Learning Center
Plans & pricing Sign in
Sign Out

Lindahl Lecture Social Interactions Crime Ghettos and Hatred


  • pg 1
									Lindahl Lecture 2: Social
   Interactions, Crime,
   Ghettos and Hatred
    Edward L. Glaeser
    Harvard University
           Plan of Lecture
• Social Interactions: Theory and
  Motivational Facts
• Measuring Social Interactions
• Measuring Segregation and its history
• Are Ghettos Good or Bad?
• Urban Crime
• Hatred and Redistribution
  The Power of Social Interactions
• The Strong Form of the Social Interactions
  Hypothesis is that we are largely formed
  by the people around us
• Skills (Kain and Spatial Mismatch)
• Preferences
• Beliefs
• Other positive interactions (e.g. reducing
  probably of being caught)
          Localities and Beliefs
• Human beings beliefs seem as much a function
  of who you listen to as truth

• Redistribution in the U.S. and Europe
  – 60 percent of Americans believe that the poor are
    lazy (26 percent of Europeans share that view);
  – 54 percent of Europeans believe that luck determines
    income (30 percent of Americans share that view).
  – But mobility rates out of poverty are higher in Europe
    and the American poor work at least as many hours
Religious Beliefs Show Both Great
     Variation and Conviction
• 75 percent of Northern Irish and 71
  Percent of Americans believe in the devil
• 33 percent of the English, 19 percent of
  the French and 11 percent of Danes share
  that view
• Gaps between NYC and Texas within the
  U.S. are at least 30 percent
• I don’t know who is right, but they can’t all
  Approaches to Measuring Social
• Correlation with group averages

• Impact of Neighborhood more generally

• Group level heterogeneity, excess

• Social Multipliers
            Group Averages
• Basic approach is to regress:
      Individual Y=a+b*Group Y
• Problem # 1: Reflection if the group effects the
  individual then the individual effects the group
• Problem # 2: Omitted Variables that impact both
  group and individual Y
• Step # 1: use exogenous characteristics of the
  group can fix this (Case and Katz, 1991)
           Other Problems
• Problem # 3: Exogenous characteristics of
  the group may directly impact the
  individual outcome (Manski)
• Problem # 4: Correlation between
  individual unobservables and group
  observables (People Sort)
• Problem # 5: General Equilibrium Effects –
  people should be indifferent across
    The Rise of the Randomized
• Moving to Opportunity (Katz, Kling,
  Liebman, Ludwig)
  – Randomized Trial Voucher experiment
  – Confusing Income and Location
  – Effects are modest on kids
• Sacerdote (2003) QJE– randomized
  roommates – big social, small academic
• Aslund, Edin and Fredriksson (2003)–
  randomized immigrants– big effects
Excess Variance: Theory (QJE 96)
• Basic statistics tells us that the standard
  deviation of a set of averages should be
  p(1-p) divided by the square root of n
• But for many things, variances are much
• In the case of crime, variance is 1500
  times crimes per capita*1-crimes per
     Three Explanations for High
      Variance of Crime Rates
• Omitted Variables (policing, etc.)
• High crimes per criminal
• Social interactions
  – Congestion of law enforcement (but areas
    spend more when there is more crime)
  – Legitimizing crime/preference formation
  – Transmitting Knowledge
  – Reducing returns to legal activities
           The Basic Model
• Individuals are either 0 or 1, either black or
• They are on a lattice and occasionally
  imitate their neighors
• Some neighbors are fixed and never
  imitate (these are people with a high
  desire for crime or innocence)
• We look at long run behavior
      Results from the Model
• Variance of a community average
  converges to a normal with variance p(1-p)
  times f(pi), where pi is the proportion of
  fixed guys in the latice
• In the uni-directional case (only imitate on
  one side) f(pi)=(2-pi)/pi
• Thus social interactions scales up the
  variance by approximately twice the
  average distance between fixed guys
    Excess Variance: Empirics
• The basic estimate of f(pi) might be 1300
  from the crime figures, but this doesn’t
  correct for omitted variables and crimes
  per criminal
• We deal with omitted variables by using all
  observables, and again calibrating
  assuming that the variance of
  unobservables is some multiple of
    Excess Variance Empirics
• We also estimate fixed effects (but this
  might eliminate true social interactions)
• We also correct for crimes per criminal
  using a variety of estimates (3-141)
• Estimates of f(pi) are consistent across
  time and when using NYC or national data
• Estimates do differ significantly with
  different crimes per criminal and fixed
• f(pi) overall for crime with crimes per
  criminal of 6 is about 100
• The more serious the crime the lower the
  f(pi)– for murder and rape it is low and for
  muders less than 4 usually
• Crimes which are perpetrated by the
  young (larceny) have high f(pi)
• Fixed effects estimates are 140 for larceny
    Issues with Excess Variance
• This approach does not avoid the
  problems with estimating social
  interactions (except # 1)
• Unobserved heterogeneity is a particular
  problem– we try to estimate this directly
  and then calibrate
• We also use functional form and the
  scaling property to test this
         The Social Multiplier
• Basic Idea– The presence of social interactions
  means that macro-elasticities will be bigger than
  micro-elasticities because the macro estimates
  include spillovers
• Modest interpretation: macro-estimates, and all
  state-level regressions are biased and
  estimating a combination of micro-coefficient
  and social multiplier

