|Home | About | Journals | Submit | Contact Us | Français|
The major government commissions on immigration and crime in the early twentieth century relied on evidence that suffered from aggregation bias and the absence of accurate population data, which led them to present partial and sometimes misleading views of the immigrant-native criminality comparison. With improved data and methods, we find that in 1904, prison commitment rates for more serious crimes were quite similar by nativity for all ages except ages 18 and 19, for which the commitment rate for immigrants was higher than for the native-born. By 1930, immigrants were less likely than natives to be committed to prisons at all ages 20 and older, but this advantage disappears when one looks at commitments for violent offenses. The time series pattern reflects a growing gap between natives and immigrants at older ages, one that was driven by sharp increases in the commitment rates of the native-born, while commitment rates for the foreign-born were remarkably stable.
The theory that immigration is responsible for crime, that the most recent “wave of immigration,” whatever the nationality, is less desirable than the old ones, that all newcomers should be regarded with an attitude of suspicion, is a theory that is almost as old as the colonies planted by Englishmen on the New England coast.—Edith Abbott in the report of the National Commission on Law Observance and Enforcement (1931:23)
Concerns about the criminality of the foreign-born were prominent in the public debate that led the federal government to become involved in regulating immigration in 1882, as they had been in the courts and in state legislatures prior to that time (National Commission on Law Observance and Enforcement 1931). In its 1911 report, the Federal Immigration Commission, known as the Dillingham Commission, concluded that federal regulation was not effectively excluding criminal aliens and proposed strengthening restrictions. Revisions to immigration law in 1910 and 1917 expanded the grounds for deportation to include some criminal acts taking place in the United States after lawful immigration. Even after the flow of immigrants had been sharply curtailed by the National Origin Quota Act of 1924, immigrants were still blamed for driving up the crime rate. In the early 1930s, the National Commission on Law Observance and Enforcement, also known as the Wickersham Commission, devoted an entire volume of its final report to the examination of the links between immigration and crime.
The view that immigration increases crime is pervasive and quite persistent, but is there any evidence to support it? Research on immigration and crime today provides no support for this view.1 The Dillingham Commission, despite its policy recommendations, found “no satisfactory evidence” that crime was more prevalent among the foreign-born than among the native population (U.S. Senate 1970b:1). The Wickersham Commission likewise found no evidence supporting a connection between immigration and increased crime. However, these early assessments were challenged by social scientists at the time, who questioned the quality and interpretation of the data, as well as by historians, who have linked trends in violent crime to the arrivals of certain immigrant groups to the United States (Gurr 1989; Lane 1989; Monkkonen 1989; Taft 1933; Van Vechten 1941).
In this article, we reevaluate the evidence on the links between crime and immigration in the early twentieth century using the same data that served as the basis for the Dillingham and Wickersham Commissions: the prison censuses. We carefully assemble population denominators from census data tabulations and microdata samples of census records, adjusting for mortality and other demographics to provide the best estimates for the prison census numbers, taking advantage of the fact that we have much richer population data available to us than the Commissions had at the time. We pay special attention to the impact of the differences in the age distributions of immigrants and natives and the aging of the immigrant population over the period. Our findings contrast with the findings of the Dillingham and Wickersham Commissions, as well as with the findings from the United States in the recent period.
In the first decade of the twentieth century, the United States experienced its “third wave” of immigration. Over 9 million immigrants entered the country between 1900 and 1910, amounting to a major demographic shock given that the U.S. population in 1900 was only 76 million. The arrival rate in 1907 reached almost 15 per 1,000 persons in the population, a number surpassed only in the early 1850s and in 1882.
Like the nineteenth-century waves of mass migration, the “third wave” provoked a significant nativist backlash. This backlash was fueled not only by the large numbers of arrivals but also by the fact that these new immigrants came predominantly from Southern and Eastern Europe rather than Northern Europe, the source of previous waves of immigrants. The “new stock” of immigrants was viewed as being inferior to the “old stock,” and anti-immigrant rhetoric was infused with racist and anti-Catholic sentiment (Jones 1980:1081). Unlike previous nativist movements, however, this one succeeded in securing legislative restrictions on immigrant arrivals. In 1897, Congress voted in favor of legislation to impose a literacy test on new arrivals only to encounter a presidential veto. Over the next 20 years, three similar bills met the same fate. Finally in 1917, Congress mustered enough votes to override President Wilson’s veto and passed the Immigration Act, which imposed a literacy test on new arrivals. The irony of the success in 1917 is that the rise in literacy in Southern and Eastern Europe in the preceding two decades had rendered the test generally ineffective in reducing the numbers of arrivals (Goldin 1994:226).
The failure of the literacy test to achieve the desired ends led to calls for quotas on the number of immigrants who could enter in a given year. In 1921, Congress passed the Emergency Quota Act, which established an annual ceiling of 357,000 new arrivals. In 1924, the National Origins Quota Act reduced that ceiling by half and also codified the bias against the “new” immigrant sources by setting the quota for each nationality at 2% of its numbers reported in the 1890 federal census (Jones 1980:1082). The result of the quota system was a sharp drop in the number of immigrant arrivals. Between 1924 and 1925, the arrival rate fell from over 6 per 1,000 to 2.5. By 1930, the immigrant arrival rate had fallen to 2 per 1,000 in the population, and it did not rise above that figure until the abolishment of the national quota system in the 1960s.
The charge that had the strongest political sway then was, as it is now, that immigrants undercut the wages of native-born workers. But another key complaint was that immigrants increased crime in the United States. Different immigrant groups were ascribed different proclivities for criminal behavior in general and different proclivities for different types of crime. Immigrants from Southern and Eastern Europe, particularly those from Italy, were accused of being more prone to violent crime than “old stock” immigrants and the native-born population (Bingham 1908).
These perceived links between immigration and crime shaped federal immigration law. The 1917 Immigration Act included a provision to deport any immigrant who had been in the United States five or fewer years and had been sentenced to at least one year in prison, and any immigrant who had been convicted of a more serious offense or prostitution, no matter the time spent in the United States. These perceptions also influenced the development of sociological theory and the emerging field of criminology.
