The Effect of Social Connectedness on Crime: Evidence from the Great Migration∗ Bryan Stuart University of Michigan [email protected]

Evan Taylor University of Michigan [email protected]

September 22, 2015

Abstract The Great Migration of millions of African Americans out of the South generated considerable variation in social connectedness across destinations. We use this variation to study the effect of social connectedness on crime from 1960-2009 across U.S. cities. Our estimates imply that a one standard deviation increase in social connectedness leads to a 12 percent reduction in the murder rate. Estimated effects are driven by cities with a high African American population share. Cities with different levels of social connectedness saw dramatically different changes in crime rates over time.



For helpful conversations and feedback, we thank Martha Bailey, Dan Black, John Bound, Charlie Brown, John DiNardo, Daniel Nagin, Seth Sanders, Jeff Smith, and Lowell Taylor; plus seminar participants at the University of Michigan.

1

Introduction

For almost 300 years, the enormous variance of crime rates across different places has puzzled social scientists and policy makers (e.g., Guerry, 1833; Quetelet, 1835). Even when accounting for differences in observed demographic and economic characteristics across space, crime rates vary tremendously (Glaeser, Sacerdote and Scheinkman, 1996). Motivated in part by this puzzle, a number of papers examine the role of social connectedness in accounting for city and neighborhood crime rates (e.g., Glaeser, Sacerdote and Scheinkman, 1996; Sampson, Raudenbush and Earls, 1997; Sampson, Morenoff and Earls, 1999). The critical challenge in assessing this hypothesis is finding plausibly exogenous variation in social connectedness across different locations. Commonly used measures, such as participation in civic or political organizations or responses to survey questions about cooperation, might be affected by crime rates, causing an important endogeneity problem. In this paper, we study the relationship between crime and social connectedness across U.S. cities from 1960-2009. Our key contribution is use of variation in social connectedness across otherwise similar cities driven by the twentieth century Great Migration of African Americans out of the South. For example, consider Decatur, Illinois and Albany, New York, two cities which are similar in the number of Southern black migrants they received, 1980 population, and 1980 African American population share.1 Nearly forty percent of Decatur’s migrants came from Brownsville, Tennessee, while the second and third largest sending birth towns account for roughly two percent each of all migrants. Albany, on the other hand, did not receive a comparably large share of its migrants from any birth town. The top three sending birth towns to Albany each account for about 2 percent of migrants. Qualitative evidence suggests that the concentration of Brownsville migrants translated into stronger community ties: even in the mid 2000’s, Brownsville high school reunions were held in Decatur (Anonymous, 1994; Smith, 2006). The small size of Brownsville (3,000 total residents in 1920) and other Southern towns makes it more likely that migrants would have known each other. We measure social connectedness using a Herfindahl-Hirschman Index of birth town 1

We provide details below.

1

to destination city population flows. Our ability to measure these long-run population flows comes from use of the Duke SSA/Medicare dataset. Variation in social connectedness comes from the tendency to follow previous migrants from one’s birth town, a central feature of the Great Migration (Stuart and Taylor, 2014). Importantly, we use variation in social connectedness conditional on the total number of migrants living in a destination. Historical context supports our central assumption that this variation is plausibly exogenous with respect to unobserved determinants of crime in 1960-2009, conditional on a variety of economic and demographic controls. Rich qualitative work suggests that early location decisions, many made in the 1910’s, were driven by employment opportunities (Scott, 1920; Bell, 1933; Rubin, 1960; Gottlieb, 1987; Grossman, 1989). If anything, it seems plausible that individuals living in a destination with their close friends and family were less likely to out-migrate in response to crime, making it more difficult to find a negative effect of social connectedness on crime. Our measure of social connectedness is positively correlated with the share of a destination’s work force employed in manufacturing, a relatively attractive sector for African American migrants (see Stuart and Taylor, 2014). Our measure of social connectedness is not correlated with crime rates from 1911-1914, before the Great Migration, or in a consistent manner with economic or demographic covariates from 1960-2000. To guide our empirical analysis, we develop a simple model in which an individual’s decision to commit crime depends on a location-specific payoff, social connectedness, and peers’ actions. The model highlights that the effect of social connectedness among African Americans with ties to the South on city-level crime rates depends importantly on the nature of peer effects. The first prediction of the model is that if social connectedness decreases crime committed by blacks with ties to the South and cross-group peer effects are not sufficiently negative, then social connectedness decreases the equilibrium city-level crime rate. Second, under plausible peer effect specifications, the effect of social connectedness on the city-level crime rate is larger in magnitude in cities with a higher African American population share. We find that higher social connectedness causes considerably lower crime rates. At the mean,

2

a one standard deviation increase in social connectedness leads to a precisely estimated 12 percent decrease in murder, the best measured crime in FBI data. The effect of replacing social connectedness in Albany with that of Decatur is a 36.6 (6.8) percent decrease in murders, a 50.3 (5.2) percent decrease in robberies, and a 35.8 (7.9) percent decrease in motor vehicle thefts. The elasticity of crime with respect to social connectedness ranges from -0.03 to -0.24 across the seven index crimes of murder, rape, robbery, assault, burglary, larceny, and motor vehicle theft, and is statistically distinguishable from zero for six of the seven crimes. The negative effect of social connectedness on crime is driven by cities with a relatively high black population share. Our empirical estimates agree with the theoretical model’s predictions. Consistent with our proposed causal mechanism, social connectedness has a particularly strong negative effect on murders attributed to African American youth, a modest negative effect on murders attributed to black adults, and much smaller and imprecisely estimated effects on murders attributed to whites. Social connectedness has large and important effects on the evolution of crime rates from 19602009. Social connectedness appears to have little effect in the 1960’s, a period of substantial migration and relatively low crime rates. From the 1970’s to 1990’s, a period of high crime rates, the effect of moving from the 75th to 25th percentile of social connectedness is an increase in murder rates of 26-56 percent. The effect of social connectedness fades in the 2000’s, a period of relatively low crime and several decades after the end of the Great Migration. One violation of our identification assumption is that connected groups of migrants tended to move to cities with low unobserved determinants of crime which persist over time; this could lead to a spurious negative correlation between social connectedness and crime. If this selection were quantitatively important, then controlling for the crime rate from 1960-1964 would affect estimates for crime after 1965. Controlling for the 1960-1964 crime rate has virtually no effect on our estimates, allaying concerns about this violation. In addition, our results are robust to different approaches of handling possibly missing data and different sample selection rules. Our theoretical model provides a mapping between the estimated effect of social connectedness on city-level crime and the effect of social connectedness on crime committed by African

3

Americans with ties to the South, which is unobserved. Using a parametrized model, we find a large range of possible values for the latter effect. This uncertainty arises because of uncertainty about the magnitude of peer effects and the large influence of peer effects on this mapping. Our study relates closely to papers which use social structure to explain neighborhood-level crime rates within a city (e.g., Sampson, Raudenbush and Earls, 1997; Sampson, Morenoff and Earls, 1999), drawing upon earlier work by Coleman (1990) and others. We differ from this literature in using variation in social connectedness across cities, as opposed to within a city. We also differ in our use of plausibly exogenous variation in social connectedness from the Great Migration. Finally, our setting allows us to study the effect of social connectedness over the course of 50 years. A number of papers examine the role of peers’ criminal behavior (Case and Katz, 1991; Glaeser, Sacerdote and Scheinkman, 1996; Ludwig, Duncan and Hirschfield, 2001; Calv´o-Armengoi and Zenou, 2004; Silverman, 2004; Kling, Ludwig and Katz, 2005; Ludwig and Kling, 2007; Bayer, Hjalmarsson and Posen, 2009; Sciandra et al., 2013; Damm and Dustmann, 2014). The peer effects studied in these papers are important for our setting, but are not our main focus. Instead, we estimate the effect of stronger social ties on criminal activity. Our paper also relates to the literature examining the consequences of ethnic and racial segregation (e.g., Borjas, 1995; Cutler and Glaeser, 1997; Cutler, Glaeser and Vigdor, 1999; Alesina, Baqir and Easterly, 1999; Alesina and Ferrara, 2000; Bertrand, Luttmer and Mullainathan, 2000). For almost a century, economists have studied the timing of the Great Migration and outcomes for migrants, plus sending and receiving areas (e.g., Scroggs, 1917; Smith and Welch, 1989; Carrington, Detragiache and Vishwanath, 1996; Collins, 1997; Boustan, 2009, 2011; Hornbeck and Naidu, 2014; Black et al., forthcoming). Recent work has emphasized the importance of social interactions in understanding the Great Migration (Chay and Munshi, 2013; Stuart and Taylor, 2014). More broadly, there is enormous interest in the causes and consequences of criminal activity and incarceration in U.S. cities, especially for African Americans (e.g., Freeman, 1999; Evans, Garthwaite and Moore, 2014; Neal and Rick, 2014)

4

2

Historical Background

In the Great Migration, nearly six million African Americans left the South from 1910 to 1970 (Census, 1979). Although migration was concentrated in certain cities, like Chicago, Detroit, and New York, other cities also experienced dramatic changes. For example, Chicago’s black population share increased from two to 32 percent from 1910-1970, while Racine, Wisconsin experienced an increase from 0.3 to 10.5 percent (Gibson and Jung, 2005). Migration out of the South increased from 1910-1930, slowed during the Great Depression decade, and then resumed forcefully until the 1970s. Figure 1 uses Census data to show how the number of African Americans in the North evolved during the twentieth century.2 As seen in panel A, the number of Southern-born African American adults living in the North increased from 300,000 in 1910 to 1.2 million in 1940, then increased to nearly 3.5 million by 1980. Panel B shows that the number of black children living in the North increased dramatically, from less than one million in 1940 to nearly three million in 1960 and over four million after 1970. The number of African American children with at least one parent born in the South follows a similar pattern as the total number of children until 1980. Relatively few of the black children living in the North were themselves born in the South, indicating low migration by children. As time passed, the earliest migrants, born in the South, had children born in the North, who eventually became adults in the North and had children there. Census data allow only an imperfect accounting for the share of African Americans with direct, first-, or second-generation ties to the South, but this number is certainly large; in 1900, nearly 90 percent of all African Americans lived in the South. Factors which contributed to the mass exodus of African Americans from the South include an increase in labor demand and decrease in immigrant labor supply in Northern cities with the onset of World War I (Scroggs, 1917; Scott, 1920; Gottlieb, 1987; Marks, 1989; Jackson, 1991; Collins, 1997), a decrease in labor demand in agriculture due to the boll weevil’s destruction of cotton crops 2 For figure 1, the South consists of the former Confederate states, while the North consists of all other states. The former Confederate states are Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, Texas, and Virginia.

5

(Scott, 1920; Marks, 1989, 1991), and generally better economic and social opportunities outside the South. Migrants tended to follow vertical routes established by railroad lines, e.g., Mississippi-born migrants predominantly moved to Illinois, and South Carolina-born migrants predominantly moved to New York (Scott, 1920; Carrington, Detragiache and Vishwanath, 1996; Collins, 1997; Boustan, 2011; Black et al., forthcoming). Labor agents, who offered paid transportation, employment, and housing, directed some of the earliest migrants, but their role diminished sharply after the 1920s (Gottlieb, 1987; Grossman, 1989).3 Qualitative historical accounts and recent quantitative work indicate that social interactions strongly affected location decisions. Initial migrants, who often migrated before World War I, chose their location primarily in response to economic opportunity. Migrants connected workers from their hometown to jobs and shelter in the North, sometimes leading to persistent population flows from the birth town to the destination city (Rubin, 1960; Gottlieb, 1987; Stuart and Taylor, 2014). For example, migrants from Houston, Mississippi had close friends or family at two-thirds of all initial destinations. Among these migrants, the most important destination characteristics were employment opportunities and the presence of family (Rubin, 1960). Using data on several hundred thousand migrants, Stuart and Taylor (2014) find that birth town-level social interactions had very large impacts on the location decisions of African American migrants.4 These social interactions mirror vertical migration patterns, established by railroad lines, and were stronger in destination cities with more manufacturing employment, a particularly attractive sector for black workers. The experience of John McCord, born in Pontotoc, Mississippi, captures many important features of early black migrants’ location decision.5 In search of higher wages, nineteen-year-old McCord traveled in 1912 to Savannah, Illinois, where a fellow Pontotoc-native connected him with a job. Two years later, McCord moved to Beloit, Wisconsin after hearing of opportunities there. 3

Gottlieb (1987) finds evidence of labor agents bringing Southern blacks to work in Pittsburgh only in the years 1916-19 and 1922-23. Grossman (1989) argues that the role of labor agents in Chicago had diminished by 1917. 4 See also Chay and Munshi (2013). 5 The following paragraph comes from work by Bell (1933), as discussed by Knowles (2010).

6

Within a week, he started as a janitor at Fairbanks Morse and Company, a manufacturer. After two years in Beloit, McCord spoke to his manager about returning home for a vacation. The manager, needing more labor, asked McCord to recruit workers during his trip. McCord returned with 18 unmarried men, all of whom worked for Fairbanks Morse and Company. Thus began a persistent flow of African Americans from Pontotoc to Beloit: among individuals born from 1916-1936, 14 percent of migrants from Pontotoc lived in Beloit’s county at old age.

