Crime 250C Spring 08. (1-4 4 follow Freeman 99) 1. Motivation. 2. Magnitude and trends. 3. Market model. 4. Prevention and policy

Transcription

1 Crime 250C Spring Motivation 2. Magnitude and trends 3. Market model 4. Prevention and policy 5. Back to gangs and insurgencies (1-4 4 follow Freeman 99)

2 1. Motivation Can incentive models help? - Heterogeneity vs. incentives in incarceration How important are supply factors, demand (returns to crime), interventions? Should labor market interventions account for effect on crime? Possible applications of: attitudes to risk, social interactions matter, strategic behavior? Can we explain trends and demographic patterns in incidence? Cost effective policies - normative constraints

3 Other fields in play Criminology Psychology of deviant behavior Sociology crime runs in families, childhood experiences, abuse Genetics? Ethnographers work on gangs Ethicists on punishment, legalization

4 2. Magnitude and Trends 1997: 13.5m crimes reported by police (5 per thousand residents) m reported in surveys National survey of drug abuse: 2.6% of adults report a felony in past year 30% of adult males arrested for a serious crime during their lifetimes, about 9% will be imprisoned. 1997: 1.7m americans incarcerated, 3.2m on probation,.7m paroled (2.9% of adults altogether, about 5% of adult males)

5 Demographic predictors 1991: 67% of inmates aged (twice population share) - 2/3 HS dropouts In US and worldwide, arrest rates increase with age, peak around 20 and then decline Arrests and convictions disproportionately young, male and black (in US) Among 25/-34 y.old dropouts 12% incarcerated Victims have the same predictors: young, male, black, less- educated Skewed intensity: 23% of delinquents commit 61% of offenses in Philadelphia study (Tracy et al 1985) Among prisoners in CA above median average is 10.6 crimes/year, below median average is 0.6 (Greenwood, 1996)

6 Age profile

7 Cost 1997: $100B in public spending on police, corrections and judicial system Private security is estimated to cost 170% as much. - increased from 62K workers to 690K workers from

8 Trends in crime rates Why the disparity in levels? Reporting rates. Trend increase in reporting rates (Bound & Bogess 1993)

9 Labor econ and trend? Donoghue and Levitt Abortion? (QJE 01) Foote and Goetz (06) - found a mistake in the coding D & L (06) admit the mistake and try a new approach Dills and Miron (06) critique the new approach Might stick with Reyes (07) which links decline in crime to cross state variation in lead exposure

10 Trends in incarceration 1997: 1.7m incarcerated, up 10x over 30 years.

11 3. Market Models Potential criminal (1-p)U(W c ) pu(s) ) > U(W) Implies: W c > W.. But that doesn t t seem to be the case in Chicago Attitudes to risk matter (risk neutrality assumed here) Outside opportunities (W) matter Punishment (S) matters Enforcement (p) matters Misses: No retaliation for defection No heterogeneity

12 Incarceration and Incapacitation Incarceration (one form of S) works two ways: 1. Deterrence (above) 2. Incapacitation U.S. experiment with incarceration in 80s and 90s indicates that supply of new criminals is fairly elastic in W c (Freeman 96), (Zimring( and Hawkins 91). Incapacitation matters: Dahl and Dellavigna 08 - between 6PM and 12AM, a one million increase in the audience for violent movies reduces violent crime by 1.1 to 1.3 percent.

13 4. Prevention and Policy Back to deterrence and the Becker model, what if you set very high S? Normative objection Community will respond by reducing cooperation with police (Akerlof( Akerlof-Yellen 94) Ehrlich (1973) found large deterrent effects Juries and judges will respond by reducing probability of conviction, negating Ehrlich s effect (Andreoni( 95)

14 Prohibition (Miron( and Zweibel 91) Estimated alcohol consumption using deaths and alcohol related Illness Enforcement failed due to noncooperation of public, and even of police (i.e., corruption) Organized crime became entrenched.

