IMMIGRATION AND CRIME IN AN ERA OF TRANSFORMATION: A LONGITUDINAL ANALYSIS OF HOMICIDES IN SAN DIEGO NEIGHBORHOODS, *

Transcription

1 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: 1 21-JUL-10 8:19 IMMIGRATION AND CRIME IN AN ERA OF TRANSFORMATION: A LONGITUDINAL ANALYSIS OF HOMICIDES IN SAN DIEGO NEIGHBORHOODS, * RAMIRO MARTINEZ, JR. Department of Criminal Justice Florida International University JACOB I. STOWELL Department of Criminal Justice and Criminology University of Massachusetts Lowell MATTHEW T. LEE Department of Sociology University of Akron KEYWORDS: immigration, social disorganization, longitudinal analysis, neighborhoods, homicide * An earlier draft of this article was presented at the 2007 annual meeting of the American Society of Criminology and the 2008 Conference on Issues Facing the Latino Community sponsored by the Institute for the Study of Ethnicity, Race, and Sexuality at the University of Washington. This research was made possible, in part, through grants by the National Institute on Drug Abuse to the Latino Drug Abuse Research Center Violence Study (DA ), the Harry Frank Guggenheim Foundation, and the National Science Foundation under grants SES and SBR to the National Consortium on Violence Research (YR3-TSRP1). Additional support for this project was provided to the first author while he was a visiting scholar at the University of Houston Center for Mexican American Studies. The authors thank the City of San Diego Police Department (SDPD), especially former Chief Jerry Sanders, Captain Jim Collins, and Lieutenants Kevin Rooney and Terry McManus, for providing access to homicide files. We also would like to thank Robert J. Bursik Jr., Glenn Deane, Richard Rosenfeld, the Editor, and the anonymous reviewers for their thoughtful comments on previous versions of this article. Points of view and conclusions are our own and do not reflect the views of any funding agency or the SDPD. Direct correspondence to Ramiro Martinez, Jr., Department of Criminal Justice, University Park, SW 8th Street, Miami, FL ( martinra@fiu.edu) American Society of Criminology CRIMINOLOGY VOLUME 48 NUMBER

2 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: 2 21-JUL-10 8: MARTINEZ, STOWELL & LEE Emerging research associated with the immigration revitalization perspective suggests that immigration has been labeled inaccurately as a cause of crime in contemporary society. In fact, crime seems to be unexpectedly low in many communities that exhibit high levels of the following classic indicators of social disorganization: residential instability, ethnic heterogeneity, and immigration. But virtually all research conducted to date has been cross-sectional in nature and therefore unable to demonstrate how the relationship between immigration and crime might covary over time. This limitation is significant, especially because current versions of social disorganization theory posit a dynamic relationship between structural factors and crime that unfolds over time. The current study addresses this issue by exploring the effects of immigration on neighborhood-level homicide trends in the city of San Diego, California, using a combination of racially/ethnically disaggregated homicide victim data and community structural indicators collected for three decennial census periods. Consistent with the revitalization thesis, results show that the increased size of the foreignborn population reduces lethal violence over time. Specifically, we find that neighborhoods with a larger share of immigrants have fewer total, non-latino White, and Latino homicide victims. More broadly, our findings suggest that social disorganization in heavily immigrant cities might be largely a function of economic deprivation rather than forms of neighborhood or system stability. Almost 70 years ago, Shaw and McKay (1969 [1942]) vividly illustrated the residential settlement patterns of new immigrants. Through their research, they provided a detailed picture of the role immigration played in shaping juvenile crime. For example, Shaw and McKay observed how immigrants settled into high-crime communities in part because of inexpensive housing and restricted access to economic resources available to newcomers. Through absorption into the occupational structure of the city and assimilation into mainstream society, immigrants 1 gradually moved into areas of lower crime and higher economic affluence (Bursik, 2006). The larger significance of Shaw and McKay s observation was that most immigrant groups initially settled into socially disrupted, high-crime communities and often were perceived incorrectly as a disproportionately crime-prone population. They argued that any direct connection between particular groups of immigrants and crime was largely spurious (Bursik, 2006: 24). Instead, they identified the causes of crime as corresponding to neighborhood structural properties, such as residential turnover and ethnic heterogeneity, both of which might be brought about by immigration 1. The notable exception was the rural migration of Blacks/African Americans.

