Journalof Economic Perspectives—Volume 18,Number 1—Winter 2004 — Pages163– 190

UnderstandingWhy Crime Fellin the 1990s:Four Factors that Explainthe Declineand Six that DoNot

Steven D.Levitt

rimefell sharply inthe United States in the 1990s, in all categories of crimeand allparts ofthe nation. Homiciderates plunged 43 percentfrom C thepeak in 1991 to 2001, reaching the lowest levels in 35 years. The FederalBureau of Investigation’s (FBI)violent and propertycrime indexes fell 34 and 29percent, respectively, over that sameperiod. These declines occurred essentiallywithout warning: leadingexperts were predicting an explosionin crime intheearly and mid-1990s,precisely the point when crimerates began to plunge. Although expertsfailed to anticipate the decline, there has beenno shortage ofhypotheses toexplain the drop incrimeafter the fact. Table1 presentsa tallyof aLexis-Nexissearch ofthe most frequently cited reasons forthe crime decline in articlesin major newspapers overthe period 1991– 2001. The single most frequent explanationgiven is the innovative policing strategies put intoplace. Thecrime declineis also frequentlyattributed to increased imprisonment, changes inthe marketfor crack cocaine, theaging of the population, toughergun controllaws, thestrong economy and increasesin the number of police. In this paper, Iattemptto sort out whycrime declined in the 1990s. I begin witha reviewof the facts. Ithen analyzethe leading explanations forwhy crime fell, lookingat possibledeterminants that changed insome substantial wayin the1990s. Most ofthe supposed explanations listedin Table1 actuallyplayed little direct role inthecrime decline, including the strong economy of the 1990s, changing demo- graphics, betterpolicing strategies, gun controllaws, concealedweapons lawsand increaseduse of thedeath penalty. Fourfactors, however,can account forvirtually y Steven D.Levittis theAlvin H.Baum Professorof Economics, University of Chicago, and ResearchFellow, AmericanBar Foundation,both in Chicago, Illinois. This article was written whilethe author was a fellowduring the 2002– 2003 academic year atCenter for Advanced Studyin the Behavioral Sciences, StanfordUniversity, Stanford, California. 164Journal of Economic Perspectives

Table 1 MediaExplanations for the Decline in Crime in the 1990s,Ranked by Frequencyof Mention

Explanation Numberof mentions

Innovative policingstrategies 52 Increasedreliance on prisons 47 Changesin crack/ otherdrug markets 33 Agingof the population 32 Toughergun control laws 32 Strongeconomy 28 Increasednumber of police 26

Allother explanations 34

Notes: Basedon a Lexis-Nexissearch of articles written aboutthe national declinein crime in leading newspapers over the period 1991–2001. Newspapers included in the tabulationare the New York Times,Washington Post, USA Today,Houston Chronicle, San Francisco Chronicle,Chicago Sun Times, Boston Globe, Atlanta Journal Constitution, MinneapolisStar Tribune and San Diego Union-Tribune. Theseare the ten largestcirculation newspapers that areincluded in Lexis-Nexis. allof the observed decline in crime: increases in thenumber of police, the rising prison population, thewaning crackepidemic and thelegalization of abortion. 1 Thus, Iconclude that thedecline in crimedoes not reallypose apuzzle,but rather, isreadily explained by the available theories. The real puzzle that stands unan- swered,I argue,is whycrime rates did not startfalling earlier. In theŽ nal section, Ioffersome tentative observations about what crimetrends might be expected in thenext decade.

DeŽning Features of the Decline in Crime inthe 1990s

Severaldifferent aspects ofthedecline in crime are particularly noteworthy: its size,breadth and persistenceacross categoriesof crime; its universality across geographicand demographicgroups; and itsunexpectedness. Iconsiderthese characteristicsin turn.

TheMagnitude of the Decline Themost remarkable feature of the crime decline in the United States was its sheermagnitude. Figure 1 presentstime series data forhomicides from 1950 to

1 Reyes(2002) offersan additionalintriguing explanation for the declinein crime: the reductionin levelsof leadin the blooddue to the eliminationof leadedgasoline and lead-basedpaints. Becauseof the highlyspeculative nature of the Reyesconjecture at the presenttime, I donot discussthis hypothesis at greaterlength, although it isclearly an areaworthy ofcontinued future research. Steven D.Levitt165

Figure 1 HomicideRate, 1950 –2001

12 s t 10 n e d i s e r

8 0 0 0 , 0 0

1 6

r e p

s e 4 d i c i m o

H 2

0 1950 1960 1970 1980 1990 2000

2001.Homicide is themost accurately measured and mostserious crime and thus providesa usefulbenchmark. Homiciderates were relatively steady at about 4 –5 per100,000 population from1950 through themid-1960s, at which point they startedrising to a peakof 10.2 per 100,000 population in1980. From 1980 to 1991, thehomicide rate  uctuated between8 –10per100,000 population. Afterthat, the homicidetrend began alarge,steady decline. Between 1991 and 2000,homicide ratesper capita fellfrom 9.8 to 5.5 per 100,000, a drop of44 percent. Since that time,homicide rates have beensteady. Thesame pattern observed for homicide is present for every major crime categoryand inboth ofthe commonly used measuresof crime in theUnited States: the FBI’sUniformCrime Reports (UCR), coveringcrimes reported to the police, and theNational Crime Victimization Survey (NCVS), a large,nationally represen- tativephone surveyof Americans. Table 2 reportsthe percentage decline between 1991and 2001.For purposes ofcomparison, thepercentage change incrimerates overthe period 1973 –1991is also shown (1973is the Ž rstyear of availability for NCVS).For the period 1991 –2001,crimes reported to the police fell between 24 and 46percent across thevarious crime categories. The reduction in criminal victimizationsin NCVS iseven greater, ranging from 45 to 58 percent. 2

2 Thecongruence between the UCRand NCVSdata forthe 1990sis heartening, giventhat the aggregate trendsin these two data sourceshave failedto track closelyone another historically, as evidencedin the Ž rst columnof the table.Between 1973 and 1991,the UCRdata suggestsharply risingcrime in most categories,whereas the victimizationdata Ž nds declinesin crime for more than half ofthe categories. Boggessand Bound(1997) arguethat apartialexplanation for the differentpatterns isthat the twodata sets measuresomewhat different crimes; for example, NCVS crimes tend tobe lessserious, even within acrimecategory. 166Journal of Economic Perspectives

Table 2 NationalTrends in Speci Ž cCategoriesof Crime

Percentagechange in crime Percentagechange in crime Crimecategory anddata source category,1973 –1991 category,1991 –2001

Crimesreported to the policefrom UCR Violentcrime 182.9 233.6 Homicide 15.4 242.9 Rape 173.4 224.8 Robbery 150.0 245.8 Aggravatedassault 1118.1 226.7 Propertycrime 138.2 228.8 Burglary 13.0 240.9 Larceny 156.7 223.2 Motorvehicle theft 149.8 234.6 Criminalvictimizations from NCVS Violentcrime 11.6 250.1 Rape 220.0 245.0 Robbery 215.5 253.3 Aggravatedassault 23.9 256.9 Simpleassault 110.7 247.0 Propertycrime 232.0 252.8 Burglary 241.3 255.6 Theft 246.5 251.6 Motorvehicle theft 116.2 258.6

Notes: Allvalues in the tableare percentage changes in crime rates. Entriesin the toppanel of the table arebased on Uniform Crime Report data collectedby the FederalBureau of Investigation. These changesare de Ž nedin terms of victimizationrates percapita. Entriesin the bottompanel are from the NationalCrime Victimization Survey. For violent crime, the reportedvalues are percentage changes in crimeper person age 12 and older.For property crime, the percentagechanges are per household. The calculationsin the tablecorrect for the redesignof NCVS that occurredin 1993. Uniform Crime Report data fromrecent years is available online at http://www.fbi.gov .NCVSdata isavailable from http:// www.ojp. usdoj.gov/bjs .

Thedecline in crime has also beenremarkable in its steady persistence. Homiciderates fell in nineof the ten years in the decade of the 1990s, with the only exceptionbeing a minorupward blipin 1992. In theprevious three decades, homicidehad neverfallen for more than threeconsecutive years. Robbery, bur- glaryand larcenyeach fellevery year between 1991 and 2000.Prior to 1991, robbery rateshad fallenin only eight of the preceding 30 years. Thedrop inU.S. crimeappears tobe unusual among countriesof the world, although deŽ nitionaland reportingdifferences across countries,as wellas thepoor qualityof crime statistics in mostcountries other than theUnited States, make such internationalcomparisons dif Ž cult. Barclay,Tavares and Siddique(2001) provide oneof the most careful cross-country comparisons ofcrime trends. That analysis reportsthat homiciderates fell 4 percenton averagein European Union (EU) memberstates between1995 and 1999,a periodover which U.S. homiciderates fell 28percent. Violent crime rose 11 percent on averagein EU countriesover that sametime period, compared to 20 percent drops inU.S. violentcrime. Burglary UnderstandingWhy Crime Fell inthe 1990s 167

Table 3 PercentageChanges in Crime Rates for Different Population Groups, 1991–2001

Homicide ViolentCrime PropertyCrime

EntireUnited States 242.9 233.6 228.8 Region Northeast 250.0 242.9 241.5 South 244.6 227.3 225.6 Midwest 232.1 231.5 223.3 West 242.7 238.1 232.0 Urban/rural Within MSAs 245.0 236.7 231.7 Citiesoutside MSAs 231.4 218.4 218.1 Rural 235.0 22.9 211.0 City size .250,000 249.2 243.2 236.1 50,000–250,000 241.4 237.5 231.9 ,50,000 234.9 222.7 223.2

Source: FBI UniformCrime Reports. Notes: Entriesin the tableare the percentagechanges in per capita crimereports between the years1991 and 2001according to the FBI UniformCrime Reports.

and motorvehicle theft fell 14 percent and rose7 percent,respectively, in theEU, whilefalling 19 and 22percent in the United States.