15 2. Rational Peasant Model Motivation Hearts and Minds Operating procedure of US and Allied Special Forces includes providing local public goods: 1. Control some territory 2. Ask population what services they want 3. Provide them 4. Ask population for information about insurgents 5. Use information to ambush or capture insurgents, allowing control of more territory 6. Repeat (1)-(5) (5) until entire country is controlled Note: a) This is less prosaic than hearts and minds b) Even the most disenfranchised population gets services

16 Akerloff-Yellen (94) Gang Model Motivation Gangs are limited by community norms of behavior e.g., if the parents complain that a member is selling drugs in a primary school, the member is disciplined by the gang now I understand that if you ain t got the community with you it s s just a matter of time before you got to close up shop. - Duck (the gang member who learned his lesson)

17 A-Y Y model with public goods (and otherwise stripped down) Community: representative agent chooses whether to snitch to police: 0 = c = 1 Gang: chooses level of crime: s = 0 Police: chase criminals and provide public goods

18 Community Utility from cooperating with police U c = (B( c sg - a c R) c where g = 0 is the level of govt. provided public goods, which complement public safety, and R is the expected value of gang retaliation, a constant. It s s linear, so at the corner solution, c=0 iff a c R = B c sg noncooperation constraint ; otherwise c=1

19 Gang Maximizes U g = (B( g A g mc) ) s where B g and A g are positive constants, m is enforcement effort set by police, s is crime. If B g A g mc = 0 when c=1, then gang will choose s so that the noncooperation constraint just binds, at s * = a c R / B c g, so s > 0 and s decreases in g. If B g A g mc > 0 when c=1, then gang chooses infinite s..police will set m high enough to avoid that.

20 Police Minimize A p s + B p m + C p g.. which yields a corner solution for m = B g / (A g (avoiding infinite crime at c=1), and an interior solution for g 2 = (A( p a c R)/(B c C m ). Note: reducing cost of public goods C m raises g & reduces s That could include reducing corruption in public good provision, or increasing the efficiency of taxation.

21 Back to Rational Peasant Analogy to Rational Peasant model of insurgency, where even the disenfranchised noncombatants are favored with public goods by Special Forces. Extensions s and g need not be complements If gangs can provide public goods can there be a race to the top in public good provision? If government can retaliate can there be a race to the bottom in extortion?

22 Clubs or Rational Peasants? The two models share defection or snitching (cooperation with authorities) In the club model a combatant defects, while in the rational peasant model a noncombatant snitches Who cares which model is relevant? - predicting tactic choice by rebels: - conventional tactics are cheaper, but share information with noncombatants (ambush, IED) while modern (club) tactics do not (suicide attack) - most organizations are not capable of club tactics because of defection constraints - targeting of police efforts - only club model relevant for domestic terrorism

23 Evidence from Iraq Why Statistics Can Help

24 1. Street Lights a public good Street lights Amara Qal'at Saleh Al-Mejar Dahuk Al-Kabi Resafa Al Sadr Kerbala Basrah Fao Suq Nassriya Erbil Al-Shoyokh Kut Adhamiya Khadamiya Al-Samawa Falluja Amedi Zakho Mada'in Al-Ka'im Najaf Tarmia Al-Rifa'i Al-Kahla Sumel Al-Hamdaniya HadithaKarkh Al-Zubair Tooz Ali Tilkaif Al-Gharbi Mosul Al-Musayab Mahmoudiya Hamza Al-Maimouna Al-KhalisAl-Muqdadiya Al-Rumaitha Al-Salman Al-Daur Al-Suwaira Kirkuk Al-Shatra Baiji Badra Al-Khidhir Ramadi Al-Hai Al-Qurna Abu Al-Khaseeb Shatt Hashimiya Hilla Al-Midaina Shaqlawa Al-Hindiya Ba'quba Soran Baladrooz Heet Al-Mahawil Kufa DiwaniyaTikrit Halabja Darbandihkan Al-Manathera Akre Sulaymaniya Penjwin Al-Arab Choman Al-Shamiya Telafar Khanaqin Makhmur Kifri Mergasur Koisnjaq Al-Shikhan Kalar Rania Al-Hawiga Samarra Afaq Chamchamal Sinjar Dokan Daquq Al-Shirqat Al-Na'maniya Sharbazher Pshdar Abu Ghraib Ain Al-Tamur Al-Chibayish Ana Street lights 02 Street lights 04 Street lights 04 Street lights are pretty durable, but some places improved and some got worse - These are measured on a scale of 1-bad to 3-good in a 2004 survey.