3 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: 3 21-JUL-10 8:19 IMMIGRATION AND CRIME, as a social process but have nothing to do with the characteristics of immigrants themselves. In other words, immigrants were not biologically predisposed to crime and their cultural traditions did not necessarily promote crime, as other criminologists of that era had argued. Decades later, most ecologically minded scholars agree that it is the structural characteristics identified by Shaw and McKay that produce socially disorganized communities by weakening intervening mechanisms of social control. Although politicians and public commentators continue to exploit stereotypes that essentialize crime as a property of immigrants themselves, or the poor more generally, the disorganization perspective focuses on types of places rather than on types of people (Lee, 2003: 11). It turns out that both kinds of explanation might require revision, at least for cities transformed by recent waves of immigration. This change is suggested by arguments that the recent homicide surge in some cities reflects their low levels of immigration, whereas immigrant destination cities have not experienced this increase (see Dale, 2007). More generally, few serious scholars today see crime as an inherent characteristic of immigrant groups. Moreover, based on empirical research findings, it stands to reason that some neighborhood factors might not disrupt communities as the initial theoretical logic suggests. This is the question being asked by researchers associated with the emerging immigration revitalization perspective (Lee and Martinez, 2002: 365; see also Lee and Martinez, 2009; Sampson, 2008; Sampson, Morenoff, and Raudenbush, 2005). These scholars are showing that immigration, ethnic heterogeneity, and residential instability often are associated with reduced crime rates. This trend is partly because crime rates today are often highest in areas of concentrated disadvantage, which is associated with the most segregated and racially/ethnically homogeneous neighborhoods in large cities (Sampson and Wilson, 1995). But positive social change is also a major factor. Researchers in a variety of other fields, such as health, mental health, and education, have spent years marveling at the immigrant paradox of high poverty combined with betterthan-expected life outcomes (cf. Rumbaut, 1999). But the discipline of criminology is just beginning to appreciate the myriad ways in which immigration in the modern era paradoxically has strengthened institutions of social control, fostered economic development, and sparked a revival of previously high-crime, inner-city neighborhoods. The present study is designed in part as a response to recent calls by scholars to address these specific limitations in the extant criminological literature (see Bursik, 2006; Martinez, 2002; Sampson and Bean, 2006; Stowell, 2007). But the point of the article is not simply to add a longitudinal analysis involving racially/ethnically disaggregated data to the existing

4 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: 4 21-JUL-10 8: MARTINEZ, STOWELL & LEE body of cross-sectional research, as important as that might be for assessing the immigration revitalization perspective. Instead, we use our unique longitudinal analysis to advance the communities and crime field by exploring a series of hypotheses drawn from the immigration and social disorganization literatures that shed new light on the processes behind ecological variations in crime. We no longer can assume contemporary communities are fundamentally similar to those studied by Shaw and McKay (1969 [1942]) and other founders of the social disorganization approach working in the early and middle part of the twentieth century. In one sense, our findings serve as a partial corrective to one aspect of this perspective. We argue that although the processes of social disorganization remain important, some factors producing it have changed with respect to immigration. The stream of immigrants currently reshaping the United States is no longer primarily of European origin. The racial/ethnic/immigrant composition of many neighborhoods has grown increasingly diverse, and the Mexican border has supplanted Ellis Island as the predominant entry point into the nation. Although this transformation is well documented by sociologists and historians, we still know very little about how immigration impacts social disorganization and, in turn, about levels of violent crime in a contemporary social context. Moreover, research has yet to quantify whether the impact of shifting levels of immigration over time disrupts border communities as many politicians and pundits speculate (for a review, see Martinez, 2002: 1 31). Admittedly, this matter is not simple, as even Shaw and McKay s (1969 [1942]) multifaceted theory of crime was developed inductively from their contextually specific data. As such, we recognize that some exceptions to the general theoretical premises might exist because local conditions eclipse broad theoretical predictions (Martinez, Lee, and Nielsen, 2004: 152). By acknowledging this principle, our attempt to assess the impact of immigration or other social processes of disorganization on crime must also consider the outcomes in relation to a potentially changing local context (Entwisle, 2007: 687; see also Nielsen, Lee, and Martinez, 2005 on the importance of local variations in the causes of crime). Therefore, our interpretations of the findings incorporate detailed knowledge of the research setting as well as an exploration of how these processes unfold over time. To address these issues, the current study employs neighborhood-level data from the city of San Diego collected from 1980 to As a major immigrant destination, this research site has undergone an intense compositional transformation since 1980, making this border city an ideal critical case (Yin, 1994) for investigating the impact of the dynamic processes thought to be associated with social disorganization. The substantive focus of our analysis is on the consequence of changing levels of immigration on