3 3.1

A Simple Model of Crime and Social Connectedness Set-up

There are two generations of agents. For simplicity, we consider a static model in which each member of the younger generation makes a single decision about whether to commit crime or not, while members of the older generation do not commit crime. There are three types of agents: blacks with ties to the South (τ = bs), blacks without ties to the South (τ = bn), and all others (τ = w). Members of the older generation have a tie to the South if they were born there. Members of the younger generation have a tie to the South if at least one of their parents, members of the older generation, was born in the South. We index members of the younger generation by i and older generation by o, and we focus on a single geographic area. Let Ci = 1 if person i commits crime and Ci = 0 otherwise. Suppose the probability that person i of type bs commits crime is

E[Ci |τi = bs, j(i) = j] = αbs +

X

bs γi,o,j + E[g bs (C−i )],

(1)

o

where j(i) denotes the birth town of i’s parents. The probability of committing crime depends on three things. First, all individuals of type bs face the common element αbs , which captures all non-social determinants of crime (e.g., due to police, employment opportunities, climate). Second, bs the intergenerational influence of older agent o on individual i is γi,o,j , reflecting the role of parents

and other adults in discouraging crime. Finally, peer influences are captured by g bs (C−i ), where 7

C−i is the vector of actions of all agents besides i; the expected peer influence for an individual from birth town j is E[g bs (C−i )] ≡ gjbs . The functional form assumptions embedded in equation (1) help motivate the measure of social connectedness used in our empirical analysis, but are not necessary for the main theoretical results. We model intergenerational influence (or social connectedness) as a function of whether the bs bs parents of person i share a common birth town with person o. In particular, we let γi,o,j = γH if bs the agents share a birth town, j(i) = j(o), and γi,o,j = γLbs otherwise, where j(o) denotes the birth

town of o. The aggregate intergenerational influence experienced by person i is a weighted average bs of the high connectedness effect (γH ) and the low connectedness effect (γLbs ),

X

bs γi,o,j

o

bs Nj,0 bs + = bs γH N0

bs Nj,0 1 − bs N0

! γLbs ,

(2)

bs where Nj,0 is the number of elders of type bs from birth town j which live in a city, and N0bs = P bs j Nj,0 is the total number of elders. If the strength of intergenerational influence depends on bs , lead to differences birth town connections, then previous migration decisions, embedded in Nj,0

in expected crime rates for agents from different birth towns. The probability that a randomly chosen person of type bs commits crime is

E[Ci |τi = bs] =

X j

bs Nj,1 N1bs

! E[Ci |τi = bs, j(i) = j],

bs where Nj,1 is the number of type bs youth with a connection to birth town j, and N1bs =

(3)

P

j

bs Nj,1

bs bs is the total number of youth. If we assume that (Nj,0 /N0bs ) = (Nj,1 /N1bs )∀j, then we obtain

bs E[Ci |τi = bs] = αbs + γLbs + (γH − γLbs )HHIbs + g¯bs ,

(4)

P bs where HHIbs = j (Nj,0 /N0bs )2 is the Herfindahl-Hirschman Index and the average peer effect is P bs g¯bs = j (Nj,1 /N1bs )gjbs . Similar relationships exist for African American youth without ties to the South and white youth. 8

HHI emerges as a natural way to measure intergenerational influence, or social connectedness, bs at the city level. The direct effect of higher HHI on the type bs crime rate is γH − γLbs . One bs reasonable case is γH < γLbs < 0, so that elders discourage youth from committing crime, and

the effect is stronger among youth and elders linked through a common birth town. Because peer effects can augment (or weaken) the intergenerational effect, we now examine the equilibrium of this model.

3.2

Equilibrium and Main Results

We use HHI to measure social connectedness and assume that peer effects depend on the average crime rate, leading to the following equilibrium, C¯ bs = F bs (αbs , HHIbs , C¯ bs , C¯ bn , C¯ w )

(5)

C¯ bn = F bn (αbn , HHIbn , C¯ bs , C¯ bn , C¯ w ) C¯ w = F w (αw , HHIw , C¯ bs , C¯ bn , C¯ w ) where C¯ τ is the average crime rate among group τ youth, and F τ is an unknown function. The equilibrium crime rate vector C¯ all ≡ [c]0 is a fixed point of equation (5). We are interested in the effect of HHIbs on equilibrium crime rates. It is straightforward to show that 

  bs 0 dC¯ all −1 bs ∂F /∂HHI 0 0 , = (I − J) dHHIbs

(6)

where I is the 3 × 3 identity matrix and J, a sub-matrix of the Jacobian of equation (5), captures the role of peer effects, D We assume the equilibrium is stable, which essentially means that peer effects are not too large.6 For example, if ∂F bs /∂ C¯ bs > 1, and there are no cross-group peer effects, then a small increase (decrease) in the crime rate among group bs leads to an equilibrium where all (no) members of group bs commit crime. In a stable equilibrium, a small change in any 6

The technical assumption underlying stability is that all eigenvalues of J lie in the interval [0, 1).

9

group’s crime rate does not lead to a different equilibrium. The effect of social connectedness on crime depends importantly on the nature of peer effects. We now discuss key results which hold for what we think are plausible models of peer effects, given our priors and existing empirical evidence.7 The first result is that the negative effect of social connectedness on crime extends to a setting with peer effects, as long as the equilibrium is stable and cross-group peer effects – i.e., off-diagonal elements of J – are non-negative.   Result 1. dC¯ all /dHHIbs ≤ 0 if ∂F bs /∂HHIbs < 0, the equilibrium is stable, and cross-group peer effects are non-negative. In a stable equilibrium with non-negative cross-group peer effects, the crime-reducing effect of social connectedness among Southern blacks is not counteracted by higher crime among other groups. Hence, equilibrium crime rates weakly decrease in Southern African American HHI. With negative cross-group peer effects, the reduction in crime rates among Southern blacks could lead to higher crime by other groups. ¯ which is a weighted average Most of our empirical work examines the city-level crime rate, C, of three group-specific crime rates, C¯ = P b [P s|b C¯ bs + (1 − P s|b )C¯ bn ] + (1 − P b )C¯ w ,

(7)

where P b is the black population share and P s|b is the share of the black population with ties to the South. Result 1 is a sufficient, but not necessary, condition to ensure that the city-level crime rate decreases in Southern black HHI. There exist situations in which cross-group peer effects are negative, but an increase in HHIbs leads to a decrease in the equilibrium city-level crime rate. The second result states that the effect of social connectedness on the city-level crime rate decreases (or, increases in magnitude) with the black population share for certain peer effect models. bs ¯ Result 2. dC/dHHI decreases with P b if the equilibrium is stable and cross-group peer effects

are non-negative and sufficiently small. 7

Throughout, when we refer to different models of peer effects, we mean different parametrizations of J.

10

Assume that the effect of HHIbs on crime does not depend on the black population share, yielding8 ¯ bs ¯ bn dC¯ w d2 C¯ s|b dC s|b dC = P + (1 − P ) − . dP b dHHIbs dHHIbs dHHIbs dHHIbs

(8)

Two jointly sufficient conditions for result 2 are (a): dC¯ bs /dHHIbs < dC¯ w /dHHIbs and (b): dC¯ bn /dHHIbs < dC¯ w /dHHIbs . If Southern black social connectedness leads to greater crime reductions among both groups of black youth, relative to white youth, then the total effect is larger (in magnitude) in cities with a higher share of black youth. The nature of peer effects determines whether condition (a) and (b) are satisfied. As a simple example, suppose there are no cross-group peer effects between black and white youth (J13 = J23 = J31 = J32 = 0). As a result, an increase in HHIbs does not affect the expected crime rate among white youth, ensuring that condition (a) holds. For condition (b) to hold, an increase in HHIbs must lead to a decrease in crime among non-Southern black youth, which will be true if peer effects between the two groups of black youth are non-negative. As shown in appendix A, the formal conditions in this example are a stable equilibrium and J21 > 0. To summarize, we expect that higher social connectedness among African Americans with ties to the South will lead to lower city-level crime rates (result 1). We also expect that the effect will be stronger in cities with a higher black population share (result 2). The relationship between the direct effect of HHIbs on crime by African Americans with ties to the South and the aggregate effect of HHIbs on city-level crime depends on the nature of peer effects, which we examine in section 7. If we relaxed this assumption, it is not clear whether we would expect, say, dC¯ bs /dHHIbs to become more or less negative in response to higher P b . The effect might decrease in magnitude (move towards zero) if the higher black population share diluted existing community ties. Alternatively, the effect might increase in magnitude if the higher black population share reinforced community ties. The former case makes result 2 less likely to hold, while the latter case makes it more likely. 8

11

4

Data

We use three different data sets in our empirical analysis. First, we measure annual city-level crime using FBI Uniform Crime Report (UCR) data for 1960-2009, available from ICPSR. UCR data contain voluntary monthly reports from individual police agencies, which we aggregate at the city-year level. Our measure of crime is the number offenses reported to police. We focus on the seven index crimes: murder and non-negligent manslaughter (“murder”), forcible rape (“rape”), robbery, assault, burglary, larceny, and motor vehicle theft. Murder is the best measured crime, with robbery and motor vehicle theft also relatively well-measured (Blumstein, 2000; Tibbetts, 2012). The UCR data also contain annual population measures from the Census Bureau.9 Starting in 1980, information on the age, sex, and race of offenders are available for index crimes resulting in arrest; we refer to this as “ASR” data below. We construct annual crime counts by summing across the different months in a year, as in Chalfin and McCrary (2014).10 Because missing observations are indistinguishable from true zeros, we drop any city-year in which any of the three property crimes (burglary, larceny, and motor vehicle theft) equal zero. We also show that our results are robust to treating zeros as Chalfin and McCrary (2014) do. Second, we use the Duke SSA/Medicare dataset to measure social connectedness. The data contain sex, race, date of birth, date of death (if deceased), and the ZIP code of residence at old age (death or 2001, whichever is earlier) from the Medicare Part B Master Beneficiary Records. In addition, the data contain individuals’ birth town from the SSA NUMIDENT file. The Duke data are ideally suited to measuring long-run birth town-to-destination city population flows, which form the basis of our social connectedness measure. We focus on individuals born from 1900-1936 in the former Confederate states.11 Finally, we use Census city data books from 1960, 1970, 1980, 1990, and 2000 to measure city9 There is considerable measurement error in annual city-level population. Our results are not sensitive to replacing observed population with predicted values from a flexible city-specific regression model. 10 We follow Chalfin and McCrary (2014) in decreasing the number of murders for year 2001 in New York City by 2,753, the number of victims of the September 11 terrorist attack. 11 Coverage rates are very high for African Americans born in the South from 1916-1936 (Black et al., forthcoming; Stuart and Taylor, 2014). We focus on African American migrants born from 1900-1936, as the older cohorts were important in establishing persistent migration patterns which generate variation in social connectedness.

12

level covariates, described in further detail below. These data are only available for cities with at least 25,000 residents in years 1960, 1980, and 1990; we apply the same restriction for years 1970 and 2000. To be in our sample, a city must have received at least 25 Southern-born African American migrants (as measured in the Duke data) and have crime and demographic data for a given year. We focus on cities in the Northeast, Midwest, and West Census regions. We use Federal Information Processing System (FIPS) place definitions throughout. In our main analysis, we exclude the 14 cities with 1980 population greater than 500,000, as we found considerable measurement error in murder counts for these cities.12 Furthermore, our measure of social connectedness does not vary much across the largest 14 cities (see figure 2, discussed below).

5

Empirical Approach

5.1

Estimation

Motivated by the model in section 3, we estimate the effect of social connectedness on city-level crime rates (result 1) and whether this effect is stronger in cities with a higher African American population share (result 2). We measure social connectedness with a Herfindahl-Hirschman Index, P 2 HHIk = j (Nj,k /Nk ) , where Nj,k is the number of migrants from birth town j which move P to destination city k, and Nk ≡ j Nj,k is the total number of migrants. HHIk is a natural way to measure social connectedness, as shown in section 3, and approximately equals the probability that two randomly chosen migrant-residents of city k share a birth town.13 Because our regression models include log HHI, log number of migrants, and log city population, our empirical results are identical when using city population as the denominator of HHI. In addition, to isolate social 12

In particular, for 1980-1989, we constructed annual murder counts using the UCR data, which are not broken down by age, race, or sex, and the ASR data, which are. In principle, both data sets should yield the same number of murders in a city. However, there are substantial discrepancies in the largest cities, as shown in appendix figure A.1. 13 The probability that two randomly chosen migrants living in city k come from the same birth town is X P[j(i) = j(i0 )] = P[j(i) = j(i0 )|j(i0 ) = j] P[j(i) = j] j

=

X Nj,k − 1 Nj,k j

Nk − 1 Nk

13

≈ HHIk

connectedness arising from a single birth town, we use the leading term of HHIk , which equals the squared migrant share of the top sending town, i.e., (N1,k /Nk )2 , where N1,k = maxj {Nj,k }. We view HHI as a useful proxy for the unobserved level of social connectedness.14 Variation in HHI, conditional on the number of migrants, is driven by migrants coming from fewer birth towns. Qualitative work provides strong evidence that birth town social ties had substantive, long-lasting effects on life in destination cities. For example, a news account from 2002 states, “Fort Wayne is Marion, Alabama, once removed” (Crowder and Spencer, 2002). Indeed, the Duke data show a strong flow of migrants from Marion, Alabama to Fort Wayne, Indiana (Stuart and Taylor, 2014, Table 2). Roughly 1,000 of Erie, Pennsylvania’s 11,600 African American residents once lived in Laurel, Alabama, while nearly half had family connections to Laurel, leading an Erie resident to say, “I’m surrounded by so many Laurelites here, it’s like a second home” (Anonymous, 1983).15 In the Duke data, a higher share of Erie black residents were born in Laurel than in Erie. The same article describes an active Erie chapter of a Laurel high school alumni association; migrants from Brownsville, Tennessee established a similar organization in Decatur, Illinois, as described in the introduction. Our primary estimating equation is,

0 Yk,t = exp[ln(HHIk )δ + ln(Nk )θ + Xk,t β] + k,t ,

(9)

where Yk,t is the number of crimes in city k in year t, HHIk and Nk are defined above, Xk,t is a vector of observed covariates which potentially explain the level of crime in a city, and k,t captures unobserved determinants of crime. We describe Xk,t in detail below. We estimate Poisson regression models because many cities report no murders in a given year. When estimating equation (9), we cluster standard errors at the city level to allow for arbitrary autocorrelation in the unobserved 14

Appendix table A.3 displays the relationship between log HHI and measures of social capital, available at the county-level using 1990 data, from Rupasingha and Goetz (2008). The raw correlations between log HHI and various measures of social capital are positive, but small and indistinguishable from zero. After controlling for the log number of migrants and state fixed effects, these correlations shrink even further. 15 The article also states that, in the 1930’s and 1940’s, a group of ministers traveled between the two locations, providing information on job opportunities in Erie.

14

determinants of crime.16 Table 1 displays summary statistics for the 19,471 city-years in our sample. Around 19 percent of city-years have zero murders. The average population size is 92,000. The average number of Southern black migrants born from 1900-1936, measured in our Duke data, is 667; this number is not directly comparable to Census population counts because the Duke data contain only one-third of individuals from these birth cohorts alive in 1960.17 On average, the top sending birth town accounts for six percent of all migrants observed in the Duke data.

