International Immigration,Internal Migration,and Homicide in Canadian Provinces

Abstract

The relationship between immigration and crime is politically charged and often fueled by the presence (or lack) of xenophobia. Many theoretical and empirical assessments of this relationship indicate that immigration does indeed lead to increased crime, but more recent (and very early) research investigating homicide calls this finding into question. The current analysis investigates the relationship between immigration and homicide using multiple measures of migration and Canadian provinces as the unit of analysis. It is found that the link between immigration and homicide is complex and dependent on the measure of migration used. Generally speaking, the results presented here are consistent with the more recent and very early research. Immigration, in and of itself, does not increase homicide. Rather it is the increase in the most criminogenic subpopulation that matters, that is young males.

Keywords

homicide immigration interprovincial migration

Introduction

The relationship between immigration and crime is a politically charged issue that tends to lead to xenophobia-inspired policies. Until very recently, most of the empirical assessments of this relationship are rather dated and based on limited data and questionable theory (Mears, 2001). Therefore, the foundation of immigration policy lacks contemporary empirical support, and the often-cited relationship between immigration and crime (the former increasing the latter) is more mythological than scientific (Hagan & Palloni, 1999). For some reason, the mythology of the relationship between immigration and crime persists despite research stating otherwise that dates back a century—see the Dillingham Commission Reports (Croxton, 1911), Wickersham Commission Reports (National Commission on Law Observance and Enforcement, 1931), J. Cohen (1931), and Shaw and McKay (1931, 1942).

Recent research on the relationship between immigration and homicide also calls this positive relationship between immigration and crime into question. Conducting analyses at the meso-level (neighborhoods) in U.S. cities along the U.S.–Mexican border, Ramiro Martinez, Jr. and colleagues find that neighborhoods with larger immigrant populations either have similar or lower levels of homicide than neighborhoods with smaller immigrant populations (Martinez, Stowell, & Cancino, 2008). Needless to say, this research has significant implications for criminological theory, immigration policy, and public opinion. If we can show with more recent data, consistently in different contexts, that immigrants are not more criminogenic than native-born populations, theoretical perspectives that follow the mythological relationship must be reconsidered. This information may then be used to inform the public. Consequently, immigration policy that recognizes immigrants do not cause increases in crime may be put into place without uninformed protest.

This article contributes to the immigration and crime empirical literature in a number of ways. First, the unit of analysis is the Canadian province, providing a Canadian perspective to the relatively small macro-based immigration and crime literature. Such an analysis allows for a replication of this recent research in another context. Second, three different types of dependent variables are used. Not only is the traditional crime rate calculated but also an alternative crime rate and the location quotient (LQ). These varying dependent variables provide greater insight into the results. Third, because of data available from Statistics Canada, this analysis is able to independently assess the impact of international immigration and interprovincial migration. Such a separation has not been shown in previous research. Fourth, and last, this analysis provides a much-needed temporal component to understand the relationship between immigration and crime. Very little longitudinal research exists on the immigration and crime relationship. As discussed in the following sections, this allows for the use of a statistical methodology (fixed effect panel estimation) that is superior to cross-sectional analysis when attempting to uncover causal relationships.

Immigration, Homicide, and the Canadian Provinces

Theoretical Expectations

The primary purpose of this article is not to develop or test any particular theoretical framework. Rather, the primary purpose is twofold: First, this article serves as a replication of recent macro-level research on the relationship between immigration and crime, and second, to investigate the importance of different measurements of crime and immigration in an effort to better understand the relationship between them. Though the current focus is not theoretical testing or development, there are a number of theoretical perspectives that provide insight into the relationship between immigration and homicide, or immigration and crime, more generally. Each of these theoretical perspectives has a long history in the criminological literature and predicts that increases in immigration lead to increases in crime. These theoretical perspectives are strain theory, subculture theory, social control theory, and social disorganization theory. Each is discussed in turn.

Strain theory refers to the opportunity structures that are available to immigrants, or any other population for that matter (Merton, 1938). Because of a lack of social networks relating to employment and/or difficulties with the recognition of foreign credentials, the legitimate opportunities for immigrants may be limited. This leads to strain on the part of immigrants, subsequently leading to illegitimate activities that may lead to homicides. Of course, this normally occurs only when there is a breakdown in the social institutions that help maintain social control (Messner & Rosenfeld, 1994, 2001), but in the case of immigrants, these social institutions may simply not be in place.

Subculture theory may be relevant because the immigrant population arrives with or develops a different set of values than the dominant interest groups in society that set the rules of law (A. K. Cohen, 1955). Consequently, there is a difference between what immigrant populations and the dominant groups in society believe to be legitimate activities. This leads to immigrant populations committing crime, again potentially leading to homicides.

Social control theory refers to our having social bonds to family, friends, work, conventional society, and so on, which prevent our committing criminal acts (Hirschi, 1969). Immigrant populations may have bonds in their new society to any family members and friends present with them. However, depending upon how long immigrants have resided in their new society, the presence of other bonds to conventional society may be less than those who are native born.

Last, and most prominently within the criminological literature, there is social disorganization theory. As outlined by Shaw and McKay (1931, 1942), areas with high degrees of unemployment, population turnover, and ethnic heterogeneity are socially disorganized and conducive to crime—see also Sampson, Raudenbush, and Earl (1997) and Sampson and Raudenbush (1999). This is because these areas are not able to set common values and goals that may be used to solve community problems (Bursik, 1988). In the case of some immigrant populations, ethnically heterogeneous areas contain people who literally cannot speak to each other because of language difficulties. Within this theoretical context, it is the neighborhood that is conducive to crime, and immigrant populations tend to (initially) locate in these socially disorganized areas (Butcher & Piehl, 1998).

Though these theoretical perspectives are well established within the criminological literature, there are counter views or alternative interpretations using these theoretical perspectives. Lee, Martinez, and Rosenfeld (2001) state that immigrant populations have strong ties to family and local labor markets that offset the effects stated previously. More specifically, these authors argue, though do not state explicitly, that there is not necessarily a different subculture or a lack of social control within immigrant populations. As such, these theories are not wrong; they simply are not applicable in this context.

With regard to social disorganization theory, Shaw and McKay (1942) never state that particular ethnic groups are more criminogenic than other ethnic groups. On the contrary, they find that once new immigrants are able to move out of socially disorganized areas, their criminal activities fall in line with native-born populations in their new neighborhoods. There is also a nonlinearity in social disorganization theory when measuring recent immigrants: As an area has more recent immigrants, its ethnic heterogeneity increases, leading to social disorganization; however, as that immigrant population grows (immigrants tend to come in waves from particular countries and settle in similar areas), ethnic heterogeneity eventually decreases because the immigrant population dominates the neighborhood and social organization may increase. This leads to ethnic enclaves (Borjas, 1999; Stark, 1991) that have been shown to increase job-market opportunities through network effects (Lazear, 1999; Portes, 1987). Last, social disorganization theory may not have “behaved” as expected because past research has not considered the effects of immigration-based population change versus population change due to internal migration. Therefore, the nuances of social disorganization theory are important to understand because it is more flexible than it appears on the surface.

Empirical Assessments of Immigration and Homicide

These theoretical differences in predicting the relationship between immigrants and crime are most easily resolved empirically. If immigrants are found to be more active in illegitimate activities, then one of the above-mentioned theoretical perspectives is likely at work. If not, theoretical expectations must be altered and/or reformulated.