TheUniversality of the Drop in Crime Thedrop ofcrime in the 1990s affected all geographic areas and demographic groups. Table3 presentsthe percentage decline in homicide, violent crime and propertycrime from 1991 –2001by region, urban/ ruraland citysize. In each of thesesubgroups and forall crime categories, the trend has beendownward. Crime declinesin the Northeast outpaced therest of thecountry, whereasthe Midwest was alaggard.The greatest percentage improvements in crime occurred within met- ropolitanstatistical areas (MSAs)and especiallyamong largecities with populations over250,000. Rural areas, particularlyon violentand propertycrime, saw much smallerdeclines in both absoluteterms and percentageterms. For instance, the homiciderate per 100,000 residents in large cities fell 12.9 per 100,000 (from 26.2 to13.3). The decline in homiciderates for cities with populations lessthan 50,000 was only1.5 (from 4.3 to 2.8). Table4 shows changes inhomicidefor the 25 most populous citiesas of1991. The Ž rstcolumn liststhe peak year for homicide by city. In almostthree-fourths of thecities, the peak occurred between 1990 and 1993.The next two columns presentthe homicide rate per 100,000 residents in thepeak year and in2001.The Ž nal column isthepercentage reduction in homicidefrom the peak year to 2001. Thecities are ordered by the percentage decline in homicide they experienced. 168Journal of Economic Perspectives

Table 4 HomicideTrends in Large U.S. Cities

Year that Percentagedecline homicide Peakhomicide rate 2001homicide rate inhomicide, City peaked (per100,000) (per100,000) peak to 2001

NewYork City 1990 30.7 8.1 273.6 San Diego 1991 14.7 4.0 272.8 Austin 1985 12.8 3.9 269.5 San Jose 1985 7.8 2.4 269.2 Seattle 1994 12.8 4.4 265.6 Jacksonville 1990 27.6 9.9 264.1 Houston 1991 36.5 13.4 263.3 San Antonio 1992 22.5 8.5 262.2 Dallas 1991 48.6 19.7 259.5 Denver 1992 19.3 7.9 259.1 Honolulu 1986 5.6 2.3 258.9 San Francisco 1993 17.5 7.7 256.0 Boston 1990 24.9 11.0 255.8 Washington, D.C. 1991 80.6 40.6 249.6 Los Angeles 1993 30.5 15.6 248.9 Columbus 1991 21.6 11.4 247.2 Philadelphia 1990 31.7 20.4 235.6 Detroit 1987 62.8 41.3 234.2 Chicago 1992 33.1 22.9 230.8 Phoenix 1994 21.5 15.3 228.8 Indianapolis 1991 19.4 14.0 227.8 Memphis 1993 32.0 24.1 224.7 Las Vegas 1991 15.7 11.9 224.2 Baltimore 1993 48.2 38.7 219.7 Milwaukee 1991 25.6 21.1 217.6

Notes: Allcities with populationgreater than 500,000in 1991 are included in the table.The peak year forhomicide is de Ž nedin homicides per 100,000 residents. For San Francisco,2001 homicide rates are not available,and the 2000homicide rates areused instead.

NewYork City, which has garneredenormous attention for its success in Ž ghting crime,leads the list with a 73.6percent reduction in homicide.A numberof other cities(San Diego,Austin, San Jose, Seattle)that have receivedfar fewer accolades, however,nearly match theNew York City experience. Even the cities near the bottomof the list have experiencedhomicide reductions of roughly 20 percent. Theuniversality of these gains arguesagainst idiosyncraticlocal factors as the primarysource of the reduction. In demographicterms, breakdowns of crime offending rates by race, gender and ageare not directlyavailable, because in many cases offendercharacteristics areunknown. In those cases wherean arrestis made, however,information is gatheredby the FBI inthe Uniform Crime Reports. Across everycrime category and everybreakdown by race, gender and age,substantial declinesin arrest rates have occurred. Steven D.Levitt169

Figure 2 ExpertForecast Made by JamesAlan Fox in 1995 vs. Actual Teen Homicide Offenders

10000

7 Actual crime pre-projection 1

– 9000 Optimistic projection 4 1

Pessimistic projection

e 8000

g Actual crime post-prediction a

s

r 7000 e d n

e 6000 f f O

e 5000 d i c i 4000 m o H

3000 f o

r

e 2000 b m

u 1000 N 0 1976 1980 1984 1988 1992 1996 2000 2004

TheUnexpectedness of the Drop in Crime Having just livedthrough an enormousreduction in crime, it is hard to reconstructjust how unexpectedsuch adeclinereally was. Evenafter the fall had begun, someof the world ’smostprominent criminologists dismissed the decline as atransitoryblip that wouldquickly be reversed. Forexample, in 1995 the U.S. AttorneyGeneral commissioned a reporton crimetrends from Northeastern professor James Alan Fox,one of the most widely quotedcriminologists in the popular press. Figure2 reproducesand extendsone of the Ž guresfrom that report.The line in Figure 2 up until1994 re  ects the observednumber of homicides by 14 –17-year-olds.The two dotted lines represent Fox’s(1996)optimistic and pessimisticfuture projections. In theoptimistic case, youth homicideswere projected to rise by about 15percent. In thepessimistic scenario, youth homicidewas goingto more than doubleover the next decade, promptingFox to say ina 1996 ScientiŽc American article, “thenext crime wave will getso bad that itwillmake 1995 look like the good olddays. ” Thedashed lineshows what actuallyhappened: juvenilehomicide rates fell by morethan 50percentin the ensuing sixyears. 3 Foxwas not alonein predicting that the1990s would be a diredecade with respectto crime. James Q. Wilson(1995, p. 507)wrote, “Just beyond thehorizon, therelurks a cloud that thewinds willsoon bringover us. Thepopulation willstart

3 Twoyears after his initialreport, in spite of sharply decliningjuvenile crime rates inthe intervening years, Fox(1997) continuedto project large increases in juvenilehomicide over the next decade.No further reportswere commissioned. 170Journal of Economic Perspectives

gettingyounger again ....Getready. ” In asymposiumon crimepublished in Journalof Economic Perspectives in1996, John DiIulio(1996, p. 8)wrote, “It is not inconceivablethat thedemographic surge of the next 10 years will bring with it young malecriminals who makethe . ..Bloodsand Cripslook tame by compari- son.” EvenPresident Clinton got into the act, stating: “We’vegotabout sixyears to turn this juvenilecrime thing around, orour country isgoing to be living with chaos” (Allpolitics,1997). In short, thecrime decline was so unanticipated that it was widelydismissed as temporaryor illusory long after it had begun.

Six Factors thatPlayed Little or No Role in the Crime Decline

Thelist of explanations offeredas towhy crime has fallenis a lengthyone. Here,I beginwith six commonly suggested and plausibletheories that inpractice do not appear importantin explaining the decline of crime rates.

1)TheStrong Economy of the 1990s Thedecade of the 1990s saw sustained economicgrowth. Real GDP percapita grewby almost 30 percent between 1991 and 2001.The annual unemploymentrate fellfrom 6.8 in 1991 to 4.8 percent in 2001. If macroeconomicperformance is an importantdeterminant of crime rates, then theeconomy could explainfalling crime. In economicmodels of crime such as Becker(1968), improvements in legiti- matelabor market opportunities make crime relatively less attractive. This predic- tionis likely to be more relevant for crimes involving direct Ž nancial motivation such as burglary,robbery and auto theft,but lessimportant for homicide, assaults and rape. On theother hand, tothe extent that activitiesthat areassociated with increasedlevels of either offending or victimization are normal goods —likealcohol consumption, frequentingnightclubs and owninga car —thelink between eco- nomicactivity and crimeis theoreticallyambiguous. Empiricalestimates of the impact of macroeconomic variables on crimehave beengenerally consistent across studies: Freeman(1995) surveys earlier research, and morerecent studies include Machin and Meghir(2000), Gould, Weinbergand Mustard (1997),Donohue and Levitt(2001) and Raphael and Winter-Ebmer (2001).Controlling for other factors, almostall of thesestudies report a statistically signiŽ cant but substantivelysmall relationship between unemployment rates and propertycrime. A typicalestimate would be that aonepercentage point increasein theunemployment rate is associated witha onepercent increase in property crime. Violentcrime, however, does not varysystematically with the unemployment rate. Studiesthat have used othermeasures of macroeconomic performance like wages oflow-income workers come to similar conclusions (Machin and Meghir,2000; UnderstandingWhy Crime Fell inthe 1990s 171

Gould, Weinbergand Mustard, 1997). 4 Based on theseestimates, the observed 2percentagepoint declinein theU.S unemploymentrate between 1991 and 2001 can explainan estimated2 percentdecline in propertycrime (out ofan observed drop ofalmost30 percent),but no change inviolentcrime or homicide. The sharp increasesin crime in the 1960s —adecadeof strong economic growth —further corroboratethe weak link between macroeconomics and crime. If theeconomy has amajor impacton crime,the likely channel isnot through thedirect effect estimated in thestudies noted above, but rather,indirectly through stateand localgovernment budgets. Twoof the factors that Iidentifyas most importantin reducingcrime are increased spending on policeand prisons. To the extentthat thesebudget items are affected by macroeconomic performance, one wouldexpect to observe a strongerlink between the economy and crimethan is found inthe studies above, which controlfor criminal justice variables when estimatingthe link between economic variables and crime.