• Aggressive interpretation: we can actually
  estimate spillovers
                 Basic Setup
• y=b*y+c*z+e,
  – y is community average,
  – z is an exogenous characteristic
  – e is the error term

• Within communities if you regress y on z, you
  will get the correct coefficient of c.

• Raw regressions are biased upwards because
  of the correlation between your z and community
  y, but this bias is pretty small in big places
       Social Multiplier Theory
• But taking the average in the community we find
  y=b*y+c*z+e, or y=c/(1-b)*z+e/(1-b)

• If we regress community average levels of y on
  community average levels of z, we estimate c/(1-
  b)– which is potentially much higher

• In the general model, the size of the social
  multiplier will rise with community size because
  this includes all neighbors
     Social Multiplier Empirics
• Dartmouth roommates estimates of joing a
  fraternity of drinking beer in high school–
  coefficient rises from .1 to .15 to .23
  moving from individual to floor to dorm

• With multiple z’s it may make sense to first
  estimate “c” from micro-regressions and
  then regress y on cz the resulting
  coefficient is 1/(1-b)– the multiplier
 More Social Multiplier Empirics
• With this approach, the coefficient of
  country crimes on predicted crime (using
  micro-data and demographics) is 1.7 and
  2.8 for states
• Using time series coefficients are 4.4 for
  homicides and 8.1 for overall crimes
• PUMA wages on predicted wages are 1.7–
  state wages on predict wages are 2.2
        Segregation in Cities
• American blacks are enormously
  segregated– and have been for more than
  100 years
• Other immigrant groups can also be
  segregated– Hispanics, immigrants at the
  start of the century
• Potentially, at least, this is problematic– at
  the very least it is an interesting puzzle
      Measuring Segregation
               blackstract non  blackstract
Dis  .5                  
        tracts blackstotal   non  blackstotal
             blackstract blackstract
Iso                    *            if totaltract  k
      tracts blackstotal   totaltract
            blackstract non  blackstract 
then      blacks  non  blacks       
     tracts       total             total 
           totaltotal                      blackstotal   
                                     Iso 
                                                         
k  non  blackstotal blackstotal
                                            totaltotal   
Dissimilarity over Time
                                 Isolation over Time
                                                      Figure X: Mean Isolation 1890-2000




Index of Isolation






                          1890   1900   1910   1920      1930       1940          1950       1960            1970   1980   1990   2000

                                                      Unweighted Average      Weighted by Black Population
Bigger Cities are more Segregated
                                                               Figure 5: Dissimilarity by MSA Size, 2000

                              MSAs with population under
                                  200,000 (N=123)

                           MSAs with population between
  Categorization of MSA

                           200,000 and 750,000 (N=138)

                           MSAs with population between
                           750,000 and 1,500,000 (N=30)

                          MSAs with population exceeding
                                1,500,000 (N=42)