Academic studies of the links between immigration and crime grew out of more general studies of the experiences of immigrants. In 1900, 66% of the foreign-born population in the United States lived in cities of 2,500 or more, and 38% lived in cities of 100,000 or more (U.S. Senate 1970a:139). Theories about immigrants, therefore, were largely imbedded in theories of the functioning of the urban landscape. One such theory was “social disorganization theory,” developed at the University of Chicago, which described how life in urban America weakened social bonds (Bursik 2006; Wikstrom 1998). Shaw and McKay (1942) applied social disorganization theory to delinquency, arguing that immigrant assimilation was integrally related to urban life. Immigrants initially settled in high-crime areas and then moved to less disorganized neighborhoods following success in the labor market. While new immigrant groups would be perceived to have high rates of criminality, Shaw and McKay assumed that this relationship was spurious, depending on residential location rather than nativity status. Thomas and his coauthors (1971) attributed the differences in experience across immigrant groups to the varying “endowments” people brought with them and how they intersected with social organizations that existed in the new country.
Others theorized that the process of immigration had a direct connection to criminality. One important example of this is Sellin’s (1938) emphasis of the “ culture conflict” faced by immigrants as they adjust to a new set of behavioral norms, a conflict that might lead to greater criminal activity. Furthermore, some argued that immigration might lead to greater crime even if immigrants themselves do not have higher criminal propensity. At the aggregate level, it is possible that immigration would increase the criminal activity of the native-born by displacing natives from work, promoting urban ization, and increasing “the variety of patterns of behavior” (Sutherland 1924:128).
Early theories of crime stressed the individual rather than group characteristics associated with crime. But these theories too could be used to explain higher rates of crime among immigrants. As we discuss more below, immigrants, particularly recent arrivals, were disproportionately represented in the demographic groups with the highest rates of crime: males in their late teens and 20s. Immigrants to the United States also ranked high on other well-known criminogenic factors, including poverty (Taft 1933).
At the same time, some theorists emphasized mechanisms that would lead immigration to reduce, rather than increase, crime. Sutherland (1924:124) noted such an effect: that immigrants may have developed strong respect for the law in their home countries, formed in their “homogenous and stable groups” before migrating to the more disorganized American city. Of course, the extent to which this is true might depend on both the characteristics of the home region as well as the reasons for immigration.
Despite the active public and scholarly debate about the links between immigration and crime, the empirical investigation of these links was quite limited. The most prominent and extensive efforts to gather and analyze data were conducted by governmental commissions. These commissions, though, had political agendas that influenced how they presented and interpreted the data. It is significant, for instance, that the Dillingham Commission stated its conclusion as there being “no satisfactory evidence” that crime was more prevalent among immigrants than among the native population. The failure to find a clear nativity difference in overall criminal activity led the Commission to focus on nativity differences in the types of crimes committed. The findings on this subject are stated in a much more certain tone: “From the data gathered it is evident that immigration has had a marked effect upon the nature of the crimes committed in the United States. This effect has been to increase the commission of offenses of personal violence” (U.S. Senate 1970b:2). The Commission singled out immigrants from Southern Europe, and especially those from Italy, for their involvement in homicides. The data underlying these claims, however, undermine the assuredness in which they are presented. The Commission did not find that immigrants were more likely than natives to commit or to be convicted for committing violent crimes. Rather, it found that within the incarcerated population, a higher fraction of immigrants than natives had been convicted of violent crimes. As Oscar Handlin pointed out in his scathing review of the Dillingham Commission reports, such evidence tells us only that violent crimes represented a greater share of criminal behavior for immigrants than for natives. It tells us nothing about the relative or absolute criminality of immigrants (U.S. Senate 1970a:xxxv–xxxvi). By 1931 and the Wickersham Commission, the pendulum had swung to the other side, and critics accused the Commission of presenting its findings in such a way as to portray immigrants in the best possible light (Taft 1933).
Even absent the problem of political agendas, contemporary investigations suffered from limited or poor-quality data. Only a select number of jurisdictions regularly compiled police and court records, and the nativity information in these data was often suspect (Sutherland and Van Vechten 1934). The most complete data came from special censuses of the population in penal institutions conducted by the Census Bureau, the so-called prison censuses. However, analysis of these data was complicated by the fact that most of the special censuses did not occur in the same year as a population census. We would like to look at incarceration rates—the ratio of the number of incarcerations for members of a particular group to the number of individuals in that group in the population. For a relatively stable population, like the native-born, a difference of a few years in the timing of measurement of the numerator and the denominator will not bias the constructed measure very much, if at all. But such a timing difference could substantially bias the constructed measure for immigrants, especially in the early decades of the century, when immigrant inflows were high. The Census Bureau was very concerned about this issue and for the most part shied away from presenting incarceration rates. Instead, the census reports presented comparisons of the percentage of the foreign-born in the general population at the last census to the percentage of foreign-born in the incarcerated population, accompanied by warnings that the population data may understate the immigrant population at the time of the prison census.
A more serious problem with most of the early investigations of immigration and incarceration is that they did not adequately deal with differences in the age distributions between the immigrant and native-born populations. The Census Bureau, as well as other researchers at the time, was sensitive to the fact that the relatively small fraction of young children in the immigrant population would inflate the perception of criminality for the foreign-born. Their solution was to compare the percentage foreign-born in the incarcerated population to the percentage foreign-born in the adult population, most often defined as the population aged 15 and older. This, however, led them to fall into the trap of aggregation bias in just another guise. This problem was pointed out most convincingly by C.C. Van Vechten, the Chief of the Institutional Section of the Census Bureau, who was writing in the aftermath of the finding of the Wickersham Commission that natives were twice as likely as were immigrants to be imprisoned. Van Vechten (1941) argued that this “2 for 1” advantage was primarily due to the differences in the age distributions between the two groups. The National Origins Quota Act of 1924 had sharply reduced the inflow of new migrants. As a result, the immigrant population had aged relative to the native-born population.