2)ChangingDemographics Theaging of the baby boomersrepresents a profound demographicshift. The elderlyhave extremelylow rates of both offendingand criminalvictimization. In 2001,people over the age of 65 had percapita arrestrates approximately one- Ž ftieththe level of 15 –19year-olds. Perkins (1997), using NCVSdata, reportsthat those overthe age of 65 experiencevictimization rates for serious violent crime that areless than one-tenth ofthose ofteenagers. Given that theshare ofthe elderly population increasedduring the 1990s, a purelydemographically driven decline in crimemight be expected. Twoother concurrent demographic changes, however,counterbalance the crime-relatedbene Ž ts ofan agingpopulation. First,between 1990 and 2000,the blackpopulation rosefrom 12.1 percent to 12.9 percent. For reasons that areonly partlyunderstood (Sampson and Lauritsen,1997), blacks have elevatedvictimiza- tionand offendingrates relative to other Americans, particularly for homicide, wherethe differences across racesare almost an orderof magnitude. 5 Second, in spiteof the overall aging of thepopulation, theecho ofthebaby boomis leading toa temporaryincrease in the number of teenagers and young adults. Between 1995and 2010,the number of 15 –24year-olds is projectedto increase by roughly 20percent, and theshare ofthe population betweenthe ages of 15 and 24will increasefrom 13.7 percent to 14.6 percent. (In comparison, 15 –24year-olds represented18.7 percent of the population in1980.)This agegroup has agreatly elevatedinvolvement in crime.Indeed, many ofthe dire predictions for increased

4 Thestudies cited in this paragraphuse U.S. data, exceptfor one study that focuseson the United Kingdom.Cross-county estimates also fail toshow a consistent relationshipbetween GDP percapita and crime,although high levelsof incomeinequality are correlated with high crime,as inSoares (1999). 5 Thegrowth in the Hispanicpopulation was evengreater (from 9.0 percentto 12.5 percent of the populationbetween 1990 and 2000). Forthe mostpart, however,Hispanic offending and victimization rates areonly slightly abovethose of whites. 172Journal of Economic Perspectives

crimerates in the 1990s were based inpart on theincreasing number of adolescents. Overall,these various demographic shifts probablyhad aslightameliorating effecton crime.The estimates of Levitt (1999), using an Oaxaca-decomposition approach, suggestthat overallchanges inthe age distribution may have reduced homicideand violentcrime by a fewpercent and propertycrime by as much as 5–6percent.Changes inracial composition largely offset the age-distribution beneŽ ts forhomicide and reducethe estimate somewhat for violent crime, but the propertycrime bene Ž tlargelyremains. Thus, demographicshifts mayaccount for alittlemore than one-sixthof the observed decline in propertycrime in the 1990s, but arenot an importantfactor in the drop inviolent crime.

3)BetterPolicing Strategies An enormousamount ofmedia attention has focused on thepolicing strate- giesinstituted in New York City under the leadership of Police Commissioner WilliamBratton and MayorRudy Guiliani.Their crime- Ž ghtingapproach involved increasedenforcement of nuisance activitieslike aggressive panhandling and better useof technology in identifying crime “hot spots.” Otherchanges inpolicing strategysuch as “communitypolicing, ” inwhich thepolice attempt to work more closelyas allieswith communities rather than simplyresponding to emergency calls, werewidely adopted inmany othercities in the 1990s. In Boston, an inno- vativemultiagency collaboration targeted gang violence(Kennedy, Piehland Braga, 1996). Therehave beenvery few rigorous academic studies of the impact of policing strategies.A numberof early quasi-randomized studies, discussed inWilson (1985), providedlittle evidence that communitypolicing strategies lowered crime. Indeed, itis often dif Ž cultto identify which policedepartments have adopted particular practices,when theadoptions occurredand why. In the1990s, federal funds were availableto police departments that implementedcommunity policing initiatives, leadingmany departmentsto advertise themselves as doingcommunity policing, whetheror not theyactually changed policingpractices. Due to such dif Ž culties, no compellingcross-city comparisons ofpolicing practices have beenperformed. SinceNew York City is heldup as theclear innovator inpolicingpractices, and sinceit enjoyedthe greatest crime declines of any largecity, an analysis ofthat city ’s experiencerepresents a logicalstarting point. 6 In myopinion, thereare reasons for skepticismregarding the claim that NewYork City ’spolicingstrategy is thekey to itsdecline in crime. First, the drop incrime in New York began in 1990. Crime

6 In arecentworking paper, Corman and Mocan(2002) analyze the NewYork City case using monthly time-seriesdata. They Ž nd that aggressivepolicing tactics (as proxiedby the numberof misdemeanor arrests) arenegatively associated with robberyand motorvehicle thefts, butdo not have astatistically signiŽ cant impacton other crimes. One dif Ž culty with the Cormanand Mocanapproach, however, is that they arenot ableto include reliable controls for the extentof the crackepidemic, which was steadily waning at the sametime that aggressivepolicing tactics wereput intoplace. Steven D.Levitt173

declinesof roughly 10 percentacross awiderange of offenses occurred in 1991and 1992.Guiliani, however, did not takeof Ž ceuntil 1993, at which point Brattonwas movedfrom the New York City Transit Police, where he had beenusing thesame approaches, and appointed policecommissioner. With the exception of homicide, which does declinesharply in1993,the trend in crime shows no obviousbreak after Brattonis appointed. Second, thechange inpolicingstrategies was accompanied by enormousgrowth in thesize of the police force. Between 1991 and 2001,the New YorkCity police force grew 45 percent —an increasethree times greater than the national average.I arguelater in this paperthat increasesin the number of police areeffective in reducingcrime. By myestimates, the unusually largeexpansion of thepolice force in New York City would be expected to reduce crime there by 18percent more than thenational average,even without any change inpolicing strategy.If oneadds 18percent to New York City ’scrimehomicide experience in Table2 (changing thedecline from 73.6 percent to 55.6 percent), New York City isabout averageamong largecities. Third, giventhat fewother cities in Table 2 institutedNew York City –typepolicing approaches, and certainlynone withthe enthusiasm ofNew York City itself, it is dif Ž cultto attribute the widespread declines incrime to policing strategy. Even Los Angeles and Washington, D.C., twocities notorious forthe problems they have experiencedwith their police forces (Los AngelesPolice Commission, 1996;Thompson, 1997),achieved declines in crime on par withNew York City once thegrowth in the size of New York City ’s police forceis accounted for.Fourth, NewYork City has had abortionrates among the highestof anywhere in thenation sinceabortion was legalizedthere in 1970, three yearsbefore the Supreme Court decision in Roe v. Wade (410U.S. 113[1973]) madeabortion legal nationwide. If onebelieves my arguments later in this paper that alinkexists between rates of abortionand laterrates of crime,this connection providesfurther evidence against theargument that thereis an unexplainedcrime declinein New York City that can beattributed to policing strategy. Thus, whilethe impact of policing strategies on crimeis an issueon which reasonablepeople might disagree given the lack of hard evidence,my reading of thelimited data that areavailable leads me to the conclusion that theimpact of policingstrategies on NewYork City crime are exaggerated, and that theimpact on national crimeis likely to be minor.

4)Gun ControlLaws Thereare more than 200million Ž rearmsin private hands inthe United States—morethan thenumber of adults (Cookand Ludwig,1996). Almost two- thirdsof homicides in the United States involve a Ž rearm,a fractionfar greater than otherindustrialized countries. Combining those twofacts, onemight conjec- turethat easyaccess toguns inthe U.S. maybe part ofthe explanation for our unusually high homiciderates. Indeed, themost careful study on thesubject Ž nds that higherrates of handgun ownership, which representabout one-thirdof all Ž rearms,may be a causal factorin violent crime rates (Duggan, 2001). Thereis, however,little or no evidencethat changes ingun controllaws in the 174Journal of Economic Perspectives

1990scan account forfalling crime. For example, the Brady Handgun Violence PreventionAct of 1993 instituted stricter requirements for background checks beforea gun issold. However,Ludwig and Cook(2000) report no differencein homicidetrends after the passage ofthe Brady Act in states affectedby thelaw and states that alreadyhad policiesin placethat wereat leastas stringentas those inthe BradyAct. Giventhe realities of an activeblack market in guns (Cook, Molliconi and Cole,1995), the apparent ineffectivenessof gun controllaws should not come as agreatsurprise to economists. Evenin thelate 1980s, prior to the Brady Act, only about one-Ž fth ofprisoners reported obtaining their guns through licensedgun dealers(Wright and Rossi, 1994). Gun buy-back programsare another formof public policy instituted in the 1990sthat islargelyineffective in reducing crime. First, the guns that aretypically surrenderedin gun buy-backs arethose guns that areleast likely to be used in criminalactivities. The guns turnedin will be, by de Ž nition, those forwhich the ownersderive little value from the possession ofthe guns. In contrast, those who are using guns incrimes are unlikely to participatein such programs.Second, because replacementguns arerelatively easily obtained, thedecline in thenumber of guns on thestreet may be smaller than thenumber of guns that areturned in. Third, the likelihoodthat any particulargun willbe used inacrimein agivenyear is low. In 1999,approximately 6,500 homicides were committed with handguns. Thereare approximately65 million handguns inthe United States. Thus, ifa different handgun wereused ineach homicide,the likelihood that aparticularhandgun wouldbe used tokill an individualin a particularyear is one in 10,000.The typical gun buy-back programyields fewer than 1,000guns. Thus, itisnot surprisingthat researchevaluations have consistentlyfailed to document any linkbetween gun buy-back programsand reductionsin gun violence(Callahan, Riveraand Koepsell, 1994;Kennedy, Piehland Braga, 1996;Rosenfeld, 1996; Reuter and Mouzos, 2003). Morestringent gun-control policiessuch as bans on handgun acquisition passed inWashington, D.C., in1976 and theban on handgun ownershipin Chicago in1982do not seemto have reducedcrime, either. While initial research suggesteda bene Ž cialimpact of the D.C. gun ban (Loftin,McDowall, Weirsema and Cottey,1991), when thecity of Baltimore is used as acontrolgroup, ratherthan the af uent Washington suburbs, theapparent bene Ž ts ofthe gun ban disappear (Britt,Kleck and Bordua, 1996).Although no carefulanalysis ofChicago ’s gun ban has beencarried out, thefact that Chicago has beena laggardin the nationwide homicidedecline argues against any largeimpact of the law. From a theoretical perspective,policies that raisethe costs ofusing guns inthecommission of actual crimes,as opposed totargeting ownership, wouldappear tobe a moreeffective approach toreducing gun crime(for instance, Kesslerand Levitt,1999). The most prominentof these programs, Project Exile, which providesprison sentenceen- hancements forgun offenders,however, has beenconvincingly demonstrated to be ineffectiveby Raphael and Ludwig(2003), apparently in part becauseof thesmall scaleon which itwas carriedout. UnderstandingWhy Crime Fell inthe 1990s 175

5)LawsAllowing the Carrying of Concealed Weapons Thehighly publicized work of Lott and Mustard (1997)claimed enormous reductionsin violentcrime due to concealed weapons laws. Thetheory behind this claimis straightforward: armed victims raise the costs faced bya potentialoffender. Theempirical work in support ofthis hypothesis, however,has provento be fragilealong a numberof dimensions (Black and Nagin, 1998;Ludwig, 1998; Duggan, 2001;Ayres and Donohue, 2003).First, allowing concealed weapons should have thegreatest impact on crimesthat involveface-to-face contact and occur outsidethe home where the law might affect gun carrying.Robbery is the crimecategory that mostclearly Ž ts this description,yet Ayres and Donohue (2003) demonstratethat empiricallythe passage ofthese laws is, ifanything, positively relatedto the robbery rate. More generally, Duggan (2001) Ž nds that forcrimes that appear todecline with the law change, thedeclines in crimeactually predate thepassage ofthe laws, arguingagainst acausal impactof the law. Finally,when the originalLott and Mustard (1997)data setis extendedforward in time to encompass alargenumber of additional lawenactments, theresults disappear (Ayresand Donohue, 2003).Ultimately, there appears tobe little basis forbelieving that concealedweapons lawshave had an appreciableimpact on crime.