                                                           0     0.1          0.2         0.3            0.4           0.5             0.6     0.7
                                                                                       Dissimilarity Index, 2000
 Other Facts about Segregation
• The Midwest is the most segregated region,
  followed by the Northeast, the South and the
• Segregation is highly persistent, Cleveland and
  Chicago were among the 5 most segregated
  cities for 100 years.
• Education blacks are much less segregated than
  non-educated blacks today (.69 vs. 54 in 90), but
  not in 1970 (.76 vs .74)
                        Dissimilarity and Regions
                                                               Figure 2: Changes in Dissimilarity by MSA Growth Rate

                                -0.065                                                                                                   Growth over 25% (N=71)
Categorization of MSA

                                           -0.057                                                                              Growth between 10 and 25% (N=109)

                                                    -0.05                                                                               Growth under 10% (N=90)

                                                                                        -0.034                                        Negative growth rate (N=21)

                        -0.07            -0.06         -0.05                   -0.04                 -0.03             -0.02               -0.01                    0
                                                                          Change in Dissimilarity, 1990-2000
   Understanding Segregation
• Decentralized Racism– whites are willing
  to pay more to live with whites than blacks
  are willing to pay to live with whites

• Ports of Entry– blacks want to live with
  other blacks

• Centralized Racism– whites force blacks
  to live together through violence and law
 Testing Between The Theories
• All can explain the rise, and in principle all can
  predict the decline.
• Housing price results, however, will differ by
  theory– the group that is constrained by
  prefrences or force will pay more
   – Kain and Quigley find blacks pay more in St. Louis
   – Other studies haven’t duplicated this– housing quality
   – We look at the segregation– price premium
• Also, centralized racism predicts extremely white
  census tract.
 The Distribution of Census Tracts
Year:       1940   1970   1990

0 Blacks    21.2   19.6   7.3

Less than   39.1   36.2   10.2
1% Black
1-15 %      23.9   21.2   42.5
15-50%      7.8    9.6    16.8

