Relationship Between Immigration And Crime

Cross-City Evidence on the Relationship between Immigration and Crime / 457

Cross-City Evidence on the Relationship between Immigration and Crime

Kristin F. Butcher
Anne Morrison Piehl

Abstract
Public concerns about the costs of immigration and crime are high, and sometimes overlapping. This article investigates the relationship between immigration into a metropolitan area and that area’s crime rate during the 1980s.
Using data from the Uniform Crime Reports and the Current Population Surveys, we find, in the cross section, that cities with high crime rates tend to have large numbers of immigrants. However, controlling for the demographic characteristics of the cities, recent immigrants appear to have no effect on crime rates.
In explaining changes in a city’s crime rate over time, the flow of immigrants again has no effect, whether or not we control for other city-level characteristics. In a secondary analysis of individual data from the National
Longitudinal Survey of Youth (NLSY), we find that youth born abroad are statistically significantly less likely than native-born youth to be criminally active. Immigration and crime are not only two of today’s “hot-button” issues in politics, they seem to occupy the same region of the public mind. The constant juxtaposition of the words “immigration” and “crime” in news story after news story might forge the perception of a causal link between the two issues.
Juxtaposition, however, is not the only way in which the two are linked by those in the policy arena.
A direct relationship is assumed to exist between illegal immigration and crime.1 An explicit example of this is in Proposition 187 in California. This proposition has received most attention for the connections it asserts between illegal immigration and the rising costs of public expenditures on welfare and schools. However, the proposition also highlights victimization. Section 1 of the law reads, “The People of California find and declare as follows: . . . That they have suffered and are suffering personal injury and damage caused by the criminal conduct of illegal aliens in this state.”
Although illegal immigrants are typically cited as contributing to crime, there is also a great deal of legislative attention aimed at “criminal aliens” regardless

Journal of Policy Analysis and Management, Vol. 17, No. 3, 457–493 (1998)
© 1998 by the Association for Public Policy Analysis and Management
Published by John Wiley & Sons, Inc.
CCC 0276-8739/98/03457-37

458 / Cross-City Evidence on the Relationship between Immigration and Crime of their immigration status [see Butcher and Piehl, 1998]. The 104th Congress included a large number of bills, amendments, and resolutions that took aim at this issue.2 For example in 1996, then Senator Dole sponsored the “AntiTerrorism and Effective Death Penalty Act.”3 This act is primarily a crime prevention measure, yet a large section of it is devoted to immigration issues.
Title IV of this act is entitled “Terrorist and Criminal Alien Removal and
Exclusion.” Subtitle D (“Criminal Alien Procedural Improvements”) includes myriad amendments to the Immigration and Nationality Act (currently under debate), the Violent Crime Control and Law Enforcement Act of 1994, and the
U.S. Code. The proposals are aimed at streamlining the deportation of criminal aliens and improving the working relationship among the various divisions of government that oversee immigration and crime control in order to help identify and punish criminal aliens. The 1994 crime law itself covered many of these same topics.
Politicians also frequently draw a causal relationship between immigration and crime in public statements about the policies they propose. For example, in a 1996 editorial about the immigration bills then being debated in the two houses of Congress, some versions of which sought to reduce the numbers of legal immigrants admitted, and all versions of which promised to “crack down” on illegal immigrants, Senator Alan Simpson concluded with the following:
“Finally, this legislation is not unneeded. Some have made the claim that more immigration—both legal and illegal—is what this country needs. Anyone who believes that has not been listening to taxpayers who are being adversely affected—for example, by welfare abuse, schools that are overcrowded and beset by demands for ‘multicultural’ curricula, rising crime and expensive, timeconsuming deportation procedures. Both the Senate and House bills tackle the problem in a rational, sensible and fair way, one that advances the interests of most of our citizens—and thus our national interest” [Simpson, 1996].
During his bid for the 1996 Republican presidential nomination, one of Pat
Buchanan’s television advertisements stated: “Each year millions of illegal immigrants pour across our southern border into the United States. Most come without job skills. Crime explodes. And who pays the cost of their health care, housing, welfare? You do. . . . [I will] declare a time-out on new immigration.
Secure America’s borders. And insist on one language, English, for all Americans”
[Edsall and Claiborne, 1996].
Whether implicitly or explicitly stated, the obvious message of the political rhetoric and the proposed legislation is that through laws specifically targeting immigrants, crime will be reduced. The public seems to believe this, as evidenced by a 1993 Time magazine poll showing that 59 percent of respondents believe recent immigrants “add to the crime problem” [Nelan, 1993, p. 11]. Although this has to be true in some trivial sense—if there are fewer immigrants (or a
1

The distinction between legal and illegal immigrants is often blurred in the reporting of immigration issues. The public believes that most immigrants come to the United States illegally [Nelan,
1995, p. 10]. In fact, legal immigrants far outnumber illegal. Estimates of the fraction foreign born who are in the United States illegally range from 15 percent to 24 percent [see Borjas, 1990, pp. 63–
75; Nelan, 1993, p. 10].
2
Legislative interest in this issue shows few signs of abating. The 105th Congress considered at least 50 bills that touched on crime and immigration.
3
PL 104–132, 104th Cong., id sess., 24 April 1996. The bill was cosponsored by eight other senators from both sides of the aisle: Hatch, Nickles, Thurmond, Simpson, Brown, Kyl, Gramm, and
Feinstein.