5 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: 5 21-JUL-10 8:19 IMMIGRATION AND CRIME, levels of community violence (as measured by homicide) over time. Using homicide data as a proxy for crime has its shortcomings. But we believe that the reliability of homicide data relative to other crimes, which are less comprehensively reported to police, coupled with the high level of public concern about this serious form of crime, justifies its selection as the dependent variable in this first neighborhood-level longitudinal, multivariate test of the immigration crime relationship in the modern era. LITERATURE REVIEW AND HYPOTHESES SOCIAL DISORGANIZATION AND IMMIGRATION REVITALIZATION One of Shaw and McKay s (1969 [1942]) contributions was directing attention to the consequences of distressed community conditions on levels of crime. They noted how variations in the rates of delinquency across nativity, nationality, or racial groups corresponded to the structural composition of communities. They also argued that when exposed to the same types of areas, boys of native parentage and boys of foreign parentage have similar rates of delinquency (Bursik, 2006). Because of this underlying logic, the direct effect of immigration was not a focus of the initial tests of the theory. Although social disorganization theory posits that changes in the nationality composition of a population are not accompanied by appreciable changes in relative rates of delinquency, this assumed temporal influence has not yet been examined fully (Bursik, 2006: 26). Existing studies have documented the importance of social structural factors in observed levels of criminal deviance. This finding is hardly controversial in the communities and crime literature (see Lee, Martinez, and Rosenfeld, 2001; Martinez, Stowell, and Cancino, 2008; Stowell, 2007). But social disorganization theory goes further, positing that crime is a relatively constant condition of a specific type of urban neighborhood, independent of the characteristics of its residents (Shaw and McKay, 1969 [1942]). In advancing this argument, disorganization theorists make two assumptions about neighborhood processes over time. The theory takes for granted the structural stability of such neighborhoods, which in turn suggests the relative intractability of their levels of criminal deviance over time. But the traditional assumption of ecological stability has been questioned in more recent studies of social disorganization (cf. Bursik, 1988; Bursik and Webb, 1982; Schuerman and Kobrin, 1986). The conclusion of this literature is that the stability, so carefully documented by Shaw and McKay in the first half of the twentieth century, has attenuated somewhat after World War II, a process Bursik (1984: 402) referred to as increased

6 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: 6 21-JUL-10 8: MARTINEZ, STOWELL & LEE ecological differentiation. Therefore, in this study, we test the following hypothesis: Hypothesis 1 (H1): Ecological structures such as immigration, economic disadvantage, and residential stability will vary over time. However, important work by Sampson and Morenoff (2006: 199) has demonstrated that, in the contemporary era, neighborhoods have remained remarkably stable in their relative position in economic, political, and social terms despite the inflow and outflow of individual residents, at least in Chicago. That is, secular increases in the structural covariates of homicide do not necessarily translate into differential effects in multivariate models if communities maintain the same position relative to each other. Therefore, our assessment of H1 will include an examination of the extent to which nonsecular change occurred in a context of secular change. Because longitudinal studies are costly and difficult to conduct, one of Bursik s (1988: 524) five criticisms of the social disorganization tradition was that current research conducted within that framework was based on cross-sectional data, which must assume that local communities are not undergoing a redefinition of their role in the ecological system. In other words, this work must take for granted that communities do not change over time relative to each other or in terms of overall structural conditions. The existing cross-sectional tests of the relationship between immigration (and other structural conditions) and crime, although consistent with Shaw and McKay s (1969 [1942]) early specification of social disorganization theory, might require updating at a theoretical level, particularly if nonsecular changes have occurred in a given ecological setting. Sorting through the mixed findings of this body of work is impossible because it is incapable of identifying where in the criminal career cycle a given community might be (see Schuerman and Kobrin, 1986). If we have indeed entered an era of structural (or system ) instability a debatable contention about which more research clearly is required (see Sampson and Morenoff, 2006 for a promising approach; see also Peterson and Krivo, 2005) then a valid quantitative assessment of social disorganization would have to model any changes in crime rates as they relate to possible changes in structural covariates over time. Following this logic, a longitudinal approach is the most appropriate test of this notion. The analytical methodology employed in the present study allows for just such comparisons. Specifically, our regression models focus on whether, and to what degree, temporal differences in structural and compositional factors within communities are associated with differences in levels of lethal violence. This approach is consistent with several arguments underpinning disorganization theory. As Bursik and Webb (1982: 28) pointed out, the

7 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: 7 21-JUL-10 8:19 IMMIGRATION AND CRIME, dynamic changes associated with the invasion and succession of new demographic groups is at the heart of the ecological approach to crime. Another basic assumption of classic disorganization theory is that various forms of neighborhood or residential instability (as opposed to the ecological system instability associated with nonsecular changes) weaken social control and cause crime. But in the current period of urban transformation, this begs the question, What does residential stability mean in an era of concentrated poverty in some areas and immigration revitalization in others? We have good reasons to suspect that residential stability, often defined in terms of population turnover and the proportion of vacant buildings, will have different effects in the San Diego of today compared with the Chicago of yesterday. Shaw and McKay (1969 [1942]) were concerned with an especially destructive source of instability in the industrializing cities of the Midwest the expansion of the industrial center that pushed into surrounding residential areas and caused neighborhood deterioration. But how should we conceive instability in an era when immigration often suppresses crime (cf. Lee and Martinez, 2009), fluctuating property values in some gentrifying urban neighborhoods encourage mobility as people sell their homes for profit or abandon them when prices fall, and concentrated poverty in homogenous, native-born, extremely disadvantaged areas seems to be driving urban crime rates? Unlike industrializing Chicago, large sections of San Diego transitioned from rural to suburban status during the time period of our study. Little reason exists to suspect this kind of change will increase crime rates, even though the stability of such areas (as measured by the commonly used proxy length of time at current residence ) is likely to be low. To begin to reconceptualize the meaning of neighborhood stability and instability, we investigate the impact of two variables independent of controls that have been central to the communities and crime literature: immigration and residential instability. Both capture aspects of population turnover (i.e., in-migration and out-migration) that once were thought to contribute to disorganization. The connection with residential instability is clear, as it is sensible to assume that the movement of large numbers of people in and out of a community would disrupt networks of social control and increase crime. Immigration traditionally has been thought to increase disorganization as well because the invasion of immigrants into a community increases ethnic heterogeneity and poverty, introduces language barriers, and even might initiate a process of out-migration of established residents. In sum, residential instability is perceived as a barrier that makes it difficult for members of a neighborhood to agree on collective goals and effectively control the behavior of residents. Indeed, some scholars have used the proportion of foreign born and the proportion of ethnic minorities as proxies for a subculture of weakened