50+%        7.9    13.4   23.3
Segregation and Housing Prices
• Ln(Rent)=a+b*Black+c*Black*Segregation
  +Other controls for city and unit
• In 1940, rental coefficient on black is -
  1.4(.36), interaction is 1.3(.5)
• In 1970, coefficient on black is -.42(.13)
  interaction is .38(.16)
• In 1990, coefficient on black is .15(.07)
  interaction is -.38(.1)
    Segregation and Attitudes
• The General Social Survey collects
  information on opinions on race
• In more segregated cities, blacks are more
  likely to say that they prefer white
  neighborhoods (not significant)
• In more segregated cities, whites say that
  they believe in the right to segregation and
  support a ban on interracial marriage
   Are Ghettos Good or Bad?
• Outcome=a+b*black+c*black*segregation
  +metro area and background controls
• 20-30 year old blacks in 1990 are less
  likely to work, earn less, get less
  education, are more likely to be teenage
  parents, in more segregated cities
• Effect is less true in previous decades
• Effects are large and robust
    Problems with Estimation
• Selection on omitted variables
  – One reason not to use within city variation
  – Little evidence for migrants sorting on
    observables into less segregated area
• In places where blacks do worse they
  segregate more
  – Use Rivers (Hoxby) or long-standing
    government barriers as instruments
 Why is there more crime in Cities
• Overall serious crimes have an elasticity of
• This was .28 in 1970– declining crime in
  big cities is a factor
• Correcting for more underreporting in big
  cities– the elasticity is .24
• Victimization results show a ten
  percentage point difference
           A Decomposition
• People commit crimes when B>K+PC
• B(Y,Q)=K(X)+P(Y)C
• e=(P/Q)dQ/dP=PC/(Q*dB/dQ)
  – Use Levitt’s estimate of .2 for this
• City Elasticity=e*(N/P)(dP/dN)-
     eB/PC(dB/N)+dk/dN factors
  – Lower Probilities of arrest
  – Greater Availability of targets
  – Selection into cities
   Lower Probability of Arrests in
• Congestion of law enforcement– keep
  number of crimes/cop constant but
  increase the number of suspects in a city
• Arrests per crime -- -.05
• Arrests per crime adjusted for
  underreporting-- -.13
• Police Officers per capita-- .05
• In total, can explain 10-20% of city size
 Higher Returns to Crime in Cities
• Density increases the availability of street
  targets (street crime is higher around
  apartments)– proximity is good for crooks
• Value per crime .05-.09
• Victimization Survey goes up to .11-.13
• Overall this can possibly explain another
  15 maximum of the crime-cities connection
  Social Heterogeneity in Cities
• The city size effect disappears when we
  control for percent single parent family
• What does this mean– maybe this is itself
  a result of social disfunction
• Instrument using AFDC laws interacted
  with poverty rate in 1970
• This can explain the remaining ½ of the
  city size effect
• This also reflects the social multiplier
 The Economics of Punishment
• The basic model is a positive theory of
  crime, a normative theory of punishment
• But punishment itself is less than optimal
  and worth trying to understand
• There are several different levels of choice
  – Choice of laws and rules
  – Choices by judges and prosecutors
   What do Prosecutors Maximize
   (ALER with Kessler and Piehl)
• In the U.S., in crimes which are both federal and
  state, federal attorneys essentially get to choose
  who to go after
• They chose to go after people without records,
  who are rich, white, likely to have private
  attorneys, nonviolent
  – Career concerns, i.e. skills, connections
  – Even true among possession cases
• This may be socially beneficial if you need good
  federal attorneys to beat good private attorneys,
  but may also be bad
    Sentencing for Homicides
Optimal Sentencing might minimize
    N is number of crimes
    P is probability of being caught
    C is social cost of sentence
    L is sentence length
    V is cost of crime occuring
         Optimal Sentencing
-N’(PL)*P(PCL+V)=N(PL)*PC or
eV=(1-e)PCL where e is supply elasticity or
• Facts about sentencing across murders
  – The probability of being caught fact is about
  – The supply elasticities are hard to measure,
    but seem close to right
  – Little victim effects
  Other Facts about Sentencing
• Big victim effects for homicide sentences
   – Killing a black leads to a 40 percent shorter sentence
   – Killing a woman leads to a 50 percent longer
• Perpetrator effects are quite weak
• These effects are true even when addressing
  vehicular homicides (drunks)
• The sentencing in those crimes is itself a little
• The role of emotions and redistribution in
  sentencing by Juries
    The Theory of Punishment
• Key Insight of Public Choice – design rules
  to deal with incentives by implementers
• Bureaucrats vs. Independents (English vs.
  French)– check against gov’t vs. laziness
• Judicial Codes– good for controlling
  independent judges
• Quantity controls make things easier to
         Riots (JUE, 1998)
• Generally an urban phenomenon– rural
  uprisings were easier to suppress
• Important both because of the raw
  damage and because of regime change
• Basic model is getting enough people to
  overwhelm the police
• Initial crowds
• Emotional Minority
• Organization
Basic Rioting Theory

        Costs of Crime

                           Benefits of Crime

 Number of Criminals/Rioters
           Facts about Riots
• Across countries US + India have the most riots

• Increases with ethnic heterogeneity,
  urbanization and the interactions

• Decreases in Dictatorships

• Decreases with GDP
                In the U.S.
• In the 60s and today, little correlation with
• Homeownership decreases riot
  occurrence rates in the 60s
• South had fewer riots
• In places with more cops, riots were less
• Big variable is size of the non-white
The Political Economy of Hatred
• Hatred is the willingness to pay to hurt others
  (terrorism can be a version of it)
• Also forms of race hatred (anti-Semitism, anti-
  Black behavior in the U.S. South)
• Not the same thing as discrimination
• Not automatic with fragmentation, and quite
• A combination of emotions and cognition
  Psychology in the market– rational
  entrepreneurs and less rational people
                Motivational Facts
                      Social Welfare Spending                                                  Fitted values
20          Luxembur
            Sweden Netherla

           Spain                           Uruguay

           Ireland                                                       New Zeal
         Finland           UK                                   Israel
10             Switzerl
                              Chile               Canada

                            Australi                                                          Brazil
     Iceland                                      Argentin
5                                      Barbados                                     SriLanka
                 Costa Ri
                                    Nicaragu                                        Mexico        Trinidad
                                   Paraguay                                      Bolivia          Colombia
                                                                                       Fiji Isl
                                       Honduras    Thailand                    Rep. Dom
                                                          Turkey Singapor                Guatemal
                                             Philippi                                               EcuadorPeru