5.2

Identification

The key parameter of interest in equation (9) is δ, which can be interpreted as the elasticity of the crime rate with respect to HHI, our measure of social connectedness, because we include log population as a covariate. If social connectedness reduces city-level crime, consistent with result 1 of the model, we expect δ < 0. A sufficient condition to identify δ is

k,t ⊥ ⊥ HHIk |(Nk , Xk,t )

(10)

Condition (10) states that, conditional on the number of migrants which move to k and the vector of observed covariates, HHI is independent of unobserved determinants of crime.18 Two features of the Great Migration inform the plausibility of condition (10). First is the process by which certain destinations initially received significant numbers of migrants from the same birth town. Rich qualitative work suggests that early location decisions, many made in the 1910’s, were driven by employment opportunities and had a considerable effect on later migrants’ location decisions (Scott, 1920; Bell, 1933; Rubin, 1960; Gottlieb, 1987; Grossman, 1989).19 In other 16

We also estimated standard errors clustered at the metropolitan statistical area level. The resulting standard error on δˆ is slightly smaller than when clustering at the city level. 17 The 1960 Census counts 5.7 million African Americans born in the South from 1900-1936. The Duke data contain 2.9 million individuals with these characteristics, but birth place and destination are identified for 1.9 million individuals. 18 Note that condition (10) does not guarantee identification of the other parameters (θ, β). For example, identification of θ requires exogenous variation in the the total number of migrants to each city; Boustan (2011) provides one possible strategy for such an approach. 19 For example, Scott (1920) writes, “The tendency was to continue along the first definite path. Each member of

15

work, we show that town-based social networks had an enormous impact on location decisions, and that these social interactions were strongest in destinations with attractive jobs (Stuart and Taylor, 2014). Appendix table A.1 shows that HHI, measured in the latter part of the twentieth century, is not correlated with homicide rates from 1911-1914, just before the start of the Great Migration. Although underpowered, using data from only 31 of the largest cities, the results are consistent with the qualitative evidence which indicates that early location decisions of connected migrants were not driven by crime. The second issue is the degree to which connected migrants moved out of a city in response to unobserved crime shocks, relative to unconnected migrants. Condition (10) allows the level of migration to respond to unobserved determinants of crime, so that Nk depends on k,t . The key requirement is that connected migrants do not respond differently to k,t than non-connected migrants. If, as seems plausible, connected migrants are infra-marginal with respect to their location decision, then connected migrants would out-migrate less than non-connected migrants in reponse to higher k,t , leading to a higher measure of HHI in cities with worse unobserved crime shocks. This would bias our estimate of δ upwards, making it more difficult to conclude that social connectedness reduces crime. Moreover, as shown in table 2, nearly 90 percent of African American migrants born from 1900-1936 stayed in the same county for the 5 year periods 1955-1960, 1965-1970, 1975-1980, 1985-1990, and 1995-2000. From 1965 on, between 60 and 80 percent of migrants stayed in the same house over a 5 year period. Low mobility rates among these migrants allay some concerns that out-migration after 1960 biases estimates of δ. The plausibility of condition (10) depends on the set of observed covariates Xk,t . The demographic covariates we include are log population; percent black; percent female; percent age 5-17, 18-64, and 65 and older; percent at least 25 years old with a high school degree; percent at least 25 years old with a college degree; and log city area. The economic covariates we include are log methe vanguard controlled a small group of friends at home, if only the members of his immediate family. Letters sent back, representing that section of the North and giving directions concerning the route best known, easily influenced the next groups to join their friends rather than explore new fields. In fact, it is evident throughout the movement that the most congested points in the North when the migration reached its height, were those favorite cities to which the first group had gone” (p. 69).

16

dian family income; the unemployment rate; the labor force participation rate; and manufacturing employment share.20 With only a few exceptions, discussed in the note to table 4, we observe these covariates in each decade from 1960-2000. In explaining crime in year t, we only use covariates corresponding to the decade in which t lies. We allow coefficients for all covariates besides log population to vary across decades. In addition, we include state-by-year fixed effects to account for time-varying shocks to crime or reporting behavior. Table 3 reports regressions of log HHI on the log number of migrants and a variety of city-level covariates for 234 destination cities observed in every decade from 1960 to 2009. To facilitate comparisons, we normalize all variables, separately for each column, to have mean zero and standard deviation one. Column 1 shows that the log number of migrants explains 71 percent of the variation in log HHI, and log HHI and the log number of migrants are negatively correlated. Adding state indicator variables, in column 2, explains more of the variation in HHI but does not affect the relationship between HHI and the number of migrants. Column 3 adds a variety of demographic and economic covariates measured in 1960. The coefficient on manufacturing employment share is positive and statistically significant at the 1 percent level, consistent with the results in Stuart and Taylor (2014). Of the other 12 covariates, two are significant at the 10 percent, but not 5 percent, level. Columns 4-7 display results from using covariates measured in 1970, 1980, 1990, and 2000. Besides the manufacturing employment share, no demographic or economic covariate is significantly related with log HHI across decades. The bottom panel reports p-values from tests that demographic or economic covariates are unrelated to log HHI. We fail to reject this null hypothesis for 1960-1980, but find evidence of a relationship between social connectedness and covariates in 1990 and 2000. Social connectedness might have affected these later outcomes, so we do not view significant correlations in the later period as evidence against our identification assumption. Appendix table A.2 shows results when adding a number of covariates measured among African-Americans at the city-level; there is some evidence that log HHI is positively related to the share of African Americans with a college 20

Stuart and Taylor (2014) find that manufacturing employment share strongly predicts the strength of social interactions in location decisions among Southern black migrants.

17

education, but these relationships are not very precisely estimated. In sum, the weak relationship between social connectedness and almost all city-level covariates is consistent with our description of migrants’ location choices and our identification assumption. To better understand the variation used to identify δ, figure 2 plots log HHI and the log number of Southern black migrants. Our identification strategy exploits variation in HHI conditional on the number of migrants in a city, which is variation in the vertical dimension of figure 2. The negative correlation arises because a large number of sending towns were necessary to generate a large number of migrants, due to the small size of Southern birth towns relative to destination cities. There is very little variation in HHI conditional on the number of migrants for cities with 1980 population above 500,000. Figure 3 shows that most of the variation in social connectedness is driven by a single sending town. Sixty-seven percent of the variation in log HHI is explained by the leading term of log HHI, which equals the log squared percent of migrants from the top sending town. This is consistent with the finding of qualitative work that connected groups of migrants resulted when the right pioneer migrant got the right job at the right time (Scott, 1920; Bell, 1933).

6

Empirical Results

6.1

The Effect of Social Connectedness on Crime Rates

Table 4 illustrates the relationship between social connectedness and the murder rate. To assess the sensitivity of our results to different covariates, we first focus on the 234 cities consistently observed across all five decades; we examine the full sample below. When controlling for demographic covariates and state-by-year fixed effects in column 1, our point estimate (standard error) of δ is -0.193 (0.052). Adding economic covariates in column 2 yields similar results, -0.200 (0.045). The estimate is attenuated somewhat when replacing state-by-year with region-by-year indicators, as seen in column 3.21 Column 4 shows that when only using the leading term of HHI, the estimate is -0.066 (0.022). In sum, we find a statistically significant, negative relationship between social 21

We use four Census regions.

18

connectedness, measured by HHI or its leading term, and the murder rate in cities from 1960-2009. We prefer the specification in column 2, which controls for demographic and economic covariates and includes state-by-year indicator variables. Table 5 shows that the manner in which we control for the number of Southern black migrants, or the size and social connectedness of other population groups, has relatively small effects on the relationship between HHI and crime. Column 1 contains the estimates in column 2 of table 4 to facilitate comparisons. As seen in column 2 of table 5, not controlling for the log number of Southern black migrants leads to an estimate of δ of -0.306 (0.045).22 In column 3, we control for the log number of Southern black migrants, as before, and add four covariates for log HHI and the log number of native white and foreign migrants. The point estimate on African American migrant log HHI is -0.146 (0.047). In column 4, we replace the log number of Southern black migrants variable with ten indicator variables, one for each decile of the Southern black migrant distribution. Controlling more flexibly for the number of migrants has very little effect on the results. Table 6 extends the analysis to all index crimes. Across six of the seven crimes (all except larceny), we find a negative and statistically significant relationship between social connectedness and offenses reported to police. These regressions include 488 cities, some of which are not in the sample across all years. As seen in column 1, our estimate of the elasticity of the murder rate with respect to HHI is -0.153 (0.036). Besides murder, we emphasize results for robbery and motor vehicle theft, which are relatively well measured. The estimates for robbery and motor vehicle theft are -0.235 (0.035) and -0.149 (0.041), relatively close to the result for murder. In sum, we find very strong evidence that higher social connectedness decreases crime, consistent with result 1 of the model.23 To better understand the magnitudes implied by the estimates in table 6, consider two examples which replace a city’s HHI with that from a more connected, but otherwise comparable city. We first consider Middletown, OH and Beloit, WI, which are comparable in terms of the 1980 popula22

For identification purposes, we strongly prefer the specification which controls for the log number of migrants. We estimate this regression to demonstrate that the strong relationship between HHI and the number of migrants does not account for the negative coefficient on log HHI. 23 Appendix table A.4 displays results for all covariates from the regressions in table 6.

19

tion, percent black, and the number of migrants received.24 However, HHI in Beloit (0.057) is over four times as large as in Middletown (0.014). The effect of replacing Middletown’s HHI with that of Beloit is a 19.3 (4.1) percent decrease in murders, a 28.1 (3.6) percent decrease in robberies, and an 18.9 (4.7) percent decrease in motor vehicle thefts. The second example we consider is even more extreme: HHI in Decatur, IL (0.118) is almost twenty times larger than that of Albany, NY (0.006).25 The effect of replacing Albany’s HHI with that of Decatur is a 36.6 (6.8) percent decrease in murders, a 50.3 (5.2) percent decrease in robberies, and a 35.8 (7.9) percent decrease in motor vehicle thefts. These examples demonstrate that social connectedness potentially has very large effects on crime rates.

6.2

The Effect of Social Connectedness on Crime Rates and Black Population Share

Besides predicting a negative effect of HHI on crime, the model in section 3 also predicts a stronger effect in cities with a higher African American population share. Table 7 estimates the baseline specification separately for different terciles of percent black, measured in 1960. Across increasing levels of the black population share, the estimated effect of HHI on the murder rate is -0.026 (0.136), -0.049 (0.044), and -0.189 (0.069). A similar pattern exists for other crimes, including robbery and motor vehicle theft. Across all crimes, point estimates for the lowest percent black tercile are indistinguishable from zero, while point estimates for the highest percent black tercile are negative and statistically significant.26 Figure 4 shows the effect of moving from the 25th to 75th HHI percentile for different percent black terciles. In particular, we estimate a separate Poisson model for each percent black tercile. Using the value of covariates associated with the average crime rate for each tercile, we plot two different predicted murder rates: one for the 75th percentile (HHI = 0.028) and one for the 25th 24

For Middletown and Beloit, the 1980 population is 35,207 and 43,719; the 1980 percent black is 11.3 and 12.0; and the number of Southern black migrants is 376 and 407. 25 For Decatur and Albany, the 1980 population is 94,081 and 101,727; the 1980 percent black is 14.6 and 15.9; and the number of Southern black migrants is 760 and 874. 26 However, standard errors for estimates in the lowest percent black tercile are quite large, and we cannot reject equality of coefficients in the low and high terciles for murder (z = −1.07) or robbery (z = −1.43), but can for motor vehicle theft (z = −2.37).

20

percentile (HHI=0.008). There is no substantive effect of HHI at the lowest percent black tercile. At the middle tercile, increasing HHI across the interquartile range leads to 0.9 fewer murders per 100,000 population (relative to a base of around 5 murders per 100,000); the effect is 3.2 fewer murders per 100,000 population at the highest percent black tercile (relative to a base around 10 murders per 100,000). As seen in table 7, the effects at the two lowest terciles are indistinguishable from zero, while the effect at the highest tercile is different from zero. In sum, the negative effect of HHI on crime appears to be driven by cities with high levels of percent black, consistent with result 2 of the model, though relatively large standard errors temper this conclusion.

6.3

Additional Results: Age and Race of Offender, Temporal Pattern

To this point, we have presented results for city-level crime. Table 8 shows results from 19801989 data on the age and race of offenders for murders resulting in arrest.27 Column 1 shows that the elasticity of the aggregate murder rate with respect to HHI is -0.176 (0.059), similar to the result from the 1960-2009 UCR data. Columns 2-5 examine how Southern black migrant social connectedness affects crimes committed by black and white juveniles and adults. Social connectedness has the strongest effect on murders committed by black juveniles, -0.747 (0.180), and a sizable effect on murders committed by black adults, -0.276 (0.080). Point estimates for whites are considerably lower, -0.117 and -0.018, and indistinguishable from zero, though the standard errors are relatively large. It appears that the relationship between Southern black HHI and crime is strongest for African Americans, consistent with a causal effect of social connectedness on city-level crime. To facilitate discussion of the temporal pattern of the effect of HHI on crime, figure 5 plots the time series of crime rates for all seven index offenses and murder for our sample. The murder rate was relatively flat from 1960-1964, then increased dramatically from 1965-1980 before 27

ASR data quality appears to fall considerably in the 1990s. For example, from 1980-1992, on average, 33 Illinois cities report valid ASR crime data (as determined by non-zero reports of burglary, larceny, and assault). There are 11 such cities in 1993, and 10 in 1994. After 1995, only Chicago reports valid crime data. On the other hand, from 1989 to 1990, there are 52 more cities for which we have crime data. To balance the decline in data quality with the changing sample composition, we focus on 1980-1989.