8 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: 8 21-JUL-10 8: MARTINEZ, STOWELL & LEE social control (cf. Schuerman and Kobrin, 1986: 71, 78). The very presence of such groups constitutes a leading edge of movement into emerging high-crime areas (Schuerman and Kobrin, 1986: 78). In other words, when immigrants and ethnic minorities move into a neighborhood, disorganization and increased crime are the inevitable byproducts. But as we have discussed, this thesis has been challenged by the immigration revitalization perspective, a part of the more general immigrant paradox (Lee and Martinez, 2006: 91; Rumbaut, 1999) literature, which finds that immigration produces unexpectedly positive outcomes across a range of domains, including health, mental health, educational attainment, and crime. It is likely that immigration and residential stability are related. Immigration might reduce stability, immigrants might settle in unstable areas, or both processes might occur simultaneously. We test the following bivariate relationship to clarify the extent to which immigration might be associated with instability over time: Hypothesis 2 (H2): Immigration is associated with instability. Because we examine this relationship over time, we can assess whether immigration precedes instability or whether immigrants are settling into already unstable areas. But our primary theoretical interest is the extent to which immigration and residential stability emerge as significant predictors of homicide in a multivariate longitudinal analysis. Drawing on the immigration revitalization perspective and our knowledge of San Diego, it is our contention that immigration will be associated with decreased homicide over time, controlling for residential stability and other structural covariates of homicide: Hypothesis 3 (H3): Controlling for other factors, immigration decreases homicide. HOMICIDE AND THE RACIAL/ETHNIC INVARIANCE HYPOTHESIS To this point, our hypotheses have addressed important issues of stability and change over time and across neighborhoods that are central to the communities and crime literature, particularly with regard to social disorganization and immigration revitalization. We question whether ecological structures vary over time (H1) and the extent to which immigration is associated longitudinally with residential instability and homicide in the contemporary era (H2 and H3). But another important debate persists in the literature that must be engaged the racial/ethnic invariance hypothesis. If structural sources of crime are similar for all groups, then it is unnecessary to test hypotheses for specific groups such as Latinos, non-latino Whites, and non-latino Blacks (Krivo and Peterson, 2000; McNulty,

9 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: 9 21-JUL-10 8:19 IMMIGRATION AND CRIME, ). As with other major theories of crime (Krivo and Peterson, 2000: 556), the social disorganization perspective argues that levels of neighborhood crime are not associated with the nativity or nationality characteristics of their inhabitants. But previous research has provided mixed support for this notion. On the one hand, the weight of the empirical evidence documents the fact that community characteristics shape crime. On the other hand, it is less clear whether levels of crime endure regardless of the underlying community nativity, racial, and ethnic composition. Researchers comparing neighborhood levels of violent crime by nativity, race, and ethnicity have demonstrated both similarities and differences in the criminal victimization of specific groups and the covariates of victimization, depending on local context (Peterson and Krivo, 2005: 347). A recent authoritative review of the literature sided with the invariance hypothesis but noted that only two studies were methodologically sophisticated enough to test adequately the invariance hypothesis and concluded that the state of knowledge needs substantial improvement (Peterson and Krivo, 2005: 347). Much of the previous work on racial invariance has focused on Black and White differences (Peterson and Krivo, 2005). But recent research indicates that immigrants, as well as Latinos and Asians, regardless of nativity status, fare better than expected on a host of social and economic factors, including violent crime rates, than native-born African Americans with similarly high levels of disadvantage (see Lee and Martinez, 2002, 2006; Martinez and Lee, 2000: 486; Sampson and Bean, 2006: 20). For example, of the three largest racial/ethnic groups in the United States, overall homicide rates are highest for Blacks, followed by Latinos, and then by Whites, and predictors of homicide have similarities and differences at the community level (Peterson and Krivo, 2005). This pattern has prompted some researchers to conclude immigrants and Latinos are less violent, even more so when they live in concentrated immigrant areas (Sampson and Bean, 2006: 21). In one recent study, Phillips (2002) found that structural conditions fully explain the gap between White and Latino homicide rates but only account for half of the difference between White and Black rates. Most relevant to the present study are previous findings conducted in San Diego. The city of San Diego is a major entry point into the United States and is a border city with a large and impoverished Latino (heavily Mexican-origin) population comprising the largest racial/ethnic minority group in the city. Analyzing levels of lethal violence between the period, researchers found that homicide among Latinos in San Diego generally follows the same pattern as among Blacks and Whites in terms of the predictive power of concentrated disadvantage (Lee, 2003; Lee, Martinez, and Rosenfeld, 2001; Martinez, 2002; Martinez, Lee, and