Generally speaking, immigrants are not particularly criminogenic (Butcher & Piehl, 1998; Hagan & Palloni, 1998; Martinez & Lee, 2000; Reid, Weiss, Adelman, & Jaret, 2005). There are two factors at work here in understanding why this is the case. First, if immigrant populations come from impoverished nations, their frame of reference is important because the “poor” neighborhood they now live in may be relatively better than their life before. Because of this different perspective, a greater value placed on work, and perceived opportunities for economic advancement, immigrants commit less crime (Kao & Tienda, 1995; Zhang & Sanders, 1999). Second, some immigrant populations are better educated than native-born populations (Alba & Nee, 1997). Therefore, assuming there is no difficulty with the recognition of credentials, these immigrants are in a better position than native-born populations to obtain (well-paying) legitimate employment. Though not explicitly investigating the relationship between immigration and crime, Andresen (2006b) finds that the relationship between recent immigrants and crime is negative. In this particular context (Vancouver, Canada), the recent immigrant population is of high socioeconomic status and settles in the wealthier areas of the city where crime was low and remained low—see Ley (1999) and Ley and Smith (2000) for a discussion of the high socioeconomic status of recent immigrants to British Columbia.

Most of the recent research on the relationship between immigration and crime is undertaken by Ramiro Martinez, Jr. and colleagues (Lee et al., 2001; Martinez, 2000; Martinez et al., 2008; Martinez & Lee, 2000; Nielsen et al., 2005; Stowell & Martinez, 2007). Dominantly through an analysis of Latino and African American homicide, this research has repeatedly shown that recent immigrants have either a statistically insignificant or negative relationship with homicide. As such, this branch of research challenges the standard theoretical perspectives of the relationship between immigration and crime.

At macro levels of analysis, there is far less research. Butcher and Piehl (1998) and Reid et al. (2005) show that recent immigration has no statistical relationship with either property or violent crime. In addition, Reid et al. show that particular ethnic groups (Asians) have negative relationships with crime—this is similar to the result found by Andresen (2006b). In a national (U.S.) study undertaken by Lauritsen (2001), increases in immigration to suburban areas lead to increases in crime, whereas increases in immigration to urban areas lead to decreases in crime.

Perhaps the most compelling recent macro-level research is that of Ousey and Kubrin (2009) and Stowell, Messner, McGeever, and Raffalovich (2009). Both of these studies investigate the relationship between immigration and violent crime, finding a negative relationship. The compelling aspect of these studies is the research design—longitudinal. As these authors note, cross-sectional research is limited in testing theory or relationships between variables, more generally. In a cross-sectional analysis, statistically significant relationships imply that the distribution of one variable is related to the distribution of another variable. Little, however, can be said regarding causation. In a longitudinal analysis, one investigates the relationship between the changes in one variable relative to changes in another variable. This allows for more definitive statements to be made regarding causality. As such, the substantive findings in this macro-level research is that the negative relationship between immigration and violent crime is confirmed using a method that can be used to make more definitive statements regarding causality. This type of research design is used in the current analysis.

Needless to say, the evidence is overwhelmingly in favor of a negative relationship between immigration and crime. At the very least, one can say that the relationship between immigration and crime put forth by most theories (positive) is not monolithic. At the very most, these theories are simply wrong regarding the relationship between immigration and crime. As such, these old theories either need to be updated to correspond with known contemporary facts (Nielsen et al., 2005; Stowell & Martinez, 2007) or it must be acknowledged that these theories are no longer relevant for understanding the relationship between immigration and crime.

Another Perspective: Routine Activity Theory

The incongruence between much of the theoretical predictions on the relationship between immigration and crime and the more recent empirical evidence forces one to consider alternative perspectives. That alternative perspective is routine activity theory. Routine activity theory posits that a crime may only occur when a motivated offender and a suitable target converge in time and space without the presence of a capable guardian (L. E. Cohen & Felson, 1979). Understanding changes in crime when considering routine activity theory does not require complex changes in the motivation of offenders. Rather, routine activity theory is a probabilistic theory: Each of these convergences has a given probability of a criminal event taking place; if there is an increase or decrease in the convergences of potential offenders and potential suitable targets, there is a corresponding increase or decrease in the number of actual criminal events.

This approach to understanding crime proves to be successful in a number of empirical studies considering a variety of units of analysis. Routine activity theory proves to be instructive for understanding long-term crime trends at the macro level (L. E. Cohen & Felson, 1979), variations in neighborhood crime rates (Andresen, 2006a, 2006b), and individual victimization based on lifestyle factors (L. W. Kennedy & Forde, 1990). Needless to say, routine activity theory is fungible to a number of criminogenic situations.

The most fundamental aspect of routine activity theory is that the greater the number of convergences of motivated offenders and suitable targets without a capable guardian, the greater the number of criminal events. In the current analysis, it is posited that the presence of motivated offenders is operating as an explanatory variable, not immigration, per se. Specifically, large portions of the immigrant population may be part of the subpopulation that is most prone to extreme violence, for example, young males (Boyd, 2000; Courtwright, 1996a, 1996b). It has long been known in the criminological literature that young males are most often the offenders and victims of violent crime, across time and places (Courtwright, 1996a; Quetelet, 1842). In his analysis of violent crime, particularly in the United States, Courtwright (1996b) goes into great detail regarding the significance of young male populations and violent crime, especially homicide. Courtwright (1996b) outlines how high proportions of young males in frontier America led to very high levels of homicide. He also explicates how such increases in young males help explain the crime increase during the 1960s and 1970s: Men avoided, delayed, or terminated marriages and spent less time with spouses and children. These increases in young (unmarried) males spending less time in the relatively protective environment of the home is consistent with the routine activity theory explanation for the crime boom of the 1960s and 1970s. As such, such an explanation may prove instructive for the relationship between immigration and crime (homicide).

As discussed in the following section, Canadian data on immigration allow for a distinction between immigration, generally speaking, and immigration of young males, specifically. Therefore, the analysis in this article has the purpose of providing a simpler explanation for variations in homicide that relate to immigrants; rather than appealing to underlying social forces, it is simply the increased presence of young males.

Data and Method

Though the evidence states that immigrant populations do not commit more crime than native-born populations, there are aspects of immigration that may alter the socioeconomic urban structure that in turn alter criminogenic factors. For example, a large immigrant population may be low-skilled and flood the labor market, negatively affecting labor markets for native-born populations (Grogger, 1998), leading to an increase in crime. Alternatively, a large immigrant population may lead to the rejuvenation of stagnating (urban) economies (Lee et al., 2001), leading to a decrease in crime. Consequently, meso-level (neighborhood) and macro-level (metropolitan areas, states/provinces, and nations) analyses need to be undertaken.

To build on the macro-level analysis of Reid et al. (2005), this article uses the Canadian province as the unit of analysis. This choice is instructive for a number of reasons. First, because of data availability, it fills a gap in the literature on the use of panel data (cross sections measured across time) that is needed to better understand causal linkages (Ousey & Kubrin, 2009; Stowell et al., 2009)—all data are for the years 1986 to 2005, 200 observations. Second, at the provincial level, though homicide is still a relatively rare event, its presence across space and time allows for the calculation of multiple homicide measurements and more traditional empirical methods. Third, and last, Statistics Canada not only counts the number of immigrants for each province but also the number of interprovincial migrants and their respective ages. Therefore, differences between international and internal migration may be obtained, as well as differences between migrants in general and young male migrants—young males are the most criminogenic population, particularly for homicide (Boyd, 2000; Hirschi & Gottfredson, 1983).

Canadian Homicide

As with most industrialized societies, homicide is a relatively rare criminal event in Canada consisting of 0.02% of criminal events known to police (Li, 2007; Silver, 2007). Moreover, the Canadian homicide rate has been in a relatively steady decline since the mid-1970s, being at or less than 2 homicides per 100,000 for the past 15 years—this is one third the homicide rate in the United States, on average. Despite this downward trend, there is significant variation in the Canadian homicide rate across the Canadian provinces; for decades, the Western provinces have had homicide rates above the national average (Giffen, 1965). Despite numerous attempts to understand these regional differences (Daly, Wilson, & Vasdev, 2001; Giffen, 1965, 1976; Hartnagel, 1978, 1997; L. W. Kennedy, Silverman, & Forde, 1991), they are still not understood. However, having such regional variation in homicide (and migration) is instructive for understanding the relationship between immigration and homicide at an aggregate level.