21

falling from 1981-1984. From 1985-1993, murder increased again, before declining sharply from 1993-2000 and remaining relatively flat thereafter. The pattern of the index crime rate is broadly similar.28 Panel A of figure 6 examines the effect of HHI on the evolution of murder rates. In particular, for each five year period from 1965-1969 to 2005-2009, we estimate a Poisson regression model and take the level of covariates associated with the average crime rate. As discussed in more detail below, we control for the average crime rate from 1960-1964 in these regressions.29 We then plot the murder rate associated with the 75th and 25th percentiles of HHI. By construction, the two series intersect for the period from 1960-1964. There is virtually no effect of HHI from 19651969. In the 1970’s, a period of rising murder rates, the less connected city sees a considerably larger increase in murder. Murder rates remain 26-56 percent higher for the less connected city through the mid 1990s. After 1995, when murder rates declined nationally, the less connected city sees a sharper decline, until murder rates converge in the 2000’s. Panel B shows qualitatively similar results for the motor vehicle theft rate. Individuals most likely to commit murder in 1970 were born around 1950 to mothers born around 1925.30 The individuals affected by social connectedness in the 1970’s are children of postwar migrants and grandchildren or great-grandchildren of the earliest group of migrants. What could explain the insignificant effect of HHI on crime in the late 1960’s? First, there was substantial migration from South to North in the 1950’s and 1960’s, as seen in figure 1, which could explain the lack of effect if exposure to a stable community throughout one’s life is important in reducing crime. Another possibility is that the individuals committing crime in the 1960’s, a period of relatively low crime, were infra-marginal and not susceptible to intergenerational influence. One explanation for the diminishing effect of HHI in the 2000’s is that connectedness dissipated with time, as older generations of migrants died and younger generations moved to different locations. Another possibility is that the individuals who selected into crime during a low crime period 28

Patterns from our sample are comparable to national patterns of UCR data. The time series of crime rates can depend on the data source (e.g., Boggess and Bound, 1997). 29 Figure A.2 displays results when we do not control for the average crime rate from 1960-1964. 30 Among black males, the highest offending rate for murder is between ages 18-24 (Fox, 2000).

22

were not susceptible to social influence. Our results do not distinguish between these different explanations.

6.4

Robustness

If, contrary to our identification assumption (10), connected groups of migrants tended to locate in cities with low unobserved determinants of crime, and these unobserved determinants of crime persisted over time, then our estimate of δ is biased downwards. Figure 7 examines this concern by reporting estimates from models with and without controlling for the 1960-1964 log average crime rate.31 Estimates of the effect of HHI on murder and motor vehicle theft are not affected by controlling for the 1960-1964 crime rate, providing no evidence against our identification assumption. In panel A of appendix table A.5, we include the 14 largest cities which are excluded from the main analysis. The elasticity of HHI with respect to the murder rate increases slightly from -0.153 (0.036) to -0.145 (0.036). There are other minor changes, but the qualitative conclusions from table 6 remain. In panel B, we estimate negative binomial models for the sample excluding the largest cities. Point estimates for all crimes tend to increase. For example, the murder rate elasticity increases to -0.086 (0.033), but is still negative and statistically significant. We prefer the Poisson model because it requires fewer assumptions to generate consistent estimates of δ (e.g., Wooldridge, 2002). The key assumptions are condition (10) and a properly specified conditional mean function. Appendix table A.6 displays results when we exclude arguably extreme crime counts, which could be due to measurement error. Following Chalfin and McCrary (2014), panel A displays results when we exclude years which are less than 1/6 or greater than six times the mean number of crimes for each city. Panel B applies a similar rule, but for each city’s median number of crimes. In both cases, results are similar to those from table 6. Finally, our primary sample only contains cities with at least 25 migrants, measured in the 31

Controlling for the average log crime rate is unattractive because many cities report zero murders in a given year.

23

Duke data, and at least 25,000 residents, measured in Census data. When we only include cities with at least 50,000 residents, the murder rate elasticity decreases from -0.153 (0.036) to -0.190 (0.040). When we additionally limit the sample to cities with at least 50 migrants, the murder rate elasticity is -0.209 (0.042). When we restrict attention to cities with at least 50,000 residents and at least 100 migrants, the estimate is -0.230 (0.043).32 Our key conclusions do not depend on this sample selection rule.

7

Connection between Empirical Estimates and Model

We now use the model of section 3 to connect the estimated effect of HHI on city-level crime to the effect of HHI on crime committed by blacks with ties to the South. Equations (6) and (7) imply that the elasticity of the city-level crime rate with respect to Southern black HHI can be written

  δ = εbs sbs P b (P s|b m11 + (1 − P s|b )m21 ) + (1 − P b )m31 ,

(11)

bs ¯ ¯ is the parameter estimated in our regression models, εbs ≡ where δ ≡ (dC/dHHI )(HHIbs /C)

(∂F bs /∂HHIbs )(HHIbs /F bs ) captures the direct effect of HHI on the Southern black crime rate, sbs ≡ C¯ bs /C¯ is the ratio of the Southern black crime rate to the overall crime rate, P b is the black population share, and P s|b is the share of blacks with ties to the South. Peer effects are captured by (1 − J22 )(1 − J33 ) − J23 J32 det(M ) J23 J31 + J21 (1 − J33 ) = det(M ) J21 J32 + J31 (1 − J22 ) = det(M )

m11 = m21 m31

(12)

where M = I − J, m = M −1 , and the elements of m are given by equation (12). Equation (11) provides a mapping between δ and εbs . Given εbs , we can use equation (6) to solve for the effect of 32

The number of cities in our sample falls from 488 to 186, 158, and 134 with these restrictions.

24

a change in Southern black HHI on crime rates for each group.33 Our central point estimate of δ, reported in table 6 for murder, is -0.153. We set the black population share P b = 0.13 and the fraction of the black population with ties to the South P s|b = 0.6.34 We do not observe the share of crimes committed by blacks with ties to the South. In the ASR data, half of the murders resulting in arrest are attributed to African Americans. If blacks with and without ties to the South commit equal amounts of crime, then sbs = (0.5)/(0.13) ≈ 3.8. We make a number of simplifying assumptions regarding peer effects. First, we assume that own-group peer effects are equal across all three groups.35 Second, we assume that cross-group effects are symmetric in terms of elasticities.36 Third, we assume that cross-group effects between whites and both groups of African Americans are the same. These assumptions can be written as J11 = J22 = J33 , J12 = J21 , J13 = J23 , and J31 = J32 . Letting Eab denote the elasticity form of Jab , these can be written as E11 = E22 = E33 , E12 = E21 , and E13 = E23 = E31 = E32 . We summarize here what existing estimates imply about J and provide more detail in appendix B. Damm and Dustmann (2014) estimate the effect of municipality crime rates on criminal convictions for refugees age 15-21 in Denmark. They find that a one percentage point increase in the local conviction rate leads to a 0.32 percentage point increase in the annual conviction rate, suggesting on-diagonal values of J close to 0.3. Their estimates also suggest off-diagonal elements of J near 0. Estimates from Case and Katz (1991), using data on Boston youth from 1989, suggest on-diagonal values of J close to 0.1. Ludwig and Kling (2007) find no evidence that neighborhood violent crime rates affect violent crime arrests among Moving to Opportunity experimental participants age 15-25. Finally, estimates in Glaeser, Sacerdote and Scheinkman (1996) suggest on-diagonal elements of J close to 0.5 for murder and close to 1 for robbery and motor vehicle theft, among other crimes. 33

¯ bs HHIbs ¯ bn HHIbs dC dC bs ¯ bs = ε m11 , dHHIbs C ¯ bn dHHIbs C HHIbs bs ¯ bs and ε is due to peer effects. C

In particular: ¯ bs

¯ bs

C = εbs m21 C ¯ bn , and

¯ w HHIbs dC ¯w dHHIbs C

¯ bs

= εbs m31 C ¯ w . The difference C

dC between dHHI bs 34 The black population share in our sample is 0.13 in 1980. From the Duke data, the average share of African American migrants born in the South is 0.6. 35 We are aware of no evidence suggesting that own-group peer effects differ for white versus black youth. 36 Given the differences in crime rates between blacks and non-blacks, we believe that assuming symmetric crossgroup elasticities is more realistic than assuming symmetric cross-group linear effects (J).

25

The above estimates are not necessarily comparable to each other or our setting, as they rely on different contexts, identification strategies, data sources, and crime definitions. We posit that reasonable values of on-diagonal elements of J (own-group peer effects) are between 0 and 0.5, and off-diagonal elements of J (cross-group peer effects) are likely small, but could be sizable for African Americans with and without ties to the South. For purposes of illustration, we parametrize the cross-race effects so that elasticities equal 0 or 0.1. Table 9 maps our central estimate, the effect of social connectedness on city-level crime rates, to the effect on crime rates of various groups under different peer effect models.37 In particular, we consider a one standard deviation decrease in social connectedness, which increases the total crime rate by 11.9 percent. This implies an increase in the crime rate of blacks with ties to the South of between 39.8 percent, when there are no cross-group peer effects (column 1), and 16.6 percent, when peer effects operate across all groups (column 7). The crime rate of blacks without ties to the South increases by between 0 and 20.8 percent, while the crime rate of whites increases by between 0 and 8.1 percent. Depending on the parametrization, up to 82 percent of the effect on blacks with ties to the South is driven by peer effects. Given the existing empirical evidence, we place some emphasis on the results in columns 3 and 4, which imply increases in the Southern black crime rate of 33.8 and 24.5 percent, and modest increases in the crime rate of blacks without ties to the South.

8

Conclusion

This paper studies the effect of social connectedness on crime across U.S. cities from 1960-2009. The key contribution is use of variation in social connectedness across otherwise similar cities driven by the Great Migration. Our measure of social connectedness is driven by the tendency to follow the location decisions of previous migrants from one’s birth town. We find that higher social connectedness causes considerably lower crime rates. At the mean, a one standard deviation 37

The chosen values of J are feasible given the stable equilibrium assumptions made. In addition, result 2 holds for all of these parametrizations.

26

increase in social connectedness leads to a precisely estimated 12 percent decrease in murder. We also find evidence that the effect of social connectedness is stronger in cities with a higher African American population share. Our empirical results agree with the predictions of our theoretical model. Social connectedness has large and important effects on the evolution of crime rates from 19602009. Social connectedness appears to have little effect in the 1960’s, a period of substantial migration and relatively low crime rates. From the 1970’s to 1990’s, a period of high crime rates, the effect of moving from the 75th to 25th percentile of social connectedness is an increase in murder rates of 26-56 percent. The effect of social connectedness fades in the 2000’s, a period of relatively low crime and several decades after the end of the Great Migration. Several mechanisms could underlie the negative effect of social connectedness on crime. In future work, we hope to examine the effect of social connectedness on other outcomes of interest, including educational attainment, marriage, and fertility.38

References Alesina, Alberto, and Eliana La Ferrara. 2000. “Participation in Heterogeneous Communities.” Quarterly Journal of Economics, 115(3): 847–904. Alesina, Alberto, Reza Baqir, and William Easterly. 1999. “Public Goods and Ethnic Divisions.” Quarterly Journal of Economics, 114(4): 1243–1284. Anonymous. 1983. “Blacks in Pennsylvania Town Recall Southern Past.” The Baytown Sun. Anonymous. 1994. “North, in Search of Work: Economics, Jim Crow Drove Migration from Tennessee.” Herald & Review. Bayer, Patrick, Randi Hjalmarsson, and David Posen. 2009. “Building Criminal Capital Behind Bars: Peer Effects in Juvenile Corrections.” Quarterly Journal of Economics, 124(1): 105–147. Bell, Velma Fern. 1933. “The Negro in Beloit and Madison, Wisconsin.” Master’s diss. University of Wisconsin. Bertrand, Marianne, Erzo Luttmer, and Sendhil Mullainathan. 2000. “Network Effects and Welfare Cultures.” Quarterly Journal of Economics, 115(3): 1019–1055. Black, Dan A., Seth G. Sanders, Evan J. Taylor, and Lowell J. Taylor. forthcoming. “The Impact of the Great Migration on Mortality of African Americans: Evidence from the Deep South.” American Economic Review. 38

To do so, we would benefit from restricted access Census data, in order to calculate outcomes at the city-level for African Americans.

27

Blumstein, Alfred. 2000. “Disaggregating the Violence Trends.” In The Crime Drop in America. , ed. Alfred Blumstein and Joel Wallman, 13–44. New York:Cambridge University Press. Boggess, Scott, and John Bound. 1997. “Did Criminal Activity Increase During the 1980s: Comparisons Across Data Sources.” Social Science Quarterly, 18: 725–39. Borjas, George J. 1995. “Ethnicity, Neighborhoods, and Human-Capital Externalities.” American Economic Review, 85(3): 365–390. Boustan, Leah Platt. 2009. “Competition in the Promised Land: Black Migration and Racial Wage Convergence in the North, 1940-1970.” Journal of Economic History, 69(3): 756–783. Boustan, Leah Platt. 2011. “Was Postwar Suburbanization ‘White Flight’? Evidence from the Black Migration.” Quarterly Journal of Economics, 125(1): 417–443. Calv´o-Armengoi, Antoni, and Yves Zenou. 2004. “Social Networks and Crime Decisions: The Role of Social Structure in Facilitating Delinquent Behavior.” International Economic Review, 45(3): 939–958. Carrington, William J., Enrica Detragiache, and Tara Vishwanath. 1996. “Migration with Endogenous Moving Costs.” American Economic Review, 86(4): 909–930. Case, Anne C., and Lawrence F. Katz. 1991. “The Company You Keep: The Effects of Family and Neighborhood on Disadvantaged Youths.” NBER Working Paper 3705. Census, United States Bureau of the. 1979. “The Social and Economic Status of the Black Population in the United States, 1790-1978.” Current Population Reports, Special Studies Series P-23 No. 80. Chalfin, Aaron, and Justin McCrary. 2014. “Are U.S. Cities Underpoliced? Theory and Evidence.” Chay, Kenneth, and Kaivan Munshi. 2013. “Black Networks After Emancipation: Evidence from Reconstruction and the Great Migration.” Coleman, James S. 1990. Foundations of Social Theory. Cambridge, MA:Harvard University Press. Collins, William J. 1997. “When the Tide Turned: Immigration and the Delay of the Great Black Migration.” Journal of Economic History, 57(3): 607–632. Crowder, Carla, and Thomas Spencer. 2002. “Perry County Natives Find Success Far from Home.” The Birmingham News. Cutler, David M., and Edward L. Glaeser. 1997. “Are Ghettos Good or Bad?” Quarterly Journal of Economics, 112(3): 827–872. Cutler, David M., Edward L. Glaeser, and Jacob L. Vigdor. 1999. “The Rise and Decline of the American Ghetto.” Journal of Political Economy, 107(3): 455–506. Damm, Anna Piil, and Christian Dustmann. 2014. “Does Growing Up in a High Crime Neighborhood Affect Youth Criminal Behavior?” American Economic Review, 104(6): 1806–1832. Evans, William N., Craig Garthwaite, and Timothy J. Moore. 2014. “The White/Black Educational Gap, Stalled Progress, and the Long-Term Consequences of Crack Cocaine Markets.” Fox, James Alan. 2000. “Demographics and U.S. Homicide.” In The Crime Drop in America. , ed. Alfred Blumstein and Joel Wallman, 288–317. New York:Cambridge University Press. Freeman, Richard B. 1999. “The Economics of Crime.” In Handbook of Labor Economics. Vol. 3C, , ed. Orley Ashenfelter and David Card, 3529–3571. Amsterdam:North Holland. Gibson, Campbell, and Kay Jung. 2005. “Historical Census Statistics on Population Totals by Race, 1790 to 1990, and by Hispanic Origin, 1790 to 1990, For Large Cities and Other Urban Places in the United States.” U.S. Census Bureau Population Division Working Paper No. 76. 28