6)IncreasedUse of Capital Punishment In the1980s, a totalof 117 prisoners were put todeath inthe United States. That numbermore than quadrupled to478 in the 1990s. The debate over the effectivenessof the death penaltyas adeterrenthas beenongoing for three decades. Ehrlich(1975, 1977) presented early evidence arguing in favor of a deterrenteffect. A numberof critics demonstrated the sensitivity of the Ehrlich Ž ndings toseemingly minor changes inspeci Ž cation (Forst,Filatov and Klein, 1978;Passell and Taylor,1977; Leamer, 1983; Cameron, 1994).A seriesof more recentstudies that incorporatedata fromthe 1990s, however, have tendedto once again Ž nd adeterrenteffect (Dezhbakhah, Rubin and Shepherd,2002; Mocan and Gittings,2003). Largelylost in this debate,however, are two important facts (Katz, Levittand Shustorovich, 2003).First, given the rarity with which executionsare carried out in this country and thelong delays in doing so, arationalcriminal should not be deterredby the threat of execution. Despite increases in capital punishment in recentyears, the likelihood of beingexecuted conditional on committingmurder isstillless than 1in200.Even among those on death row,the annual executionrate isonly 2 percent,or twice the death ratefrom accidents and violenceamong all Americanmen. Amongthe subsample ofindividuals engaged in illegal activities, thedeath ratesare likely to bemuch higher.Levitt and Venkatesh(2000) report a death rateof 7 percentannually forstreet-level drug sellers in the gang they analyze. Kennedy, Piehland Braga(1996) estimate violent death ratesto be 1–2percentannually among allgang membersin Boston. It ishard tobelieve the fearof execution would be a drivingforce in a rationalcriminal ’scalculus in modernAmerica. Second, eventaking as givenvery large empirical estimates of the 176Journal of Economic Perspectives

deterrentimpact of the death penalty —such as Ehrlich ’s(1975)classic estimateof sevenmurders deterred per execution or Mocan and Gittings(2003) estimate of six murdersdeterred per execution —theobserved increase in thedeath penaltyfrom 14executions in 1991 to 66 in 2001 would eliminate between 300 and 400 homicides,for a reductionof 1.5 percent in the homicide rate, or less than one-twenty- Ž fth oftheobserved decline in thehomicide rate over this timeperiod. Moreover,any deterrenteffect from such executionscannot explainthe decline in othercrimes. Given the way the death penaltyis currentlypracticed in the United States,it is extremely unlikely that itexerts signi Ž cant in uenceon crimerates.

Four Factors That Explain theDecline in Crime

Having arguedthat many common explanations forthe decline in crime are unlikelyto hold thetrue answers, Inow turn tofourfactors that did, inmyreading ofthe evidence, play a criticalrole in the crime reduction of the 1990s: the increasingnumber of police, the skyrocketing number of prisoners, the ebbing of thecrack epidemic and legalizationof abortion in the 1970s.

1)Increasesin the Number of Police Policeare the Ž rstline of defense against crime.More than $60billion is spent each yearon policing.Studies on theconnection betweenthe number of police and crimein the 1970s and 1980s,as surveyedby Cameron (1988), tended to Ž nd an insigniŽ cant ornegative correlation, because these studies typically failed to account forthe endogeneity problem. The political response to rising crime is to hiremore police, so thenumber of police affect the amount ofcrime, but the amount ofcrimealso affectsthe number of police.A numberof recent studies have addressed this endogeneityproblem with a widevariety of identi Ž cation strategies and reachedthe conclusion that morepolice are associated withreductions in crime. Marvelland Moody (1996),using a “Granger-causality ” approach on panel data forU.S. states and largeU.S. citiesdemonstrate that increasesin thenumber ofpolice are associated withreductions in crime in the future. They estimate elasticitiesof crimewith respect to thenumber of policeof approximately 20.30. Corman and Mocan (2000)use high-frequency time-series data toreduce the endogeneityproblem. They argue that thegovernment response to rising crime occurs onlywith a lagof a numberof months, so that endogeneitywill be less importantwhen analyzingmonthly data than annual data. Usingtime-series data forNew York City, Corman and Mocan Ž nd elasticitiesranging from 20.29 to 21.385across crimecategories, with a medianvalue of 20.452.Levitt (1997) uses thetiming of mayoral and gubernatorialelections as an instrumentfor police hiring.Politicians disproportionately increase the size of the police force in ad- vance ofelections, but electionsare unlikely to impact crime directly, making electionsa plausibleinstrumental variable. Using this identi Ž cation strategy, Steven D.Levitt177

elasticityestimates range between 20.05 and 21.98across crimecategories, with a medianvalue of 20.79(although seeMcCrary, 2002, which points out that cor- rectlycomputed standard errorsmake those point estimatesstatistically insigni Ž - cant). Levitt(2002) Ž nds much moreprecisely estimated elasticities of 20.43 to 20.50when using changes inthe number of Ž reŽ ghtersas an instrumentfor changes inthe number of police. Fire Ž ghtersare a logicalinstrument for police because spending on thesetwo types of protective services tend tobe similarly affectedby localbudgets and preferences,but Ž reŽ ghtersare unlikely to exert any direct in uenceon crimerates. Thenumber of police of Ž cersper capita, which istracked by the FBI and reportedannually inthe Uniform Crime Reports, increased by 50,000 – 60,000 ofŽ cers,or roughly 14 percent, in the 1990s. Although this increasewas greater than inpreviousdecades, itwas smallerthan mighthave beenexpected given the 1994omnibus crimebill, which, byitself, had promisedan extra100,000 new police ofŽ cerson thestreets. Using an elasticityof crime with respect to the number of police of 20.40,the increase in police between 1991 and 2001can account fora crimereduction of 5 – 6percentacross theboard. Theincrease in policecan thus explainsomewhere between one- Ž fth and one-tenth ofthe overall decline in crime. Whetherthis investmentin police has beena cost-effectiveapproach toreduc- ingcrime is a differentquestion. As noted above, annual expenditureson police areapproximately $60 billion, so thecost ofthe 14 percent increase in police (assuming marginalcost isequal to average cost, which islikelyto be a reasonable approximation)is $8.4 billion a year.The bene Ž ts ofcrime reduction are more difŽ cultto quantify. Themost commonly used estimatesof the cost ofcrime to victims(for example, Miller, Cohen and Rossman, 1993)places the costs ofcrime at roughly$500 billion annually inthe early 1990s. Given the sharp declinesin crime,today ’sestimateswould likely be substantially lower —perhaps $400billion in currentdollars. If theincrease in police reduced crime by 5 – 6percent,then the correspondingbene Ž tofcrime reduction is $20 –25billion, well above the esti- matedcost. Thus, at leastto a crude Ž rstapproximation, the investment in police appears tohave beenattractive from a cost-bene Ž tperspective.

2)TheRising Prison Population The1990s was aperiodof enormous growth in the number of people behind bars, as demonstratedin Figure 3. Aftermany decades ofrelatively stable impris- onment rates,the prison population beganto expand inthe mid-1970s. By 2000, morethan twomillion individuals were incarcerated at any point intime,roughly fourtimes the number locked up in1972.Of that prison population growth,more than half tookplace in the 1990s. The increase in prisonerscan beattributed to a numberof factors, themost important of which werethe sharp risein incarceration fordrug-related offenses, increasedparole revocation and longersentences for those convictedof crimes (Kuziemko and Levitt,2003). Thetheory linking increased imprisonment to reduced crime works through twochannels. First,by lockingup offenders,they are removed from the streets and 178Journal of Economic Perspectives

Figure 3 AdultIncarceration Rate, 1950 –2001

500

450 0

0 400 0 , 0

0 350 1

r e

p 300

d e

n 250 o s i r

p 200 m i

s 150 t l u

d 100 A 50 0 1950 1960 1970 1980 1990 2000 unable tocommit further crimes while incarcerated. This reductionin crime is known as theincapacitation effect.The other reason prisons reducecrime is deterrence—theincreased threat of punishment induces forward-lookingcriminals not tocommit crimes they otherwise would Ž nd attractive.Empirical estimates of theimpact of incarceration on crimecapture both ofthese effects. Theevidence linking increased punishment tolower crime rates is verystrong. Typicalestimates of elasticities of crime with respect to expected punishment range from 2.10 to 2.40, withestimates of theimpact on violentcrime generally larger than those forproperty crime (Marvell and Moody, 1994;Spelman, 1994; Levitt, 1996;Donohue and Siegelman,1998). But most of these estimates are based on simplecorrelations. Given the clear endogeneity between crime and imprisonment (when crimeis rising,the prison population willalso riseif expected punishment percrime is held constant), onemight suspect that such correlationsestimates wouldunderstate the true impact of imprisonmenton crime.Indeed, Levitt(1996) obtains estimatesat thehigh end ofthe range when using prison overcrowding litigationas an instrumentfor the size of the prison population. Courtdecisions in prison overcrowdinglawsuits are a plausibleinstrument for the prison population becausethese decisions have alargeimpact on thegrowth rates in state prison inmates, but thereis littlereason tobelieve that such litigationaffects crime rates, exceptthrough theimpact on thenumber of people incarcerated. Surveys of prison inmatesyield estimates of reductions in crimesdue to incarceration that are consistent withthe econometric studies (DiIulio and Piehl,1991). Usingan estimateof the elasticity of crimewith respect to punishment of 2.30 forhomicide and violentcrime and 2.20for property crime, the increase in incarcerationover the 1990s can account fora reductionin crime of approximately UnderstandingWhy Crime Fell inthe 1990s 179