Homicide Data and Measurement

All variables used in the analysis are obtained from Statistics Canada’s Socio-Economic Information Management System (CANSIM). In Canada, there are four types of Criminal Code offences that constitute homicide: first-degree murder, second-degree murder, manslaughter, and infanticide; all four offences are included in the homicide counts. Homicide is measured in three forms: the traditional homicide rate, an adjusted homicide rate, and the homicide LQ. Neither of these latter two calculations have been used previously in this literature.

The traditional homicide rate is the standard crime measurement of the number of homicides per 100,000 population—total population of the unit of analysis. Because of its widespread use, this measurement is used here. However, this form of the crime rate has a relatively long history of being shown to be problematic in measuring crime (see Boggs, 1965; Harries, 1991). Though most often considered problematic at the neighborhood level, there are difficulties at the provincial (and even national) level as well.

Considering the homicide rate as a measure of violence in society, using the total population of the unit of analysis as the normalizing variable in a crime rate implicitly assumes that all members of that population are at equal risk of offending and/or victimization. This is simply not the case. It has long been known that youth are far more active in criminal activities than adults as both offenders and victims (Hirschi & Gottfredson, 1983). In addition, the vast majority of homicides are committed by young males (Boyd, 2000; Courtwright, 1996a, 1996b). Because of these known facts regarding homicide, an adjusted homicide rate is also analyzed in the following using the number of young males (aged 15-29 years) as the normalizing variable. As shown by Andresen, Jenion, and Jenion (2003), the use of such a normalizing variable drastically alters the nature of the homicide trend at the national level; the same is true here at the provincial level. With this adjusted homicide rate being informed by known facts relating to homicide, if the statistical results are significantly different from the traditional homicide rate, the results from the adjusted homicide rate are taken as correct.

It should be noted that using the total number of homicide incidents in the numerator and a different normalizing variable leads to a potential mismatch in the calculations for this dependent variable. However, it is well known that the total number of homicide incidents is overrepresented by young males as offenders. As such, changing the normalizing variable is more likely to (partially) correct the mismatch in the traditional calculation. This is why the research in this area takes this approach, see Boggs (1965), Harries (1991), and Andresen (2006a).

The LQ is a measure that has been used in economic geography since the 1940s to measure employment or industrial specialization (Isard et al., 1998; Miller, Gibson, & Wright, 1991). Brantingham and Brantingham (1993, 1995, 1998) introduced the LQ into criminological research more than a decade ago, but its adoption has been both sparse and rather recent—Rengert (1996), Hirschfield and Bowers (1997), Andresen (2007, 2009), McCord and Ratcliffe (2007), Ratcliffe and Rengert (2008), and Robinson (2008) are recent examples that use the LQ in the analysis of crime.

The lack of adoption of the LQ in criminological research is surprising because it provides a useful interpretation of criminal activity. Rather than measuring the risk of offending or victimization, per se, the LQ is a measure of specialization. In the current context, the LQ measures the percentage of crimes that homicides represent in a province relative to the percentage of crimes that homicides represent in Canada as a whole. This provides a measurement of under-, over-, or expected representation of homicide in Canadian provinces. For example, in 2006, Ontario had a low homicide rate relative to the rest of Canada (below the national average), that may lead the police to believe that homicide is not a problem in Ontario. However, based on the LQ, Ontario is one of four provinces that exhibits an overrepresentation of homicide in its crime mix relative to the rest of Canada (Alberta, Saskatchewan, Manitoba, and Ontario); in fact, Ontario is the only central or eastern province with such an overrepresentation. Consequently, homicide is a problem within Ontario, relative to other crime classifications.

The LQ is included in the analysis because of its connection to routine activity theory. As posited by routine activity theory, if there is an increase in the number of convergences, there is a corresponding increase in criminal events. However, these convergences do not occur in all places equally: there are hot spots and cold spots of crime (Chainey & Ratcliffe, 2005). The use of the LQ allows for any specializations in homicide at the provincial level to be tested against theoretically identified covariates: immigrants, in general (social disorganization theory), and young males who happen to be immigrants (routine activity theory). In the context of social disorganization theory, where there are more immigrants, there should be more crime; and in the context of routine activity theory, where there are more young males (partially because of young male immigrants) there should be more crime.

The LQ is calculated as follows:

L Q = \frac{\frac{C_{i n}}{C_{t n}}}{\frac{\sum_{n = 1}^{N} C_{i n}}{\sum_{n = 1}^{N} C_{t n}}}

(1)

where C_in is the count of homicides in province n, C_tn is the count of all crimes in province n, N is the total number of provinces, and all crimes include all Criminal Code offences, less traffic. If the LQ is equal to one, the province has its proportional share of homicide; if the LQ is greater than one, it has a disproportionately larger share of homicide; if the LQ is less than one, the province has a disproportionately smaller share of homicide. Specifically, if a province has a LQ of 1.20, that province has 20% more homicides than expected given the percentage of homicide in Canada as a whole—that province then “specializes” in homicide. Miller et al. (1991) provide the following classifications that are useful for interpreting the LQ: very underrepresented areas, 0 ≤ LQ ≤ 0.70; moderately underrepresented areas, 0.70 < LQ ≤ 0.90; average represented areas, 0.90 < LQ ≤ 1.10; moderately overrepresented areas, 1.10 < LQ ≤ 1.30; and very overrepresented areas, LQ > 1.30.

The use of these three measures of homicide allows for greater identification of the relationship between immigration and crime. The traditional and adjusted homicide rates provide similar analyses to those done in previous research. The LQ, however, provides new insight into this relationship. Rather than asking whether immigration increases the volume of homicide, when used in a statistical analysis, the LQ asks whether immigration affects the specialization (under-, average, or overrepresentation) of homicide.

Migration Data

As stated previously, one of the contributions to the immigration and crime literature this article provides is measuring migration in a number of different ways. First, there is the distinction between external migration (international immigration) and internal migration (interprovincial migration). This distinction allows for the identification of differential effects from two different forms of migration. Is it international immigration that matters or simply migration in general that has a relationship with homicide? Second, to consider the importance of motivated offenders (routine activity theory), there is a distinction between total migration and the migration of young (15-29 years old) males. This distinction allows for the identification of differential effects from the total migrant population and the migrant subpopulation of young males. Is it migrants that lead to changes in homicide or simply changes in the population of (motivated) young males? Third, and last, there is the distinction between gross migration and net migration. Is it the number of “new” people that enter a province that affect homicide or the net change in the number of people?

This leads to a total of eight migration variables. Each of these variables is measured as recent, an aggregate of the most recent 5 years, and a percentage of the total population. In the statistical analysis given in the following sections, each of these eight migration variables is analyzed individually as well as collectively. Such a statistical methodology allows for the identification of false inference when only one form of migration is measured.

Control Variables

To isolate the independent effect of the migration variables on homicide, a number of control variables are used. These control variables revolve around the presence of youth, economic deprivation, and the risk of apprehension; as youth are the most criminogenic subpopulation (Hirschi & Gottfredson, 1983), economic deprivation leads to more illegitimate opportunities than not (Blau & Blau, 1982; Daly et al., 2001; Shaw & McKay, 1942; Wells & Rankin, 1991), and the risk of apprehension decreases criminal choices (Avio, 1979; Ehrlich, 1975, 1977; Katz, Levitt, & Shustorovich, 2003). Youth are measured using the percentage of 15- to 29-year-old males. Economic deprivation is measured using the Gini coefficient, GDP per capita, the percentage of the population living in low income, the unemployment rate, and the divorce rate; these variables are commonly used to measure economic deprivation in the social disorganization theory literature. The risk of apprehension is captured by corrections expenditures per capita, police expenditures per capita, officers per capita, and the number of incidents per police officer. A trend variable is included to control for any trends in the data.