Glaeser, Edward L., Bruce Sacerdote, and Jos´e A. Scheinkman. 1996. “Crime and Social Interactions.” Quarterly Journal of Economics, 111(2): 507–548. Gottlieb, Peter. 1987. Making Their Own Way: Southern Blacks’ Migration to Pittsburgh, 19161930. Urbana:University of Illinois Press. Grossman, James R. 1989. Land of Hope: Chicago, Black Southerners, and the Great Migration. Chicago:University of Chicago Press. Guerry, Andr’e-Michel. 1833. Essai sur la Statistique Morale de la France. Paris:Crochard. Hornbeck, Richard, and Suresh Naidu. 2014. “When the Levee Breaks: Black Migration and Economic Development in the American South.” American Economic Review, 104(3): 963–990. Jackson, Blyden. 1991. “Introduction: A Street of Dreams.” In Black Exodus: The Great Migration from the American South. , ed. Alferdteen Harrison, xi–xvii. Jackson:University Press of Mississippi. Kling, Jeffrey R., Jens Ludwig, and Lawrence F. Katz. 2005. “Neighborhood Effects on Crime for Female and Male Youth: Evidence from a Randomized Housing Voucher Experiment.” Quarterly Journal of Economics, 120(1): 87–130. Knowles, Lucas W. 2010. “Beloit, Wisconsin and the Great Migration the Role of Industry, Individuals, and Family in the Founding of Beloit’s Black Community 1914 - 1955.” Ludwig, Jens, and Jeffrey R. Kling. 2007. “Is Crime Contagious?” Journal of Law and Economics, 50(3): 491–518. Ludwig, Jens, Greg J. Duncan, and Paul Hirschfield. 2001. “Urban Poverty and Juvenile Crime: Evidence from a Randomized Housing-Mobility Experiment.” Quarterly Journal of Economics, 116(2): 655–679. Marks, Carole. 1989. Farewell, We’re Good and Gone: The Great Black Migration. Bloomington:Indiana University Press. Marks, Carole. 1991. “The Social and Economic Life of Southern Blacks During the Migrations.” In Black Exodus: The Great Migration from the American South. , ed. Alferdteen Harrison, 36– 50. Jackson:University Press of Mississippi. Neal, Derek, and Armin Rick. 2014. “The Prison Boom & The Lack of Black Progress after Smith & Welch.” Quetelet, Adolphe. 1835. Sur l’Homme et le Developpement De Ses Facultes. Paris:Bachelier. Rubin, Morton. 1960. “Migration Patterns of Negroes from a Rural Northeastern Mississippi Community.” Social Forces, 39(1): 59–66. Rupasingha, Anil, and Stephan J. Goetz. 2008. US County-Level Social Capital Data, 19902005. The Northeast Regional Center for Rural Development, Penn State University, University Park, PA. Sampson, Robert J., Jeffrey D. Morenoff, and Felton Earls. 1999. “Beyond Social Capital: Spatial Dynamics of Collective Efficacy for Children.” American Sociological Review, 64(5). Sampson, Robert J., Stephen W. Raudenbush, and Felton Earls. 1997. “Neighborhoods and Violent Crime: A Multilevel Study of Collective Efficacy.” Science, 277(918). Sciandra, Matthew, Lisa Sanbonmatsu, Greg J. Duncan, Lisa A. Gennetian, Lawrence F. Katz, Ronald C. Kessler, Jeffrey R. Kling, and Jens Ludwig. 2013. “Long-term Effects of the Moving to Opportunity Residential-mobility Experiment on Crime and Delinquency.” Journal of Experimental Criminology, 9(4): 451–489. Scott, Emmett J. 1920. Negro Migration During the War. New York:Oxford University Press. Scroggs, William O. 1917. “Interstate Migration of Negro Population.” Journal of Political Econ29

omy, 25(10): 1034–1043. Silverman, Dan. 2004. “Street Crime and Street Culture.” International Economic Review, 45(3): 761–786. Smith, James P., and Finis Welch. 1989. “Black Economic Progress After Myrdal.” Journal of Economic Literature, 27(2): 519–564. Smith, Sheila. 2006. “All-class Reunion Recalls Decatur’s Ties to Brownsville, Tenn.” Herald & Review. Stuart, Bryan A., and Evan J. Taylor. 2014. “Social Interactions and Location Decisions: Evidence from U.S. Mass Migration.” Tibbetts, Stephen G. 2012. Criminological Theory: The Essentials. Los Angeles:SAGE Publications. Wooldridge, Jeffrey M. 2002. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA:MIT Press. Table 1: Summary Statistics, Crime and Social Connectedness, 1960-2009

Mean

S.D.

Q1

Offenses reported to police per 100k population Murder 6.659 9.527 1.655 Rape 29.142 30.067 9.986 Robbery 210.835 250.265 65.814 Assault 1,126.425 1,090.556 288.811 Burglary 1,237.043 1,002.575 662.547 Larceny 32,44.092 1,952.688 2,002.587 Motor Vehicle Theft 581.007 522.095 258.130 Population 91,904.023 93,349.786 39,258 S. Black Migrant HHI 0.020 0.016 0.008 Log S. Black Migrant HHI -4.227 0.783 -4.857 Top Sending Town Share 0.061 0.040 0.036 S. Black Migrants 666.761 1425.218 54

Q3

Fraction Zero

8.532 39.621 264.716 1,609.119 1,618.099 4,202.917 736.334 102,652 0.028 -3.579 0.074 624

0.191 0.071 0.003 0.005 0.000 0.000 0.000 -

Notes: Each observation is a city-year. HHI and migrant counts are calculated among all individuals born in the former Confederacy states from 1900-1936. Data on rape is only available starting in 1964. Sample is restricted to cities with population less than 500,000 in 1980. N = 19,471 Sources: FBI UCR, Duke SSA/Medicare dataset

30

Table 2: Mobility Rates, Blacks born in the Confederacy from 1900-1936, living in the North Compared to 5 years ago, percent living in Same State Census Year (1)

Same House (2)

Same County (3)

Different County (4)

1960 1970 1980 1990 2000

39.00 57.40 75.45 78.25 80.21

47.86 30.24 19.81 15.05

2.57 3.27 2.11 2.10

Unknown (5)

Different State (6)

Abroad (7)

0.70 19.09 -

10.02 5.15 2.56 2.62 2.23

0.56 0.34 0.08 0.04 0.41

Notes: In 1970, 2.91% moved to unknown place. For 1990 and 2000, column 3 equals the percent living in the same PUMA. Sources: Census IPUMS

31

Table 3: Social Connectedness among Black Migrants and City Covariates Dependent variable: Log HHI, Southern black migrants Year of Covariates

Log number Southern black migrants Log population

(1)

(2)

1960 (3)

1970 (4)

1980 (5)

1990 (6)

2000 (7)

-0.842*** (0.035)

-0.841*** (0.039)

234 0.707

x 234 0.746

-0.865*** (0.064) 0.033 (0.061) 0.026 (0.050) 0.026 (0.047) -0.191 (0.127) -0.163 (0.106) -0.055 (0.084) -0.057 (0.109) 0.128* (0.069) -0.031 (0.048) 0.005 (0.080) 0.114* (0.059) 0.019 (0.024) 0.208*** (0.058) x 234 0.772

-0.879*** (0.070) 0.025 (0.065) 0.015 (0.057) -0.020 (0.059) 0.133 (0.203) 0.091 (0.207) 0.146 (0.146) -0.061 (0.113) 0.074 (0.063) 0.009 (0.060) -0.015 (0.084) 0.127 (0.078) 0.068 (0.053) 0.154** (0.061) x 234 0.764

-0.854*** (0.076) 0.012 (0.073) 0.028 (0.072) -0.008 (0.074) 0.182 (0.232) 0.175 (0.239) 0.268 (0.190) -0.174* (0.093) 0.057 (0.050) 0.016 (0.064) 0.003 (0.086) 0.013 (0.070) 0.025 (0.090) 0.130** (0.056) x 234 0.761

-0.772*** (0.081) -0.026 (0.084) -0.017 (0.065) -0.017 (0.078) 0.497** (0.234) 0.527** (0.247) 0.501*** (0.181) -0.052 (0.074) 0.098 (0.061) 0.021 (0.074) -0.211** (0.086) -0.011 (0.078) 0.098 (0.098) 0.145*** (0.047) x 234 0.763

-0.610*** (0.057) 0.034 (0.063) -0.026 (0.044) 0.000 (0.043) 0.189 (0.229) 0.310 (0.249) 0.284* (0.159) -0.043 (0.062) 0.034 (0.049) -0.031 (0.060) -0.051 (0.052) 0.041 (0.048) -0.031 (0.040) 0.133*** (0.036) x 234 0.769

0.14 0.19

0.70 0.24

0.27 0.99

0.03 0.05

0.002 0.13

Percent black Percent female Percent age 5-17 Percent age 18-64 Percent age 65+ Percent 25+ with HS Percent 25+ with college Log area, square miles Log median family income Unemployment rate Labor force participation rate Manufacturing employment share State fixed effects Observations Adjusted R2

p-value: Wald test that parameters equal zero Demographic covariates Economic covariates

Notes: All variables have been normalized, separately for each regression, to have mean zero and standard deviation one. There are 234 city-level observations. Sample restricted to cities with population less than 500,000 as of 1980. See note to table 4 for definitions of demographic and economic covariates. Heteroskedastic-robust standard errors in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01 Sources: FBI UCR, Census City databook, Duke SSA/Medicare dataset

32

Table 4: Murder and Social Connectedness among Black Migrants, 1960-2009 Dependent variable: number of murders reported to police (1) (2) (3) (4) Log HHI, S. black migrants Log number S. black migrants Log squared population share, top sending town Demographic covariates Economic covariates State-Year Dummies Region-Year Dummies Log-likelihood

-0.193*** (0.052) 0.150*** (0.031)

-0.200*** (0.045) 0.156*** (0.030)

-0.133** (0.055) 0.148*** (0.032)

x

x x x

x x

x -29,441

-28,905

x -31,507

0.219*** (0.028) -0.066*** (0.022) x x x -29,004

Notes: Results estimated by a Poisson regression model. There are 11,568 city-year observations and 234 cities in all columns. Sample restricted to cities with population less than 500,000 as of 1980. Demographic covariates include log population (annual), percent black (1960, 1970, 1980, 1990, 2000), percent age 5-17, 18-64, and 65+ (1960, 1970, 1980, 1990, 2000), percent female (1970, 1980, 1990, 2000), percent of population at least 25 years old with a high school degree (1960, 1970, 1980, 1990), percent of population at least 25 years old with a college degree (1960, 1970, 1980, 1990), and log of area in square miles (1960, 1970, 1980, 1990, 2000). Economic covariates include log median family income (1960, 1970, 1980, 1990), unemployment rate (1960, 1970, 1980, 1990, 2000), labor force participation rate (1960, 1970, 1980, 1990, 2000), and manufacturing employment share (1960, 1970, 1980, 1990). For decades in which a covariate is not available, we use the relevant covariate from the adjacent decade. Standard errors, clustered at the city level, are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01 Sources: FBI UCR, Duke SSA/Medicare dataset, Census City databook

33

Table 5: Robustness, Murder and Social Connectedness among Black Migrants, 1960-2009 Dependent variable: number of murders reported to police (1) (2) (3) Log HHI, S. Black Migrants Log number of S. black migrants Log HHI and number, native whites and foreign-born No control for number of S. black migrants 10 dummies for number of S. black migrants Log-likelihood

-0.200*** (0.045) x

-0.306*** (0.045)

-0.146*** (0.047) x x

(4) -0.190*** (0.046)

x -28,905

-29,134

-28,803

x -28,662

Notes: Results estimated by a Poisson regression model. The dummy variables for number of Southern black migrants correspond to deciles. Regressions include covariates from column 2 of table 4. See note to Table 4 for description of variables and sample. Standard errors, clustered at the city level, are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01 Sources: FBI UCR, Duke SSA/Medicare dataset, Census City databook

Table 6: Crime and Social Connectedness among Black Migrants, 1960-2009 Dependent variable: number of offenses reported to police

Log HHI, S. black migrants Log number S. black migrants Log-likelihood City-year observations Cities

Murder (1)

Rape (2)

Robbery (3)

Assault (4)

Burglary (5)

Larceny (6)

Motor Vehicle Theft (7)

-0.153*** (0.036) 0.156*** (0.022) -44,761 19,471 488

-0.070** (0.034) 0.071*** (0.026) -88,820 18,279 488

-0.235*** (0.035) 0.153*** (0.027) -341,731 19,471 488

-0.133*** (0.041) 0.076*** (0.029) -1,668,135 19,471 488

-0.074*** (0.024) 0.058*** (0.018) -1,112,264 19,471 488

-0.028 (0.031) 0.042* (0.024) -2,588,533 19,471 488

-0.149*** (0.041) 0.048* (0.029) -1,034,348 19,471 488

Notes: Results estimated by a Poisson regression model. Regressions include covariates used in column 2 of table 4. Standard errors, clustered at the city level, are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01 Sources: FBI UCR, Duke SSA/Medicare dataset, Census City databook

34

Table 7: Crime and Social Connectedness, by Percent Black Tercile, 1960-2009 Dependent variable: number of offenses reported to police Percent Black Tercile Low Medium High

Murder (1)

Rape (2)

-0.026 (0.136) -0.049 (0.044) -0.189*** (0.069)

-0.134 (0.140) -0.151 (0.092) -0.210*** (0.074)

Robbery (3)

Assault (4)

0.025 -0.170 (0.165) (0.135) -0.182*** -0.172** (0.067) (0.085) -0.232*** -0.264*** (0.072) (0.077)