12percent for the Ž rsttwo categories and 8percentfor property crime, or about one-thirdof the observed decline in crime. 7 Annual expenditureson incarcerationtotal roughly $50 billion annually. Combiningthis spending Ž gurewith the cost ofcrime to victims and elasticities noted above, expenditureson prisons appear tohave bene Ž ts that outweighthe directcosts ofhousing prisoners,subject tothree important caveats. First,a dollar spent on prisons yieldsan estimatedcrime reduction that is20percent less than a dollarspent onpolice,suggesting that onthemargin, substitution towardincreased policemight be the ef Ž cientpolicy. Second, itseemsquite plausible that substantial indirectcosts areassociated withthe current scale of imprisonment, such as the adversesocietal implications of imprisoning such alargefraction of young African Americanmales. Finally, given the wide divergence in thefrequency and severityof offendingacross criminals,sharply decliningmarginal bene Ž tsofincarceration are apossibility.In otherwords, thetwo-millionth criminal imprisoned is likely to imposea much smallercrime burden on societythan the Ž rstprisoner. Although theelasticity of crime with respect to imprisonment builds in some declining marginalreturns, the actual drop offmay be much greater.We do not have good evidenceon this point. Thesecaveats suggestthat furtherincreases in imprison- mentmay be less attractive than thena ¨ve cost beneŽ tanalysis wouldsuggest.

3)TheReceding Crack Epidemic Beginningin 1985,the market for crack cocaine grewrapidly. Crack cocaine isproduced byheating a mixof powder cocaine and bakingsoda. Theresulting precipitatetakes the form of airy nuggets. Extremelysmall quantities of this compound, when smoked,produce an intense,short-lived high. Theemergence of crackcocaine representedan importantdevelopment both becauseit facilitated thesale of cocaine bythe dose fora retailprice of $5 –$10and becausethe extreme high associated withcrack proved to be popular among consumers. Crackfre- quentlysold inopen-air markets with youth gangs controllingthe retail distribu- tion. Thecrack cocaine tradeproved highly lucrative for gangs, leadingto violence as rivalgangs competedto sell the drug (Levitt and Venkatesh, 2000). Anumberof authors have highlightedthe extreme patterns inviolence that stronglysuggest an importantrole for crack (most notably, Blumsteinand Rosen- feld,1998; Cook and Laub, 1998).Figure 4, forinstance, presentshomicide rates byage and racefor the period 1976 –2000.Beginning in 1985, homicide rates for blackmales under the age of 25 began asteepascent, morethan triplingin less than adecade, beforeonce again fallingdramatically to levels slightly above those

7 In the light ofthe estimateslinking increased incarceration to lower crime, it isperhaps surprising that the risingprison population of the 1980sdid not inducea commensuratedecline in crime in that period.Among adults, crimerates werein fact steadilyfalling throughout the 1980s.These declines, however,were masked by sharply risingyouth crimein the 1980s.These increases in juvenilecrime appearto be duein part tothe crackepidemic (discussed below), as wellas tofalling punishments in the juvenilejustice system over this sametime period (Levitt, 1998). 180Journal of Economic Perspectives

Figure 4 HomicideOffending Rates by Race and Age

350

White male, 14 –24 300 White male, 25 1 Black male, 14 –24 Black male, 25 1

0 250 0 0 , 0 0 1 200 r e p

s e d

i 150 c i m o

H 100

50

0 1976 1980 1984 1988 1992 1996 2000 ofthe pre-crack era. In starkcontrast, thehomicide rates of older black males continued on along-termsecular decline. Young whitemales also experienceda short-run increasein homicide in the late 1980s, but both thebase ratesand the increasesfor whites are much lower.The concentration and timingof the homicide spikeamong theyoung blackmales, which coincideswith the rise and fallof the crackmarket, is suggestive of crack cocaine playinga criticalrole. Despitethe seeming importance of crackin explainingthe crime  uctuations since1985, remarkably little quantitative research on thesubject goesbeyond pictureslike Figure 4. Goldstein,Brownstein, Ryan and Bellucci(1997) carried out in-depth analysis ofhomicides in New York City in 1988. They classify over 25percent of these homicides as crack-related.Almost all of these homicides involvedcrack distribution, rather than homicidesinduced bythe psychopharmo- logicaleffects of crack or the need of drugusers to obtain money.The lack of clear indicatorsof the importance of crack markets across timeand space has madeit difŽ cultto study thepossible links to crime. I have not seena study that linksthe ebbingof the crack epidemic to falling crime, and Iam awareof onlytwo studies that attemptto relate the rise of crack cocaine toincreased crime. Cork (1999), using city-leveldata, Ž nds that sharp increasesin juvenilecrack arrest rates happen coincidentwith or shortly in advance ofsimilarly sharp increasesin juvenile gun homicides.Grogger and Willis(2000) use cocaine-related emergency room visits and asurveyof police chiefs in an attemptto identify when crack Ž rstenters a city. Theythen usedifferences-in-difference sbetweenthe central city and thesurround- ingsuburbs toidentify the impact of crack. Theyconclude that crimerates in a Steven D.Levitt181

broad rangeof categories are elevated about 10percent in thesecentral cities due tothe arrival of crack, although theirstudy isnot particularlyuseful for calculating thecontribution of the waning crackepidemic on crimein the 1990s. 8 Although theresearch is limited, I nonethelessbelieve that crackhas quite likelyplayed an importantrole in the decline in homicidein the 1990s, at leastfor homicide.Figure 4 suggestsan alternativestrategy for estimating that number, which isprobablyinferior to Grogger and Willis(2000), but morerelevant to the questionat hand. Underthe assumption that thespike in homicide by young black malesis driven primarily by crack, then onecan useolder black males, young white malesor even all people except young blackmales as potentialcontrol groups. Sincecrack was not whollycon Ž ned toyoung blackmales, one would expect that any ofthese estimates would likely represent a lowerbound. Alternatively,one can calculatethe differences-in-differences between young and oldblacks and whites. As crackebbed from 1991 to 2001, young blackmales experienced a homicide declineof 48 percent, compared with 30 percentfor older black males, 42 percent foryoung whitemales and 30percent for older white males. Depending on which controlgroup one views as mostreasonable, the estimated impact of crack on homicidescommitted by young blackmales ranges from 6 to18 percent.Given that young blackmales commit about one-thirdof homicides, this translatesinto a reductionof 2 – 6percentin overallhomicides in the1990s due to crackreceding. Based on strongprior beliefs and thelikelihood that theseestimates represent lowerbounds, Iviewthe 6 percentnumber as plausible,which wouldmean that the declineof crackexplains about 15percent of the fall of homicide. With respect to crimesother than homicide,however, my priors suggest smaller impacts: perhaps 3percent(10 percent of the observed decline) for violent crime and no impacton propertycrime.

4)TheLegalization of Abortion TheU.S. SupremeCourt ’s Roe v. Wade decisionin 1973 may seem like an unlikelysource of the decline in crimein the1990s, but agrowingbody ofevidence suggestsan importantrole for legalized abortion in explaining falling crime rates twodecades later.The underlying theory rests on twopremises: 1) unwanted

8 Although the Groggerand Willis(2000) identi Ž cationstrategy iscleverand the bestwork to date on the subject, Iremainskeptical of the estimatesfor a numberof reasons. First, centralcities and suburbs have verydifferent baseline levels of crime,which makes the differences-in-differencesapproach more difŽ cultto justify (although the authors dotheir best to address this issuein the paper). Second,the assumptionthat crackonly in  uencedcentral cities and not suburbsis unlikely to be accurate. In percentageterms, homicide rates ofsuburbanblacks increased as muchas centralcity blacksduring the timeperiod when crack was emerging(the sameis true for whites in suburbs and centralcities). Thus, the identiŽ cationin this methodis coming primarily from differences in the racialcomposition of centralcities and suburbs,rather than throughdifferences between blacks in citiesand suburbs,which wouldseem to be a moreplausible source of variation under the assumptionthat crackis onlyin central cities. 182Journal of Economic Perspectives

childrenare at greaterrisk for crime, and 2)legalizedabortion leads to areduction inthenumber of unwanted births. Withrespect to the Ž rstpremise, the negative impact of adverse home envi- ronmentsgenerally on eventualcriminality, the evidence is quite clear (Loeber and Stouthamer-Loeber,1986; Sampson and Laub, 1993).A numberof studies have lookedat cases ofwomenliving in jurisdictions in which governmentalapproval to have an abortionwas required,who sought tohave an abortion, but weredenied theability to do so (Dagg,1991; David, Dytrych, Matejcek and Schuller,1988). For example,Dagg (1991) reports that thesewomen overwhelmingly kept their babies, ratherthan givingthem up foradoption, but that theyoften resented the unwanted children.These children who wereborn because theirmothers were denied an abortionwere substantially more likely to be involved in crime, even when control- lingfor the income, age, education and health ofthe mother. Theevidence for the second premisethat legalizedabortion leads to a reduc- tionin the number of unwanted birthsalso appears compelling.Levine, Staiger, Kane and Zimmerman(1999) Ž nd that legalizationof abortion is associated with roughlya 5percentdrop inbirthrates. They estimated that thedrop inbirthswas roughlytwice as greatfor teenage and nonwhitemothers as itwas forthe nonteen, whitepopulation. Consistent withthis Ž nding, thenumber of children put upfor adoption felldramatically after abortion became legal (Stolley, 1993). Also consis- tentis the decline in infanticide in the United States that isdocumented by Sorenson, Wiebeand Berk(2002) following the legalization of abortion. Indeed, otherstudies have documentedimprovements in a widerange of outcomes for those born at atimeof legalized abortion, includingreduced infant mortality, childhood povertyand growingup ina single-parenthousehold (Gruber,Levine and Staiger,1999), as wellas lowerrates of drug usage (Charles and Stephens, 2002). Donohue and Levitt(2001) report a numberof pieces of evidence consistent witha causal linkbetween legalized abortion and crime,a hypothesis that tomy knowledgewas Ž rstarticulated in Bouza (1990).The Ž vestates that allowed abortionin 1970 (three years before Roe v. Wade)experienceddeclines in crime ratesearlier than therest of thenation. Stateswith high and lowabortion rates in the1970s experienced similar crime trends for decades untilthe Ž rst cohorts exposedto legalized abortion reached the high-crime ages around 1990.At that point, thehigh-abortion states saw dramaticdeclines in crime relative to the low-abortionstates overthe next decade. Themagnitude of thedifferences in the crimedecline between high- and low-abortionstates was over25 percent for homicide,violent crime and propertycrime. For instance, homicidefell 25.9percent in high-abortion states between1985 and 1997compared to an increase of4.1 percent in low-abortion states. Paneldata estimatescon Ž rmthe strong negativerelationship between lagged abortion and crime.An analysis ofarrest rates byage reveal that onlyarrests of those born afterabortion legalization are affected bythe law change. UnderstandingWhy Crime Fell inthe 1990s 183