Statistical Methodology

The statistical specification used in the analysis is shown in Equation 2:

y_{i j t} = α + γ_{j} + \sum_{m = 1}^{8} β_{m} x_{m} + \sum_{k = 1}^{10} λ_{k} Z_{k} + ε_{i j t},

(2)

where y_ijt is homicide measure i in province j at time t, α is the common intercept, γ _j represents fixed effect for province j, β _m is the vector of parameter estimates for the m migration variables, x_m is the matrix of m migration variables, λ _k is the vector of parameter estimates for the k control variables, z_k is the matrix of k control variables, and ϵ _ijt is the iid error term. Final model selection is undertaken using a general-to-specific methodology (testing down) and a 10% significance level; a 10% significance level is used to avoid omitted variable bias because when a 5% level is used to remove statistically insignificant variables, the estimated parameters of the remaining variables change, sometimes significantly. White’s heteroskedastic-consistent standard errors are used for all statistical significance testing. The Godfrey (1978) test for autocorrelation in the residuals (stationarity), a more general and powerful form of the Durbin–Watson statistic, found no evidence of autocorrelation in the analyses.

The analysis below is undertaken using a fixed effects panel specification. Such a statistical specification is not commonly used in empirical studies that are criminological in nature, but those empirical studies find interesting results: Nagin and Paternoster (1991) use a panel specification to distinguish between two theories of the impact of prior delinquency on future delinquency; Levitt (2001) reverses the relationship between unemployment and crime analyzing U.S. states; and Worrall and Pratt (2004) show that high school dropouts, welfare rates, unemployment rates, and the percentage of African Americans are insignificantly related to property crime in California.

Incorporating a temporal component into the analysis of cross sections (provinces) allows for greater flexibility in statistical modeling because of decreased multicollinearity and more degrees of freedom (Greene, 2000; Hsiao, 2003). Despite this increased flexibility, cross-sectional heterogeneity may be present. This makes the use of ordinary least squares (OLS) problematic when analyzing panel data (Baltagi, 2001); cross-sectional units are expected to respond to independent variables in the same manner (same slope parameters), but the levels of homicide are different across the provinces (different intercepts). Moulton (1986, 1987) outlines the statistical issues that emerge from cross-sectional heterogeneity, but generally speaking, cross-sectional heterogeneity leads to statistical bias because the variation from cross-sectional heterogeneity is forced into the error term (Wooldridge, 2002). As such, statistical methods specific to panel data are required: random or fixed effects estimation.

When employing either fixed or random effects in statistical estimation, variables are added to the regression specifications that account for cross-sectional heterogeneity: fixed effects estimation adds group-specific constant terms and random effects estimation adds a group-specific error term (Greene, 2000). There are potentially a large number of group-specific constant terms for fixed effect estimation such that it loses degrees of freedom and corresponding efficiency. This often makes random effects estimation a better alternative. Though random effects estimation may appear to be a better alternative, it is not always appropriate because it assumes the group-specific error term is not correlated with other variables when the specification or statistical bias occurs (P. E. Kennedy, 2008). This assumption is often violated when using sociodemographic and socioeconomic measurements. Therefore, this assumption must be tested using the Hausman (1978) test.

Regardless of the results of testing for fixed versus random effects specification, fixed effects estimation has a particular advantage: it estimates the “short run effects” from changes in the independent variables (P. E. Kennedy, 2008, p. 287). In an analysis that uses cross sections representing only a single year, the results represent the long run equilibrium relationships. As such, if an analysis is concerned with the changes in independent variables, such as the impact of increases in immigrant populations on homicide, fixed effects estimation is the appropriate statistical specification. Such a specification allows for a representation of the impact of actual immigrant populations rather than a representation of where the immigrant populations happen to reside. In addition to this methodological reason for using fixed effects estimation, statistical testing shows that random effects estimation is biased in the current context. Consequently, considering the current interest and statistical testing, fixed effects estimation is the appropriate statistical specification.

Results

Descriptive Results

The descriptive statistics for the three homicide measures are shown in Table 1. The values shown are for all years under analysis. Detailed year-by-year tables are available on request. Immediately apparent, when considering the mean, is the east-to-west increase in the traditional homicide rate; as one moves from Newfoundland to Quebec to British Columbia, there is a definite increase in the traditional homicide rate. The adjusted homicide rate also exhibits this trend, though the levels of the adjusted homicide rate are much higher because of the different normalizing variable used. In fact, the ranking of provincial homicide rates is exactly the same for both the traditional and adjusted homicide rates. At this stage of the analysis, these two rates are not expected to have significantly different results.

Table 1.

Homicide Descriptive Statistics, by Province.

	Minimum	Maximum	M	SD
Homicide rate per 100,000 population
Newfoundland	0.00	2.14	0.90	0.53
Prince Edward Island	0.00	2.20	0.71	0.61
Nova Scotia	0.85	2.58	1.71	0.64
New Brunswick	0.67	2.75	1.48	0.54
Quebec	1.32	3.10	2.05	0.46
Ontario	1.34	2.35	1.69	0.29
Manitoba	2.28	4.31	3.27	0.69
Saskatchewan	1.28	4.34	2.83	0.76
Alberta	1.84	3.49	2.50	0.49
British Columbia	2.06	3.79	2.84	0.49
Homicide rate per 100,000 young males
Newfoundland	0.00	21.77	7.55	4.90
Prince Edward Island	0.00	21.72	6.56	5.84
Nova Scotia	7.92	25.43	15.53	5.22
New Brunswick	6.23	20.55	12.72	3.78
Quebec	13.07	25.32	18.51	3.08
Ontario	10.93	19.52	15.06	2.03
Manitoba	21.48	40.00	28.78	5.72
Saskatchewan	11.85	39.03	25.28	7.10
Alberta	15.88	29.38	20.68	3.69
British Columbia	20.10	33.99	26.09	4.40
Homicide location quotient
Newfoundland	0.00	1.32	0.64	0.35
Prince Edward Island	0.00	1.38	0.41	0.36
Nova Scotia	0.45	1.39	0.86	0.26
New Brunswick	0.43	1.49	0.88	0.29
Quebec	0.82	1.63	1.24	0.19
Ontario	0.72	1.14	0.92	0.11
Manitoba	0.90	1.48	1.19	0.19
Saskatchewan	0.46	1.33	0.96	0.21
Alberta	0.75	1.24	1.02	0.11
British Columbia	0.71	1.09	0.90	0.11

The homicide LQ descriptive statistics point to an interesting fact. In Canada, the provinces exhibit an increasing trend in homicide rates when moving east to west, but there is no such trend using the homicide LQ. Therefore, though the levels of the homicide rate are greater in western Canada than eastern Canada, as a proportion of all crime, there is very little variation across the country—aside from Quebec and Manitoba, there is no significant specialization in homicide.

Turning to the independent variables, the descriptive statistics and bivariate correlations for the entire data set are reported in Tables 2 and 3, respectively—detailed tables at the provincial level are available on request. With regard to the bivariate correlations, there are no surprising relationships to report. None of the control variables have any correlation coefficients that are high enough to cause any concern, but there are two bivariate correlations between migration variables that are rather high, 0.99: immigrant - net immigrants and immigrants, young males - net immigrants, young males. Most often, based on such high correlation coefficients or their corresponding variance inflation factors, two of these variables would be removed from the regression analysis. This removal is due to the understanding that multicollinearity leads to bias. This is not the case. In fact, the removal of relevant variables because of multicollinearity likely leads to omitted variable bias. In the presence of multicollinearity, least squares estimation is unbiased; in fact, least squares estimation is still the best unbiased linear estimator. The only issue that arises is that the variances of the estimated parameters become inflated, potentially leading to statistically insignificant results (P. E. Kennedy, 2008). As shown in the following, there is enough independent variation within immigrants and net immigrants for both to be statistically significant in one of the models. As such, none of the independent variables are excluded from the regression analysis, a priori, to avoid omitted variable bias. One rule of thumb that has emerged in the econometrics literature is as follows: “Don’t worry about multicollinearity if the t statistics are all greater than 2” (P. E. Kennedy, 2008, p. 196). In other words, if the variables of concern have statistically significant results, there is no concern in the analysis.