This point can be seen most starkly in Figure 1, which presents the age distributions for the foreign- and native-born white male populations. By 1930, as shown in the top panel, more than half of the foreign-born population was over the age of 30, and more than 10% were 65 or older. These differences in age distributions can have a profound effect on perceptions of relative criminality because age is so strongly correlated with criminal behavior. Crime rates peak for males in late adolescence and decline rather precipitously thereafter. Comparisons of native and immigrant criminality in 1930 that do not control for these differences in age distributions will necessarily bias the results in favor of immigrants.
Such aggregation bias also plagues any consideration of the changes over time. One interpretation of the contrasting findings of the Dillingham and Wickersham Commissions is that the incarceration rate among immigrants was falling relative to that of the native-born between 1900 and 1930. But the discussion in the previous paragraph offers an alternative interpretation: perhaps the immigrant population was simply getting older over the first few decades of the century. The bottom panel of Figure 1 presents the age distributions of the native- and foreign-born white male populations in 1910. Here, the bulge in the foreign-born population is in the late 20s and 30s, reflecting the large immigrant inflows between 1900 and 1910. Even if there had been no change in age-specific incarceration rates over the period, we would expect to see the incarceration rate of the foreign-born population as a whole decrease because, over time, a smaller fraction were in the high-incarceration age groups.
The weaknesses of the contemporary investigations of crime, incarceration, and immigration would seem to make this a ripe area for historical analysis. But to date, only limited work has been done. The research that has been conducted, though, challenges to some extent the conclusions of the Dillingham and Wickersham commissions. A number of historians have argued that U.S. trends in violent crime were strongly correlated with new immigrant arrivals (Gurr 1989). Monkkonen (1989), using data from newspaper accounts and coroners’ records from the 1850s, claimed that New York City’s homicide rates would have been a third or possibly even two-thirds lower had it not been for the city’s large immigrant population (p. 91). Lane (1989) likewise found that Italian immigrants were disproportionately involved in homicides in Philadelphia in the early twentieth century. It is difficult, however, to discern what these studies tell us more generally. First, these studies provide data only on a limited number of jurisdictions. Second, they tell us only about the involvement of immigrants in homicides, which account for a very small fraction of all crime.
Our contribution to this literature is to reanalyze the data used by the Dillingham and Wickersham commissions, with the modern advantages of detailed population data and the computing power to analyze them. First, we address the basic question of whether, after differences in age distributions are controlled for, immigrants were more likely than the native-born to be imprisoned in the early decades of the twentieth century. We then examine how immigrant incarceration patterns varied with time spent in the United States and country of origin. These investigations allow us to consider the theories put forward by criminologists at the time about the factors influencing immigrant involvement in crime.
The basic question we would like to ask is, were immigrants more or less likely than the native-born to commit crimes? We can never, however, address this question directly because we cannot observe criminality per se, but rather things that are recorded, like crime reports, arrests, and convictions. Crime measured by any of these types of data will necessarily understate criminal activity, and there are reasons to believe that all of these types of data may over- or understate relative immigrant involvement in crime: immigrants may have been less likely to report victimization, or racial prejudice on the part of the police or courts may have made them more likely to be arrested and convicted of crimes. The degree and even the direction of these biases are difficult to evaluate because of the complex path from crime commission to incarceration, with discretion operating at each of the many steps in between. Differential treatment by nativity may be due to factors associated with immigrant status per se (such as policies toward increased detention of noncitizens) or due to factors that are correlated with immigrant status (such as living in a large city or being impoverished or unfamiliar with court processes). Analysts at the time detailed the many reasons that the foreign-born might be disadvantaged in court: from not understanding the court system, to lack of adequate translation services, to being easy targets for scams (Claghorn 1971). Many of these mechanisms would suggest large immigrant disadvantages for minor and nuisance offenses. But it was also suggested that lack of confidence in the rule of law and/or nonresponsiveness of police to foreign-born complainants might cause some conflicts to escalate to violent crime due to the lack of involvement of the police and courts in early stages of conflict. Unfortunately, even in modern crime data, obtaining reliable evidence on the magnitudes of these effects is difficult.
The prison population data collected by the Census Bureau have the advantages of (1) having quality information on nativity and related factors, (2) containing a census of all inmates, making the study not only nationally representative but actually based on national coverage, (3) having served as the primary source of the findings of the Dillingham and Wickersham Commissions, and (4) being comparable to some contemporary research on related questions.
The disadvantage of using the prison census data is that they record events that take place, in the terminology used above, several steps after the criminal acts themselves. The discretion involved in each intervening step means that the difference in the incarceration rates of immigrants and natives is a noisy measure of the difference in the crime rates of the two groups. Modern studies of crime often use police records, but unfortunately, we cannot use police records to look at nativity differences in crime in the early twentieth century. Both Commissions considered and rejected police records due to incompleteness or poor quality. By the late 1920s, experts began to prefer a measure closer to the point of crime commission (Maltz 1977). This produced the movement to structure the collection of such data under the Uniform Crime Reports, which would provide some standardization of criteria. These data were not collected until 1930, and voluntary participation by police departments was initially low. The prison census data may not be immune from racial and ethnic prejudice, but analysts at the time believed court actions were a better reflection of behavior than police actions because each step in the criminal justice process provides the opportunity to check earlier steps for such bias (Maltz 1977; National Commission on Law Observance and Enforcement 1931).2
In 1904, the Census Bureau conducted its first special enumeration of prisoners separate from the population census.3 Data were collected on the population in penal institutions on June 30, 1904, as well as on all commitments to these institutions between January 1, 1904, and December 31, 1904. The Census Bureau conducted similar special enumerations in 1910 and again in 1923. In 1926, the Census Bureau began annual counts but limited their scope to state and federal facilities.4
Despite being commonly referred to as “prison censuses,” the focus of the data collection, as well as the bulk of the analysis by the Census Bureau, was on commitments to prisons rather than on the prison population at a given moment in time. All of the censuses collected and presented data on what we might call the “stock” of prisoners on particular date, but most of the detailed tables and breakdowns pertain to commitments, or the “flow,” into prisons over a given time period. This focus became more pronounced over time; by the 1930 prison census, only one of the 54 tables pertained to the prison population on a given date, and one more reported the average daily prison population.