Anumberof other studies con Ž rmalinkbetween abortion and crime. 9 Reyes (2002)reports somewhat smaller, but stillsubstantial estimatesof abortion on crimeusing U.S. data. Sen(2002) Ž nds alinkbetween abortion and crimein Canadian data that mirrorsthe U.S. experience.Pop-Eleches (2002) documents theeffects of an unexpectedabortion ban imposedin Romania in1966. Extra- polatingthe conservative estimates of Donohue and Levitt(2001) to cover theperiod 1991 –2000,legalized abortion is associated witha 10percent reduction inhomicide, violent crime and propertycrime rates, which wouldaccount for 25–30percent of the observed crime decline in the 1990s.

Summarizingthe Evidence Thebasic conclusions ofthe analysis forthe 1990s are summarized in Table5. Each rowof thetable corresponds toapossibleexplanatory factor discussed inthe paper. The Ž rstthree columns representmy estimatesof the impact that factorhas had on homicide,violent crime and propertycrime, respectively. The Ž nal column ismyassessment ofhow speculativethe estimates in therow are. The bottom three rowsof the table present the estimated cumulative impact of the listed factors and theactual declinein crime according topolice reports and victimizationsurveys. Notethat thebottom three rows match up relativelyclosely, particularly for the reportedcrime data. Thedeclines in victimization are larger than thepredicted declines.For the most part, however,the decline in crime in the 1990s is not a mystery.The observed crime patterns can bereconciled in a straightforward manner withwhat weknow about thefactors that in  uencecrime. 10

Extending the Analysis to the Period 1973 –1991

Theemphasis ofthis articleis onexplainingthe decline in crime in the1990s. It isalso interesting,however, to explore the extent to which changes intheset of factors Ianalyzecan explainthe observed patterns ofcrimein earlierperiods. The year1973, when data fromthe NCVS Ž rstbecomes available, serves as alogicaltime periodin which tobegin the analysis. Table6 summarizesthe results of theanalysis forthe period 1973 –1991. The

9 Only onepublished study, Joyce(2003), has challengedthe hypothesislinking abortion and crime, arguingthat the linkbetween abortion and crimeis not presentfor the period1985 –1990.In areplyto Joyce,however, Donohue and Levitt(2003) demonstratethat forthe precisecohorts that Joyce Ž nds no reductionin crimein 1985 –1990,lifetime rates ofcriminalityare indeed reduced. Joyce ’sresultsappears tobe dueto the fact that crackhit the high-abortionstates, inparticularNew York and California,earlier and harderthan otherstates. 10 Thewestern European experience lends further supportto a causalrole of these four factors inthe U.S. crimereduction. Unlike the United States, Europehas experiencedrelatively small increases in policeand prisonpopulations. Nor was crackever a majorfactor inEurope. Although legalrules regardingabortion vary widelyacross Europe, no western European country has an abortionrate close tothe Americanrate. Giventhe absenceof these catalysts forcrime reduction, it isnot surprisingEurope has not matchedthe U.S. crimedecline. 184Journal of Economic Perspectives

Table 5 Summarizingthe Estimated Contribution of Various Factors to the Decline in Crimein the 1990s

Percentagechange in crime that thisfactor accounts for over the period 1991–2001:

Violent Property Certaintylevel Factor Homicide crime crime ofestimated impact

Strongeconomy 0 0 22 High Changingdemographics 0 22 25 High Betterpolicing strategies 21 21 21 Low Guncontrol laws 0 0 0 Medium Concealedweapons laws 0 0 0 High Increasedusage of capital punishment 21.5 0 0 Medium Increasesin the numberof police 25.5 25.5 25.5 Medium Increasesin the prisonpopulation 212 212 28 High Thedecline of crack 26 23 0 Low Legalizedabortion 210 210 210 Medium

Total ofall factors considered 236 233.5 231.5

Actual changein UCR reportedcrime 243 234 229 Actual changein NCVS victimization — 250 253

Notes: Theestimated impacts reported in the tableare based on the discussionpresented throughout the text ofthis article.The last columnof the tablegives my appraisalof how speculative the estimatesare foreach of the factors considered. structureof Table 6 parallelsthat ofTable 5. Theone factor that dominatesall othersin termsof predictedimpact on crimein this earlierperiod is thegrowth in theprison population. Between1973 and 1991,the incarceration rate more than tripled,rising from 96 to 313 inmates per 100,000 residents. By my estimates, that should have reducedviolent crime and homicideby over 30 percentand property crimeby more than 20percent.Note that this predictedimpact of incarceration is much largerthan forthe latter period. Therise of crack cocaine servedas astrongcounterbalancing forceto rising prison populations inthis earlierperiod, especially for homicide. Between 1984 (roughlythe date at which crack Ž rstappeared) and 1991,homicide rates of young blackmales nearly doubled. Thegrowth in homicide was much smallerfor other groups: 30,40 and 7percentincreases for older black males, young whitemales and olderwhite males, respectively. (Between 1973 and 1984,homicide trends across thesegroups weregenerally similar.) If homiciderates of black males had mirrored that ofthe rest of the population between1984 and 1991,overall homicide rates wouldhave been16 percentlower in 1991,which Itakeas abestguess of the impact ofcrack over this timeperiod. Following the same logic used earlier,the implied impactof crackon violentcrime is half as largeand theimpact on propertycrime negligible. Steven D.Levitt185

Table 6 Summarizingthe Estimated Contribution of Various Factors to the Decline in Crime, 1973–1991

Percentagechange in crime that this factor accounts for overthe period 1973 –1991:

Factor Homicide Violentcrime Propertycrime

Strongeconomy 0 0 2 Changingdemographics 4 1 26 Betterpolicing strategies 0 0 0 Guncontrol laws 0 0 0 Concealedweapons laws 0 0 0 Increasedusage of capital punishment 0 0 0 Increasesin the numberof police 23 23 23 Increasesin the prisonpopulation 235 235 224 Therise of crack 16 8 0 Legalizedabortion 22 22 24

Total ofall factors considered 220 231 235

Actual changein UCR reportedcrime 5 83 38 Actual changein NCVS victimization — 2 232

Notes: Theestimated impacts reported in the tableare based on the sameset ofassumptionsunderlying Table5, butapplied to the observedchanges in the factors indicatedover the period1973 –1991.

Theestimated impact of adding morepolice and legalizedabortion is much smallerin the earlier decades than inthe 1990s. Between 1973 and 1991,the numberof police per capita grewonly 7 percent(compared with 14 percentin the 1990s).By the year 1991, the proportion of the population at-riskfor criminal activitythat had beenexposed to abortion in utero was stillsmall. Abortion was legalizedin most of the country in1973, implying that abortionexposure was primarilylimited to those 17and underin 1991. Thus, whileincreased police and abortionexposure worked to reduce crime in theperiod 1973 –1991,these factors playeda much morelimited role than was thecase inthe 1990s. Thefactors that wereunimportant inexplaining crime changes inthe 1990s generallyalso had littleimpact in the early period. The lone exception to this conclusion istheimpact of demographicshifts, which servedto increase homicide ratesin the1970s and 1980sslightly, but toreduceproperty crime. The differential impactof demographics on homicideand propertycrime is driven by two factors. First,between 1973 and 1991,there was adrop intheshare ofteenagers relative to young adults. Teenagersdisproportionately engage in property crime, whereas young adults areat greaterrisk for homicide and violence.Second, increasesin the fractionof the population that isblack have amuch largerimpact on homicide than propertycrime since the racial gap inhomiciderates is much greaterthan it isforproperty crime. In summary, thefactors Iexaminecumulatively predict crime declines 186Journal of Economic Perspectives

between1973 and 1991of between 20 and 35percent. Essentially all of this predictedreduction is attributable to increased incarceration; the other factors largelycounterbalance oneanother. Comparingthese predictions to theobserved patternin reported crime (as shown inTable 2 and also presentedin the penul- timaterow of this table),the hypothesized crime declines failed to materialize.For allcrime categories, crimes reported to the FBI actuallyincreased between 1973 – 1991,in somecases sharply (reportedviolent crime rose 83 percent). Comparing thepredictions to NCVS victimizationtrends, thenumbers are much closer. Although asubstantial gap betweenpredicted violent crime and victimization remains,the magnitude of the gap isonlyone-third as largeas forreported crime. Forproperty crime, the predicted and actual victimizationare nearly identical. Thus, incontrast tothe1990s, the actual crimeexperience in the1973 –1991 periodis not wellexplained by the set of factors analyzedin this paper. There appears tobe a substantial unexplainedrise in crime over the period 1973 –1991. Becausereported crime and victimizationdata show so littlecongruence prior to the1990s, it is dif Ž cultto know the true magnitude of the residual.