10 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: JUL-10 8: MARTINEZ, STOWELL & LEE Nielsen, 2004). Yet Latinos experienced less violence and victimization than Blacks (and even Whites at times) given the relatively high levels of disadvantage (see also Peterson and Krivo, 2005). Furthermore, some predictors of Latino violence, especially community nativity composition, are unique, bringing in new issues that transcend race. Notably, the concentration of immigrants in San Diego also seems to tell a different story with respect to violence than the concentration of African Americans in cities beyond the Sunbelt (Lee, Martinez, and Rosenfeld, 2001; Martinez, 2002). As with the other issues we have reviewed, much of this research is cross sectional, which hinders our ability to make temporal predictions in the present study. But taking all existing research into consideration, some of which focuses on immigration and some on Latinos, we agree with Krivo and Peterson (2000: 557) that similarity of conditions for different racial and ethnic groups do not exist in the vast majority of places, including San Diego. Given the previous research, we expect that some factors (such as economic deprivation) will exhibit invariant effects on the homicide levels of different racial and ethnic groups, whereas other factors (such as immigration) are more likely to affect coethnics, such as Latinos, than African Americans. We therefore offer the following hypothesis: Hypothesis 4 (H4): Effects on homicide of all structural factors except immigration are racially/ethnically invariant over time. ACCOUNTING FOR TEMPORAL ORDER Our first four hypotheses were motivated by theoretical concerns. But the problem of temporal ordering has also been a methodological theme of this article. We have argued in favor of a longitudinal approach that captures the dynamic processes at the heart of the neighborhood ecology of crime. Cross-sectional findings with regard to racial invariance or immigration might be confounded by the issue of time-ordering. Our over-time analysis seeks to overcome this limitation. But it is also important to determine the impact of immigration in racial/ethnic-specific models of changes in lethal violence, net of a given community s previous homicide levels. After all, it is possible that the negative immigration crime relationship could be the result of immigrants moving into low-homicide areas. This effect might be especially prominent for Latinos in San Diego but less so for Blacks. We assess this aspect of the immigration crime debate by including a temporally lagged count of homicides in our final set of total and group-specific models. Hypothesis 5 (H5): The immigration effect remains robust despite the inclusion of a temporally lagged count of homicides.

11 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: JUL-10 8:19 IMMIGRATION AND CRIME, THE CURRENT SETTING: SAN DIEGO AND THE COMPLICATED STORY OF NEW IMMIGRATION The time period that we have selected for this study represents a strategic era of transformation in which to test our hypotheses because (as we will show) San Diego experienced a dramatic increase in the foreign-born population during this time. But unraveling the immigration crime link in San Diego is complicated by the reality that the city s Mexican- and Asian-origin populations are a study in contrasts both consist of old and new arrivals. Members of the two dominant immigrant groups also might reside in neighborhoods with varying levels of community disadvantage, affluence, and violence. Latinos outnumber Asians by a nearly 2:1 ratio, and Filipinos comprise nearly half of the Asian population (Lee and Martinez, 2006). With its historical origins in pre-1845 Mexico, Latinos originally moved in from various interior and border regions across Mexico, but after annexation, later arrivals also traveled from El Paso and other parts of the southwestern United States. In addition, migrants from Central and South America also moved through Mexico on their way to America. San Diego s much-maligned twin city Tijuana serves as the gateway through which a disproportionate number of border crossers (both legal and illegal) enter the United States (see Nevins, 2002). Latinos are the largest ethnic minority group in the city, but migrants born abroad represent almost half of the Mexican-origin population (47 percent). Most professional positions are occupied by non-latino Whites, whereas a smaller share of the professional employment is of Mexican or Asian origin. On the whole, the local Latino middle class is smaller and less influential than in other Latino immigrant destination cities such as San Antonio, El Paso, or Miami. So the ethnic enclave effect that has been hypothesized to buffer crime in some other immigrant destination cities is not as prominent in San Diego. Consider also that San Diego currently has only one Latino city council member and has not had a Latino mayor at least since California was annexed into the United States. This research setting therefore represents a more stringent test of the immigration revitalization perspective, again suggesting that it is well suited as a critical case study (Yin, 1994). Despite their underrepresentation in professional occupations and local politics, it is possible that over time, as the Latino and Asian populations in San Diego expanded, the concentration of immigrants might help to buffer neighborhoods against crime and delinquency (see Sampson and Bean, 2006). As Sampson (2008) pointed out, the beneficial effects of immigration as a social process reduce crime for nonimmigrant groups as well.