Some explanation for this focus is given in the report of the 1923 prison census. Data on the incarcerated population on a given date, it was argued, was useful for assessing the costs of institutional care for different types of offenders but not for studying criminality. It was pointed out that an increase in the prison population could occur without an increase in the number of crimes being committed; longer sentences would increase the number of individuals incarcerated on any given date. Commitments over a specified period of time were viewed as a better index of criminality. Increases or decreases in commitments may not be exactly proportional to increases and decreases in criminal activity, but it was argued, “other things being equal,” an increase in the number of commitments for a particular offense was directly related to an increase in convictions for that offense, which was likely related to the frequency that offense was committed (U.S. Department of Commerce 1926: 4–5).
This focus on the flow rather than the stock of prisoners contrasts sharply with studies on incarceration in the current period. The flow measure may give a better approximation for crime rates than the stock of prisoners; however, flows are dominated by more common, but less serious, crimes. The 1910 data allow us to analyze the differences between the “stock” and “flow” of inmates across jurisdiction levels and across crime types. The top panel of Table 1 shows how the number enumerated in an institution differs from the flow of new commitments in 1910 for several demographic groups. Overall, the 479,787 commitments are 4.3 times the 111,498 present at a point in time in prisons, jails, and workhouses. Looking across demographic groups, the results show that women have a much higher ratio of flow to stock than men (7.6 to 4.1), and foreign-born whites have a higher rate than native-born whites (5.1 to 4.6). Similar patterns are evident for the flow out (discharges) relative to the stock.5 These numbers indicate that the foreign-born will generally look relatively worse with a flow measure than with a stock measure. Also note that commitment data may have more measurement error, as the rates of “unknown nativity” and “offense unknown” are much higher in the commitment data than in the enumerated data.
The bottom panel of Table 1 reports the enumerated population and the commitment flow for 1910 by offense (offenses with fewer than 1,000 enumerated were suppressed in the table). Commitment numbers are dominated by the less serious crimes of disorderly conduct and vagrancy, which make up 65% of all commitments. Homicides, which are 13% of the enumerated population, are less than 1% of the flow into this broad set of institutions. Thus, when one considers commitments to all penal institutions, the conclusions will be driven by high-volume offenses that are punished by short stays in confinement. These lesser crimes may be treated quite differently in different states.
However, the picture changes when we narrow our focus to commitments to state and federal prisons, which generally house the most serious offenders, as shown in the final column of Table 1. Although state laws varied, typically only those sentenced to a year or more would be placed in a state, rather than local, facility. Accordingly, state and federal prisons accounted for the majority of the sentenced inmate population at any given moment in time. Of the population incarcerated on January 1, 1910, 53% were in state or federal prisons.
Commitments to prisons were dominated by more serious offenses. The distribution of offenses for such commitments was, in fact, very similar to the distribution of offenses for the enumerated population. The only notable difference between these two distributions is the much smaller share in prison commitments of less serious offenses like disorderly conduct. The data on prison commitments are also of higher quality than those of commitments more generally. In fact, the percentage of observations with missing information on nativity is smaller for prison commitments than for the enumerated population.6
Commitments for more serious offenses are a better measure of “criminality” as usually perceived, and these are the offenses that impose the greatest costs to society. Moreover, convictions for minor offenses—such as disorderly conduct—reflect, to a much greater extent than those for serious offenses, the choices made by law enforcement officials. Focusing on serious offenses also makes our findings more comparable to studies of the recent period, which look at the prison population at a given moment in time. We examine nativity differences in commitments for minor offenses, but most of our discussion and analysis focuses on commitments for more serious crimes.
Another challenge for our analysis is the tremendous variation across the published prison census reports in how the data are presented and even how population subgroups are defined. Our strategy is to exploit the strengths of each prison census. Ideally, we would like to have data on prison commitments by gender, age, nativity, and offense, but such detailed breakdowns are available only for the annual prison censuses starting in 1926, and these provide data only on state and federal prisons. The 1923 prison census, however, provides breakdowns by gender, age, nativity, and jurisdiction, allowing us to compare state prisons to municipal and county jails. As shown in Table 1, the breakdown by jurisdiction roughly coincides with the division between more serious and less serious offenses. The 1904 prison census does not provide breakdowns by jurisdiction, but it does separate commitments for “major” and “minor” offenses by gender, age, and nativity. “Major offenses” included “all crimes that are universally held to be of a grave nature,” and included all person offenses, the most aggravated offenses against chastity, perjury, counterfeiting, arson, burglary, forgery, embezzlement, and serious cases of larceny and other offenses when punished by imprisonment of more than one year (U.S. Department of Commerce and Labor 1907:28). We treat commitments for major offenses in 1904 as roughly comparable to commitments to prison in the later censuses. Unfortunately, the 1910 prison census, despite presenting more tables than any other prison census report, does not provide data that specify type of offense or jurisdiction separately by gender, age, and nativity. So for the most part, we leave the 1910 data out of our analysis.
We limit our attention to males ages 18 and older. As can be seen in Table 1, females during this period had very low incarceration rates. Most commitments of females were for prostitution and generally involved short sentences. Juveniles, too, had very low rates of incarceration. But more problematic for this study, the treatment of juveniles in the criminal justice system varied greatly across states and over time. A 15-year-old convicted of robbery might be committed to a juvenile facility in one state but to a jail in another state. Although some of the prison censuses do provide information on commitments to juvenile facilities, many of these commitments are for minor, juvenile-specific offenses like truancy, as well as for other noncrime reasons, like having deceased or incapacitated parents.