Cross-City Evidence on the Relationship between Immigration and Crime / 459

smaller population in general) the number of crimes is likely to be lower; if a criminal alien is deported, she or he will not be in the United States to commit another crime—this article is an investigation of whether this is true in any deeper sense. Is there any hard evidence that the large inflow of immigrants during the 1980s has adversely affected crime rates in the United States? Is there any evidence that immigrants are more likely to commit crimes than the native born? This article is not meant to be an explicit investigation of the efficacy of any particular policy proposition. Rather, it investigates whether targeting immigrants is likely to be a powerful policy lever through which society can materially alter crime.
Because immigration is geographically concentrated, the public debate about immigration is often framed in terms of expenditures, that is, who bears the costs of public services for immigrants.4 The same is true about the criminal justice impact of immigration. The public is concerned not only with criminal victimization, but also about the costs associated with the enforcement and punishment of crimes.
If immigrants have an adverse impact on criminal justice expenditures, it must be through one of four avenues.5 First, immigrants may be more likely to commit crimes than natives, or commit crimes that are more costly to society.
Second, immigrants may have an adverse impact on crime by crowding natives out of the legal sector. If immigrants adversely affect natives’ legal alternatives by taking jobs or overburdening the welfare system, low-skilled natives may increase their involvement in criminal activity. Both of these effects imply that crime rates would be higher in areas with heavy immigrant concentrations, other things being equal. Third, immigrants may be more likely to be apprehended or convicted than natives. This may occur if immigrants do not have the same knowledge of the legal system as natives or because they are in the criminal justice system for immigration violations. Obviously, immigrants are “at risk” for this kind of apprehension while natives are not. Finally, immigrants may serve longer terms than natives, either because they are given longer sentences or because they are less likely to be paroled [McShane, 1987].
We investigate the first two of these possible avenues through which immigrants might affect the criminal justice system. We exploit the fact that

4

Previous research on the effect of immigrants on public expenditures has yielded controversial and conflicting results. Accounting exercises that try to determine whether immigrants pay as much in taxes as they take out in services are often sensitive to the assumptions made. For example, Simon [1984, 1989] and Clark and Passel [1993] both find that immigrants by and large
“pay their own way.” On the other hand, Huddle reports that immigrants represented a net cost to
Texas taxpayers of $4 billion in 1992 because of expenditures on education, health care, and other services [Berke, 1994]. Other studies take a more direct approach, asking whether immigrants participate in programs that cost money. Borjas and Trejo [1991] report that immigrants are, on average, more likely to participate in the welfare system than are natives. However, Borjas and
Trejo [1991] and Borjas [1994] also find that immigrants are less likely to participate in welfare than demographically comparable natives. In the crime context, we will investigate whether immigrants are more likely to engage in crime, in the raw means and controlling for demographic characteristics. 5
Both legal and illegal immigrants may affect expenditures on government services. Although the press and politicians tend to focus on the effect of illegal immigrants, economic research tends to focus on the effect of both legal and illegal immigrants as the relevant concern for public policy. It should also be noted that most publicly available data sources, including the Current Population
Survey (CPS) and the National Longitudinal Survey of Youth (NLSY) (used in this article), do not allow one to distinguish between legal and illegal immigrants.
460 / Cross-City Evidence on the Relationship between Immigration and Crime immigration is a geographically concentrated phenomenon and use data from the Current Population Survey (CPS) and the Uniform Crime Reports (UCR) to compare crime in immigrant-intensive cities to other cities. This comparison demonstrates whether immigrants themselves appear to be disproportionately likely to commit crimes and whether they appear to cause native-born people to engage in criminal activity. Because the two effects cannot be separately identified using this strategy, we also use data on individuals from the 1980
National Longitudinal Survey of Youth (NLSY) to see whether immigrants are more likely to report criminal activity than natives, in the raw means and controlling for their characteristics.
Although we find that the flow of recent immigrants is positively correlated with the level of crime, it has no effect on one-year changes in the crime rate.
This result holds whether or not we control for changes in other city-level variables, such as demographics or labor market indicators. To the extent that we can test it with our data, it does not appear that immigrants assimilate
into crime, because analysis of longer changes yields the same null result. Analysis of individual-level data shows that immigrants are less likely than the native born to report committing any crime. When these estimates are adjusted using demographic characteristics, the difference between immigrants and natives increases. Much of the current policy debate centers on curtailing the number of legal immigrants admitted each year. Neither the cross-city nor the individual-level data support the use of the crime issue to justify cutting the flow of new immigrants. The findings have less clear implications for discussions of changing the composition of immigrants admitted [Chiswick, 1995]. Because little of the variation in changes in city crime rates can be explained by demographics, it seems unlikely that changing the composition of immigrants will make an appreciable impact on crime. The individual analysis, however, suggests that favoring women and those with higher levels of education could reduce the average criminality among immigrants to a level even further below the average among natives.
The article is organized as follows: The first section describes our data from the Current Population Surveys (CPSs) and the Uniform Crime Reports (UCR) and discusses how we construct data for metropolitan areas. Next, we present our methodological approach. The following section relates city crime rates to the total fraction immigrant in a city. We then present regression results for city-level crime rates controlling for the fraction of the population composed of recent immigrants and other city-level characteristics. The fifth section critiques and summarizes our city-level analysis. Finally, we use the NLSY to investigate differences in criminal activity between native-born and foreign-born individuals, and we offer our conclusions.