       0                                   .2                          .4                                .6       .8
                                                             Racial Fractionalization
            Welfare within U.S.
                     afdc maximum benefit for a thre                              Fitted values


                                                        New York
                Rhode Is
                   Wisconsi                       Michigan
      New Hamp

           Oregon                                New Jers
            Iowa     Kansas
         Utah                                                                            Maryland
      North Da
      South Da                                        Illinois
      Montana Nebraska
       Wyoming Colorado                                            Virginia
                         Oklahoma    Ohio                    Delaware
             Arizona          Indiana Missouri    Florida
          New Mexi                                                            North Ca         Georgia
             West Vir
                                                        Arkansas                                         South Ca
                                           Texas        Tennesse                                            Louisian

                                                                                          Alabama                        Mississi
      .00298                                                                                                           .355608
                                                      % pop. black, 1990
Anti Americanism around the World
• Please tell me if you have a very
  favorable, somewhat favorable, somewhat
  unfavorable or very unfavorable opinion of
  the United States
• Vietnam 4% very vs. 8% for France and
  Canada (27 vs. 34 and 27 including
• Argentina 23 % very vs. 3% Guatemala
  and 2% Honduras (49 vs. 13 and 5)
Anti Americanism and Islam
                     Highly Unfavorable to U.S.                        Fitted values
 59                                                                                                          Pakistan



      Bolivia                                                                                 Banglade

       South Af
                     India                                                             Mali

       Mexico                                 Tanzania
      South Ko
      SlovakiaUganda Russia
        Peru Kenya  Bulgaria       Cote d'I                  Nigeria
      Guatemal             Ghana
      Honduras                                                                                  Uzbejist
         0                                                                                                      99.8
                                                muslims as % pop 1980 w ce95
   Emotional Roots of Hatred
• Asdasd

• Chemicals

• Ultimatum Games

• Murders, Riots, Gangs and Vengeance
      The Formation of Hatred
• Hatred is always and everywhere formed by
  stories of past and future atrocities
  – Tales of Blacks raping white girls in the South
  – Tales of Jews killing Jesus, drinking children’s blood,
    and the Protocols of the Elders of Zion
  – America and the Child in Barcelona
• Often these stories need repetition more than
  – Or sometimes they are true, but that group isn’t
    particularly guilty
The Role of Political Entrepreneurs
• White conservatives in the south pushed
  race hatred in the 1880s and 1890s
• Right wing politicians in Europe (Lueger,
  Schonerer, Hitler) pushed anti-Semitism
• So did the Czar (Ochrana gave us the
• Today, anti-Americanism is also used by
  different political groups
              The Model
• The setting two parties competing for
  votes and sending out messages of hate
• Voters may investigate, form opinions
  about the dangers of an out-group
• Voters choose between the candidate
• In-group voters choose whether or not to
  isolate themselves from the out-group
• Contact with out-group might occur
     Key decisions recursively
• Will in-group members isolate themselves
  – Only when they have heard a hate creating message
    and not investigated
• Which candidate will voters support
  – They vote their pocketbook except
  – When they have heard a hate creating message and
    not investigated and then they are more likely to favor
    policies that hurt the minority
• Will voters investigate the message
  – If it will change their isolation behavior
• Will politicians send hate creating messages
  – If it will increase votes
     Key Comparative Statics
• Private individuals are more likely to
  investigate hate-creating messages when
  – The minority group is large
  – The minority group isn’t segregated
  – The potential harm is large and the gains from
    self-protection are high
• Politicians are more likely to send a hate
  creating message when
  – This message is unlikely to be investigated
 Comparative Statics on Supplying
• Hatred will be more likely to be supplied
  when the group is potentially more of a
  threat (holding search constant)
• Hatred is more likely when the group is
  different along the policy-relevant
• The right pushes hate against poor
  minorities the left against rich minorities
             Other Results
• With two issues the key is whether the
  group is policy relevant
• More extremism on the issue that the out-
  group is different on leads to more hatred
• Hating the haters can be an effective
• Policies related to migration or segregation
  are natural complements to hatred
   Understanding why race hatred
    rises between 1870 and 1910
• In the 1880s, depression created fertile ground for the
  first American party, the Populists, committed to
  redistribution from rich to poor.
• “More important to the success of Southern Populism
  than the combination with the West or with labor was the
  alliance with the Negro… Populists of other Southern
  states followed the example of Texas, electing Negroes
  to their councils and giving them a voice in the party
  organization.” (Woodward)
• “I have no words which can portray my contempt for the
  white men, Anglo-Saxons, who can knock their knees
  together, and through their chattering teeth and pale lips
  admit that they are afraid the Negroes will ‘dominate us.’”
           But the response was
• “Alarmed by the success that the Populists were
  enjoying with their appeal to the Negro voter, the
  conservatives themselves raised the cry of ‘Negro
  domination,’ and white supremacy, and enlisted the
  Negrophobe elements”