25 Who improved? 2 Amara Qal'at S Al-Mejar Dahuk Al Resaf street light Kerbala Basrah Kut Suq Fao Al-S Nassriya Erbil Al Sadr Adhamiya Khadamiy Falluja Al-Samaw Al-Ka'im Amedi ZakhoMada'in Tarmia Najaf Al-Rifa' Al-Hamda Sumel Karkh Al-Kahla Haditha Al-Zubai Tooz Ali Al-G Tilkaif Al-Suwai Mosul Al-Musay Mahmoudi Kirkuk Al-Maimo Al-Daur Al-Shatr Hamza Al-Khali Al-Salma Baiji Al-Rumai Al-Muqda Al-Khidh Ramadi Al-Hai Al-Qurna Tikrit DarbandiDiwaniya Al-Hindi Halabja Shaqlawa Kufa Abu Al-K Al-Midai Shekhan Ba'quba Akre Sulayman Soran Hashimiy Al-Mahaw Al-Manat Balad Penjwin Shatt Al-Hawig Hilla Baladroo Daquq Telafar Kalar Al-Shami Al-Na'ma Choman Khanaqin Afaq Al-Shirq Dokan Koisnjaq Heet Makhmur Sharbazh Samarra Chamcham Rania Mergasur Pshdar Kifri Sinjar Al-Shikh Abu Ghra Ain Al-T Badra street light in Dec 2002

26 Street lights by themselves don t Balad predict much Al-Daur Mosul Al-Hamdaniya Abu Ghraib Al-Muqdadiya Mahmoudiya Tikrit Tarmia Ramadi Al-Ka'im Falluja Heet Al-Mahawil BaladroozBaiji Al-Maimouna Haditha Karkh Khanaqin Al-Musayab Mada'in Kifri Samarra Al-Hawiga Telafar Khadamiya Abu Al-Khaseeb Adhamiya Al Sadr Amara Shatt Hilla Hashimiya Al-Midaina Ba'quba Al-Qurna Al-Arab Al-Khalis Makhmur Mergasur Ain Choman Koisnjaq Al-Tamur Soran Shaqlawa Al-Hindiya Al-Salman Afaq Sinjar Al-Shirqat Daquq Al-Shamiya Al-Shikhan Al-Manathera Akre Kufa Diwaniya Kirkuk Al-Suwaira Al-Khidhir Al-Rumaitha Tooz Pshdar Sharbazher Chamchamal Rania DokanHalabja Al-Na'maniya Penjwin Kalar Sulaymaniya Darbandihkan Al-Hai Al-Shatra Hamza BadraAli Tilkaif Al-Zubair Al-Kahla Al-Gharbi Al-Rifa'i SumelZakho Najaf Amedi Al-Samawa Nassriya Erbil Suq Fao Kerbala Kut Basrah Al-Shoyokh Al-Mejar Dahuk Resafa Qal'at Al-Kabi Saleh Al-Chibayish Street lights 04 Ana Incident per 1000 persons Fitted values Mean of Sig. Acts in 04 is 0.60 per 1000 per year Regression slope is (not stat. significant)

27 Multivariate regression: 2004 e( p_inc X ) Incidents 04 vs Street lights 04 (conditioning on street lights 02) e( p_inc X ) Incidents 04 vs Street lights 02 (conditioning on street lights 04) e( street_light X ) coef = , (robust) se = , t = e( street_light_02 X ) coef = , (robust) se = , t = 3.02 All statistically significant, F= - conditional on service provision after invasion, good services under Saddam predict violence vs. coalition forces - conditional on service provision under Saddam, post-invasion service provision predicts less violence vs. coalition forces

28 Could it be reverse causality? Violence could be bad for street lights. so look at the timing: Street Lights and Violence: 2004, 05, 06, 07 e( p_inc X ) Incidents vs Street Lights 04 Incidents e( street_light X ) coef = , (robust) se = , t = (conditioning on street lights 02) e( p_inc X ) Incidents e( street_light X ) coef = , (robust) se = , t = e( p_inc X ) Incidents 2006 e( p_inc X ) Incidents e( street_light X ) coef = , (robust) se = , t = Interpretation: Granger causality runs from lights in 02 and 04 to subsequent violence? Tactical implication: Location of escalation was predictable based on 02 and 04 data e( street_light X ) coef = , (robust) se = , t = -3.67