DATA AND DESCRIPTIVE STATISTICS
Crime Rates
Our data on crime rates come from the Uniform Crime Reports (UCR), which are collected by the Federal Bureau of Investigation (FBI). The crimes are reported by police jurisdictions and then aggregated up to metropolitan area
(MA) level by the FBI. The numbers include all “offenses known to law

Cross-City Evidence on the Relationship between Immigration and Crime / 461

enforcement” in eight categories: homicide, forcible rape, robbery, aggravated assault, burglary, larceny/theft, motor vehicle theft, and arson.
These crimes are referred to as “index” offenses. The first four crimes are categorized as violent crimes; the others are property crimes. Crime rates are reported as index crimes per 100,000 in population [U.S. Department of Justice, Federal Bureau of Investigation, 1988]. The most common property crimes are larceny and theft, which make up 63 percent of the offenses. Burglaries are 24 percent of property crimes. Among violent crimes, aggravated assaults constitute 60 percent of the offenses and robbery 35 percent.
Using the UCR may cause us to overlook some important types of crime. For example, drug dealing, simple assaults, fraud, vandalism, and weapons violations are not included in the UCR measure of index crimes. If some of these crimes are disproportionately committed by immigrants, then our analysis of the relationship between fraction immigrant in a city and the index crime rate will not pick up the total effect of immigrant inflows on criminal
activity.6
On the other hand, focusing on the UCR index crimes has definite benefits.
First, the FBI works to make sure that the data are consistent across cities. In fact, data are missing in some years because the crimes were not reported in the fashion required by the FBI. Secondly, by focusing on index crimes, we will capture the offenses that are generally considered the most serious. Thus, although they may not provide ideal measures of criminal activity, index crime rates are at least consistent across years and metropolitan areas and include the most virulent types of crimes.7
Currently there is widespread concern about increasing crime. Although crime rates have skyrocketed since the 1960s,8 they were relatively steady during the
1980s. Figure 1 plots index, property, and violent crime rates (weighted by population) from 1979 to 1992 for the 43 metropolitan areas used in our subsequent analysis.9 One can see from Figure 1 that the majority of crimes are property crimes. The index crime rate was at a peak in the early 1980s, and has declined or remained steady since then. In contrast to the steady overall crime rates, violent crime rose through the 1980s. In the analysis that follows, we will study overall crime (dominated by property crimes) and violent crimes separately.
Immigrant Characteristics
It is well documented that recent immigrants are less skilled than both earlier immigrants and natives [see, for example, Borjas, 1990]. The first two columns in Table 2 show characteristics of recent immigrants compared to the rest of

6

The UCR are often criticized for reporting errors. However, a priori there is no reason to think that this will bias our results on the effect of immigration on crime rates.
7
It would be interesting to perform the same analysis for drug crimes that we are able to do for the index crimes. However, it is difficult to obtain measures of the number of drug crimes, because they are not generally reported. Data on drug arrests are available; however, they are an unreliable indicator of drug crimes. The number of drug arrests is as much a function of the allocation of resources as of the underlying problem.
8
The overall crime rate per 100,000 went from approximately 2000 in 1960 to approximately 6000 in 1990 [Maguire, Pastore, and Flanagan, 1993, pp. 354–358, Table 3.122].
9
See Table 1 for a list of the metropolitan areas included.

462 / Cross-City Evidence on the Relationship between Immigration and Crime
Table 1. Metropolitan areas (MAs).
Akron, OH
Albany, NY
Anaheim, CA
Atlanta, GA
Baltimore, MD
Birmingham, AL
Boston, MA
Buffalo, NY
Chicago, IL
Cincinnati, OH
Cleveland, OH
Columbus, OH
Dallas, TX
Denver, CO
Detroit, MI
Gary, IN
Greensboro, NC
Houston, TX
Indianapolis, IN
Kansas City, MO
Los Angeles, CA
Miami, FL

Milwaukee, WI
Minneapolis, MN
Nassau County, NY
Newark, NJ
New Orleans, LA
New York, NY
Norfolk, VA
Paterson, NJ
Philadelphia, PA
Pittsburgh, PA
Portland, OR
Rochester, NY
Sacramento, CA
San Bernardino, CA
San Diego, CA
San Francisco, CA
San Jose, CA
Seattle, WA
St. Louis, MO
Tampa, FL
Washington, DC

Notes: This table lists the complete set of metropolitan areas (MAs) that are consistently identified by the Current Population Survey (CPS) from 1979 to 1990. In the figures and tables in this article not all of these MAs are included in all years because there are missing crime data for some
MAs in some years.