As noted above, calculating commitment rates requires accurate population data that correspond with the timing of the prison censuses. Although all of the annual censuses of state and federal prisons starting in 1926 present similar data, we choose to use the 1930 prison data so that we can use the data from the 1930 federal population census to construct the denominators. In order to examine the 1904 and 1923 prison data, we must construct population estimates. The writers of the 1904 prison census report simply used data from the 1900 population census. However, the 1900 data understate the size of the foreign-born population relative to that of the native-born in 1904, particularly in the younger age categories, due to the high immigrant inflows of the period. In order to capture this inflow, we construct population estimates for 1904 using microdata from the 1910 population census made available through the Integrated Public Use Microdata Series (IPUMS; Ruggles et al. 2008). The 1910 census collected data on the year of immigration, which we use to identify the foreign-born who had arrived in the United States by 1904. We then age the population backward to 1904 and adjust for mortality to construct population estimates by age and nativity.7 Using these population estimates rather than the 1900 population data lowers the commitment rates for 18- to 19-year-olds by 18%. Taking into account the immigration flows between 1900 and 1904, therefore, significantly alters the constructed age profile of the incarceration rates of the foreign-born population. We use the same procedure to construct population estimates for 1923 using the 1930 IPUMS data set.8
In Table 2, we present commitment rates per 100,000 population by age, race, and nativity. Note that the age categories are not of consistent width; because of the importance of age in the study of crime outcomes, we report age at the lowest level of aggregation available. The data for 1904 are commitments for “major offenses,” and the data for 1923 and 1930 are commitments to state and federal prisons.9
The highest commitment rates, by far, are among black Americans. This phenomenon persists today and is the subject of a long literature (e.g., Blumstein 1982; Sampson and Lauritsen 1997; Western and Pettit 2000). Besides the issue of potential racial bias in enforcement and conviction rates, the comparison of the incarceration patterns of blacks and the foreign-born in this period is complicated by the very different geographical distributions of these populations. The immigration-crime debate in the early twentieth century was framed in terms of the comparison of foreign-born with native-born whites. We choose to emphasize the same comparison. However, as is easily seen by looking at the last column in Table 2, if the comparison group were all natives instead of native whites, the relative performance of immigrants would appear much better.
The 1904 data for both groups exhibit the steep age-crime curve familiar to criminologists. The peak for foreign-born males, however, is earlier (at 18–19 rather than 20–24) and higher than that of the native-born. By age 30, though, the rates are quite comparable across nativity groups.10
In 1923, as in 1904, the foreign-born had higher commitment rates to state and federal prisons than natives at younger ages but lower rates at older ages—age 35 and older in this case. In contrast, in 1930, the commitment rates of foreign-born white males were lower at each age than for the native-born, in some cases substantially lower. Detailed results are shown in the bottom panel of Table 2. The comparison of native-born to foreign-born whites as presented in the 1930 data is potentially misleading, however. In both the 1930 population census and prison census, people born in Mexico or of Mexican descent were classified simply as “Mexican.” In earlier population censuses and special enumerations of penal institutions, Mexicans were classified as “whites.”11 Therefore, Mexican immigrants are excluded from the 1930 data on “foreign-born whites” in spite of forming a sizable part of the foreign-born.12 Individuals of Mexican descent who were born in the United States were also excluded from the “native-born white” category, but the impact of this exclusion is smaller given the size of this group relative to the size of the native-born population as a whole. Given the constraints of how the data were reported, the only way we could include those of Mexican descent was to substantially broaden the definition of foreign-born. The category “foreign-born white plus all other races” includes Mexicans as well as Native Americans, Chinese, Japanese, and all others deemed not white and not black. Mexicans account for the vast majority of this “other race” group. But about a third of all males over the age of 15 identified as “Mexican” in the population census were born in the United States. Adding this group to the foreign-born muddles the comparison by nativity, but it at least can give us a sense of the degree to which the exclusion of Mexicans influences the incarceration patterns we observe.
Adding “other races” to the foreign-born category generally increases its commitment rate by 30%. This substantially narrows the difference between native and foreign-born commitment rates but changes the sign of this difference only for 18-year-olds. In 1930, the age-adjusted rates of prison commitment are quite similar for the native and the (broadly defined) foreign-born at younger ages, but by age 20, the foreign-born appear to be somewhat less likely to be committed to state or federal prison.
The picture that emerges from Table 2 contrasts with that pieced together from the Commission and Census Bureau reports of the period. The general pattern of native incarceration rates increasing relative to those of immigrants over time is the same, but the starting and ending points are different. In 1904, young immigrants had higher rates of commitments for serious offenses than did natives of the same ages. Moreover, the change in commitment rates over time for this group was also much smaller than for older immigrants. The relative decline of the incarceration of the foreign-born is really a story about a growing gap between natives and immigrants at older ages.
Age-specific commitment rates clearly tell us more than could aggregated rates about differences between natives and immigrants, but they cannot tell us how differences in the age distributions of the two groups affected the perceptions of the relative criminality of the two groups. Table 3 summarizes the impact of the different and changing age distributions of the foreign- and native-born populations on aggregate commitment rates. Using our population estimates, the overall commitment rate for the native-born was 77 per 100,000 in 1904, and nearly double that in 1930 at 140 per 100,000. For the foreign-born, the rate was somewhat lower than natives in 1904 at 69 and falling to 52 by 1930.13
To demonstrate the impact of the different age distributions, we report counterfactual aggregate rates by applying the nativity-specific commitment rates by age to three different fixed age distributions: the native-born distribution in the census year, the foreign-born distribution in the census year, and the 1930 native-born distribution. As shown in the bottom of Table 3, in each case for the native-born, the commitment rate increases dramatically from 1904 to 1930 (though it is somewhat less pronounced using the foreign-born age distribution). However, the shift in the estimates for the foreign-born is dramatic. In 1904, both alternate age standardizations show commitment rates for immigrants that are higher than the unstandardized value, and in fact are higher than those of the native-born. Even in a period of large inflows of immigrants, the age distribution of the foreign-born advantaged them in aggregate commitment rates. Had the Dillingham Commission performed such calculations using the 1904 data, it would have concluded that immigrants were more criminal than natives. The age-standardized commitment rates for immigrants in 1923 and 1930 are basically the same as that of 1904. The decline in the overall commitment rate for immigrants observed in the top panel of Table 3 is due to the aging of the foreign-born population. From these calculations, it is clear that the growing gap between the foreign-born and native-born incarceration rates comes from the doubling of the commitment rate of the native-born over this time period.