Table 2.

Descriptive Statistics, Independent Variables.

	Minimum	Maximum	M	SD
Immigrants (%)	0.28	6.13	1.99	1.63
Immigrants, young males (%)	0.04	0.54	0.17	0.10
Net immigrants (%)	0.27	5.91	1.90	1.56
Net immigrants, young males (%)	0.02	0.28	0.09	0.05
Interprovincial (%)	1.40	12.15	7.41	2.90
Interprovincial, young males (%)	0.01	3.00	0.86	0.82
Net interprovincial (%)	−1.28	0.62	−0.19	0.39
Net interprovincial, young males (%)	0.16	2.59	1.21	0.60
Young males (%)	9.8	14.7	11.4	1.2
Gini coefficient	34.7	44.0	39.9	1.9
GDP per capita	17,977	62,152	29,925	7,093
Low income (%)	5.1	18.9	11.8	2.7
Unemployment rate	3.9	20.1	10.1	3.7
Divorce rate	119.2	405.4	237.3	51.9
Corrections expenditures per capita	22.7	69.1	39.7	10.6
Police expenditures per capita	80.8	235.7	155.0	34.6
Officers per capita	136.8	209.2	174.9	18.0
Incidents per officer	31.0	84.0	51.0	13.4

Table 3.

Bivariate Correlations, Independent Variables

	X1	X2	X3	X4	X5	X6	X7	X8	X9	X10	X11	X12	X13	X14	X15	X16	X17	X18
X1	1	0.55***	0.99***	0.56***	−0.35***	0.20***	0.56***	−0.19***	−0.21***	0.56***	0.51***	0.32***	−0.47***	0.45***	0.18**	0.59***	0.38***	0.28***
X2		1	0.54***	0.99***	0.06	0.32***	0.55***	0.06	0.36***	0.09	0.42***	0.26***	−0.42***	0.80***	0.18**	0.29***	0.30***	0.40***
X3			1	0.55***	−0.35***	0.20***	0.56***	−0.19***	−0.22***	0.56***	0.51***	0.32***	−0.47***	0.44***	0.18**	0.59***	0.37***	0.28***
X4				1	0.06	0.32***	0.55***	0.06	0.35***	0.09	0.42***	0.26***	−0.42***	0.80***	0.18**	0.29***	0.30***	0.40***
X5					1	0.63***	0.07	0.90***	0.23***	−0.31***	−0.04	−0.35***	0.12	0.03	0.08	−0.73***	−0.64***	0.50***
X6						1	0.65***	0.84***	0.17**	−0.08	0.24***	−0.22***	−0.18***	0.38***	−0.20***	−0.24***	−0.29***	0.44***
X7							1	0.25***	−0.16**	0.36***	0.59***	−0.02	−0.55***	0.63***	−0.15**	0.40***	0.21***	0.32***
X8								1	0.21***	−0.29***	−0.04	−0.38***	0.13	0.10	−0.12	−0.66***	−0.62***	0.41***
X9									1	−0.63***	−0.24***	0.01	0.23***	0.35***	−0.02	−0.26***	−0.07	−0.04
X10										1	0.64***	0.34***	−0.51***	0.03	0.09	0.55***	0.26***	0.17**
X11											1	−0.04	−0.71***	0.33***	0.20***	0.58***	0.21***	0.30***
X12												1	0.03	0.19***	−0.13	0.25***	0.32***	0.07
X13													1	−0.41***	−0.38***	−0.63***	−0.58***	−0.42***
X14														1	−0.01	0.30***	0.38***	0.37***
X15															1	0.16**	0.24***	0.50***
X16																1	0.79***	−0.09
X17																	1	0.05
X18																		1

Note: X1 = immigrants (%); X2 = immigrants, young males (%); X3 = net immigrants (%); X4 = net immigrants, young males (%); X5 = interprovincial (%); X6 = interprovincial, young males (%); X7 = net interprovincial (%); X8 = net interprovincial, young males (%); X9 = young males (%); X10 = Gini coefficient; X11 = GDP per capita; X12 = low income (%); X13 = unemployment rate; X14 = divorce rate; X15 = corrections expenditures per capita; X16 = police expenditures per capita; X17 = officers per capita; X18 = incidents per officer.

Correlation is statistically significant at the 5% level. ***Correlation is statistically significant at the 1% level.

Of particular importance for the subsequent analysis is the provincial variation of immigration growth and the percentage of the population that is composed of immigrants. With regard to growth, the greatest growth in immigrant populations is in Quebec, Ontario, and British Columbia. In addition, the eastern provinces have had just as much immigrant population growth as the other western provinces. The percentage of the population that is composed of recent immigrants shows a similar pattern, aside from the eastern provinces being significantly lower than the other provinces. This is of particular importance because the highest migrant populations are not simply in the provinces that have the highest homicide rates. If this was the case, even when using fixed effects estimation, the inferential results may lead to spurious results.

Inferential Results

Before the various results for the migration variables are covered, the results for the control variables are discussed. For all three homicide measurements, the control variables always have the same sign for the estimated parameters (positive), but variable retention does vary for each of the final models. Young males, the Gini coefficient, and low income have the expected relationships with homicide, but the remaining variables are expected to have negative relationships with homicide based on more traditional theories relating immigration and crime. For example, increases in average income (GDP per capita) are expected to decrease homicide. However, this result is easily understood in the context of routine activity theory. As discussed by L. E. Cohen and Felson (1979), increases in income lead to increases in criminal opportunity because people are more frequently away from the relatively protective environment of the home and purchase more goods for subsequent theft. This is also consistent with the economics of crime literature (Ehrlich, 1975, 1977). The interpretation is similar for the criminal justice variables representing corrections expenditures per capita, police expenditures per capita, officers per capita, and criminal incidents per officer: The provinces that have the greatest increases in homicide, or crime in general, have had the greatest increases in criminal justice resources, based on need. In other words, police and corrections do not cause crime.

The results for the traditional homicide rate are presented in Table 4. When only one migration variable is included in the statistical model at a time, only the percentages of immigrants and net immigrants remain statistically significant after the general-to-specific statistical testing. Important to note is that the estimated parameters for both these migration variables are positive. However, the magnitudes of these estimated parameters (0.15 and 0.16) are rather low, implying that although there is a positive relationship between immigration and homicide, that relationship is not particularly strong: A 1% increase in immigrants or net immigrants increases the traditional homicide rate by 0.15 or 0.16, respectively.

Table 4.

Regression Results, Homicide Rate per 100,000 Population.

	Initial migration variable(s)
	Immigrants (%)	Net immigrants (%)	All
Immigrants (%)	0.15***
Net immigrants (%)		0.16***	0.20***
Interprovincial, young males (%)			−0.69***
Net interprovincial, young males (%)			1.05***
Young males (%)	0.15*	0.15*	0.27***
Gini coefficient	0.08*	0.08*
GDP per capita	0.00***	0.00***	0.00***
Low income (%)	0.06**	0.06**	0.09***
Corrections expenditures per capita	0.01*	0.01*
Officers per capita	0.02***	0.02***	0.03***
Incidents per officer	0.04**	0.04***	0.04***
Trend	−0.05***	−0.05**
Adjusted R²	0.78	0.78	0.78

Note: All statistical significance are based on White’s heteroskedastic-consistent variance–covariance matrices.

Indicates 10% statistical significance. **Indicates 5% statistical significance. ***Indicates 1% statistical significance.