the population in the March 1985 CPS for the 43 cities in our analysis.10 Recent immigrants have lower levels of education, lower wages, and lower employment probabilities than the rest of the population. In addition, immigrants are more likely to be Hispanic, male, and young. Thirty-three percent of the recent immigrants in 1985 were between the ages of 15 and 24. Only 21.3 percent of the rest of the population fell into this category.
Criminal Characteristics
Recent immigrants have demographic characteristics similar to those of criminal offenders. Those who commit crimes are disproportionately male, young, poorly educated, and nonwhite. The last column of Table 2 shows the characteristics of those incarcerated in 1991 from the Survey of Inmates of State Correctional
Facilities. The population of inmates is overwhelmingly male, Hispanic or black, and poorly educated. The mean age (32.0) is the same as the mean age among the recent immigrant population. The percentage aged 15 to 24 is lower than in the immigrant population, but one should note that these figures are for the

Recent immigrants in the 1985 March CPS are defined as those living abroad in March 1980.
Butcher and Card [1991, p. 293, Table 1] show comparisons between these data and the 1980 census.
At least 85 percent of those living abroad five years ago are estimated to be immigrants. The characteristics of recent immigrants in the 1980 census and the 1985 CPS are shown to be very similar.
10

Cross-City Evidence on the Relationship between Immigration and Crime / 463

7500
7000

UCR Crime rates per 100,000

6500
6000
5500
5000
4500
4000
3500
3000
2500
2000
1500
1000
500
1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992
Year
Violent Crime Rate

Overall Crime Rate

Property Crime Rate

Figure 1. Average crime rate for metropolitan areas by year (see table 1 for a list of the metropolitan areas in this analysis).

stock of inmates, not the flow into correctional facilities. Among arrestees in
1991, 36.8 percent were between the ages of 16 and 24. Those age groups account for 13.1 percent of the U.S. population [Maguire, Pastore, and Flanagan, 1993,
p. 424, Table 4.3].
On average, inmates of state prisons have 1.6 fewer years of education than the population at large. The importance of education as a predictor of incarceration has been noted by Piehl [1994]: Among young (age 20 to 24) black males, those without high school degrees are nearly four times more likely to be institutionalized as those who have finished high school; for whites, male high school dropouts are 5.5 times more likely to be institutionalized.
Thus, one potential reason for a link between immigration and crime rates is that recent immigrants have demographic characteristics similar to criminal offenders. Even if immigrants are no more likely to commit crimes than similar natives, a large inflow of immigrants may add to a city’s population of young, male, poorly educated minorities. Borjas [1990] argues that the relevant question for public policy is not whether immigrants are less successful in the United
States than demographically comparable natives, but whether immigrants disproportionately add to the population that is less successful in the United
States. We will address this issue later in the crime context.
Metropolitan Area Level Data
The majority of our analysis uses a regression framework to explain metropolitan area crime rates during the 1980s. In some cases, we control for metropolitan area characteristics aggregated from the CPS data on individuals living in these

464 / Cross-City Evidence on the Relationship between Immigration and Crime
Table 2. Characteristics of natives, immigrants, and inmates.
March 1985 CPSa
Natives and early immigrantsc Percentage of total population
Citizen
Female
Hispanic
Black
Mean age
Age 15–24
Mean education
Workingd
Log wage
Standard deviation of log wages

97.4%
—
52.7%
8.4%
14.3%
41.1
21.3%
12.4
64.3%
2.05
0.66

Recent immigrants 2.6%
—
47.0%
35.5%
8.5%
32.3
33.0%
11.3
57.8%
1.74
0.64

1991 Survey of inmatesb
Inmates
—
95.6%
5.45%
16.7%
47.3%
32.0
21.9%
10.8
62.2%
—
—

a
March 1985 Current Population Survey (CPS). The sample contains individuals age 15 and older in 43 metropolitan areas. Recent immigrants are defined as individuals living abroad in
March 1980. b 1991 Survey of Inmates of State Correctional Facilities. These numbers represent the characteristics of those in correctional facilities in 1991, not the characteristics of those entering correctional facilities during that time. c This column contains everyone except recent immigrants. d In the CPS, “working” is the percentage reporting valid wages for the previous year. Means are weighted using CPS sampling weights. In the inmate survey, working is the percentage reporting employment in the week prior to arrest.

cities. The Office of Management and Budget (OMB) defines MAs for use by government agencies to ensure comparability. Metropolitan areas are defined in terms of entire counties, except in New England. Conditions for identification as an MA include population size and employment conditions [OMB, 1993].
Definitions of MAs are altered each June to reflect growth, and substantial changes are made following the decennial censuses.11 For example, many changes were made in MA standards as of June 1983 to reflect the 1980 census.
Most important to our analysis is the introduction of tighter standards for the inclusion of outlying counties in MAs. Standards for the status “central city” were also tightened at that time. The implications of these changes for our results will be investigated in the fifth section of the article.
Our sample of cities includes the 43 metropolitan areas identified by the CPS throughout the 10-year period from 1979 to 1990. More MAs are available in the later years due to population changes and changes in the confidentiality restrictions of the CPS. However, we use the cities that were consistently available throughout the period. Nonetheless, due to missing crime data in some years, our working sample usually includes fewer than 43 metropolitan areas.
Most of our data are from the 1981–1984 and 1986–1990 March CPSs. We

11

The information in the remainder of this paragraph comes from a document from the U.S.
Bureau of the Census [1992].