12 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: JUL-10 8: MARTINEZ, STOWELL & LEE DATA AND METHODS In the current research, we analyze administrative neighborhood-level panel data for the three most recent decennial census years. These data are unique among neighborhood-level studies of immigration because they allow us to examine social processes over time, a subject that has generated much speculation but relatively little scholarship. Data for the dependent variable the number of homicide victims were obtained from records maintained by the San Diego Police Department (SDPD) homicide investigation unit. Homicide information was gathered directly from the incident report written by the lead investigator on the case, including information about the location of the homicide, narratives, and other information about the incident. Each report was retrieved and copied by hand by the lead author at the SDPD and shipped back to the home office. The homicide files were read and coded by a trained staff of coders following a coding report used by other city-level homicide researchers (see Martinez, Stowell, and Cancino, 2008). To maximize the stability of our regression estimates, the dependent variable for each panel year is the number of killings pooled during a 5- year period. Specifically, for each of the panel years, the dependent variable includes the count of homicides for the 2 years leading up to and after the given census. In other words, crime data for each of the time points under investigation are as follows: , , and The use of pooled data is common in existing community-level research, primarily because the outcome of interest is a rare event (Lee, 2003; Lee, Martinez, and Rosenfeld, 2001; Martinez, Stowell, and Cancino, 2008; Stowell, 2007; see also Mears and Bhati, 2006: 21). All homicides were geocoded and referenced using a San Diego base street map for the year The data on killings were aggregated into census tracts, thus permitting the matching of the total number of homicides and racial/ethnic specific (non-latino White, non-latino Black, and Latino) homicides to tract-level data compiled in the Neighborhood Change Data Base (NCDB) (GeoLytics, 2002). The NCDB contains data on population characteristics in addition to a variety of other social-structural correlates of lethal violence. This data source is particularly useful for longitudinal analyses because the neighborhood-level variables have been normalized to tract boundaries used in the 2000 Census, which avoids problems associated with boundary changes during the 20-year period being studied A potential limitation to geocoding all incidents to 2000 streets is that changes to the city s street layout might produce disparities in the percent of homicides for which we can obtain geographic coordinates across the three time periods (see Paulsen and Robinson, 2004). This issue does not seem to be a problem in this

13 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: JUL-10 8:19 IMMIGRATION AND CRIME, Following the example of previous research on this topic, the independent variables are constructed using decennial census data. The variables to be included in the regression models include a series of social structural indicators informed by neighborhood-level examinations of social disorganization theory, and particularly the work of Sampson and colleagues (see Sampson and Morenoff, 2004; Sampson, Morenoff, and Raudenbush, 2005). In the current study, quantifying the impact of immigration on lethal violence is of most substantive interest. To obtain a more comprehensive understanding of the underlying processes, we will use a general measure of immigration, defined as the percent of the population born outside of the United States. The measure of economic disadvantage used in the present study is an index that includes the following dimensions of community economic viability: percent of the population living in poverty, percent of families receiving public assistance, and percent of families with children younger than age 18 that are headed by females. It is necessary to control for levels of economic deprivation because neighborhood disadvantage has been shown to be a strong predictor of levels of criminal violence (see Lee, Martinez, and Rosenfeld, 2001; Sampson, Morenoff, and Raudenbush, 2005). As discussed, social disorganization theory also highlights residential stability (or instability) as a factor that promotes (or undermines) networks of informal social control in communities. To account for neighborhood settlement patterns, we created a neighborhood stability index that includes the percent of households that are owner-occupied and the percent of persons who have lived in the same house for more than 5 years. Additionally, as a proxy for neighborhood affluence, we constructed a measure of the percent of those aged 16 years and older employed in a professional or managerial capacity. It is important to note that each of these occupational classifications primarily comprise non-service-sector jobs, including engineers, financial planners, lawyers, and scientists. We believe that it is important to include this measure, as previous research has documented the ability of neighborhood affluence to buffer against violent crime (see Sampson and Bean, 2006). As a measure of the level of informal social control in neighborhoods, the models will include a measure of the relative presence of adults per child defined as the ratio of adults (persons aged 18 years and older) to children (persons aged 17 years and younger). case, however, as we had a successful match rate of around 95 percent for each of the three periods, a level well above the recommended acceptable hit rate of 85 percent (Ratcliffe, 2004).