Burglary (5)

Larceny (6)

Motor Vehicle Theft (7)

0.021 (0.086) -0.104** (0.049) -0.155*** (0.041)

-0.016 (0.081) 0.000 (0.046) -0.180** (0.071)

0.193 (0.175) -0.173** (0.083) -0.276*** (0.092)

Notes: The percent black cutoffs are 0.021 and 0.073. Regressions include covariates used in column 3 of table 4. Percent black is measured in 1960. Standard errors, clustered at the city level, are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01 Sources: FBI UCR, Duke SSA/Medicare dataset, Census City databook

Table 8: Social Connectedness and Murder, by Race and Age of Offender, 1980-1989 Dependent variable: number of murder arrests of age/race group Black White All (1) Log HHI, S. black migrants

-0.176*** (0.059) Log number S. black migrants 0.091** (0.044) Log-likelihood -9,260

Adult (2) -0.276*** (0.080) 0.524*** (0.064) -5,238

Juvenile (3)

Adult (4)

-0.747*** -0.117 (0.180) (0.074) 0.390*** 0.056 (0.135) (0.056) -1,676 -6,911

Juvenile (5) -0.018 (0.178) 0.119 (0.100) -2,305

Notes: Poisson regressions include covariates used in column 3 of table 4. Standard errors, clustered at the city level, are in parentheses. There are 3,586 city-year observations and 418 cities in the sample. * p < 0.1; ** p < 0.05; *** p < 0.01 Sources: FBI UCR, Duke SSA/Medicare dataset, Census City databook

35

Table 9: Connection between Model and Empirical Estimates

Peer effect parametrization J11 = J22 = J33 (own-group) J12 = J21 (cross-group, black) J13 = J23 (cross-race, white on black) J31 = J32 (cross-race, black on white) Implied peer effect elasticities E11 = E22 = E33 (own-group) E12 = E21 (cross-group, black) E13 = E23 (cross-race, white on black) E31 = E32 (cross-race, black on white)

(1)

(2)

(3)

0 0 0 0

0.25 0 0 0

0 0 0 0

0.25 0 0 0

(4)

(5)

(6)

(7)

0.25 0.25 0.2 0.2 0 0.67 0 0.015

0.5 0 0 0

0.5 0.4 0 0

0.5 0.4 0.67 0.015

0.25 0.2 0 0

0.5 0 0 0

0.5 0.4 0 0

0.5 0.4 0.1 0.1

0.25 0.2 0.1 0.1

Percent increase in crime rate due to one standard deviation decrease in social connectedness Total crime 11.9 11.9 11.9 11.9 11.9 11.9 11.9 White crime 0.0 0.0 0.0 6.3 0.0 0.0 8.1 Black crime 23.9 23.9 23.9 17.6 23.9 23.9 15.8 Northern black crime 0.0 0.0 9.0 7.1 0.0 20.8 14.5 Southern black crime 39.8 39.8 33.8 24.5 39.8 25.9 16.6 Direct effect of HHI 39.8 29.8 23.5 16.9 19.9 4.7 3.0 Peer effect 0.0 9.9 10.2 7.6 19.9 21.3 13.6 Notes: Table 9 presents the effects of a one standard deviation decrease in log HHI (-0.78). See text for details.

36

0

2

Population, millions 4 6 8

10

Figure 1: Number of African Americans in the North

1900

1910

1920

1930

1940

1950 Year

1960

1970

1980

1990

2000

1980

1990

2000

Adults in North Adults in North, born in South

0

1

Population, millions 2 3 4

5

(a) Adults

1900

1910

1920

1930

1940

1950 Year

1960

1970

Children in North Children in North, parents born in South Children in North, born in South

(b) Children

Note: The South here consists of former Confederate states, and the North consists of all other states. The different groups are mutually exclusive.

37

Log HHI, Southern black migrants -3 -2 -7 -6 -5 -4

Figure 2: Social Connectedness and Number of Southern Black Migrants

Linear fit: -0.40 ( 0.01), R2 = 0.69

4

6 8 10 Log number of Southern black migrants

25-75k, 0-.05 75-150k, 0-.05 150-500k, 0-.05

25-75k, .05-.15 75-150k, .05-.15 150-500k, .05-.15

25-75k, .15-1 75-150k, .15-1 150-500k, .15-1

>500k, 0-.05

>500k, .05-.15

>500k, .15-1

12

Notes: Figure contains cities in sample in 1980. Cities are classified by their 1980 population and African American population share.

38

Log HHI, Southern black migrants -4 -6 -3 -5 -2

Figure 3: Top Sending Town Accounts for Most of Variation in HHI

Linear fit: 0.58 ( 0.02), R2 = 0.67

-8

-6 -4 Log Squared Percent of Migrants from Top Sending Town

25-75k, 0-.05 75-150k, 0-.05 150-500k, 0-.05

25-75k, .05-.15 75-150k, .05-.15 150-500k, .05-.15

25-75k, .15-1 75-150k, .15-1 150-500k, .15-1

>500k, 0-.05

>500k, .05-.15

>500k, .15-1

-2

Notes: Figure contains cities in sample in 1980. Cities are classified by their 1980 population and African American population share.

39

4

Average crime per 100k population 8 12 6 10

14

Figure 4: Social Connectedness and Murder Rate, by Percent Black Tercile

1

2 Percent black tercile, 1960 HHI at 75th percentile

3

HHI at 25th percentile

Notes: To construct figure 4, we estimate a separate Poisson model for each percent black tercile, and take the value of covariates associated with the average crime rate for each tercile. We then plot two different predicted crime rates: one for the 75th percentile (HHI = 0.028) and one for the 25th percentile (HHI=0.008). Poisson regression models are estimated using the covariates in column 3 of table 4. The percent black cutoffs are 0.021 and 0.073.

40

10 20

05 20

00 20

95 19

19

90

85 19

80 19

75 19

70 19

65 19

19

60

4

8 10 6 Murders per 100,000 Population

12

UCR Index Offenses per 100,000 Population 2000 4000 6000 8000 10000

Figure 5: Evolution of Crime Rates Over Time

Year Index Offenses

Murder

Notes: Index offenses include murder, rape, robbery, aggravated assault, burglary, larceny theft, and motor vehicle theft. Sample is limited to cities in our analysis sample with less than 500,000 residents in 1980.

41

4

Average crime per 100k population 8 6 10

12

Figure 6: Social Connectedness and the Evolution of Crime Rates Over Time

1960

1965

1970

1975

1980 1985 Year

HHI at 75th percentile

1990

1995

2000

2005

HHI at 25th percentile

200

Average crime per 100k population 800 1000 400 600

(a) Murder

1960

1965

1970

1975

1980 1985 Year

HHI at 75th percentile

1990

1995

2000

2005

HHI at 25th percentile

(b) Motor Vehicle Theft

Notes: For each five year period from 1965-1969, 1970-1974, etc., we estimate a Poisson regression model and take the level of covariates associated with the average crime rate. Besides the covariates used in column 2 of table 4, we control for the log average crime rate from 1960-1964. We then plot the murder rate associated with the 75th and 25th percentiles of HHI. By construction, the two series intersect for the period from 1960-1964.

42

-.6

Coefficient on Log HHI -.4 0 -.2

.2

Figure 7: Importance of Controlling for Average 1960-1964 Crime Rates

1965

1970

1975

1980

1985 Year

1990

Control for 1960-64 crime

1995

2000

2005

No control

-.6

Coefficient on Log HHI -.4 0 -.2

.2

(a) Murder

1965

1970

1975

1980

1985 Year

1990

Control for 1960-64 crime

1995

2000

2005

No control

(b) Motor Vehicle Theft

Figure 7 shows point estimates and 95-percent confidence intervals from estimating a separate regression for years 1965-1969, 1970-1974, and so on, for models with covariates used in column 2 of table 4, with and without also controlling for the log average crime rate from 1960-1964.

43

Online Appendix A

Additional Theoretical Details

As noted in the text, two jointly sufficient conditions for result 2 are (a): dC¯ bs /dHHIbs < dC¯ w /dHHIbs and (b): dC¯ bn /dHHIbs < dC¯ w /dHHIbs . Assuming that ∂F bs /∂HHIbs < 0, conditions (a) and (b) are equivalent to m11 > m31 and m21 > m31 (these variables are defined in section 7). Simple algebra demonstrates that condition (a) is satisfied if and only if (1 − J22 )(1 − J33 − J31 ) > J32 (J21 + J23 ),

(A.1)

which is true if cross-group peer effects are small enough. Similarly, condition (b) is satisfied if and only if J21 (1 − J33 − J32 ) > J31 (1 − J22 − J23 ).

(A.2)

If black peers have stronger effects on each other than Southern blacks and whites, J21 > J31 ≥ 0, inequality (A.2) is satisfied if 1 − J33 − J32 > 1 − J22 − J23 , which reduces to J22 > J33 under the assumption that J32 = J23 . If there are no cross-group peer effects between black and white youth, J13 = J23 = J31 = J32 = 0, then 1 − J22 (1 − J11 )(1 − J22 ) − J12 J21 J21 = (1 − J11 )(1 − J22 ) − J12 J21 =0

m11 =

(A.3)

m21

(A.4)

m31

(A.5)

The stable equilibrium assumption implies that J22 ∈ [0, 1) and (1 − J11 )(1 − J22 ) > J12 J21 , ensuring that condition (a) holds. Condition (b) additionally requires J21 > 0. B

Details on Peer Effect Parametrization

Damm and Dustmann (2014) estimate the effect of municipality crime rates on convictions for criminal behavior of refugees in Denmark. For males, they find that a one percentage point increase in the local crime rate leads to a 7-13 percent increase in the probability of conviction between ages 15-21 (Table 3, also see p. 1820). Given an average conviction rate of 46 percent, this translates into a 3-6 percentage point increase in the probability of conviction; we will take the midpoint of 4.5. For females, there is no effect. Note that these results pertain to a 7 year period. Combining the sex effects and creating an annual rate, a one percentage point increase in the local crime rate leads to a 4.5 / (7 · 2) ≈ 0.32 percentage point increase in the annual conviction rate. This suggests diagonal elements of J close to 1/3. Damm and Dustmann (2014) find that, beyond the impact of the overall local conviction rate, the conviction rate of co-nationals has an additional impact while the conviction rate of immigrants from other countries does not (Table 7). This suggests that i

cross-group peer effects might be small, though the nature of cross-group interactions could differ considerably across the setting studied by Damm and Dustmann (2014) and here. Case and Katz (1991) report that a one percent increase in the neighborhood crime rate leads to a 0.118 percent increase in a Boston youth’s self-reported propensity of committing a crime within one year (Table 10). These are evaluated at the sample mean. 17 percent of youths commit a crime within the past year. This suggests a value of diagonal of J close to 0.1. In their preferred estimates, Ludwig and Kling (2007) find no evidence that neighborhood violent crime rates affect violent crime arrests among MTO participants age 15-25 (Table 4). They also note that “... estimates for the overall effects of MTO mobility assignments are not directly informative about whether crime is contagious, because MTO moves change multiple neighborhood characteristics simultaneously, which could have offsetting effects” (p. 494). In the model of Glaeser, Sacerdote and Scheinkman (1996), there are two types of agents arranged along circle. Fixed agents are not affected by neighbors when deciding whether to commit crime, while compliers perfectly imitate their neighbor. In this model, a small increase in the rate of crime has no effect on fixed agents, but has a one-for-one effect on compliers; hence, the diagonal elements of J equal the probability that an agent is a complier. Using the notation of Glaeser, Sacerdote and Scheinkman (1996), the probability that an agent is a fixed type is π. In Table IIA, the authors report estimates of f (π) = (2 − π)/π. The diagonal element of J is thus 1 − π = 1 − 2/(1 + f (π)). Using UCR murder data across cities for 1970 and 1985, Glaeser, Sacerdote and Scheinkman (1996) report estimates of f (π) between 2 and 4.5, implying estimates of the diagonal of J between 1/3 and 2/3. Because the authors do not differentiate between different population groups, this implicitly represents a population weighted average of J11 , J22 , and J33 .39 For simplicity, we take this evidence to suggest that J11 = J22 = J33 = 1/2 for murder. For robbery and motor vehicle theft, the authors estimate f (π) in the range of 37-155 and 141-382, suggesting diagonal elements of J very close to 1.