Cross-City Evidence on the Relationship between Immigration and Crime / 465

leave out March 1980 and March 1985 because the migration variables refer to a different time period than the rest of the March samples. We use the March migration supplements to define recent immigrants as those who report living abroad one year previously. In 1980 and 1985 the migration questions refer to the individual’s residence five years ago.
We also use a measure of the stock of foreign born in each city. These numbers are calculated from the November 1979 and the November 1989 CPSs. These two installments of the CPS include supplements that ask questions about country of birth.12 In this case we define immigrants as those born abroad.
Comparable questions about country of birth were asked only sporadically during the 1980s.13 In this analysis we concentrate on recent immigrants (lived abroad one year ago) and the stock of immigrants in 1979 and 1989 (born abroad). In addition to the immigrant variables, in some specifications we include other characteristics of the population over age 14 in each of the cities.14 The variables fraction black, fraction Hispanic, mean age, mean education, and labor market information are from the March CPSs.15 These city-level values are calculated by taking the weighted (using CPS sampling weights) mean of the relevant variable for each metropolitan area.
Table 3 presents the means of the city-level variables used in the analysis that follows. Note that fraction recent immigrant (within the last year) is a very small number for most cities. It ranges from a high of 0.075 (Miami in 1981) to a low of 0 (various cities, various years). There may be difficulties with defining recent immigrants in this way: The fraction immigrant arriving in the previous year may be too small in some cities to be accurately measured by the CPS.16
This concern about measurement error motivates part of the analysis in the fifth section of the article.
A disaggregated analysis could be performed for 10-year changes using the
1980 and 1990 censuses. However, using several cross sections of data from the
CPS allows us to answer questions the census would not. The 1980 to 1990 changes would cause us to miss important developments affecting crime in urban areas during the decade. For example, use of crack cocaine is widely purported to have increased violent and property crimes in urban areas, yet its usage in many cities began, peaked, and declined all within the 1980s. Our data will allow us to identify any contemporaneous links between immigration and crime which the censuses would not.
12

The November 1979 CPS includes the “Origin and Foreign Language Supplement.” The November 1989 CPS includes the “Language, Immigration, and Emigration Supplement.”
13
In addition to November 1979 and 1989, this information is available in the April 1983, June
1986, and June 1988 CPSs.
14
Unlike studies of labor markets where only the working age population is relevant, we include most teenagers and all the elderly in our sample. As cited earlier, teenagers make up a substantial portion of those who commit crimes. In addition, all segments of the population are potential victims of crime, though the probability of victimization declines with age past age 24 [U.S. Department of Justice, Bureau of Justice Statistics, 1994]. For these reasons only children under the age of 15 are absent from our sample.
15
The hourly wage variable is calculated using reports of yearly wages and salaries, weeks worked, and usual hours per week. See Card and Lemieux [1996, pp. 328–330] for a discussion of the use of this wage variable versus the hourly wage rate variable available in the outgoing rotation group samples of the CPS.
16
In addition, the numbers of immigrants arriving in one year are too small to allow disaggregation by education, age, sex, or race. The March CPSs do not report country of origin until 1994.

466 / Cross-City Evidence on the Relationship between Immigration and Crime
Table 3. Means for metropolitan areas (MAs).
Variable

Mean

Variable

Mean

Fraction new immigrants

0.006
(0.0004)

Fraction employed

0.651
(0.003)

Mean age

41.05
(0.101)

Fraction immigrant 1979

0.084
(0.004)

Fraction female

0.526
(0.001)

Fraction immigrant 1989

0.097
(0.006)

Fraction black

0.131
(0.005)

Overall crime rate
(per 100,000 people)

6373
(95.21)

Fraction Hispanic

0.064
(0.005)

Log overall crime rate

8.725
(0.015)

Mean education

12.38
(0.026)

Violent crime rate

737.0
(19.65)

Fraction central city

0.356
(0.008)

Log violent crime rate

6.498
(0.025)

Change in overall crime rate

-61.36
(27.54)

Population

2,356,654
(97,817)

Log population

14.473
(0.032)

Change in log overall crime rate

-0.010
(0.004)

Mean wage

9.487
(0.101)

Change in violent crime rate 14.51
(4.26)

Mean log wage

2.032
(0.010)

Change in log violent crime rate

0.017
(0.006)

Wage dispersiona

2.038
(0.027)

Observationsb

339

Source: Calculations from Current Population Surveys (CPSs) and Uniform Crime Reports
(UCR) for years 1981–1984, 1986–1990.
Note: Standard errors are in parentheses. a 90th–10th percentile of log wages. b There are 333 observations for the variables measured as changes.