29 Street Lights and Violence: 2004, 05, 06, 07 OLS results for Incidents per 1000 persons, Dependent variable: Incident per 1000 persons Street lights (0.72) (1.18) (2.39) (2.88) Street lights (0.64) (1.08) (2.24) (2.81) Constant (0.55) (1.12) (2.28) (2.14) R squared Sample size Average incident per 1000 persons

30 Is it literally Street lights? Total Number of Incidents Mar Dec Hour of the day Total Number of Incidents

31 Is it literally Street lights? OLS results for Day vs Night Incidents per 1000 persons, Dependent variable: Incident per 1000 persons Log of incident per 1000 persons Street lights (1.44) (2.07) (2.33) (2.68) Street lights X night (1.64) (2.24) (3.71) (4.13) Street lights (1.42) (2.06) (1.81) (2.08) Street lights 02 X night (1.57) (2.20) (2.89) (3.17) Night (= 1) (1.52) (1.62) (1.88) (2.25) Constant (1.36) (1.48) (1.30) (1.53) R squared Sample size Average day-time incident per 1000 persons Average night-time incident per 1000 persons

32 Is it literally Street lights? e( p_inc X ) Incidents 04 vs Street lights 04 (conditioning on street lights 02) e( p_inc X ) Incidents 04 vs Garbage collection 04 (conditioning on street lights 02) Similar results if public garbage collection is used instead of 04 street lights e( street_light X ) coef = , (robust) se = , t = e( pub_garbage X ) coef = , (robust) se = , t = e( p_inc X ) Incidents 04 vs Street lights 02 (conditioning on street lights 04) e( p_inc X ) Incidents 04 vs Road quality 04 (conditioning on street lights 04) Similar results if road quality is used instead of 02 street lights e( street_light_02 X ) coef = , (robust) se = , t = e( road_qual X ) coef = , (robust) se = , t = 3.48 Implication: It looks like Past (current) public service provision predicts (reduced) violence

33 Recap: Data supports both club and rational peasant models The next steps: Models have an additional prediction about tactics and geography: - In areas where noncombatants do not cooperate with government we expect conventional attacks, - while in areas where noncombatants cooperate, suicide attacks and other tactics which do not share information with noncombatants are required.

34 Where did these data come from? Capable of pulling classified data from military sources, declassifying it, and making it available for civilian researchers. In this case, the SIGACTs and survey data come from the US Central Command - don t ask for these two datasets until we ve examined, cleaned and written a paper