When all of the migration variables are initially included in the statistical specification, however, the results change. Net immigrants is still statistically significant and positive (0.20), but immigrants is no longer statistically significant. More importantly, two of the interprovincial migration variables are statistically significant: interprovincial young males and net interprovincial young males. Curiously, these variables have opposite signs for their estimated parameters. A 1% increase in interprovincial young males leads to a decrease in the traditional homicide rate by −0.69, whereas a 1% increase in net interprovincial young males leads to an increase in the traditional homicide rate by 1.05. The interpretation of these results is as follows: Provinces that have increases in the in-migration of young males from other provinces have decreases in the traditional homicide rate, but provinces that have a net increase in young males from interprovincial migration have increases in traditional homicide rates. Therefore, it is not immigration per se, but increases in the proportion of the most criminogenic subpopulation that matter, young males. In addition, the magnitudes of the estimated parameters for the interprovincial migration variables are 4 and 5 times the magnitude of the net immigration estimated parameters, indicating that interprovincial migration is more important than international immigration.

Table 5 shows the statistical results for the adjusted homicide rate. When only one migration variable is included in the statistical model at a time, not only are the percentages of immigrants and net immigrants statistically significant after the general-to-specific statistical testing, but so are the percentages of young male immigrants and young male net immigrants. The estimated parameters for the adjusted homicide rate are significantly greater in magnitude than the traditional homicide rate results. This is simply because of the different normalizing variable used in the adjusted homicide rate calculation. As such, the sign and relative magnitudes of the estimated parameters are important here.

Table 5.

Regression Results, Homicide Rate per 100,000 Young Males.

	Initial migration variable(s)
	Immigrants (%)	Immigrants, young males (%)	Net immigrants (%)	Net immigrants, young males (%)	All
Immigrants (%)	1.38***				−26.97*
Immigrants, young males (%)		−15.11**
Net immigrants (%)			1.47***		29.61*
Net immigrants, young males (%)				−28.18**
Interprovincial, young males (%)					−6.31***
Net interprovincial, young males (%)					10.84***
Young males (%)					1.31***
Gini coefficient	0.72*		0.72*
GDP per capita	0.00***	0.00***	0.00***	0.00***	0.00**
Low income (%)	0.41*	0.92***	0.41*	0.93***	0.80***
Corrections expenditures per capita		0.20***		0.20***
Police expenditures per capita	0.09*		0.09*		0.08*
Officers per capita	0.13**	0.18***	0.13**	0.18***	0.21***
Incidents per officer	0.36***	0.30***	0.37***	0.30***	0.34***
Trend	−0.45***	−0.26***	−0.45***	−0.26**
Adjusted R²	0.77	0.77	0.77	0.77	0.78

Note: All statistical significance are based on White’s heteroskedastic-consistent variance–covariance matrices.

Indicates 10% statistical significance. **Indicates 5% statistical significance. ***Indicates 1% statistical significance.

Interestingly, young male immigrants and young male net immigrants decrease the homicide rate by large magnitudes: A 1% increase in young male immigrants and young male net immigrants lead to decreases in the adjusted homicide rate of −15.11 and −28.18, respectively. Clearly this is a counterintuitive result. Given that the young males variable is not statistically significant in this specification (removed through statistical testing), there is likely statistical bias present.

As with the traditional homicide rate, when all of the migration variables are initially included in the statistical specification, the results change. Young male immigrants and young male net immigrants are no longer statistically significant, but interprovincial young males and net interprovincial young males are statistically significant. Contrary to the results for the traditional homicide rate, the impacts of international immigrants are of greater magnitude than the interprovincial migrants. There is a consistency within the current model’s result: The estimated parameters for immigrants and interprovincial young males are negative, whereas the estimated parameters for net immigrants and net interprovincial young males are positive—net changes lead to increases in the adjusted homicide rate.

As stated previously, these latter results are taken to be more representative of the macro-level relationship between immigration and homicide because the normalizing variable is considered a better population at risk than the traditional homicide rate. In addition, the interpretation of the estimated parameters is in line with expectations. If there are more people entering a province, young males or not, and the count of homicides remains relatively constant, the homicide rate is expected to fall—the normalizing variable in the homicide rate (denominator) increases, whereas the counts of homicide (numerator) does not change. However, it is the net change (increase) in the most criminogenic subpopulation (young males) that leads to increases in the homicide rate.

The discussion now turns to the homicide LQ (Table 6). Unlike the traditional and adjusted homicide rates, when the migration variables are entered individually using the homicide LQ, none of the variables remain statistically significant. When all of the migration variables are initially entered, however, a number of the migration variables remain statistically significant.

Table 6.

Regression Results, Homicide Rate Location Quotient.

	Initial migration variable(s)
	All
Immigrants (%)	−2.13***
Net immigrants (%)	2.28***
Interprovincial, young males (%)	−0.25**
Net interprovincial (%)	0.29*
Net interprovincial, young males (%)	0.32*
GDP per capita	0.00**
Low income (%)	0.03**
Police expenditures per capita	0.01***
Trend	−0.03***
Adjusted R²	0.52

Note: All statistical significance are based on White’s heteroskedastic-consistent variance–covariance matrices.

Indicates 10% statistical significance. **Indicates 5% statistical significance. ***Indicates 1% statistical significance.

An interesting result with the homicide LQ is the consistency with the adjusted homicide rate. The percentages of immigrants and interprovincial young males have negative estimated parameters, and the percentages of net immigrants, net interprovincial migrants, and net interprovincial young males have positive estimated parameters. In addition, the magnitudes of the estimated parameters are greater for the international immigration variables than for the interprovincial migration variables, indicating that international immigration is more important than interprovincial migration. The general interpretation of this result is that increases in migrants lead to decreased specialization in homicide, but net increases in migrants lead to increased specialization in homicide. Therefore, if increased in-migration decreases homicide specialization and increased net migration increases homicide migration, it is not the migrating populations that affect homicide specialization, but the net change in the most criminogenic subpopulation, young males, and the population, more generally, as outlined by research on violence in frontier America (Courtwright, 1996b).

Discussion and Conclusions

The relationship between immigration and crime is a contentious one that has occupied criminological research for nearly a century. Much, but not all, of the early research states that new immigrant populations increase crime through a number of different modes. However, the more contemporary research, concentrating on homicide, calls such conclusions into question. Rather, it appears as though new immigrant populations have either a statistically insignificant or negative relationship with homicide.

This article contributes to the more recent empirical research on the relationship between immigration and crime through an aggregate analysis of the relationship between international immigration, interprovincial migration, and homicide in Canadian provinces using multiple measures of homicide and a panel data set. The general research finding is that the relationship between immigration and homicide is complex and there is no direct evidence of immigrant populations causing increases in homicide. Rather, increases of in-migration (both international and interprovincial) lead to a decrease in homicide; it is the net increase in young males that generally leads to an increase in homicide—routine activity theory. This result echoes the seminal research of David T. Courtwright (1996a, 1996b) on the positive relationship between young males and homicide. Therefore, the results presented here are consistent with those of Ramiro Martinez, Jr. and colleagues, albeit at a macro level of analysis.

These conclusions are based on the statistical results when all migration variables are initially included in the statistical specification. It is clear that when only one migration variable is included in the analysis, the results suffer from statistical bias. This is similar to the statements made by Nielsen et al. (2005) and Stowell and Martinez (2007) that recognize different ethnic immigrant populations have different relationships with violence. Similarly, different forms of migration have different relationships with homicide. Consequently, all necessary migration variables must be included in any analysis. In addition, it appears as though the traditional homicide rate may be problematic. The results for the traditional homicide rate are significantly different from the adjusted homicide rate and homicide LQ. This is likely because of improper measurement. As such, the measure chosen to represent homicide (or any other crime for that matter) is critical for an appropriate analysis to be undertaken.