Conclusions

Crimefell sharply and unexpectedlyin the 1990s. Four factors appear to explainthe drop incrime: increased incarceration, more police, the decline of crackand legalizedabortion. Otherfactors oftencited as importantfactors driving thedecline do not appear tohave playedan importantrole: the strong economy, changing demographics, innovativepolicing strategies, gun lawsand increaseduse ofcapital punishment. In starkcontrast, thecrime experience between 1973 and 1991is not wellexplained by thefactors identi Ž edinthis paper. Thereal puzzle in myopinion, therefore,is not whycrime fell in the 1990s, but whyit did not start fallingsooner. An additional conclusion fromthis analysis isthat thesimplistic accounts of whycrime fell offered by so-called experts to themedia can bequitemisleading. Of theeight reasons mostfrequently cited in newspapers, Iconclude that onlythree ofthe factors aretruly important. Afourth factorI considerimportant, legalized abortion, does not receivea singlemention. To the extent that theallocation of resourcesfor reducing crime is in  uenced bysuch pronouncements, themoney mayvery well be misspent. Doesthe analysis ofthis paperprovide any guidance as tothe likely  uctua- tions incrime rates over the next decade? Given the historical volatility in crime  uctuations and thefailure of the factors examinedto explain crime prior to the 1990s,the honest answerto that questionmay be “not much.” Nonetheless,the past mayshed somelight on thefuture. Of thefour factors that Ibelieveto account for much ofthe recent crime decline, only rising numbers of police of Ž cers and legalizedabortion are likely to be continuing contributorsto future crime declines. Giventhat thenumber of police has beengrowing for four decades, itseems UnderstandingWhy Crime Fell inthe 1990s 187

plausiblethat this trendwill continue unless localgovernment budget problems becomeextreme, contributing perhaps a5percentreduction in crime over the nextdecade. Withrespect to abortion, asubstantial fractionof the criminally active population today was born priorto legalization. As that population continues toage and isreplacedby new cohorts born afterlegalization, some ongoing reductions in crimemight be expected, amounting toperhaps a5 –10percent cumulative crime reductionover the period 2001 –2010.Prison populations areunlikely to continue togrow unchecked and have stabilizedin recent years. With many states facing difŽ cult Ž scal constraints, itwould appear likelythat prison populations may actuallyfall over the next decade, givinga slightimpetus to crime. Crack-related violenceappears tohave reacheda steadystate, withlittle future crime reduction tobe expected through that channel. Iseelittle reason tobelieve that factors that were not in uentialon crimein the preceding three decades willbe in  uential in thenext ten years. Withrespect to new threats looming on thehorizon, onepotential contributor to increasedcrime rates that does not appear tobe adequately appreciated is the coming ofage of so-called “crackbabies, ” and moregenerally, those who spent theirearly childhood yearsin families and neighborhoods ravagedby crack. Theevidence regard- ingthe direct physiological impact of exposure to crack in utero suggeststhat adverse effectsare short-lived, providing reason foroptimism (Frank, Augstyn, Knight, Pelland Zuckerman, 2001).On theother hand, thehome environment of such babies,or childrenraised in crack-af  ictedareas more generally, may be quite dif Žcult, inducing acriminogeniceffect. Overall, however, it appears that continued crimedeclines over thenext decade remain a realisticpossibility, albeit on ascalethat islikely to be much moremoderate than inthe recent past. y Iwouldlike to thank Brad De Long,John Donohue, Victor Fuchs, Andrei Shleifer, Timothy Taylorand Michael Waldman for their comments andadvice on this paper. This research was fundedby theNational Science Foundation.

References

Allpolitics. 1997. “ClintonUnveils Flurry of sonsof Criminal Justice Statistics 1999. ” Home Plans toFight Crime. ” Availableat http:// OfŽceStatistical Bulletin. May, 6/01. images.cnn.com/ALLPOLITICS/1997/02/19/ Becker,Gary. 1968. “Crimeand Punishment: clinton.crime/ . An EconomicApproach. ” Journalof Political Econ- Ayres,Ian andJohn Donohue. 2003. “Shoot- omy. 76:2, pp. 169 –217. ing Down the ‘MoreGuns, LessCrime ’ Hypoth- Bennett,William, John DiIulio and John esis.” StanfordLaw Review .55:4, pp. 1193 –312. Walters. 1996. Body Count. NewYork: Simon and Barclay, Gordon,Cynthia Tavares and Arsa- Schuster. laanSiddique. 2001. “International Compari- Black,Dan and Daniel Nagin. 1998. “Do 188Journal of Economic Perspectives

‘Right-to-Carry ’ LawsDeter Violent Crime? ” Jour- “Carrots,Sticks, and BrokenWindows. ” NBER nalof Legal Studies .27:1, pp. 209 –19. WorkingPaper No. 9061. Blumstein,Alfred and Richard Rosenfeld. Dagg,Paul K. 1991. “ThePsychological Se- 1998. “ExplainingRecent Trends in U.S. Homi- quelaeof Therapeutic Abortion —Denied and cide Rates.” Journalof Criminal Law and Criminol- Completed.” AmericanJournal of Psychiatry . May, ogy. 88:4, pp. 1175 –216. 148:5, pp. 578 – 85. Boggess,Scott andJohn Bound. 1997. “Did David,Henry, Zdenek Dytrych, Zdenek Mate- CriminalActivity IncreaseDuring the 1980s? jcekand Vratislav Schuller. 1988. BornUnwanted: ComparisonsAcross Data Sources. ” Social Science DevelopmentalEffects of Denied Abortion. New York: Quarterly. 78:3, pp. 725 –39. Springer. Bouza,Anthony V. 1990. ThePolice Mystique: Dezhbakhsh,Hashem, Paul Rubin and Joanna An Insider’sLook atCops, Crime, and the Criminal Shepherd. 2002. “DoesCapital Punishment JusticeSystem .NewYork: Plenum Press. Havea DeterrentEffect? New Evidence from Britt, Chester,Gary Kleckand David Bordua. Post-MoratoriumPanel Data. ” Mimeo,Emory 1996. “AReassessmentof the D.C. GunLaw: University. SomeCautionary Notes on the Useof Inter- DiIulio,John. 1996. “HelpWanted: Econo- ruptedTime Series Designs for Policy Assess- mists, Crime,and PublicPolicy. ” Journalof Eco- ment.” Lawand Society Review. 30:2, pp. 361 – 80. nomicPerspectives. Winter, 10, pp. 3 –24. Callahan,Charles, Frederick Rivera and DiIulio,John and Anne Piehl. 1991. “Does ThomasKoepsell. 1994. “MoneyFor Guns: Eval- PrisonPay? The Stormy National Debate over uationof the SeattleGun Buy-Back Program. ” the Cost-Effectivenessof Imprisonment. ” Brook- PublicHealth Reports. 109:4, pp. 472 –77. ings Review.Fall, pp. 28 –35. Donohue,John and Steven Levitt. 2001. “Le- Cameron,Samuel. 1988. “TheEconomics of galizedAbortion and Crime. ” QuarterlyJournal of CrimeDeterrence: A Surveyof Theoryand Evi- Economics. 116:2, pp. 379 –420. dence.” Kyklos. 41:2, pp. 301 –23. Donohue,John and Steven Levitt. 2003. “Fur- Cameron,Samuel. 1994. “AReviewof the therEvidence That LegalizedAbortion Lowered EconometricEvidence on the Effect of Capital Crime:A Replyto Joyce. ” Journalof Human Re- Punishment. ” Journalof Socio-Economics . Spring/ sources.Forthcoming. Summer,23, pp. 197 –214. Donohue,John and Peter Siegelman. 1998. Charles,Kerwin Co Ž andMelvin Stephens Jr. “AllocatingResources among Prisons and Social 2002. “AbortionLegalization and Adolescent Programsin the Battle against Crime. ” Journal of SubstanceAbuse. ” NBERWorking Paper No. Legal Studies . 27:1, pp. 1– 43. 9193. Duggan,Mark. 2001. “MoreGuns, More Cook,Philip and John Laub. 1998. “The Un- Crime.” Journalof Political Economy. 109:5, pp. precedentedEpidemic in Youth Violence, ” in 1086–114. Crimeand Justice: An Annual Review of Research. Ehrlich,Isaac. 1975. “TheDeterrent Effect of MichaelTonry, ed.Chicago: University of Chi- CapitalPunishment: AQuestionof Life and cagoPress, pp. 26 –64. Death.” AmericanEconomic Review. 65:3, pp. 397 – Cook,Philip and Jens Ludwig. 1996. Guns in 417. America:Results of a ComprehensiveSurvey of Ehrlich,Isaac. 1977. “CapitalPunishment and GunOwnership and Use .Washington: Police Deterrence:Some Further Thoughts and Foundation. Evidence.” Journalof Political Economy. 85:4, pp. Cook,Philip, Stephanie Molliconi and 741– 88. ThomasCole. 1995. “RegulatingGun Markets. ” Forst,Brian, Victor Filatov and Lawrence Journalof Criminal Law and Criminology. 16:1, pp. Klein. 1978. “TheDeterrent Effect ofCapital 59 –92. Punishment: An Assessmentof the Estimates, ” in Cork,Daniel. 1999. “ExaminingSpace-Time Deterrenceand Incapacitation: Estimating the Effects Interaction inCity-Level Homicide Data: Crack ofCriminal Sanctions on Crime Rates. A.Blumstein, Marketsand the Diffusionof Guns Among D.Naginand J.Cohen,eds. Washington, D.C.: Youth.” Journalof Quantitative Criminology . 15:4, NationalAcademy of Sciences,pp. 336 –60. pp. 379– 406. Fox, James. 1996. “Trendsin Juvenile Vio- Corman,Hope and H. Naci Mocan. 2000. “A lence:A Reportto the United States Attorney Time-SeriesAnalysis ofCrime,Deterrence, and Generalon Currentand FutureRates ofJuvenile DrugAbuse in New York City. ” AmericanEconomic Offending.” Washington, D.C.: Bureauof Justice Review.June,90, pp. 584 – 604. Statistics. Corman,Hope and H. Naci Mocan. 2002. Fox, James. 1997. “Trendsin Juvenile Vio- Steven D.Levitt189