14 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: JUL-10 8: MARTINEZ, STOWELL & LEE ESTIMATION PROCEDURE The multivariate regression equations in the present study are estimated using a fixed-effects analytical procedure, a method permitting us to capitalize on the panel structure of the data. More specifically, the fixedeffects method allows us to introduce a temporal dimension by modeling the effect of change. Indeed, the fixed-effects estimation procedure focuses exclusively on within-unit changes. Following the description in Allison (2005: 93), we estimate the following regression equation: logl it = m i + bx it + gz i + a i (1) where logl it is the log-linear function of the dependent variables, m i is the regression intercept essentially the average change in lethal violence across years x it represents the time-varying predictor variables, z i represents factors that do not vary over time, and a i is the control for unobserved fixed effects. An attractive feature of the fixed-effects model is that it controls for all stable characteristics of neighborhoods without requiring all indicators to be included in the model. As Allison (2005: 2 3) contended, a benefit of being able to control for stable characteristics is that the fixed-effects procedure effectively eliminates potentially large sources of bias, which are typically associated with model misspecification. The production of unbiased regression estimates, Johnson (1995: 1,068) concluded, allows for stronger and more valid inferences about the effects than can be made in many alternative methods (see also Argue, Johnson, and White, 1999). The focus on within-unit changes also makes fixed-effects estimation attractive for substantive reasons. Specifically, the notion that levels of criminal violence in some areas are resistant to change is one key proposition advanced by social disorganization theory. Indeed, Shaw and McKay (1969 [1942]) perceived crime as a relatively stable characteristic of certain communities, one that is not affected by demographic or compositional changes. Fundamental to this claim is the notion that successive changes in [neighborhood] nativity and nationality compositions will not have appreciable impacts on levels of crime (1969 [1942]: 315). Furthermore, although some contemporary researchers hint at the fact that immigration might have a stabilizing influence on communities, the longitudinal relationship remains largely unknown (Martinez, Stowell, and Cancino, 2008). Because fixed-effects methods focus exclusively on the degree to which changes in demographic or social structural factors in a given neighborhood are associated with changes in levels of lethal violence, the regression results will cast new light on this under-analyzed aspect of disorganization theory. Despite their strengths, some limitations persist in the fixed-effects regression techniques that bear mentioning. Specifically, the small portion

15 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: JUL-10 8:19 IMMIGRATION AND CRIME, of suburban neighborhoods with zero homicides (approximately 4 7 percent) in both the overall and race/ethnic specific analyses are automatically dropped from the analysis. Because the regression model changes, areas in which no homicides occurred for any of the three periods under investigation effectively are dropped from the likelihood function because no within-neighborhood variability can be analyzed (see Allison, 2005: 82). Substantively, the omission of such neighborhoods is not problematic for the purposes of this study, as we are interested in the dynamic link between immigration and lethal violence. In other words, the neighborhoods in which no homicides were committed will not contribute to a better understanding of the temporal changes in levels of lethal violence. 3 Because the dependent variable consists of counts, the fixed-effects models will be estimated using a Poisson-based analytical approach. In the estimation procedure, we include the logged population as the regression exposure factor, which essentially translates the outcome variable into a rate during the estimation procedure. The diagnostics indicated that overdispersion (variance greater than the mean) is a problem in these data; thus, Poisson-based negative binomial regression is used (Osgood, 2000). In the following analyses, homicide counts are regressed on the independent variables. RESULTS DESCRIPTIVE AND BIVARIATE RESULTS A key objective of the current study is to advance current knowledge on immigration and crime by introducing a longitudinal analytical design. Any study of change assumes that the analytical units have undergone observable changes during the period under investigation. In other words, if the average neighborhood structural and demographic composition remains stable over time, then it would be unnecessary to move beyond the use of cross-sectional data. Our first hypothesis was that ecological structures such as immigration, economic disadvantage, and residential stability would vary over time. The evidence indicates that between In a separate analysis (available upon request), we examined the structural and demographic compositions of these zero homicide communities. Clearly areas did experience some structural changes over time as they moved from rural to suburban communities. We do recognize that it is possible that the levels of other forms of violent and nonviolent criminal behavior in these areas also varied during this period. The limitation of available neighborhood-level longitudinal data prevents us from using broader measures of crime. Despite these limitations, we believe the strengths of the method, namely, that fixed-effects models produce unbiased estimates and, because they model temporal changes, outweigh these drawbacks.