Despite being consistent with recent research on the relationship between immigration and homicide, there are two primary limitations within the current study. First, the unit of analysis is not equivalent to those used in this recent research—this is discussed further under directions for future research. Second, though this study has incorporated both cross-sectional and temporal data and the corresponding panel data methods, identifying any (lack of) causal relationships is difficult. As pointed out by Butcher and Piehl (1998), much of the research and media attention on the relationship between immigration and crime does not differentiate between crimes actually committed by immigrants versus native-born populations. The current study suffers from this issue and there is a clear need for a micro-level longitudinal analysis.

Some of the directions for future research are specific to Canada, but there are also directions to be undertaken in a U.S. context. With regard to the United States, macro-level analyses have been undertaken that have results consistent with the work of Ramiro Martinez, Jr. and colleagues (Reid et al., 2005). However, because of the results presented here, macro-level analyses using U.S. states as the unit of analysis are in order. In addition, if possible, based on data availability, it is important for future analyses to incorporate different measures of migration, both international and internal. In the Canadian context, the immigration–homicide relationship must be analyzed at other scales. Studies in the United States have used metropolitan areas and neighborhoods (census tracts) as the unit of analysis, and there is a need for equivalent analyses to be undertaken in a Canadian context to confirm or deny the generalizability of the U.S. results.

Footnotes

Acknowledgements

The author would like to thank three anonymous reviewers whose comments significantly improved the quality of this manuscript.

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

References

Alba

R. D.

Nee

(1997). Rethinking assimilation theory for a new era of immigration. International Migration Review, 31, 826-874.

Andresen

M. A.

(2006a). Crime measures and the spatial analysis of criminal activity. British Journal of Criminology, 46, 258-285.

Andresen

M. A.

(2006b). A spatial analysis of crime in Vancouver, British Columbia: A synthesis of social disorganization and routine activity theory. Canadian Geographer, 50, 487-502.

Andresen

M. A.

(2007). Location quotients, ambient populations, and the spatial analysis of crime in Vancouver, Canada. Environment and Planning A, 39, 2423-2444.

Andresen

M. A.

(2009). Crime specialization across the Canadian provinces. Canadian Journal of Criminology and Criminal Justice, 51, 31-53.

Andresen

M. A.

Jenion

G. W.

Jenion

M. L.

(2003). Conventional calculations of homicide rates lead to an inaccurate reflection of Canadian trends. Canadian Journal of Criminology and Criminal Justice, 45, 1-17.

Avio

K. L.

(1979). Capital punishment in Canada: A time-series analysis of the deterrent hypothesis. Canadian Journal of Economics, 12, 647-676.

Baltagi

B. H.

(2001). Econometric analysis of panel data (2nd ed.). Chichester, UK: John Wiley.

Blau

J. R.

Blau

P. M.

(1982). The cost of inequality: Metropolitan structure and violent crime. American Sociological Review, 47, 114-129.

10.

Boggs

S. L.

(1965). Urban crime patterns. American Sociological Review, 30, 899-908.

11.

Borjas

(1999). Heaven’s door: Immigration policy and the American economy. Princeton, NJ: Princeton University Press.

12.

Boyd

(2000). The beast within: Why men are violent. Vancouver, BC: Greystone Books.

13.

Brantingham

P. L.

Brantingham

P. J.

(1993). Location quotients and crime hot spots in the city. In Block

C. R.

Dabdoub

(Eds.), Workshop on crime analysis through computer mapping, Proceedings (pp. 175-197). Chicago, IL: Criminal Justice Information Authority.

14.

Brantingham

P. L.

Brantingham

P. J.

(1995). Location quotients and crime hot spots in the city. In Block

C. R.

Dabdoub

Fregly

(Eds.), Crime analysis through computer mapping (pp. 129-149). Washington, DC: Police Executive Research Forum.

15.

Brantingham

P. L.

Brantingham

P. J.

(1998). Mapping crime for analytic purposes: Location quotients, counts and rates. In Weisburd

McEwen

(Eds.), Crime mapping and crime prevention (pp. 263-288). Monsey, NY: Criminal Justice Press.

16.

Bursik

R. J.

(1988). Social disorganization and theories of crime and delinquency: Problems and prospects. Criminology, 26, 519-551.

17.

Butcher

K. F.

Piehl

A. M.

(1998). Cross-city evidence on the relationship between immigration and crime. Journal of Policy Analysis and Management, 17, 457-493.

18.

Chainey

Ratcliffe

(2005). GIS and crime mapping. Chichester, UK: John Wiley.

19.

Cohen

A. K.

(1955). Delinquent boys: The culture of the gang. Glencoe, IL: Free Press.

20.

Cohen

(1931). Report on crime and the foreign born: Comment. Michigan Law Review, 30, 99-104.

21.

Cohen

L. E.

Felson

(1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44, 588-608.

22.

Courtwright

D. T.

(1996a). Violence in America. American Heritage, 47, 36-47.

23.

Courtwright

D. T.

(1996b). Violent land: Single men and social disorder from the frontier to the inner city. Cambridge, MA: Harvard University Press.

24.

Croxton

F. C.

(1911). Statistical review of immigration, 1820-1910: Distribution of immigrants, 1850-1900. Washington, DC: United States Government Printing Office.

25.

Daly

Wilson

Vasdev

(2001). Income inequality and homicide rates in Canada and the United States. Canadian Journal of Criminology, 43, 219-236.

26.

Ehrlich

(1975). The deterrent effect of capital punishment: A question of life or death. American Economic Review, 65, 397-417.

27.

Ehrlich

(1977). Capital punishment and deterrence: Some further thoughts and additional evidence. Journal of Political Economy, 85, 741-788.

28.

Giffen

P. J.

(1965). Rates of crime and delinquency. In McGrath

W. T.

(Ed.), Crime and its treatment in Canada (pp. 59-90). Toronto, Ontario, Canada: Macmillan.

29.

Giffen

P. J.

(1976). Official rates of crime and delinquency. In McGrath

W. T.

(Ed.), Crime and its treatment in Canada (2nd ed., pp. 66-110). Toronto, Ontario, Canada: Macmillan.

30.

Godfrey

L. G.

(1978). Testing against general autoregressive and moving average error models when the regressors include lagged dependent variables. Econometrica, 46, 1293-1302.

31.

Greene

W. H.

(2000). Econometric analysis (4th ed.). Upper Saddle River, NJ: Prentice Hall.

32.

Grogger

J. T.

(1998). Immigration and crime among young black men: Evidence from the national longitudinal survey of youth. In Hamermesh

D. S.

Bean

F. D.

(Eds.), Help or hindrance? The economic implications of immigration for African Americans (pp. 322-341). New York, NY: Russell Sage.

33.

Hagan

Palloni

(1998). Immigration and crime in the United States. In Smith

J. P.

Edmonston

(Eds.), The immigration debate (pp. 307-387). Washington, DC: National Academy Press.

34.

Hagan

Palloni

(1999). Sociological criminology and the mythology of Hispanic immigration and crime. Social Problems, 46, 617-632.

35.

Harries

K. D.

(1991). Alternative denominators in conventional crime rates. In Brantingham

P. J.

Brantingham

P. L.

(Eds.), Environmental criminology (pp. 147-165). Prospect Heights, IL: Waveland Press.

36.

Hartnagel

T. F.

(1978). The effect of age and sex composition of provincial populations on provincial crime rates. Canadian Journal of Criminology, 20, 28-33.

37.

Hartnagel

T. F.

(1997). Crime among the provinces: The effect of geographic mobility. Canadian Journal of Criminology, 39, 387-402.

38.

Hausman

J. A.

(1978). Specification tests in econometrics. Econometrica, 46, 1251-1271.

39.

Hirschfield

Bowers

K. J.

(1997). The development of a social, demographic and land use profiler for areas of high crime. British Journal of Criminology, 37, 103-120.

