at the sample in Panel A that includes all households, the results suggest that households living 19 It is also intuitive to think that expenses on security devices will increase. However, there is not a clear way to disentangle these expenses from the consumption section. 20 I include in this category more visible expenditures or expenditures that could potentially increase the probability of being victimized like: meals out, recreation, communication, clothes, domestic appliances, transportation, furniture, ceremonies and vacations and semi-durable goods. 28

30 in municipalities in which violence increased to the 25 th percentile of the homicide distribution and up decreased their total expenditure by around 10%. Disaggregating the sample by the gender of the household head we can see that the effect is similar for male and female household headed households when the municipality switched to the 25 th -49 th percentile of the distribution, but the effect is larger and more significant for household with a male household head when the switch is to the top 50 th percentile. In order to explore whether this reduction on expenditure is due to a reduction in total income in the household or is the reflection of a strategy taken by the household to reduce victimization we can look at different consumption categories, which could potentially shed light on these potential mechanisms. Column 2 shows the results for expenditure on food. Panel A shows that households affected by violence have reduced their expenditure on food around 6% but this negative and significant effect holds only for households with male household heads (Panel C). The effect on the category of conspicuous consumption shows a similar story (Column 3). The effects are negative and significant when looking all the households and households with male households heads. The negative effect on the category of conspicuous consumption may be an indication of a strategy taken by the household to minimize the probability of being targeted by criminal groups, or it can be a reflection of a negative shock on income as luxury goods may be those most quickly sacrificed in the household budget. Future research will explore potential mechanisms and sub-samples to explain these results. Column 4 shows no significant effects for any of the sub-samples and column 5 shows negative and significant effects on expenditure on education for households with female household heads living in municipalities where the violence increased the most. In the previous section we found that some selfemployed women are leaving the labor market, and earnings of self-employed women working in commerce activities decrease with violence. If the negative effect of income at the household level increases the likelihood of joining the labor market for young kids of school age the long term impact of this negative shock could be permanent. With this in mind, Brown and Velásquez (2013) assesses the extent to which a violent environment may alter the educational attainment, cognitive scores, time allocation, and employment behavior of children and young adults. 7. ROBUSTNESS CHECKS Measuring the impact of crime on economic outcomes imposes important challenges. Crime might be endogenous to the economic activity, which makes it difficult to identify a causal relationship between crime and, for example, labor market outcomes. The increasing trend of crime in Mexico is evident from 2007, but from 2002 to 2005 the homicide rate was very stable. As a robustness check, we estimate the same model discussed in the main specification, differencing this time individual and household observations measured in 2005 (MxFLS2) and in 2002 (MxFLS1), and using the same measures of violence 29

31 (MxFLS2 and MxFLS3) as in the main specification. We should not expect to find significant results for the variable of violence since future violence should not predict economic outcomes between 2002 and Table 10 shows the results for labor outcomes. Panel A provides the results for self-employed women. Recall that the results for self-employed women in the main specification showed a negative effect on labor participation and on hours worked, but a positive effect on hourly earnings conditional on receiving positive income. The results for self-employed women in Table 10 do not show any significant results and the sign and magnitude of the coefficients do not follow the same pattern as those from the main specification. The results in Panel B show the placebo test for employee women. In the main specification there was no evidence of effects of violence on labor force participation of women working in this sector (and the coefficients were close to zero), the coefficient on the hours worked is negative but not significant and the effects on hourly earnings positive (although significant only at the 10% level). The results in Panel B of Table 11 show only significant and positive effects for the probability of labor force participation. The effects, although significant, do not follow the same pattern as found in the same specification. These results suggest that municipalities where crime increased the most may have been on a positive trajectory of formal employment of women. Panel C and D show the results for males. The results in the main specification suggest that men do not change their labor participation but the total earnings diminish for men living in municipalities that switched to the highest percentile of the homicide rate distribution, but the effect is larger for selfemployed men. The results in Panel C of Table 11 show no evidence of these effects when looking at labor outcomes measured before the increasing wave of violence. The results in Table 12 show the results of the placebo test for the PCE outcomes. The estimates are almost all non-significant at the 5% level with only the number of significant coefficients as would be expected by chance. 8. DISCUSSION AND FUTURE WORK High levels of crime may modify the context in which individuals operate, affecting behaviors at the individual and household level (Gáfaro, Ibáñez and Calderón, 2011). In particular, they can have an effect on the labor markets available to individuals. The effects of crime might be negative if the fear of being victims of crime increases the cost of participating in the labor market or if investment decreases in a region, which in turn diminishes job opportunities. 30