•    “In Georgia and elsewhere the propaganda was
    furthered by a sensational press that played up and
    headlined current stories of Negro crime, charges of
    rape and attempted rape, and alleged instances of
    arrogance … already cowed and intimidated, the race
    was falsely pictured as stirred up to a mutinous and
    insurrectionary pitch” (both from Woodward)
       Decline of Race Hatred
• Tom Watson by 1906 said the black man “grows
  more bumptious on the street, more impudent in
  his dealings with white men, and then, when he
  cannot achieve social equality as he wishes,
  with the instinct of the barbarian to destroy what
  he cannot attain to, he lies in wait, as that
  dastardly brute did yesterday near this city, and
  assaults the fair young girlhood of the south...”
• Key lessons– strategic, related to policy
  relevance, not related to truth
    Anti-Semitism in 19th Century
• Political, not religious, and big in Russia,
  Germany, Austria, mixed in France
• Not in U.S., U.K., Italy or Spain
• Key ideological divide in the first country is king
  vs. constitutionalism (Kaiser in 1871)
• The Austrian empire “a political system so
  flagrantly out of step with the spirit of the times
  needed at least one strong ideological ally; this
  ally by a process of elimination could only be the
  Church.” (Kann)
     19th Century Anti-Semitism
• Cohn (1956) wrote “the Right (conservative,
  monarchical, ‘clerical’) maintained that there must be a
  place for the Church in the public order; the Left
  (democratic, liberal, radical) held that there can be no
  (public) Church at all.”

• And “Jews supported the Left, then, not only because
  they had become unshakeable partisans of the
  Emancipation, but also because they had no choice; as
  far as the internal life of the Right was concerned, the
  Emancipation had never taken place, and the Christian
  religion remained a prerequisite for political
  If Jews are on the left, then right
      wing anti-Semitism follows
• “from Stoecker to Hitler, rightists rarely
  attempted to refute socialism, preferring to cite
  the high percentage of intellectuals of Jewish
  origin among socialist publicists as proof of its
  subversion” (Weiss, 1996).
• In 1892, the conservative party platform
  embraced anti-Semitism and pledged to “do
  battle against the many-sided aggressive,
  decomposing, and arrogant Jewish influence on
  the life of our people” (Weiss, 1996, p. 116).
   Russia, Austria and France
• In Russia, the Czar used anti-Semitism to
  build up support for his pro-Church regime
• In Austria, anti-Semitism was actually
  used against the Emperor by Lueger
• In France, the right wing tried (Dreyfus)
  but were defeated by left wing strength
  – Zola describes the War Office that convicted
    Dreyfus as a “nest of Jesuits” prone to
    “inquisitorial and tyrannical methods.”
            Italy and Spain
• Spain’s easy– no Jews post 1492 (or at
  least 1600)
  – There was some anti-Masonic hatred that
    played a similar role (also in U.S.)
• Italy is more interesting– the modern state
  was founded on expropriation of the Pope
  – As such, the king and everyone in politics was
  – As such, there was no church in politics, and
    Jews weren’t policy relevant
            U.S. and U.K.
• Divine right monarchies and church and
  state issues were settled long before the
  19th century
• As a result, Jews weren’t particularly
  policy relevant and occupied both sides of
  the political aisle
• Disraeli and Judah Benjamin

To top