POLICY QUESTIONS AND METHODOLOGICAL APPROACH
There are several policy questions one might want to answer about immigration and crime, each requiring a different form of analysis. Are immigrants more likely than the native born to commit crime? Do immigrants adversely affect the crime rate? To answer the first question, one would use individual data. If such data are available, it is easy to compare the average criminality of immigrants to the average criminality of natives. The first thing we need to know for policy, however, is whether immigrants affect crime by any means, either by committing it themselves or by changing conditions such that crime by natives increases. Because we are interested in this second question, we take an aggregate approach throughout much of the article.
In the latter part of the article, we use individual data to investigate the question

Cross-City Evidence on the Relationship between Immigration and Crime / 467

of the relative propensity of immigrants to commit crime. This helps to shed light on whether an immigrant effect on crime is direct or indirect. Fully understanding the routes through which immigration may affect crime is important for a targeted policy response.
One approach to the question of the aggregate effect of immigration is to compare crime in immigrant-intensive cities to that in cities with few immigrants. If immigration is exogenous and if no other city characteristics
(demographic, economic, geographic, environmental) are important to the determination of the crime rate, or at least not in a manner that is correlated with immigration, this approach yields the “right” answer. However, it is likely that other factors are relevant to the determination of a city’s crime rate. There are two ways to respond. First, one can try to control for all those “relevant factors.” Second, one can look at changes in a city’s crime rate over time, under the assumption that many factors that determine the level of crime will be constant within a city over time. It is an easier proposition to control for factors relevant to changes in crime rates than to control for all factors relevant to the level of crime. Looking at changes in panel data is a common technique in the analysis of aggregate and individual data. It is preferred under the assumption that time-series variation is more reliable for identification of the effect of interest than cross-sectional variation. However, it is not necessary that what is true in the cross section must be true in changes, or vice versa. One drawback to this approach is that one will only be confident about the impact of changes in immigration on changes in crime. But from a policy perspective, that is the appropriate dimension. Immigration policy proposals focus on restricting the future flow of immigrants—thus, we care about the impact of the flow of immigration. Later we report results for both levels and changes, but consider the results based on the changes in immigration most relevant for making policy decisions. It is possible to supplement each of the aforementioned approaches
(individual, aggregate analysis of levels, aggregate analysis of changes) to see how the effect of immigrants on the crime rate compares to that of demographically similar natives. We present such “conditional” specifications throughout the article. Although controlling for education, race, ethnicity, and age may not be appropriate to questions about the number of immigrants to be admitted, it is relevant to questions of the appropriate mix of immigrants, for example, whether the United States should return to admission criteria based primarily on skills. Throughout, we point out where the conclusions one draws differ depending on whether controls for demographics are included.
In sum, the estimates presented in this article bear on the following: whether the number of immigrants is related to the crime rate; whether the flow of immigrants is related to the crime rate; whether the flow of immigrants is related to changes in the crime rate; whether the characteristics of the immigrant flow are related to the crime rate; and whether immigrants have a higher propensity to commit crime than the native born, both conditional and unconditional on their demographic characteristics.

CITY CRIME RATES AND IMMIGRANT STOCK
Figure 2 shows the relationships between the fraction immigrant (all foreign born) in a city and the level of overall crime for the years that begin and end our

468 / Cross-City Evidence on the Relationship between Immigration and Crime

14000
13000

Overall Crime Rate 1980

12000
11000
10000
9000
8000
7000
6000
5000
4000
3000
0

0.05

0.1

0.15 0.2 0.25 0.3 0.35 0.4
1979 Fraction Immigrant (Stock)

0.45

0.5

0.55

0

0.05

0.1

0.15 0.2 0.25 0.3 0.35 0.4
1989 Fraction Immigrant (Stock)

0.45

0.5

0.55

14000
13000

Overall Crime Rate 1990

12000
11000
10000
9000
8000
7000
6000
5000
4000
3000

Figure 2. Overall metropolitan area (MA) crime rates by fraction immigrant: 1980 and 1990.

Cross-City Evidence on the Relationship between Immigration and Crime / 469

data series. Cities with very small immigrant populations show a great deal of dispersion in crime rates. High immigration cities all have higher than average levels of crime. The graphs also show the regression lines. The coefficient on immigrant stock is statistically significant for both 1980 and 1990 (t-statistics =
4.3, 3.3, respectively). The relationship between the immigrant stock and violent crime is even stronger than with overall crime (t-statistic = 5.8 in 1980; t-statistic
= 4.4 in 1990).17
In each figure, the three cities with the highest fraction immigrant are also high crime cities. These cities, Miami, Los Angeles, and New York, account for
48 percent of the recent immigrants living in metropolitan areas in the 1985
March CPS.18 Without these three cities, the relationship between the crime rate and the foreign-born population is statistically insignificant. Although it changes the complexion of the analysis to include these three cities,19 it is difficult to see how a sensible analysis of immigration could drop the three locations that account for most of the immigrants and much of the U.S. population. In what follows, Miami, Los Angeles, and New York are always included.
The aforementioned graphs suggest that there is a positive correlation between the fraction immigrant in a city and the level of property and violent crimes.
This does not mean that immigrants are responsible for the crime rate; this positive relationship could arise for many reasons. If immigrants are in some part responsible for the crime rate, we would expect to see a change in the level of immigration associated with a change in the crime rate. Figure 3 shows the change in the fraction immigrant in a city (the stock in 1989 minus the stock in
1979) along the horizontal axis and the change in the crime rate (the 1990 level minus the 1980 level) along the vertical axis. There is a good deal of variation in both the change in fraction immigrant and the change in crime rates across cities during the decade. However, there is no relationship between the change in fraction immigrant in a city and the change in overall crime rates.20
Thus, the cities of Miami, Los Angeles, and New York cause the relationship between the level of crime and the level of immigration to be positive. One way to control for city-specific determinants of crime is to compare changes in these variables. Even in this simple, unconditional analysis there appears to be no substantial relationship between movements in the crime rate and the flow of immigrants into a city. In the analysis that follows, we investigate the effects of additional city characteristics on crime rates and changes in crime rates.