32 Measuring the impact of crime on labor market outcomes in Mexico is critical given the high incidence of crime observed since In addition OCGs have diversified their financial sources, and even if drug trafficking activities account for most of their economic resources they have been relying more on criminal activities that directly affect the civil population, like extortions and car thefts. Exploiting information from the MxFLS in a period of low levels of homicides rates and in a period of high levels of homicides this study estimates an individual fixed effect model that controls for any time-invariant unobserved heterogeneity that could affect both exposure to violence and labor market outcomes. Moreover, combining this strategy with an intent-to-treat approach, where the municipality of residence in the pre-violence period determines an individual s exposure to violence this analysis is able to shield the estimates from potential bias due to violence related endogenous migration. The results of this study show that the increasing violence in Mexico has had a negative effect on the labor market participation of self-employed women. The analysis further suggests that self-employed women working in retail and commerce activities have experienced a negative impact on their earnings, while the effect of on hourly earnings of self-employed women working in personal services and as domestic employees was positive and significant. This result can be explained, on the one hand by a positive selection of self-employed women on the labor market, or in the other hand, by higher payments made to this group of women to incentivize their labor participation. Similarly, for males, the negative impact is also stronger for the self-employed individuals. A major difference between the genders though is that for males labor participation is not significantly reduced and there is no evidence of a positive impact on their earnings. Moreover, earnings of self-employed males particularly in the manufacture industry and in the commerce industry are restricted with increased exposure to violent crime. Increasing violence in Mexico might not only affect the direct participants but also innocent civilians. If this is the case the government needs to respond to the negative shocks violence is imposing on an individual s wealth. The next stage of this research will focus on measuring the impact of crime on other measures of wealth at the household level, focusing in particular on exploring whether more vulnerable households have been more significantly affected than households with more defined safety nets. In addition we will complement these analysis by using levels of violence, instead of the distribution of violence, to define the non-linear effects in order to explore if there is a threshold of violence after which the negative effects are evident. 31

33 9. REFERENCES Abadie, A. and J. Gardeazabal (2003) The Economic Costs of Conflict: A Case-Control Study for the Basque Country. American Economic Review, 93(1), Becker, G. (1965). A Theory of Time Allocation Economic Journal. 75: Becker, G. (1968). Crime and Punishment: An Economic Approach Journal of Political Economy. 76: Beegle, K. (2005). Labor Effects of Adult Mortality in Tanzanian Households. Economic Development and Cultural Change 53(3): Bozzoli, C., Brück, T. and Wald, N. (2011), Self-Employment and Conflict in Colombia. DIW BerlinDiscussion Paper No Brown, R. and A. Velásquez (2013) The Effect of Violent Conflict in Mexico on the Human Capital Accumulation of Children and Young Adults mimeo Duke University Camacho, A. and C. Rodriguez (2013). Firm Exit and Armed Conflict in Colombia Journal of Conflict Resolution 57: 89 Castillo, J.C., Mejia D. and P. Restrepo (2013). Illegal drug markets and violence in Mexico: The causes beyond Calderon mimeo Universidad de los Andes Collier, Paul, On the Economic Consequences of Civil War, Oxford Economic Papers 51(1): Collier, P. and M. Duponchel (2010). The Economic Legacy of Civil War: Firm Level Evidence from Sierra Leone. WIDER Working Paper, 90 Deininger, K Causes and consequences of civil strife - micro-level evidence from Uganda. PolicyResearch Working Paper Series Washington D.C.: The World Bank. Dell, M. (2011). Trafficking Networks and the Mexican Drug War, Job Market Paper Díaz-Cayeros, A., B. Magaloni, A. Matanock, and V. Romero (2011): Living in Fear: Mapping the Social Embeddedness of Drug Gangs and Violence in Mexico, mimeo Fernandez, M., Ibanez, A.M. and Peña, X Adjusting the labor supply to mitigate violent shocks:evidence from rural Colombia," Policy Research Working Paper Series 5684, The World Bank. 32