Violent and property crimes frequently have different time trends, age patterns (with involvement in property crimes peaking at younger ages than for violent crimes), and geographical distributions. In addition, inmates convicted of property crimes have somewhat different criminal histories and postrelease outcomes compared with those convicted of violent crimes (Langan and Levin 2002). Only the 1930 census reports data that allow us to look at nativity differences in commitment rates for particular offenses, controlling for age. Even within commitments to prison, violent crimes are a minority. But for the foreign-born, violent crime represents a larger share of overall offenses (36% for those aged 18–24 and 28% for those aged 25–34) than among native whites (24% and 20%, respectively). The top panel of Table 4 shows the commitment rates for white males at the least aggregated level available. Here, the rate for foreign-born whites (excluding Mexicans) is almost identical to that of native whites at ages 18 to 24, then somewhat lower than that of native whites at older ages. Using the broader definition of foreign-born in order to include Mexican immigrants narrows the gap between immigrants and natives even more.
The middle panel of Table 4 shows the commitment rates for homicide, the most serious of the violent crimes. Foreign-born white males under age 35 have homicide commitment rates that are substantially below those of native whites of the same ages when the narrower definition of foreign-born is used. But when the broad definition is used, the homicide rate for the younger age groups exceeds that of natives. It is interesting to note that the age-incarceration curve is much flatter for homicide than for the other crimes. Although the foreign-born in 1930 had much lower overall prison commitment rates than natives, commitment rates for violent offenses were very similar across the two nativity groups. This is perhaps most easily seen in the rows reporting the aggregate rate using the 1930 native-born age distribution.
These findings for violent crime contrast with those measuring lesser offenses. Minor offenses, or those leading to short sentences in local facilities, include burglary and possession of stolen goods but are dominated by public order offenses like drunkenness and disorderly conduct. The commitment rates of the foreign-born exceed those of native-born whites when minor offenses are considered, as shown in the bottom panel of Table 4 for 1904. The gap in commitment rates for minor offenses is particularly large for males in their 40s and 50s. Note, too, that the age-incarceration curve is quite flat for the minor offenses, peaking in the 40s. These findings are consistent with the higher “commitment to enumerated” ratio for the foreign-born observed in 1910 (Table 1).
Higher commitment rates for minor offenses, however, may not be evidence of greater criminality among the foreign-born. Imprisonment for minor offenses depends greatly on law enforcement choices. The writers of the report on the 1904 prison census attributed the relatively large numbers of commitments for minor offenses among the foreign-born population to its concentration in major urban centers where such offenses were more likely to be punished (U.S. Department of Commerce and Labor 1907:28). The level of enforcement of these kinds of offenses varied greatly across jurisdictions. A study of city-level data from 1900 found that the arrest rate for drunkenness was positively correlated with the number of police per capita and the number of years the police department had had a merit system in place (Brown and Warner 1995:90).
But the higher commitment rate for such offenses among immigrants likely also reflects prejudicial enforcement even within particular jurisdictions. The decision to arrest someone for disorderly conduct or drunkenness is a discretionary one. There is ample anecdotal evidence that immigrants, especially those who did not speak English, were more likely to be arrested and convicted for such offenses. Jones (1976:213) recounted such a story in Destination America, his popular history of the experiences of immigrants in America. In his story, an Italian immigrant bought a candy bar and put it in his pocket. He was stopped by police because they assumed it must be a gun or a knife. Even after the police discovered it was just a candy bar, they arrested the man because, being unable to speak English, he could not explain how he got the candy bar.
Both the culture conflict and social disorganization theories predict that immigrant crime rates should decrease with time spent in the United States. Some evidence supporting these predictions is the contrasting experiences of young and old immigrants. In 1904, more than 45% of immigrants ages 18 to 24 had been in the United States fewer than five years, and the incarceration rates for these groups exceeded those of natives of the same age. In contrast, new arrivals accounted for 10% or less of the older age groups of immigrants that had incarceration rates very similar to the native-born.
Table 5 provides more direct evidence on assimilation by contrasting the distribution of time in the United States of those committed to penal institutions to the distribution in the civilian population of foreign-born white males in 1904. Although 3.4% of the civilian population had been in the United States for one year or less, a full 8.7% of those committed for major offenses had arrived that recently. This is a huge proportion of the population, especially considering that enforcement actions generally take some time to complete. Part of the explanation of the higher incarceration rates of immigrants than natives at younger ages seems to be about the challenges faced by recent arrivals in the United States. Assimilation theories, however, cannot explain the incarceration rate patterns for immigrants for minor offenses. More than half of those committed for minor offenses had been living in the United States for 15 years or more.
The discussion of time in the United States leads naturally to a consideration of the outcomes for the children of the foreign-born, a topic of great concern to the Dillingham and Wickersham Commissions. Figure 2 shows the commitment rates for major and minor offenses by parentage. These rates were calculated using population estimates constructed from the 1900 population census.14 In these graphs, those born to two native parents are compared with those with at least one foreign-born parent. For the major offenses, the commitment rates for the latter group are generally somewhat higher than for the children of native-born parents. The gap is not usually large, and both groups show the same age-crime curve with a peak in the early 20s. For minor offenses, the pattern is very different. Here, the children of foreign-born parents have very high commitment rates, particularly in their 40s and 50s, mimicking the patterns of the foreign-born themselves.
We now turn to the final “hot” issue in the study of immigrant criminality. Both of the Commissions concluded that the composition of offenses varied greatly across immigrant groups. The Dillingham Commission had singled out the Italians for their involvement in violent crimes. Lane (1989) likewise singled out the Italians as being disproportionately involved in homicides in Philadelphia in this period. None of these studies, however, considered the impact of age on the crime experiences of different immigrant groups. The age distributions varied quite a lot by country of origin, just as they differed between immigrants and natives. Figure 3 gives an indication of this variation by plotting the age distributions for the Germans, Irish, and Italians in 1904.15 The Germans and Irish were part of the “old stock” of immigrants, and by 1904, these populations were quite old, with nearly half of the population aged 50 or older. By contrast, the Italians, part of the “new stock” of immigrants, look young, with the bulk of the population in their 20s and 30s. Given what we have shown about the age distribution of crime, one would expect these differences to be observable in crime outcomes.