lence:An Update. ” Washington, D.C.: Bureauof Levitt,Steven. 1996. “TheEffect ofPrison JusticeStatistics. PopulationSize on CrimeRates: Evidencefrom Frank,D. A.,M.Augstyn,W. G.Knight,T. Pell PrisonOvercrowding Litigation. ” QuarterlyJour- andB. Zuckerman. 2001. “Growth,Develop- nalof Economics. 111:2, pp. 319 –52. ment, and Behaviorin Early Childhood Follow- Levitt,Steven. 1997. “Using ElectoralCycles ingPrenatal Cocaine Exposure: A SystematicRe- inPolice Hiring to Estimate the Effect ofPolice view.” Journalof the American Medical Association. on Crime.” AmericanEconomic Review. 87:3, pp. 285:12, pp. 1613 –625. 270–90. Freeman,Richard. 1995. “TheLabor Market, ” Levitt,Steven. 1998. “JuvenileCrime and Pun- in Crime. JamesQ. Wilsonand JoanPetersilia, ishment.” Journalof Political Economy. December, eds.San Francisco:ICS Press,pp. 171 –92. 106,pp. 1156 –185. Goldstein,Paul, Henry Brownstein, Patrick Levitt,Steven. 1999. “TheExaggerated Role Ryan andPatricia Bellucci. 1997. “Crack and ofChangingAge Structure in Explaining Aggre- Homicidein New York City: ACaseStudy inthe gateCrime Changes. ” Criminology .August, 37, Epidemiologyof Violence, ” in Crackin America: pp. 537–99. DemonDrugs and Social Justice. CraigReinarman Levitt,Steven. 2002. “Using ElectoralCycles and HarryLevine, eds. Berkeley, Calif.: Univer- inPolice Hiring to Estimate the Effect ofPolice sity ofCaliforniaPress, pp. 113 –30. onCrime: A Reply. ” AmericanEconomic Review. Gould,Eric, Bruce Weinberg and David Mus- September,92, pp. 1244 –250. tard. 1997. “CrimeRates and LocalLabor Mar- Levitt,Steven and Sudhir Venkatesh. 2000. ketOpportunities in the United States: 1979 – “An EconomicAnalysis ofa Drug-SellingGang ’s 1991.” Workingpaper. Finances.” QuarterlyJournal of Economics. August, Grogger,Jeff and Michael Willis. 2000. “The 115, pp. 755– 89. Emergenceof Crack Cocaine and the Risein Loeber,Rolf and Magda Stouthamer-Loeber. UrbanCrime Rates. ” Reviewof Economics and Sta- 1986. “FamilyFactors as Correlatesand Predic- tistics. November,82, pp. 519 –29. torsof Juvenile Conduct Problems and Delin- Gruber,Jonathan, Phillip Levine and Douglas quency,” in Crimeand Justice, An Annual Review of Staiger. 1999. “AbortionLegalization and Child Research,Volume 7. MichaelTonry and Norval LivingCircumstances: Who isthe ‘Marginal Morris,eds. Chicago: University of Chicago Child’?” QuarterlyJournal of Economics. 114:1, pp. Press,pp. 29 –150. 263–91. Loftin,Colin, David McDowall, Brian Joyce, Ted. 2003. “DidLegalized Abortion Wiersemaand Talbert Cottey. 1991. “Effects of LowerCrime? ” Journalof Human Resources . RestrictiveLicensing of Handguns onHomicide Forthcoming. and Suicidein the Districtof Columbia. ” Katz, Lawrence,Steven Levitt and Ellen Shus- NewEngland Journal of Medicine. 325:23, pp. torovich. 2003. “PrisonConditions, Capital Pun- 1615– 620. ishment, and Deterrence. ” AmericanLaw and Eco- LosAngelesPolice Commission. 1996. “In the nomicsReview. 5:2, pp. 318– 43. Courseof Change: The Los Angeles Police De- Kennedy,David, Anne Piehl and Anthony partment FiveYears after the ChristopherCom- Braga. 1996. “YouthViolence in Boston: Gun mission.” LosAngeles Board of PoliceCommis- Markets, SeriousYouth Offenders, and aUse sioners,May 30. ReductionStrategy. ” Lawand Contemporary Prob- Lott,John and David Mustard. 1997. “Right- lems. 59:1, pp. 147 –83. to-CarryConcealed Guns and the Importanceof Kessler,Daniel and Steven Levitt. 1999. “Us- Deterrence. ” Journalof Legal Studies .January, 26, ingSentence Enhancements to Distinguish be- pp. 1–68. tweenDeterrence and Incapacitation. ” Journal of Ludwig,Jens. 1998. “ConcealedGun Carrying Lawand Economics. April,42, pp. 343 – 63. Lawsand ViolentCrime: Evidence from State Kuziemko,Ilyana andSteven Levitt. 2003. “An Panel Data.” InternationalReview of Law and Eco- EmpiricalAnalysis ofImprisoningDrug Offend- nomics. September,18, pp. 239 –54. ers.” Journalof Public Economics .Forthcoming. Ludwig,Jens and Philip Cook. 2000. “Homi- Leamer,Edward. 1983. “Let’sTakethe Con cideand SuicideRates Associatedwith Imple- outof Econometrics. ” AmericanEconomic Review . mentationof the BradyHandgun ViolencePre- March, 73:1, pp. 31 – 43. ventionAct. ” Journalof the American Medical Levine,Philip, Douglas Staiger, Thomas Kane Association. 284:5, pp. 585 –91. andDavid Zimmerman. 1999. “Roev. Wadeand Machin,Stephen and Costas Meghir. 2000. AmericanFertility. ” AmericanJournal of Public “Crimeand EconomicIncentives. ” Unpublished Health. 89:2, pp. 199 –203. manuscript. 190Journal of Economic Perspectives

Marvell,Thomas and Carlisle Moody. 1994. Rosenfeld,Richard . 1996. “GunBuy-Backs: “PrisonPopulation Growth and CrimeReduc- CrimeControl or CommunityMobilization, ” in tion.” Journalof Quantitative Criminology. 10:2, pp. UnderFire: Gun Buy-Backs, Exchangesand Amnesty 109– 40. Programs. Martha R.Plotkin,ed. Washington, Marvell,Thomas and Carlisle Moody. 1996. D.C.: PoliceExecutive Research Forum, pp. “SpeciŽ cationProblems, Police Levels, and 1–28. CrimeRates. ” Criminology. November,34, pp. Sampson,Robert and John Laub. 1993. Crime 609– 46. inthe Making: Pathways and Turning Points through McCrary,Justin. 2002. “DoElectoral Cycles in Life. Cambridge:Harvard University Press. PoliceHiring Really Help Us Estimatethe Effect Sampson,Robert and Janet Lauritsen. 1997. ofPolice on Crime? Comment. ” AmericanEco- “Racialand EthnicDisparities in Crime and nomicReview. September,92, pp. 1236 –243. CriminalJustice in the United States, “ in Crime Miller,Ted, Mark Cohen and Shelli Rossman. andJustice: A Reviewof Research, Volume 21. M. 1993. “VictimCosts of ViolentCrime and Result- Tonry, ed.Chicago: University ofChicagoPress, ingInjuries. ” HealthAffairs. 12:4, pp. 186 –97. pp. 311–74. Mocan,Naci andR. KajGittings. 2003. “Get- ScientiŽc American. 1996. “ProŽ le:James Alan ting Off DeathRow: CommutedSentences and Fox—Catchinga ComingCrime Wave. ” June, the DeterrentEffect ofCapital Punish- pp. 40– 41. ment.” Journalof Law and Economics . October, Sen,Anindya. 2002. “DoesIncreased Abortion Forthcoming. Leadto Lower Crime? Evaluating the Relation- Passell,Peter and John Taylor. 1977. “The shipbetween Crime, Abortion, and Fertility. ” DeterrentEffect ofCapital Punishment: An- Unpublishedpaper, University ofWaterloo De- other View.” AmericanEconomic Review. 67:3, pp. partment ofEconomics. 445–51. Soares,Rodrigo. 1999. “Development,Crime, Perkins,Craig. 1997. “AgePatterns ofVictims and Punishment: Accountingfor the Interna- ofSerious Violent Crime. ” Departmentof Jus- tionalDifferences in CrimeRates. ” Unpublished tice, OfŽ ceof Justice Programs, Bureau of Jus- paper,University ofChicago Department of ticeStatistics SpecialReport, NCJ-162031. Economics. Pop-Eleches,Christian. 2002. “TheImpact of Sorenson,Susan, Douglas Wiebe and Richard an AbortionBan onSocio-EconomicOutcomes Berk. 2002. “LegalizedAbortion and the Homi- ofChildren: Evidence from Romania. ” Unpub- lishedpaper, Columbia University. cideof YoungChildren: An EmpiricalInvestiga- Raphael,Stephen and Jens Ludwig. 2003. tion.” Analysesof Social Issuesand Public Policy. 2:1, “PrisonSentence Enhancements: The Case of pp. 239–56. ProjectExile, ” in EvaluatingGun Policy: Effectson Spelman,William. 1994. CriminalIncapacita- Crimeand Violence. JensLudwig and PhilipCook, tion. NewYork: Plenum Press. eds.Washington, D.C.: BrookingsInstitution, Stolley,K. S. 1993. “Statistics onAdoptionin chapter 7. the United States. ” TheFuture of Children: Adop- Raphael,Stephen and Rudolf Winter-Ebmer. tion. 3:1, pp. 26– 42. 2001. “Identifying the Effect ofUnemployment Thompson,Cheryl. 1997. “DetailingFailings on Crime.” Journalof Law and Economics. 44:1, ofD.C. PoliceDepartment. ” WashingtonPost. pp. 259– 84. April 9. Reuter,Peter and Jenny Mouzos. 2003. “Aus- Wilson,James Q. 1985. ThinkingAbout Crime. tralia: AMassiveBuyback of Low-RiskGuns, ” in NewYork: Random House. EvaluatingGun Policy: Effectson Crime and Violence. Wilson,James Q. 1995. “Crimeand Public JensLudwig and PhilipCook, eds. Washington, Policy,” in Crime. JamesQ. Wilsonand Joan D.C.: BrookingsInstitution, chapter4. Petersilia,eds. San Francisco:ICS Press,pp. Reyes,Jessica Wolpaw. 2002. “The Impact 489–510. ofChildhood Lead Exposure on Crime. ” Un- Wright,James andPeter Rossi. 1994. Armed publishedpaper, Harvard Department of andConsidered Dangerous: A Surveyof Felons and Economics. TheirFirearms. NewYork: Aldine de Gruyter.