17

The figure for violent crime is not shown. It is available from the authors upon request.
Recent immigrant refers to anyone arriving in the past five years in the March 1985 CPS. We report the 1985 figure because it is the midpoint of our data series. No matter what year of our data or definition of immigrant is chosen, these three cities are by far the most immigrant intensive.
Miami, Los Angeles, and New York account for 22 percent of the rest of the population living in metropolitan areas.
19
Omitting Miami, Los Angeles, and New York, the coefficients (t-statistics) from a regression of the crime rate on a constant and the fraction of a city’s population that is foreign born are 6389
(1.347) for the 1980 overall crime rate; 675 (0.814) for 1980 violent crime; 891 (0.229) for 1990 overall crime; and 126 (0.176) for 1990 violent crime.
20
The figure for violent crime is not shown because the relationship is similar: The change in fraction immigrant does not affect the change in the violent crime rate. Without Miami, Los Angeles, and New York, the coefficients (t-statistics) are -4921 (1.058) for overall crime and -233 (0.381) for violent crime. The figure is available upon request.
18

470 / Cross-City Evidence on the Relationship between Immigration and Crime

3000

Change in Overall Crime Rate

2000

1000

0

-1000

-2000

-3000
-0.05

0

0.05

0.1

0.15

0.2

0.25

Change in Fraction Immigrant (Stock)

Figure 3. Changes in metropolitan area (MA) crime rates by changes in fraction immigrant, 1980 to 1990.

MULTIVARIATE ANALYSIS
In this section, we use data across cities and over time to analyze the effect of recent immigrants on city crime rates. In the previous section we used total stock of immigrants in a city. Here the immigrant variable of interest is the fraction of a city’s population that immigrated from abroad in the previous year. Although this is likely to be a very small fraction of any given city, it captures immigrant inflow rates. The concern over the “immigrant problem” has focused on recent arrivals. In addition to being younger and less well-educated than earlier immigrants and natives, these immigrants may lack the information and skills (especially language skills) necessary for success in legal labor market activities. It may be precisely this segment of the population that is most disruptive. We test the impact of recent immigrants on city crime rates in several ways.
Table 4 presents the effect for (log) crime rates. We first present unconditional estimates of the relationship, then condition upon available economic and demographic variables. Because geographic and infrastructure characteristics of a city that are not captured in the demographic controls might be expected to influence crime rates, we then do two things to address concerns about omitted variables.21 First, in Table 5 we control for city-level fixed effects. Second,

21

An example of such a characteristic is being a port of entry for drugs.

Cross-City Evidence on the Relationship between Immigration and Crime / 471

Table 6 investigates the growth in crime rates using one-year changes in (log) crime rates and city characteristics. Controlling for city-level fixed effects dramatically affects the results.
Table 4 presents estimated regression coefficients for several different specifications of crime rates for 1981–1984 and 1986–1990.22 The dependent variables are the log of the overall and violent crime rate per 100,000 in population.23 We include indicator variables for the year in all specifications.
Time is a very important factor in explaining crime rates—the set of year dummies is generally highly (jointly) significant when sufficient controls are included in the regressions.
The first column of Table 4 includes only the fraction new immigrant (within the last year) variable along with an exhaustive set of time dummies. As we saw in Figure 2, there is a positive correlation between the level of crime and the measure of immigration to a city. Because the left-hand side variable is the log of the crime rate per 100,000, this coefficient suggests that an increase in new immigrants of 1 percent of a city’s population would lead to a 13.2 percent increase in the overall crime rate. This coefficient is statistically significant at the 5 percent level.
To investigate the source of this effect, column 2 adds other demographic characteristics of the cities—mean age, fraction female, fraction black, fraction
Hispanic, and mean education—to the regressions. This specification addresses the question of whether, conditional on a city’s demographic characteristics, new immigrants adversely affect crime rates. Mean age is negatively and significantly related to the crime rate.24 This coefficient suggests that increasing the average age in a city by one year reduces the crime rate by 2.5 percent. The coefficient on fraction female is consistently negative in the various specifications and is sometimes statistically significant (at the 5 percent level), depending on the other variables included. The negative signs on these two variables are consistent with the previous evidence that young males are more likely than other members of the population to commit crimes.
The variables capturing the racial and ethnic composition of a city are both statistically significant. A larger proportion of the population that is African
American or Hispanic is associated with higher crime rates. The coefficient on the average education level in the city is positive and significant, implying that cities with more educated populations have higher levels of crime. Education levels within a city are negatively correlated with fraction black and Hispanic in the city. When fraction black and Hispanic are omitted from the regression, mean education is negatively and significantly correlated with crime rates.
Note that including these city characteristics drives the fraction new immigrant variable to insignificance. In fact, simply including fraction Hispanic and mean age wipes out the effect of new immigrants. This is because new immigrants tend to be both young and Hispanic. Fraction new immigrant is highly correlated with fraction Hispanic, which in turn is highly correlated with the total stock of

22

As stated earlier, 1980 and 1985 are omitted because the definition of recent immigrant is not comparable to the other years.
23
Specification results showed that the crime rate variables are log normally distributed.
24
We also tried using fraction of the population in various age cohorts. The age cohort variables were jointly significant and did not change the qualitative analysis.