Unfortunately, none of the prison censuses reported commitment data broken down by both country of origin and age. So we predict commitment rates for each immigrant group based on its age distribution in the general population and the age-specific commitment rates of the foreign-born overall, as found in Tables 2 and and4.4. These predicted commitment rates give us a sense of how much of the variation in commitment rates by country of origin can be explained by variation in the age distributions alone.
Table 6 reports the 1904 actual and predicted commitment rates to penal institutions, in total and separately for major and minor offenses, by country of origin. For minor offenses, the predicted rate barely fluctuates, due to the flat age-crime curve. But for major offenses, the predicted rate varies greatly across countries. This fluctuation should raise a big red flag regarding any comparisons of any crime-related variable across immigrant groups that do not account for age.
The data in Table 6 demonstrate that at least some of the differences in commitment rates by country of origin can be attributed to differences in age distributions. For instance, the commitment rate for major offenses for Russian immigrants, one of the “new” immigrant groups of the period, was high relative to those of the English, Germans, and Irish, but it was close to what was predicted given the age distribution. But differences in age distributions cannot explain all of the differences in crime involvement across immigrant groups. For some groups, the predicted rates far exceed the actual, meaning that the group is “outperforming” what would be expected given the age distribution of that immigrant group. For example, Norwegians and Swedes were committed for major offenses at half the rate that would be expected based on age alone. For the Irish, the predictions are lower than the actual experience. But while this difference is small for major offenses, it is huge for minor offenses.
The results are also dramatic for Mexicans. But there are reasons to believe that this may reflect, at least in part, problems in both the population and commitment data for this group. The seasonal migration of the Mexican immigrant population may make the census date count an understatement of the population “at risk” to be committed to a penal institution in a given year. At the same time, the commitment data may overstate the number of Mexican-born if institution administrators were prone to classify those of Mexican ancestry simply as “Mexicans.”16
The Italians merit special note given the attention this group has received in previous studies. Their commitment rate for major offenses was high—more than twice that of Irish immigrants and more than three times that of German immigrants. In part, this higher rate can be explained by the much younger age distribution of the Italians. The predicted commitment rate for Italians is almost twice that of Germans and the Irish. Nonetheless, the predicted commitment rate falls short of the actual rate by a considerable degree. Even taking the younger age distribution into account, Italian immigrants appear to have been disproportionately involved in more serious crimes.
Commentators at the time frequently attributed differences in criminality by ethnicity to differences in cultural predispositions (Bingham 1908). But before taking the data in Table 6 as confirming this view, it is important to keep in mind that immigrant groups differed in characteristics other than age that would also be expected to affect criminal involvement. Reflecting differences in population characteristics in the countries of origin and the self-selection of migrants from those populations, immigrant groups in the United States varied greatly in skill and educational levels and, more generally, economic resources. Table 7 presents data on average manufacturing wages, literacy, English proficiency, time in the United States, and percentage urban by country of origin.17 The countries are listed in descending order by the ratio of the actual commitment rate for major offenses by the predicted commitment rate.
Mexico is at the top of this list, with an actual commitment rate five times that predicted given the age distribution. But Mexicans are also at the extremes of the distributions of the other presented characteristics: they had the lowest average wages in manufacturing, the lowest literacy rate, and the lowest rate of English proficiency. Italian immigrants, for whom the ratio of the actual to predicted commitment rate was 1.7, had the second lowest rates of literacy and English proficiency as well as the second lowest level of manufacturing wages. At the bottom of the list are immigrants from Scandinavian countries who, in contrast to those from Mexico and Italy, were highly skilled and had among the highest average wages in manufacturing. Consistent with a standard economic model of crime, the criminality of immigrant groups, at least as measured by commitment rates, was strongly correlated (in ranks and in levels) with the economic opportunities of those groups.
Whether immigrants were more prone to crime than the native-born depends on how one defines crime. Our ability to measure this depends on the quality of the data as well as how discretion is used in the criminal justice system. This article relies on data from those punished for criminal activity, not direct observation of crime. The foreign-born were more likely than natives to be incarcerated for minor offenses. It is unclear, however, whether this should be interpreted as evidence of more criminal activity on the part of the foreign-born. Most commitments for minor offenses are for things like vagrancy and drunkenness, and arrests and prosecutions of such offenses depend greatly on the choices made by law enforcement officials. These choices will vary greatly across jurisdictions and even within jurisdictions across population groups. Whether these offenses result in incarceration will also depend on the economic and social resources of the offender.
The age profile of incarceration for minor offenses runs counter to the standard adjustment and “culture conflict” theories that predict elevated crime rates among immigrants. The prison commitment rates for these offenses are highest for men in their 40s, most of whom were not recent arrivals to the United States. The higher rates of incarceration for minor crimes also carried over to the so-called second generation of immigrants who were born and raised in the United States.
When the focus turns to major crimes, the gap between the native- and foreign-born narrows dramatically. For 1904, the prison commitment rates by age for the two nativity groups are quite similar, with the exception of 18- and 19-year-olds. This exception is noteworthy, though, because almost half of the foreign-born in this age group were recent arrivals in the United States. This, together with the finding that recent arrivals were disproportionately represented among prison commitments for major offenses, is at least suggestive evidence that adjustment or culture conflict issues were a factor in this period.