35 What can we learn from CERP? (Commanders Emergency Response Program) CERP is spent on public service provision, among other things CERP per capita and sig. acts: Mosul But CERP is also Balad Al-Hamdaniya Balad Al-Daur directed disproportionately at the most troublesome areas Al-Muqdadiya Al-Daur Mahmoudiya Mahmoudiya Al-Hamdaniya Mosul Tarmia Al-Hamdaniya Tarmia Balad Ramadi Al-Muqdadiya Al-Daur Al-Daur Haditha Mosul Tikrit Heet Ana Al-Rutba Mosul Balad Mahmoudiya Falluja Tikrit Al-Hamdaniya Al-Ka'im Haditha Ramadi Abu Tarmia Al-Hawiga Ghraib Ana Haditha Al-Hawiga Mada'in Baiji Karkh Baladrooz Falluja Abu Ghraib Khanaqin Baladrooz Ramadi AnaAl-Muqdadiya Al-Rutba Abu Al-Hawiga Falluja Ghraib Heet Tikrit Karkh Al-Ka'im Tarmia Al-Muqdadiya Al-Ka'im Baiji Al-Mahawil Abu Heet Kifri Mada'in Al-Khaseeb Khanaqin Ramadi Al-Rutba Ana Falluja Khadamiya Al-Shirqat Telafar Adhamiya Baiji Tikrit Khadamiya Mahmoudiya Heet Al-Mahawil Al-Musayab Haditha Khanaqin Mada'in Al-Maimouna Baladrooz Makhmur Amara Al-Salman Al-Ba'aj Telafar Al-Zubair Sinjar Tilkaif Kerbala Kifri Kirkuk Al-Hawiga Badra Daquq Sadr Baiji Karkh Mada'in Al-Musayab Al-Mahawil Adhamiya Khadamiya Basrah Ba'quba Abu Al-Khalis Khanaqin Baladrooz Al-Mahawil Al-Zubair Resafa Sadr Mergasur ErbilKoisnjaq Al-Arab Al-Khaseeb Ain Al-Tamur Al-Salman Najaf Al-Shirqat KirkukDaquq Samarra Al-Musayab Adhamiya Basrah Sadr Al-Suwaira Hashimiya Kufa Afaq Sinjar Resafa Abu Al-Khaseeb Ain Al-Khalis Al-Tamur Al-Maimouna Al-Hindiya Akre Makhmur Shekhan Al-Zubair Telafar Tooz Al-Khalis Samarra Ba'quba Al-Musayab Tooz Al-Ka'im Al-Ba'aj Resafa Sadr Afaq Diwaniya Al-Shirqat Basrah Hamza Badra Shaqlawa Tilkaif Sinjar Akre Kifri Al-Ba'aj Daquq Al-Salman Al-Midaina Zakho Choman Soran Ain Al-Maimouna Ali Al-Manathera Kufa Sumel Amara Al-Samawa Najaf Hilla Dahuk Al-Mejar Al-Khidhir Al-Rumaitha Amedi Al-Kahla FaoShatt Shaqlawa Akre Al-Shikhan Qal'at Al-Qurna Al-Shamiya Diwaniya Kirkuk Shekhan Hatra Al-Gharbi Tilkaif Al-Tamur Telafar Daquq Chamchamal Afaq Darbandihkan Dokan Halabja Kalar Penjwin Pshdar Rania Sharbazher Sulaymaniya Al-Chibayish Al-Shatra Nassriya Al-Suwaira Al-Rifa'i Suq Kut Hamza Al-Na'maniya Al-Hai Kerbala Mergasur Al-Shoyokh Saleh Hashimiya Al-Arab Al-Kabi Sinjar Badra Shekhan Badra Dokan Al-Hindiya Makhmur Al-Zubair Al-Salman Abu Ghraib CERP spending per capita Al-R Incident per 1000 persons Fitted values

36 Fao Choman Penjwin Ali Al-G Ain Al-T Ana Mergasur Al-Rutba Shekhan Badra CERP per capita and sig. acts (zoom on CERP pc <100) sig acts / 1000 Linear prediction 15 Haditha Ramadi Al-Muqda Al-Hamda Mahmoudi 10 Falluja Tikrit 5 Al-Hawig Baladroo Karkh 0 Basrah Samarra Tooz Al-Suwai Ba'quba Diwaniya Amara Baiji Al-Ka'im Al-Maimo Al-Chiba Al-Kahla Darbandi Halabja Al-Mejar Afaq Al-Khidh Al-Manat Al-Qurna Al-Samaw Qal'at S Dokan Chamcham Al-Shami Al-Midai Al-Na'ma Al-Rumai Al-Rifa' Al-Hindi Al-Shatr Kalar Kufa Kut Pshdar Rania Sharbazh Sulayman Zakho Najaf Hilla Nassriya Suq Al-S Dahuk Sumel Hashimiy Soran Erbil Al-Khali Mosul Al Sadr Al Resaf Al-Shirq Shatt Al Al-Ba'aj Kirkuk Amedi Al-Hai Khadamiy Adhamiya Sinjar Mada'in Al-Musay Hatra Makhmur Shaqlawa Al-Mahaw Telafar Al-Zubai Al-Shikh Hamza Kerbala Akre Khanaqin Abu Al-K Daquq Kifri Tilkaif CERP per capita

37 Can Hearts & Minds Be Bought? Huge investment ~ $41B Uncertain impact Reconstruction spending IRRF CERP/CHRRP Community characteristics WFP ILCS

38 How to use CERP efficiently? To answer that research question we need access to more data In particular, we need a way to filter out the reverse causality due to CERP spending being directed at locations expected to be dangerous. One solution: Troop rotation data (declassified in the same way these data have been)