39

For this same reason, this paper does not provide evidence on off-diagonal elements of J.

ii

Table A.1: Early Twentieth Century Homicide Rates and Social Connectedness Dependent variable: log HHI, Southern black migrants (1) (2) Log homicide rate, 1911

0.074 (0.139)

0.074 (0.105)

0.01 41

-0.185*** (0.059) 0.39 41

Log homicide rate, 1912 Log homicide rate, 1913 Log homicide rate, 1914 Log number, Southern black migrants R2 Cities p-value

(3)

-0.048 (0.300) -0.199 (0.289) 0.237 (0.269) 0.136 (0.261) -0.186** (0.074) 0.45 31 .87

Notes: The last row contains the p-value from the F-test that all log homicide rate coefficients equal zero. Heteroskedastic-robust standard errors in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01

iii

Table A.2: Social Connectedness among Black Migrants and City Covariates, Including CitySpecific Covariates for Blacks Dependent variable: Log HHI, Southern black migrants Year of Covariates 1970 1980 1990 2000 (1) (2) (3) (4) Log number Southern black migrants Log population Percent black Percent female Percent age 5-17 Percent age 18-64 Percent age 65+ Percent 25+ with HS Percent 25+ with College Log area, square miles Log median family income Unemployment rate Labor force participation rate Manufacturing employment share City-specific covariates for blacks Percent female Percent age 5-17 Percent age 18-64 Percent age 65+ Percent 25+ with HS Percent 25+ with College Unemployment rate Observations Adjusted R2

-0.854*** (0.070) 0.032 (0.072) 0.008 (0.057) -0.033 (0.061) -0.017 (0.227) -0.072 (0.231) 0.041 (0.161) 0.027 (0.128) 0.028 (0.074) 0.004 (0.063) -0.040 (0.095) 0.148* (0.082) 0.055 (0.055) 0.183*** (0.065)

-0.820*** (0.076) 0.035 (0.076) 0.030 (0.074) 0.026 (0.078) 0.203 (0.252) 0.243 (0.270) 0.272 (0.208) -0.129 (0.102) -0.003 (0.053) -0.031 (0.071) -0.003 (0.087) -0.030 (0.083) 0.014 (0.089) 0.155*** (0.060)

-0.787*** (0.086) 0.015 (0.086) 0.007 (0.071) -0.005 (0.092) 0.453* (0.251) 0.498* (0.264) 0.446** (0.198) 0.012 (0.089) -0.030 (0.079) -0.015 (0.078) -0.181* (0.097) -0.084 (0.090) 0.081 (0.103) 0.152*** (0.052)

-0.628*** (0.077) 0.052 (0.067) -0.033 (0.051) 0.020 (0.059) 0.300 (0.264) 0.386 (0.290) 0.329* (0.181) -0.013 (0.076) -0.020 (0.067) -0.031 (0.065) -0.048 (0.065) 0.018 (0.046) -0.023 (0.044) 0.137*** (0.038)

0.005 (0.041) 0.100 (0.081) 0.130 (0.086) 0.060 (0.055) -0.143** (0.071) 0.102* (0.052) -0.057 (0.047) 234 0.767

-0.095 (0.060) 0.099 (0.106) 0.039 (0.116) 0.055 (0.069) -0.065 (0.067) 0.099* (0.059) 0.048 (0.068) 234 0.761

0.002 (0.075) 0.159 (0.141) 0.217 (0.162) 0.105 (0.080) -0.117 (0.073) 0.106 (0.071) 0.112* (0.060) 234 0.765

0.064 (0.058) -0.084 (0.138) -0.003 (0.169) 0.001 (0.080) -0.032 (0.056) 0.040 (0.062) 0.077** (0.034) 234 0.771

iv

Table A.2: Social Connectedness among Black Migrants and City Covariates, Including CitySpecific Covariates for Blacks Dependent variable: Log HHI, Southern black migrants Year of Covariates 1970 1980 1990 2000 (1) (2) (3) (4) p-value: Wald test that parameters equal zero Demographic covariates 0.946 Economic covariates 0.125 Black covariates 0.0228

Notes: See note to table 3.

v

0.515 0.970 0.332

0.290 0.273 0.312

0.0354 0.649 0.176

Table A.3: Social Connectedness and Other Measures of Social Capital

Associational density

Dependent variable: Log HHI, Southern black migrants (3) (4) (5) (6) (7)

(1)

(2)

0.0863 (0.0531)

0.0523 (0.0480)

Social capital index

0.0639 (0.0523)

-0.000219 (0.0497)

Social capital composite index

0.0547 (0.0513)

Log number migrants State fixed effects City-level observations Counties R-squared

507 234 0.007

-0.854*** (0.0333) x 507 234 0.738

507 234 0.004

-0.856*** (0.0327) x 507 234 0.737

507 234 0.003

-0.00851 (0.0465) -0.856*** (0.0327) x 507 234 0.737

(8)

0.104 (0.0877) -0.0217 (0.0883)

0.0943 (0.0585) -0.0660 (0.0614)

507 234 0.008

-0.855*** (0.0333) x 507 234 0.739

vi

Notes: All variables are normalized to have mean zero and standard deviation one. See Rupasingha and Goetz (2008) for definitions of associational density and social capital indices, which are measured at the county level using data from 1988 and 1990. The correlation between the social capital index and the social capital composite index is 0.99. Sample limited to cities with at least 25,000 population in each decade and which received at least 25 Southern black migrants. Standard errors, clustered at the county level, are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01 Sources: Duke SSA/Medicare dataset, Rupasingha and Goetz (2008)

Table A.4: Crime and Social Connectedness, 1960-2009, Full Results

Log HHI, S. black migrants Log number S. black migrants Log population Percent black, 1960 Percent black, 1970

vii

Percent black, 1980 Percent black, 1990 Percent black, 2000 Percent age 5-17, 1960 Percent age 18-64, 1960 Percent age 65+, 1960 Percent age 5-17, 1970 Percent age 18-64, 1970 Percent age 65+, 1970

Murder (1)

Rape (2)

Robbery (3)

Assault (4)

Burglary (5)

Larceny (6)

Motor Vehicle Theft (7)

-0.153*** (0.036) 0.156*** (0.022) 0.936*** (0.054) 2.163*** (0.707) 1.954*** (0.231) 1.628*** (0.169) 1.563*** (0.208) 1.911*** (0.225) -1.185 (7.839) 8.473 (7.551) -8.324 (6.133) -6.536** (2.934) -3.689 (2.684) -3.872* (2.270)

-0.070** (0.034) 0.071*** (0.026) 0.825*** (0.041) 2.764*** (0.605) 2.502*** (0.245) 1.526*** (0.159) 0.681*** (0.210) 0.099 (0.229) -13.086* (7.508) 3.790 (6.727) -15.295*** (5.568) -9.224*** (2.972) -4.750* (2.527) -6.965*** (2.201)

-0.235*** (0.035) 0.153*** (0.027) 1.113*** (0.052) 2.327*** (0.564) 1.533*** (0.225) 1.188*** (0.193) 0.741*** (0.198) 0.435* (0.229) 0.181 (6.558) 7.659 (5.650) -2.585 (5.578) -7.348** (3.007) -3.575 (2.728) -4.031* (2.394)

-0.133*** (0.041) 0.076*** (0.029) 0.865*** (0.049) 3.030*** (0.638) 0.924*** (0.296) 0.599** (0.264) 0.193 (0.236) -0.108 (0.215) -13.048** (6.376) -0.251 (5.687) -16.191*** (4.532) -7.082 (4.435) -7.855* (4.093) -5.103 (3.257)

-0.074*** (0.024) 0.058*** (0.018) 0.938*** (0.030) 1.121* (0.628) 0.947*** (0.166) 0.335** (0.138) 0.095 (0.163) 0.162 (0.172) 0.141 (4.529) 6.360 (4.473) -6.200 (4.008) -3.900** (1.810) -4.782*** (1.574) -2.849** (1.445)

-0.028 (0.031) 0.042* (0.024) 0.865*** (0.042) 0.304 (0.522) 0.107 (0.252) -0.144 (0.237) -0.075 (0.295) -0.395 (0.258) -5.845* (3.549) -2.227 (3.290) -7.613*** (2.559) -2.303 (2.203) -3.039 (1.935) -1.652 (1.643)

-0.149*** (0.041) 0.048* (0.029) 1.259*** (0.052) 1.066* (0.626) 1.214*** (0.266) 0.887*** (0.234) 0.615** (0.280) 0.894*** (0.238) 4.297 (4.284) 7.497** (3.753) -4.324 (4.397) -0.127 (3.292) 1.906 (2.720) -0.910 (2.690)

Table A.4: Crime and Social Connectedness, 1960-2009, Full Results

Percent age 5-17, 1980 Percent age 18-64, 1980 Percent age 65+, 1980 Percent age 5-17, 1990 Percent age 18-64, 1990

viii

Percent age 65+, 1990 Percent age 5-17, 2000 Percent age 18-64, 2000 Percent age 65+, 2000 Percent female, 1960 Percent female, 1970 Percent female, 1980 Percent female, 1990 Percent female, 2000

Murder (1)

Rape (2)

Robbery (3)

Assault (4)

Burglary (5)

Larceny (6)

Motor Vehicle Theft (7)

-8.708*** (2.882) -9.819*** (2.143) -5.293** (2.371) -18.609*** (4.275) -15.241*** (2.986) -11.575*** (3.403) -5.029 (4.997) -6.084 (3.780) -4.481 (3.867) 11.316 (8.645) 0.771 (1.964) -1.944 (2.050) -3.906 (2.642) 4.504 (2.936)

-11.155*** (2.909) -8.592*** (2.070) -8.116*** (2.305) -9.317** (4.031) -7.697*** (2.657) -6.764** (3.006) -9.105* (4.970) -6.173 (4.076) -6.461* (3.774) 16.723** (7.681) 2.040 (2.046) -1.061 (2.333) -2.217 (2.806) 2.342 (2.299)

-3.798 (3.998) -4.218 (2.829) -0.670 (3.211) -7.868* (4.054) -4.805* (2.544) -3.855 (3.056) -3.025 (4.128) -2.287 (3.441) -1.707 (3.154) 1.939 (5.920) -0.611 (2.200) -1.874 (2.551) 0.203 (3.662) -0.285 (2.515)

-13.352*** (4.584) -11.983*** (3.256) -8.644** (3.591) -8.890** (4.335) -7.927*** (3.069) -6.828** (3.236) 0.133 (3.964) -1.382 (3.090) 0.335 (2.967) 9.052 (6.084) -5.422* (2.862) -4.234 (3.058) -1.387 (2.504) 4.193** (1.890)

-6.536** (2.676) -6.362*** (1.866) -4.252** (1.881) -4.868* (2.595) -6.174*** (1.835) -3.739** (1.888) 6.327* (3.329) 5.205* (2.674) 5.767** (2.529) 13.710** (6.393) -0.447 (1.351) 1.422 (1.578) 0.498 (2.113) -0.938 (1.570)

0.774 (3.974) -0.498 (2.283) 2.534 (3.615) 0.789 (3.248) -0.083 (2.478) 1.799 (2.154) 3.028 (3.969) 2.289 (3.105) 3.151 (3.027) 6.250 (4.130) -0.615 (1.468) -2.984 (2.111) -1.828 (2.128) 0.585 (1.787)

11.741*** (4.218) 9.055*** (3.157) 10.310*** (3.326) 5.983 (5.096) 5.925* (3.155) 5.448 (3.741) 9.048* (5.082) 9.302** (4.032) 7.886** (3.729) 6.715 (5.565) 0.710 (2.582) -1.328 (3.108) 3.999 (4.038) -1.551 (2.850)

Table A.4: Crime and Social Connectedness, 1960-2009, Full Results

Percent 25+ with HS, 1960 Percent 25+ with HS, 1970 Percent 25+ with HS, 1980 Percent 25+ with HS, 1990 Percent 25+ with HS, 2000

ix

Percent 25+ with College, 1960 Percent 25+ with College, 1970 Percent 25+ with College, 1980 Percent 25+ with College, 1990 Percent 25+ with College, 2000 Log area, square miles, 1960 Log area, square miles, 1970 Log area, square miles, 1980 Log area, square miles, 1990

Murder (1)

Rape (2)

Robbery (3)

Assault (4)

Burglary (5)

Larceny (6)

Motor Vehicle Theft (7)

-1.335 (1.172) -2.526*** (0.572) -2.334*** (0.534) -1.902*** (0.467) -1.417*** (0.494) 3.665 (2.685) -0.387 (0.773) 0.400 (0.483) -0.317 (0.414) 0.066 (0.444) -0.037 (0.075) 0.046 (0.053) 0.107** (0.052) 0.103** (0.048)

-0.136 (1.456) -1.419*** (0.490) -0.411 (0.471) 1.573*** (0.464) 2.740*** (0.529) 5.580*** (1.926) 1.244** (0.598) 0.091 (0.481) -0.505 (0.372) -1.025** (0.486) 0.301*** (0.071) 0.269*** (0.040) 0.277*** (0.038) 0.188*** (0.040)

-0.049 (0.938) -1.821*** (0.565) -1.496** (0.655) -1.337** (0.526) -0.706 (0.545) 0.303 (2.218) -0.294 (0.758) 0.227 (0.590) -0.033 (0.369) -0.037 (0.436) -0.108 (0.083) -0.134** (0.053) -0.099* (0.055) -0.122** (0.053)

-0.132 (1.076) -3.071*** (0.606) -1.172* (0.669) 1.122** (0.490) 1.554*** (0.453) 2.998 (2.115) 2.032*** (0.676) 0.239 (0.687) -0.604* (0.354) -0.244 (0.414) 0.075 (0.070) 0.124*** (0.046) 0.127*** (0.044) 0.114*** (0.042)

0.172 (0.941) -0.834*** (0.322) -1.416*** (0.336) 0.840** (0.359) 1.450*** (0.363) 3.963** (1.877) 1.458*** (0.362) 0.857*** (0.322) 0.710*** (0.274) -0.072 (0.329) 0.051 (0.053) 0.067*** (0.024) 0.094*** (0.026) 0.088*** (0.029)

-0.347 (0.786) -0.139 (0.382) -1.352** (0.549) 0.532 (0.416) 1.013*** (0.383) 4.125*** (1.343) 1.470*** (0.382) 1.368*** (0.394) 0.964*** (0.273) 0.735** (0.320) 0.072 (0.046) 0.095** (0.038) 0.139*** (0.034) 0.126*** (0.036)

-1.302 (0.807) -2.531*** (0.630) -1.203* (0.626) -1.137* (0.633) -0.645 (0.593) 3.218 (2.351) 0.191 (0.786) -1.284* (0.711) -1.430*** (0.536) -2.089*** (0.600) -0.159*** (0.061) -0.212*** (0.047) -0.171*** (0.051) -0.046 (0.051)

Table A.4: Crime and Social Connectedness, 1960-2009, Full Results

Log area, square miles, 2000 Log median family income, 1960 Log median family income, 1970 Log median family income, 1980 Log median family income, 1990

x

Log median family income, 2000 Unemployment rate, 1960 Unemployment rate, 1970 Unemployment rate, 1980 Unemployment rate, 1990 Unemployment rate, 2000 Labor force participation rate, 1960 Labor force participation rate, 1970 Labor force participation rate, 1980

Murder (1)

Rape (2)

Robbery (3)

Assault (4)

Burglary (5)

Larceny (6)

Motor Vehicle Theft (7)

0.076 (0.049) -2.346*** (0.823) -0.354 (0.294) -0.742*** (0.215) -0.500* (0.261) -1.285*** (0.185) 1.675 (6.951) -0.775 (1.704) 1.616 (1.317) 6.890*** (2.141) -1.346 (1.571) 0.370 (2.358) 0.920 (1.120) 2.930*** (1.013)

0.128*** (0.040) -1.448* (0.855) -0.977*** (0.291) -1.548*** (0.241) -1.972*** (0.237) -2.163*** (0.194) 11.853** (5.074) -2.006 (1.637) 2.145* (1.141) 0.995 (1.710) -1.439 (1.370) -2.653 (1.787) 1.209 (0.901) 3.443*** (0.950)

-0.184*** (0.048) -1.083 (0.690) -0.238 (0.364) -0.945*** (0.340) -1.001*** (0.311) -1.229*** (0.147) 8.602** (4.223) 0.911 (2.179) -0.623 (1.518) 2.575* (1.443) -2.334* (1.277) 3.354 (2.240) 2.542** (1.243) 3.153** (1.348)

0.096** (0.041) -1.587** (0.725) -0.057 (0.367) -0.462 (0.359) -1.346*** (0.252) -1.758*** (0.173) 7.455* (4.155) 1.344 (2.117) 2.919* (1.528) 0.677 (1.632) 0.503 (0.930) 1.940 (1.949) 3.695*** (1.375) 3.177** (1.500)

0.068** (0.028) -1.946*** (0.606) -0.741*** (0.198) -0.379* (0.216) -1.220*** (0.161) -1.341*** (0.149) 5.890 (4.794) -0.556 (1.291) 2.198** (0.986) 3.168** (1.244) 2.198** (1.079) 0.926 (1.646) 2.058*** (0.616) 2.165*** (0.668)

0.109*** (0.039) -1.059*** (0.408) -0.791*** (0.198) -0.812*** (0.236) -1.460*** (0.187) -1.228*** (0.155) 5.256* (3.028) 0.062 (1.294) 2.799*** (0.899) -0.898 (1.646) 1.869* (1.121) 1.596 (1.210) 1.881** (0.740) 4.094*** (1.129)

-0.122*** (0.043) -1.045* (0.617) 0.686* (0.375) 0.089 (0.356) -0.215 (0.375) -0.645*** (0.222) 4.755 (3.775) 1.344 (2.232) 0.982 (1.761) 2.165 (2.530) -0.839 (1.091) 1.677 (1.769) 1.075 (1.257) 1.408 (1.361)

Table A.4: Crime and Social Connectedness, 1960-2009, Full Results

Labor force participation rate, 1990 Labor force participation rate, 2000 Manufacturing employment share, 1960 Manufacturing employment share, 1970 Manufacturing employment share, 1980

xi

Manufacturing employment share, 1990 Manufacturing employment share, 2000 City-year observations Log-likelihood Cities

Notes: See note to table 6.