By 1930, the foreign-born were less likely than natives to be incarcerated for more serious crimes, as evidenced by their lower commitment rates at every age to state and federal facilities. This change from 1904 may reflect the impact of changes in immigration law and their impact on the selection of immigrant arrivals, or perhaps more likely, the sharp drop-off in the numbers of those arrivals and hence the much smaller share of recent arrivals in the foreign-born population. The threat of deportation for criminal behavior, another change in immigration law in this period, may also have had a deterrent effect on immigrant crime. A particularly interesting finding, though, is that the lower rate of incarceration for the foreign-born is due entirely to this group’s lower rate of incarceration for nonviolent crimes. Incarceration rates for violent crimes were very similar for the two nativity groups for all ages.
Aggregation bias and the absence of accurate population data meant that analysts at the time missed these important features of the immigrant-native incarceration comparison. A key contribution of this article is the demonstration of how differences in the age distributions between immigrants and natives could distort perceptions of relative incarceration rates. We show that aggregation bias plagues the data even for periods in which immigrant arrivals were high. The large differences by age in incarceration risk can cause even small differences in age distributions across populations to have large effects on aggregate incarceration rates. Although aggregation bias will most acutely affect comparisons of immigrants and natives on measures related to crime, due to the steep age profile, it will also affect to varying degrees the comparisons of immigrants and natives on any measure for which there are strong age effects. And as demonstrated in the last section of the article, age effects must also be considered in any comparison of immigrants by country of birth.
With the more complete population data available to us today, we have shown that the relative decline of the incarceration of the foreign-born is really a story about a growing gap between natives and immigrants at older ages. This growing gap was driven by sharp increases in the commitment rates of the native-born, while commitment rates for the foreign-born were remarkably stable. None of these features were apparent in the aggregate crime rates that provided the empirical basis for the policy debate at the time.
The authors thank Rutgers University’s Research Council for financial support of this project. We also thank the participants of the NBER Summer Institute, Northwestern University Economic History Workshop, the Western Economic Association annual meeting, the University of Maryland Crime and Population Dynamics Workshop, and the University of Colorado Economics Seminar for their helpful comments and suggestions.
1.A growing research literature about crime and immigration in late twentieth-century United States is finding, using a variety of data and methods, that immigrants today generally have lower rates of crime than natives (Butcher and Piehl 1998, 2007; Hagan and Palloni 1999; Martinez and Rosenfeld 2001; Sampson, Morenoff, and Raudenbush 2005).
2.Judges and juries surely were swayed by nativist views, but such views likely had a larger impact on arrests because arrests are dominated by low-level incidents with wide discretion. We do not have direct evidence on bias in punishment, but some suggestive support comes from a number of studies that found that court outcomes did not vary by nativity (Lane 1989:71; National Commission on Law Observance and Enforcement 1931:171).
3.Earlier efforts were hampered by the failure to define clearly the population of interest and by incomplete data collection (U.S. Department of Commerce 1926:5).
4.All of these censuses were restricted to individuals who had been “sentenced.” Individuals who were detained in facilities awaiting trial or sentencing were not included in the enumeration of the prison population or commitments.
5.Results available from the authors.
6.The analyses presented in this article treat missing data on nativity as being random. We believe that this assumption does not greatly bias our results, and if it does, probably disadvantages the foreign-born relative to the native-born. For commitments to state facilities, information on nativity is missing in less than 0.2% of cases. The main source of missing nativity data seems to be lax recordkeeping for high-volume, minor offenses (U.S. Department of Commerce 1918:123).
7.We adjust for mortality using the age-specific death rates for white males in 1910 in death registration states presented in Linder and Grove (1947:186, Table 9). We tried a series of alternative mortality rates including the nativity-specific death rates for 1910, the death rates for 1900 and 1920, and the death rates for 1910 multiplied by a factor of 1.5 or 0.5. Using these alternative mortality rates had little effect on the results. For no age category did it change the sign of the difference between the commitment rates of natives and the foreign-born.
8.For the 1923 estimates, we adjust for mortality using the age-specific death rates for white males in 1930 in death registration states as presented in Linder and Grove (1947:186, Table 9).
9.In 1923, data on commitments were collected for only the first 6 months of the year, whereas the data for 1904 and 1930 were collected for the entire calendar year. We doubled the numbers reported in the census in our calculations of the rates in Table 2 to make them comparable.
10.Since the population data used in the denominator of the 1904 commitment rates are constructed from samples, the presented rates are estimates that are subject to sampling error. The sampling error bounds are larger for the immigrant data than for the native due to the smaller numbers of immigrants in the samples, but for both groups the bounds are small. For instance, the 95% confidence interval for the 20- to 24-year-old rate is 139–142 for natives and 156–165 for immigrants. Most importantly, the confidence intervals for the commitment rates of young immigrants and natives do not overlap.
11.Mexicans were separated from whites in the penal institution data starting with the 1926 census of state and federal prisons.
12.Mexican-born males accounted for approximately 5% of the foreign-born male population in 1930 (U.S. Department of Commerce 1933:577, Table 8, and 586, Table 15).
13.The rise in the commitment rate for natives could reflect an increase in criminal activity, but it could also reflect changes in reporting and investigation practices (Lane 1989).
14.To construct these population estimates, we used published data from the 1900 census (U.S. Census Office 1902: Table XVI, pp. xxxvi–xxxix) and aged the population forward to 1904. We adjusted for mortality using the age-specific death rates for white males in 1900 in death registration states as presented in Linder and Grove (1947: Table 9, p. 186). We also calculated commitment rate data for natives by parentage using population estimates constructed from the 1910 IPUMS as for the previous analyses of the 1904 data. These data exhibited the same patterns and yielded the same conclusions as those presented in the article.
15.These age distributions were constructed from the 1910 IPUMS data set as described above.
16.Taylor raised such concerns relating to the data on Mexicans presented in the Wickersham Commission report. Taylor argued that the tendency of prison officials to classify persons of Mexican ancestry as simply “Mexicans” likely would have been offset by foreign-born individuals falsely claiming U.S. nativity to avoid deportation. However, given immigration law in 1904, foreign-born individuals would not have had such an incentive to misreport their nativity (National Commission on Law Observance and Enforcement 1931:200–201).
17.The data on manufacturing wages come from a survey conducted in 1908 as part of the Dillingham Commission. The other data were calculated using the 1910 IPUMS data set.