16 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: JUL-10 8: MARTINEZ, STOWELL & LEE and 2000, many San Diego neighborhoods did shift structurally, thus supporting our first hypothesis. Table 1 contains the descriptive statistics for the lethal violence and neighborhood indicators that will be included in subsequent multivariate analyses. Table 1. Descriptive Statistics Total homicide rate White homicide rate Black homicide rate Latino homicide rate Percent foreign born 15.61(10.64) 19.83(12.41) 23.95(13.44) Disadvantage index (standardized).41 (3.17).39 (3.27).40 (2.98) Stability index.16 (1.86).14 (1.92).31 (2.01) Adult/child ratio 5.44(12.83) 8.35(35.22) 6.02(13.71) Percent professional employment (8.98) (9.24) 25.23(11.90) NOTES: Standard deviations in parentheses. The total homicide rate is the number of homicides per 100,000. The White, Black, and Latino homicide rates are calculated as the number of racial/ethnic group-specific homicides per 100,000 group-specific persons. N = 297. As a study focusing on neighborhood levels of homicide, the descriptive analyses provide a broad overview of the degree to which patterns of lethal violence have changed during the period under investigation. Establishing general trends in homicide levels across all census tracts is important because it will offer some insight into whether the patterns observed in San Diego during the past two decades are in some way exceptional. The descriptive analyses will reveal whether lethal violence in San Diego followed the rise and fall in the national homicide rates during this period or whether a disparate trend emerged (see Blumstein and Wallman, 2000). The same is true for levels of race/ethnic specific homicides. It is possible that the relative risk of homicide victimization in San Diego differs from broader trends. For example, previous research has shown that Latinos generally are victimized at higher rates than non-latino Whites but not as high as non-latino Blacks (see Martinez, 2002; Martinez and Lee, 2000). We will also be able to identify whether divergent racial/ethnic trends emerge in homicides (i.e., levels decreased for one group and increased for another). Turning our attention to patterns of lethal violence, we note that the temporal variations in both overall and race/ethnic specific homicide rates are consistent with broader patterns. The data indicate that overall homicide rates, as well as those for specific race/ethnic groups, surged during the 1980s, peaked, and then declined sharply throughout the 1990s (see Blumstein and Wallman, 2000; Sampson and Bean, 2006). The uniformity

17 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: JUL-10 8:19 IMMIGRATION AND CRIME, of the rise and decline is notable with the exception of White rates, which did not rise as the overall and group-specific rates in 2000 fell to approximately one third of the corresponding 1980 values, thus mirroring patterns observed in previous research (Martinez, 2002; Peterson and Krivo, 2005). The descriptive findings highlight contours observed in previous studies, namely, that the Latino homicide rate for the city is higher than that for Whites but lower than the comparable rate for Blacks. Furthermore, despite the changes in the underlying risk of homicide victimization between 1980 and 2000, this general pattern was relatively stable during the entire period under investigation. Although these results show both consistencies and differences in lethal violence over time, it is not clear whether or to what degree they are influenced by neighborhood structural changes and immigration. The large increase in the size of the foreign-born population is also evident in the descriptive results. By 2000, nearly a quarter of the population in the average neighborhood in San Diego was born outside of the United States, which is an increase of 53 percent in two decades. Furthermore, the observed change in the average community nativity composition was the most substantial relative to the other neighborhood indicators of interest. The lone exception is reflected in the fact that average levels of neighborhood disadvantage remained relatively stable over time. The results also show that in 2000, communities had lower levels of residential stability than they did in the 1980 census. The disparity in the size of the adult and child populations also varied over time, peaking in Finally, the presence of individuals employed in professional occupations experienced a sharp increase, growing by more than 30 percent in two decades. Based on these descriptive findings, it is clear that neighborhood conditions were not static during the period under investigation. This finding suggests that the assumption of structural (or system) stability inherent in the cross-sectional analyses that have been conducted to date is problematic. Consistent with the rationale for our analytical approach, a longitudinal design does seem to be required to get to the heart (Bursik and Webb, 1982: 28) of the processes involved in shaping community levels of crime. However, as we have discussed, a key issue is whether neighborhoods maintained their relative positions to each other against a backdrop of secular change (Sampson and Morenoff, 2006). To address this issue, we examined correlations and scatterplots for our variables over time. The results (available upon request) indicate that neighborhoods remained generally, but not totally, stable relative to each other amid the temporal changes described in table 1. This trend was especially true for economic disadvantage and residential stability, as the correlations for these variables were always greater than.78, regardless of whether we compared 1980 with 1990, 1990 with 2000, or 1980 with A moderate degree of

18 \\server05\productn\c\cry\48-3\cry308.txt unknown Seq: JUL-10 8: MARTINEZ, STOWELL & LEE consistency was found for our immigration measure (r =.63) and homicide rates (r =.60) from 1980 to Figures 1 and 2 present scatterplots for these two key variables. So the question of positional stability which we refer to as system stability to distinguish it from residential stability can be answered as follows: 1) The relative position of neighborhoods in terms of disadvantage and residential stability remained largely the same, and 2) more variation was found over time when neighborhoods were compared in terms of immigration and homicide rates. In other words, some system instability occurs at the ecological level with regard to our key predictor variable of interest (immigration) as well as our outcome variable, although many neighborhoods also maintained their relative positions. Figure 1. Neighborhood Change in Percent Foreign Born, % Foreign-Born % Foreign-Born 1980 Before we turn to the multivariate models, we first must address our second hypothesis, which states that immigration is associated with residential (not system) instability. Because our indicator of stability includes a measure of length of residence in a given housing unit, the fact that stability declined dramatically over time is attributable, in part, to the influx of immigration the city experienced during this period. It is also a function of the transformation of previously rural areas into suburbs, as well as dramatically fluctuating real estate prices that provided the incentive to sell property or to abandon them. To examine the relationship between stability and immigration, we examined the bivariate correlations for these variables (available upon request). The correlations provide support for