472 / Cross-City Evidence on the Relationship between Immigration and Crime
Table 4. Regression coefficients: Log metropolitan area (MA) overall crime rate and log MA violent crime rate.
Dependent variable:
Log MA overall crime rate
(1)

Dependent variable:
Log MA violent crime rate

(2)

(3)

(4)

(5)

(6)

Fraction new immigrants

13.230**
(1.973)

-1.782
(2.093)

-1.066
(2.017)

21.380**
(3.330)

-0.499
(3.260)

-1.038
(3.136)

Mean age

———
———

-0.025**
(0.007)

-0.011
(0.010)

———
———

0.018
(0.012)

0.013
(0.015)

Fraction female

———
———

-0.811
(0.718)

-1.853** ———
(0.723)
———

-1.592
(1.119)

-2.061
(1.125)

Fraction black

———
———

1.071**
(0.175)

1.305** ———
(0.189)
———

2.747** 2.372**
(0.273)
(0.294)

Fraction Hispanic

———
———

1.692**
(0.168)

2.174** ———
(0.310)
———

2.689** 1.632**
(0.262)
(0.482)

Mean education

———
———

0.098**
(0.034)

0.270** ———
(0.047)
———

0.049
(0.053)

0.056
(0.073)

Fraction central city

———
———

———
———

0.132
(0.079)

———
———

———
———

0.180
(0.123)

Log population

———
———

———
———

0.072** ———
(0.023)
———

———
———

0.196**
(0.036)

Mean log wage

———
———

———
———

-0.663** ———
(0.156)
———

———
———

-0.345
(0.242)

Wage dispersiona

———
———

———
———

-0.060** ———
(0.029)
———

———
———

-0.018
(0.045)

Fraction employed

———
———

———
———

0.047
(0.347)

———
———

———
———

-0.134
(0.540)

Fraction immigrant, 1979

———
———

———
———

-0.535
(0.375)

———
———

———
———

0.849
(0.583)

Year dummies (p-value)b

0.641

0.003

0.028

0.500

0.376

0.150

R-square

0.1515

0.4152

0.4972

0.4464

0.4774

0.5683

339

339

339

339

339

339

Observations

Source: Calculations from Current Population Surveys (CPSs) and Uniform Crime Reports
(UCR) for years 1981–1984, 1986–1990.
Notes: Standard errors are in parentheses. See Table 1 for a list of the metropolitan areas included. a
90th–10th percentile of log wages. b P-value from an F-test of the joint significance of an exhaustive set of year dummies.
** Statistically significant at the 5 percent level.

immigrants in a city’s population.25 Thus, this specification indicates that, controlling for the average age and fraction Hispanic in a city, a higher fraction new immigrant in the city has no effect on the crime rate.

Cross-City Evidence on the Relationship between Immigration and Crime / 473

Table 5. Regression coefficients: Log metropolitan area (MA) overall crime rate and log MA violent crime rate with MA fixed effects.
Dependent variable:
Log MA overall crime rate

Dependent variable:
Log MA violent crime rate

(1)

(2)

(3)

(4)

(5)

(6)

Fraction new immigrants

-0.774
(0.899)

-0.518
(0.917)

-0.040
(0.935)

-0.247
(1.148)

-0.367
(1.180)

-0.033
(1.200)

Mean age

———
———

-0.006
(0.005)

-0.004
(0.006)

———
———

-0.008
(0.007)

-0.008
(0.007)

Fraction female

———
———

0.171
(0.342)

0.208
(0.352)

———
———

0.316
(0.440)

0.410
(0.452)

Fraction black

———
———

0.460**
(0.206)

0.548**
(0.224)

———
———

0.094
(0.265)

0.140
(0.287)

Fraction Hispanic

———
———

0.467
(0.266)

0.601** ———
(0.272)
———

0.470
(0.342)

0.474
(0.350)

Mean education

———
———

0.032
(0.030)

0.042
(0.032)

———
———

-0.009
(0.039)

-0.034
(0.041)

Fraction central city

———
———

———
———

-0.221** ———
(0.109)
———

———
———

-0.118
(0.140)

Log population

———
———

———
———

0.080
(0.058)

———
———

———
———

-0.001
(0.075)

Mean log wage

———
———

———
———

-0.015
(0.092)

———
———

———
———

0.180
(0.118)

Wage dispersiona

———
———

———
———

-0.012
(0.029)

———
———

———
———

-0.083**
(0.038)

Fraction employed

———
———

———
———

0.041
(0.235)

———
———

———
———

0.025
(0.302)

Year dummies (p-value)b