Murder (1)

Rape (2)

Robbery (3)

Assault (4)

Burglary (5)

Larceny (6)

Motor Vehicle Theft (7)

2.570*** (0.986) 0.593 (0.422) -0.011 (0.749) -0.042 (0.297) 0.562* (0.300) 0.268 (0.348) 0.254 (0.389) 19,471 -44761 488

3.078*** (1.001) 1.103*** (0.362) 1.330* (0.782) 0.412 (0.298) 0.069 (0.276) 0.233 (0.358) 1.034** (0.435) 18,279 -88820 488

3.141** (1.391) 1.136*** (0.377) 0.922* (0.522) 0.116 (0.339) 0.223 (0.376) 0.349 (0.374) 0.036 (0.364) 19,471 -341731 488

2.017** (0.950) 1.360*** (0.291) 1.849*** (0.616) 0.072 (0.426) -0.112 (0.456) 0.214 (0.415) 0.682 (0.420) 19,471 -1668135 488

2.240*** (0.741) 0.214 (0.293) 0.372 (0.538) -0.022 (0.196) -0.327 (0.254) 0.330 (0.312) 0.615** (0.314) 19,471 -1112264 488

2.890*** (0.832) 1.238*** (0.315) 0.097 (0.367) -0.267 (0.234) -0.905** (0.413) -0.332 (0.409) 0.348 (0.305) 19,471 -2588533 488

1.412 (1.509) 0.312 (0.469) 0.165 (0.508) -0.532* (0.318) 0.030 (0.451) -0.008 (0.454) -0.134 (0.502) 19,471 -1034348 488

Table A.5: Robustness to Sample and Model, Crime and Social Connectedness among Black Migrants, 1960-2009 Dependent variable: number of offenses reported to police Murder (1)

Rape (2)

Robbery (3)

Burglary (5)

Larceny (6)

Motor Vehicle Theft (7)

-0.120*** (0.027) 0.026 (0.021) -1,504,904 20,160 502

-0.102*** (0.030) 0.008 (0.023) -3,056,894 20,160 502

-0.225*** (0.040) 0.007 (0.031) -1,361,620 20,160 502

-0.010 (0.032) 0.054** (0.021) -155341 19,471 488

-0.082* (0.046) 0.069** (0.031) -125045 19,471 488

Assault (4)

A: Poisson Model, Including Big Cities Log HHI, Southern black migrants

xii

-0.145*** (0.036) Log number Southern black migrants 0.186*** (0.026) Log-likelihood -50,484 City-year observations 20,160 Cities 502

-0.149*** (0.036) 0.070** (0.028) -108,014 18,913 502

-0.178*** -0.189*** (0.039) (0.042) 0.182*** 0.117*** (0.030) (0.031) -462,988 -2,322,973 20,160 20,160 502 502

B: Negative Binomial Model, Excluding Big Cities Log HHI, Southern black migrants

-0.086*** (0.033) Log number Southern black migrants 0.160*** (0.022) Log-likelihood -42151 City-year observations 19,471 Cities 488

-0.030 (0.033) 0.087*** (0.023) -63021 18,279 488

-0.096** (0.040) 0.194*** (0.028) -101753 19,471 488

-0.058 (0.037) 0.084*** (0.027) -132359 19,471 488

-0.023 (0.027) 0.063*** (0.019) -135772 19,471 488

Notes: Compare these results to panel A of table 6. Regressions include covariates used in column 3 of table 4. Standard errors, clustered at the city level, are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01 Sources: FBI UCR, Duke SSA/Medicare dataset, Census City databook

Table A.6: Robustness to Crime Data, Crime and Social Connectedness among Black Migrants, 1960-2009 Dependent variable: number of offenses reported to police Murder (1)

Rape (2)

Robbery (3)

Assault (4)

Burglary (5)

Larceny (6)

Motor Vehicle Theft (7)

-0.024 (0.031) 0.046* (0.024) -2,340,694 19,327 488

-0.144*** (0.041) 0.050* (0.029) -984,318 19,209 488

-0.024 (0.031) 0.046* (0.024) -2,328,663 19,303 488

-0.153*** (0.041) 0.046 (0.029) -957,348 19,251 488

A: Trim Crime Counts Below 1/6 or Above 6 times City Mean Log HHI, Southern black migrants

xiii

-0.107*** (0.033) Log number Southern black migrants 0.160*** (0.021) Log-likelihood -37,709 City-year observations 15,552 Cities 487

-0.064* -0.231*** -0.126*** (0.035) (0.035) (0.041) 0.071*** 0.157*** 0.080*** (0.026) (0.026) (0.029) -73,979 -306,933 -1,485,669 16,193 18,424 15,769 488 488 488

-0.070*** (0.024) 0.059*** (0.018) -1,017,245 19,320 488

B: Trim Crime Counts Below 1/6 or Above 6 times City Median Log HHI, Southern black migrants

-0.144*** (0.032) Log number Southern black migrants 0.151*** (0.021) Log-likelihood -37,285 City-year observations 16,133 Cities 488

-0.075** -0.234*** -0.127*** (0.035) (0.035) (0.042) 0.067*** 0.155*** 0.082*** (0.026) (0.026) (0.030) -72,946 -303,605 -1,483,895 16,306 18,441 15,746 487 488 488

-0.083*** (0.022) 0.053*** (0.017) -885,429 19,303 488

Notes: For each city, we calculate the mean and median number of crimes across 1960-2009. Results in panel A exclude city-year observations which are below 1/6 of the city mean or above 6 times the city mean, as in Chalfin and McCrary. Results in panel B instead use the median. Regressions include covariates used in column 3 of table 4. Standard errors, clustered at the city level, are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01 Sources: FBI UCR, Duke SSA/Medicare dataset, Census City databook

0

Total Number of Murders, ASR 500 1000 1500

2000

Figure A.1: Comparison of crime from different data sets

0

500 Total Number of Murders, UCR 1980 1989

1000

1984 45 degree line

0

Total Number of Murders, ASR 50 100 150 200

250

(a) All cities

0

50

100 150 Total Number of Murders, UCR 1980 1989

200

1984 45 degree line

(b) Cities with 1980 population below 500,000

The UCR data contain the total number of murders per police agency. To construct a similar measure from the ASR data, we calculate the sum of murders committed by adult whites, adult blacks, adult other races, juvenile whites, juvenile blacks, and juvenile other races.

xiv

-.5

Coefficient on Log HHI 0

.5

Figure A.2: Social Connectedness and Crime Across Years

1960

1965

1970

1975

1980 1985 Year

1990

1995

2000

2005

1990

1995

2000

2005

-.6

-.4

Coefficient on Log HHI 0 -.2

.2

(a) Murder

1960

1965

1970

1975

1980 1985 Year

(b) Motor Vehicle Theft

Figure A.2 shows point estimates and 95-percent confidence intervals from estimating a separate regression for years 1960-1964, 1965-1969, and so on.

xv

The Effect of Social Connectedness on Crime ...

Sep 22, 2015 - ... he started as a janitor at Fairbanks Morse and Company, a manufacturer. ..... in a city, which is variation in the vertical dimension of figure 2.

439KB Sizes 2 Downloads 166 Views

Recommend Documents

The Effect of Social Connectedness on Crime
Apr 5, 2017 - tative evidence suggests that Southern birth town networks ..... flows for African Americans with ties to the South.10 The direct effect of social.

The Effect of Social Connectedness on Crime: Evidence from the ... - IZA
Aug 23, 2017 - HD007339) and an NICHD center grant (R24 HD041028) to the Population Studies Center at the University of Michi- gan. .... Historical accounts point to the importance of migrants who were well connected in their birth town and who worke

On the Effect of Connectedness for Biobjective Multiple ...
a polynomial expected amount of time for a (1+1) evolutionary algorithm (EA) ... objective long path problems, where a hillclimbing algorithm is outperformed by.

The Effect of Social Entitlement Programs on Private ...
Email addresses: [email protected] (Kristopher Gerardi), ..... children. This combination would create an upward bias on the estimates of β1 and γ1,. 12 ..... B., Jimenez, E. 2004. How responsive are private transfers to income?

The Effect of School Starting Age Policy on Crime
and Dhuey (2012) explain their positive effect of school starting age on adult wages, ... Bayer, Hjalmarsson, and Pozen (2009) find not only that social networks ...

The Effect of Crossflow on Vortex Rings
The trailing column enhances the entrainment significantly because of the high pressure gradient created by deformation of the column upon interacting with crossflow. It is shown that the crossflow reduces the stroke ratio beyond which the trailing c

The Effect of Crossflow on Vortex Rings
University of Minnesota, Minneapolis, MN, 55414, USA. DNS is performed to study passive scalar mixing in vortex rings in the presence, and ... crossflow x y z wall. Square wave excitation. Figure 1. A Schematic of the problem along with the time hist

On the Effect of Bias Estimation on Coverage Accuracy in ...
Jan 18, 2017 - The pivotal work was done by Hall (1992b), and has been relied upon since. ... error optimal bandwidths and a fully data-driven direct plug-in.

On the Effect of Bias Estimation on Coverage Accuracy in ...
Jan 18, 2017 - degree local polynomial regression, we show that, as with point estimation, coverage error adapts .... collected in a lengthy online supplement.

The effect of mathematics anxiety on the processing of numerical ...
The effect of mathematics anxiety on the processing of numerical magnitude.pdf. The effect of mathematics anxiety on the processing of numerical magnitude.pdf.

The effect of mathematics anxiety on the processing of numerical ...
The effect of mathematics anxiety on the processing of numerical magnitude.pdf. The effect of mathematics anxiety on the processing of numerical magnitude.pdf.

The effect of ligands on the change of diastereoselectivity ... - Arkivoc
ARKIVOC 2016 (v) 362-375. Page 362. ©ARKAT-USA .... this domain is quite extensive and has vague boundaries, we now focused only on a study of aromatic ...

The Effect of Recombination on the Reconstruction of ...
Jan 25, 2010 - Guan, P., I. A. Doytchinova, C. Zygouri and D. R. Flower,. 2003 MHCPred: a server for quantitative prediction of pep- tide-MHC binding. Nucleic ...

The influence of street lighting on crime and fear of crime
of conventional crime prevention wisdom. These ideas, usually ...... discussions with the police computer system managers, it was found that police beat allocations were ... At the end of each working day these files were downloaded onto a ...

The Effect of Social Norms on Sodium Intake in Young ...
We also predicted that participants in the social norm conditions would be more likely to report attempting to lower their sodium intake in the week following the experiment than participants in the control condition. METHOD. Participants. The study

Effect of earthworms on the community structure of ...
Nov 29, 2007 - Murrell et al., 2000). The development and application of suitable molecular tools have expanded our view of bacterial diversity in a wide range ...

The effect of Quinine on Spontan.Rhythmic contrac. of Rabbit Ileal ...
The effect of Quinine on Spontan.Rhythmic contrac. of Rabbit Ileal smoot. musc..pdf. The effect of Quinine on Spontan.Rhythmic contrac. of Rabbit Ileal smoot.

Effect of Torcetrapib on the Progression of Coronary ...
29 Mar 2007 - additional use of these data to understand the mechanisms for adverse cardiovascular outcomes observed in the suspended torcetrapib trial. Methods. Study Design. The Investigation of Lipid Level Management Us- ing Coronary Ultrasound to

Effect of Torcetrapib on the Progression of Coronary ...
Mar 29, 2007 - Pinnacle Health at Harrisburg Hospital, ... of Lipid Level Management to Understand Its Im- ...... College of Cardiology Task Force on Clin-.

An examination of the effect of messages on ...
Feb 9, 2013 - regarding promises rather than testing guilt aversion under double-blind procedures or discriminating among various models of internal motivation. (5) In CD, messages were sent before As made their decisions, and Roll choices were made

An examination of the effect of messages on ... - Springer Link
Feb 9, 2013 - procedure to test the alternative explanation that promise keeping is due to external influence and reputational concerns. Employing a 2 × 2 design, we find no evidence that communication increases the overall level of cooperation in o

25 Effect of the Brazilian thermal modification process on the ...
25 Effect of the Brazilian thermal modification process ... Part 1: Cell wall polymers and extractives contents.pdf. 25 Effect of the Brazilian thermal modification ...

The Effect of the Internet on Performance, Market ...
May 19, 2017 - are not the most popular ones, without affecting other movies. .... studies the impact of various policy, economic, and social changes, .... net users–where Internet users are people with access to the worldwide network. ..... on the