40.

Hirschi

(1969). Causes of delinquency. Berkeley: University of California Press.

41.

Hirschi

Gottfredson

(1983). Age and the explanation of crime. American Journal of Sociology, 89, 552-584.

42.

Hsiao

(2003). Analysis of panel data (2nd ed.). Cambridge, UK: Cambridge University Press.

43.

Isard

Azis

I. J.

Drennan

M. P.

Miller

R. E.

Saltzman

Thorbecke

(1998). Methods of interregional and regional analysis. Aldershot, UK: Ashgate Publishing Limited.

44.

Kao

Tienda

(1995). Optimism and achievement: The educational performance of immigrant youth. Social Science Quarterly, 76, 1-19.

45.

Katz

Levitt

S. D.

Shustorovich

(2003). Prison conditions, capital punishment, and deterrence. American Law and Economic Review, 5, 318-343.

46.

Kennedy

L. W.

Forde

D. R.

(1990). Routine activities and crime: An analysis of victimization in Canada. Criminology, 28, 137-152.

47.

Kennedy

L. W.

Silverman

R. A.

Forde

D. R.

(1991). Homicide in urban Canada. Canadian Journal of Sociology, 16, 397-410.

48.

Kennedy

P. E.

(2008). A guide to econometrics (6th ed.). Malden, MA: Blackwell.

49.

Lauritsen

J. L.

(2001). The social ecology of violent victimization: Individual and contextual effects in the NCVS. Journal of Quantitative Criminology, 17, 3-32.

50.

Lazear

(1999). Culture and language. Journal of Political Economy, 107, S95-S126.

51.

Lee

M. T.

Martinez

Jr. Rosenfeld

(2001). Does immigration increase homicide? Negative evidence from three border cities. Sociological Quarterly, 42, 559-580.

52.

Levitt

S. D.

(2001). Alternative strategies for identifying the link between unemployment and crime. Journal of Quantitative Criminology, 17, 377-390.

53.

Ley

(1999). Myths and meanings of immigration and the metropolis. Canadian Geographer, 43, 2-19.

54.

Ley

Smith

(2000). Relations between deprivation and immigrant groups in large Canadian cities. Urban Studies, 37, 37-62.

55.

(2007). Homicide in Canada, 2006. Ottawa, Ontario: Statistics Canada, Canadian Centre for Justice Statistics.

56.

Martinez

Jr. (2000). Immigration and urban violence: The link between immigrant Latinos and types of homicide. Social Science Quarterly, 81, 363-374.

57.

Martinez

Jr. Lee

M. T.

(2000). Comparing the context of immigrant homicides in Miami: Haitians, Jamaicans, and Mariels. International Migration Review, 34, 794-812.

58.

Martinez

Jr. Stowell

J. I.

Cancino

J. M.

(2008). A tale of two border cities: Community context, ethnicity, and homicide. Social Science Quarterly, 89, 1-16.

59.

McCord

E. S.

Ratcliffe

J. H.

(2007). A micro-spatial analysis of the demographic and criminogenic environment of drug markets in Philadelphia. Australian & New Zealand Journal of Criminology, 40, 43-63.

60.

Mears

D. P.

(2001). The immigration-crime nexus: Toward an analytic framework for assessing and guiding theory, research, and policy. Sociological Perspectives, 44, 1-19.

61.

Merton

R. K.

(1938). Social structure and anomie. American Sociological Review, 3, 572-682.

62.

Messner

S. F.

Rosenfeld

(1994). Crime and the American dream. Belmont, CA: Wadsworth.

63.

Messner

S. F.

Rosenfeld

(2001). Crime and the American dream (3rd ed.). Belmont, CA: Wadsworth.

64.

Miller

M. M.

Gibson

L. J.

Wright

N. G.

(1991). Location quotient: A basic tool for economic development studies. Economic Development Review, 9, 65-68.

65.

Moulton

B. R.

(1986). Random group effects and the precision of regression estimates. Journal of Econometrics, 32, 385-397.

66.

Moulton

B. R.

(1987). Diagnostics for group effects in regression analysis. Journal of Business and Economic Statistics, 5, 275-282.

67.

Nagin

D. S.

Paternoster

(1991). On the relationship of past to future participation in delinquency. Criminology, 29, 163-289.

68.

National Commission on Law Observance and Enforcement. (1931). Report on the enforcement of the Prohibition Laws in the United States. Washington, DC: United States Government Printing Office.

69.

Nielsen

A. L.

Lee

M. T.

Martinez

Jr. (2005). Integrating race, place, and motive in social disorganization theory: Lessons from a comparison of Black and Latino homicide types in two immigrant destination cities. Criminology, 43, 837-872.

70.

Ousey

G. C.

Kubrin

C. E.

(2009). Exploring the connection between immigration and violent crime rates in U.S. cities, 1980-2000. Social Problems, 56, 447-473.

71.

Portes

(1987). The social origins of the Cuban enclave economy of Miami. Sociological Perspectives, 30, 340-372.

72.

Quetelet

L. A. J.

(1842). A treatise on man and the development of his faculties. Edinburgh, Scotland: W. & R. Chambers.

73.

Ratcliffe

J. H.

Rengert

G. F.

(2008). Near repeat patterns in Philadelphia shootings. Security Journal, 21, 58-76.

74.

Reid

L. W.

Weiss

H. A.

Adelman

R. M.

Jaret

(2005). The immigration-crime relationship: Evidence across US metropolitan areas. Social Science Research, 34, 757-780.

75.

Rengert

G. F.

(1996). The geography of illegal drugs. Boulder, CO: Westview Press.

76.

Robinson

J. B.

(2008). Crime and regeneration in urban communities: The case of the big dig in Boston, Massachusetts. Built Environment, 34, 46-61.

77.

Sampson

R. J.

Raudenbush

S. W.

(1999). Systematic social observation of public spaces: A new look at disorder in urban neighborhoods. American Journal of Sociology, 105, 603-651.

78.

Sampson

R. J.

Raudenbush

S. W.

Earl

(1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277, 918-924.

79.

Shaw

C. R.

McKay

H. D.

(1931). Social factors in juvenile delinquency [National Commission on Law Observance and Enforcement, Report on the Causes of Crime, Volume II.] Washington, DC: U.S. Government Printing Office.

80.

Shaw

C. R.

McKay

H. D.

(1942). Juvenile delinquency and urban areas: A study of rates of delinquency in relation to differential characteristics of local communities in American cities. Chicago, IL: University of Chicago Press.

81.

Silver

(2007). Crime statistics in Canada, 2006. Ottawa, ON: Statistics Canada, Canadian Centre for Justice Statistics.

82.

Stark

(1991). The migration of labor. Oxford, UK: Blackwell.

83.

Stowell

J. I.

Martinez

Jr. (2007). Displaced, dispossessed, or lawless? Examining the link between ethnicity, immigration, and violence. Aggression and Violent Behavior, 12, 564-581.

84.

Stowell

J. I.

Messner

S. F.

McGeever

K. F.

Raffalovich

L. E.

(2009). Immigration and the recent violent crime drop in the United States: A pooled, cross-sectional time-series analysis of metropolitan areas. Criminology, 47, 889-928.

85.

Wells

L. E.

Rankin

J. H.

(1991). Families and delinquency: A meta-analysis of the impact of broken homes. Social Problems, 38, 71-93.

86.

Wooldridge

J. M.

(2002). Econometric analysis of cross section and panel data. Cambridge, MA: MIT Press.

87.

Worrall

J. L.

Pratt

T. C.

(2004). On the consequences of ignoring unobserved heterogeneity when estimating macro-level models of crime. Social Science Research, 33, 79-105.

88.

Zhang

Sanders

(1999). Extended stratification: Immigrant and native differences in individual and family labor. Sociological Quarterly, 40, 681-704.