34 Frankenberg, E., J. Smith and D. Thomas (2003). Economic Shocks, Wealth and Welfare The Journal of Human Resources. 38(2): Gáfaro, M., A.M. Ibáñez, and V. Calderón (2011). Forced Migration, Female Labor Force Participation, and Intra-household Bargaining: Does Conflict Empower Women?. Documentos CEDE No Bogotá: Universidad de los Andes. Guerrero-Gutiérrez, E. (2011a): Security, Drugs, and Violence in Mexico: A Survey," 7th North AmericanForum, Washington D.C. Guerrero-Gutiérrez, E. (2011b). La raíz de la violencia Nexos, June, Available online: Guerrero-Gutiérrez, E. (2012a). 2011: La dispersión de la violencia Nexos, February, Available online: Guerrero-Gutiérrez, E. (2012b). Epidemias de Violencia Nexos, July, Available online: Guerrero-Gutiérrez, E. (2012c). La estrategia fallida " Nexos, December, Available online: Hoeffler, A. and M. Reynal-Querol (2003) Measuring the costs of conflict. Oxford Working Paper Kondylis, F. (2007). Conflict Displacement and Labour Market Outcomes in Post!War Bosnia and Hersegovina Journal of Development Economics 93(2): Mejia, D. and Restrepo P. (2010). "Crime and Conspicuous Consumption," Documentos CEDE , Universidad de los Andes Molzahn, C., V. Rios and D. Shirk (2012). Drug Violence in Mexico: Data and Analysis Through 2011 Trans-Border Institute Joan B. Kroc School of Peace Studies University of San Diego Pshiva, R. and G. Suarez (2010) Capital Crimes: Kidnappings and Corporate Investment in Colombia in The Economics of Crime: Lessons for and from Latin America, edited by Rafael Di Tella, Sebastian Edwards and Ernesto Schargrodky, Chicago: University of Chicago Press. Rios, V. and D. Shirk (2011). Drug Violence in Mexico: Data and Analysis Through 2010 Trans- Border Institute Joan B. Kroc School of Peace Studies University of San Diego 33

35 Robles, G., Magaloni, B., and Calderon G. (2013) The Economic Costs of Drug-Trafficking Violence in Mexico mimeo Rubalcava, Luis and Teruel, Graciela (2006). User s Guide for the Mexican Family Life Survey First Wave, Shemyakina, Olga N., "The labor market, education and armed conflict in Tajikistan," Policy Research Working Paper Series 5738, The World Bank. Shirk, D. (2011): Drug Violence and State Responses in Mexico, Manuscript: University of San Diego. Thomas, D., Beegle, K. and Frankenberg, E. (2003) Labor Market Transitions of Men and Women During an Economic Crisis: Evidence from Indonesia. in Women in the Labour Market in Changing Economies: Demographic Issues, B. Garcia, R. Anker, and A. Pinnelli (eds.), Oxford University Press 34

36 10. Tables and Figures Figure 1. INEGI - Monthly Homicide Rate (per 100,000) : : Month of Occurrance Figure 2. INEGI and National Public Security System - Monthly Homicide Rate (per 100,000) Month of Occurrance INEGI SNSP

37 Figure 3. Timing of MxFLS and Monthly Homicide Rate 2.5 MxFLS1 MxFLS2 MxFLS3 INEGI - Monthly Homicide Rate (per 100,000) Month of Occurrance