DOES MOVIE VIOLENCE INCREASE VIOLENT CRIME?gdahl/papers/movies-and-violence.pdf · DOES MOVIE...

DOES MOVIE VIOLENCE INCREASE VIOLENT CRIME?∗

GORDON DAHL AND STEFANO DELLAVIGNA

Laboratory experiments in psychology find that media violence increases ag-gression in the short run. We analyze whether media violence affects violent crimein the field. We exploit variation in the violence of blockbuster movies from 1995to 2004, and study the effect on same-day assaults. We find that violent crimedecreases on days with larger theater audiences for violent movies. The effect ispartly due to voluntary incapacitation: between 6 P.M. and 12 A.M., a one mil-lion increase in the audience for violent movies reduces violent crime by 1.1% to1.3%. After exposure to the movie, between 12 A.M. and 6 A.M., violent crime isreduced by an even larger percent. This finding is explained by the self-selectionof violent individuals into violent movie attendance, leading to a substitutionaway from more volatile activities. In particular, movie attendance appears toreduce alcohol consumption. The results emphasize that media exposure affectsbehavior not only via content, but also because it changes time spent in alterna-tive activities. The substitution away from more dangerous activities in the fieldcan explain the differences with the laboratory findings. Our estimates suggestthat in the short run, violent movies deter almost 1,000 assaults on an aver-age weekend. Although our design does not allow us to estimate long-run ef-fects, we find no evidence of medium-run effects up to three weeks after initialexposure.

I. INTRODUCTION

Does media violence trigger violent crime? This question isimportant for both policy and scientific research. In 2000, theFederal Trade Commission issued a report at the request of thepresident and the Congress, surveying the scientific evidence andwarning of negative consequences. In the same year, the Amer-ican Medical Association, together with five other public-health

∗Eli Berman, Sofia Berto Villas-Boas, Saurabh Bhargava, David Card,Christopher Carpenter, Ing-Haw Cheng, Julie Cullen, David Dahl, Liran Einav,Matthew Gentzkow, Edward Glaeser, Jay Hamilton, Ethan Kaplan, Lawrence F.Katz, Lars Lefgren, Ulrike Malmendier, Julie Mortimer, Ted O’Donoghue, AnnePiehl, Mikael Priks, Bruce Sacerdote, Uri Simonsohn, and audiences at LondonSchool of Economics, Ohio State, Princeton University, Queens University, RAND,Rutgers New Brunswick, the Institute for Labor Market Policy Evaluation in Up-psala, UC Berkeley, UC San Diego, University of Tennessee Knoxville, Universityof Western Ontario, University of Zurich, Wharton, the Munich 2006 Conferenceon Economics and Psychology, the NBER 2006 Summer Institute (Labor Stud-ies), the 2006 SITE in Psychology and Economics, the IZA Conference on Per-sonnel and Behavioral Economics, the 2008 IZA/SOLE Transatlantic Meeting ofLabor Economists, and at the Trento 2006 Summer School in Behavioral Eco-nomics provided useful comments. We would like to thank kids-in-mind.com andthe-numbers.com for generously providing their data. Scott Baker and ThomasBarrios provided excellent research assistance.

C© 2009 by the President and Fellows of Harvard College and the Massachusetts Institute ofTechnology.The Quarterly Journal of Economics, May 2009

677

678 QUARTERLY JOURNAL OF ECONOMICS

organizations, issued a joint statement on the risks of exposure tomedia violence (American Academy of Pediatrics et al. 2000).

The evidence cited in these reports, surveyed by Andersonand Buschman (2001) and Anderson et al. (2003), however, doesnot establish a causal link between media violence and violentcrime. The experimental literature exposes subjects in the lab-oratory (typically children or college students) to short, violentvideo clips. These experiments find a sharp increase in aggressivebehavior immediately after the media exposure, compared to acontrol group exposed to nonviolent clips. This literature providescausal evidence on the short-run impact of media violence on ag-gressiveness, but not whether this translates into higher levels ofviolent crime in the field. A second literature (e.g., Johnson et al.[2002]) shows that survey respondents who watch more violentmedia are substantially more likely to be involved in self-reportedviolence and crime. This second type of evidence, although indeedlinking media violence and crime, has the standard problems ofendogeneity and reverse causation.

In this paper, we provide causal evidence on the short-runeffect of media violence on violent crime. We exploit the naturalexperiment induced by time-series variation in the violence ofmovies shown in the theater. As in the psychology experiments,we estimate the short-run effect of exposure to violence, butunlike in the experiments, the outcome variable is violent crimerather than aggressiveness. Importantly, the laboratory and fieldsetups also differ due to self-selection and the context of violentmedia exposure.

Using a violence rating system from kids-in-mind.com anddaily revenue data, we generate a daily measure of national-levelbox-office audience for strongly violent (e.g., Hannibal), mildlyviolent (e.g., Spider-Man), and nonviolent movies (e.g., RunawayBride). Because blockbuster movies differ significantly in violencerating, and movie sales are concentrated in the initial weekendsafter release, there is substantial variation in exposure tomovie violence over time. The audience for strongly violent andmildly violent movies, respectively, is as high as 12 million and25 million people on some weekends, and is close to 0 on others(see Figures Ia and Ib). We use crime data from the NationalIncident Based Reporting System (NIBRS) and measure violentcrime on a given day as the sum of reported assaults (simple oraggravated) and intimidation.

DOES MOVIE VIOLENCE INCREASE VIOLENT CRIME? 679

0

2

4

6

8

10

12

14

16

Jan-

95

Jul-9

5

Jan-

96

Jul-9

6

Jan-

97

Jul-9

7

Jan-

98

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Jan-

03

Jul-0

3

Jan-

04

Jul-0

4

Weekend

Wee

ken

d a

ud

ien

ce (

in m

illio

ns

of

peo

ple

)

Nov 9 1996Ransom

July 26 1997Air Force One

July 8 2000Scary Movie

Feb 10 2001Hannibal

Dec 13 1997Scream 2 July 21 2001

Jurassic Park 3

Nov 27 1999End of Days

July 19 2003Bad Boys II

Feb 28 2004Passion of the Christ

Mar 20 2004Dawn of the Dead

FIGURE IaWeekend Theater Audience of Strongly Violent Movies

0

5

10

15

20

25

Jan-

95

Jul-9

5

Jan-

96

Jul-9

6

Jan-

97

Jul-9

7

Jan-

98

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Jan-

03

Jul-0

3

Jan-

04

Jul-0

4

Weekend

Wee

ken

d a

ud

ien

ce (

in m

illio

ns

of

peo

ple

) May 25 1996

Twister

Dec 27 1997Titanic June 12 1999

Austin Powers 2 Aug 4 2001Rush Hour 2

May 4 2002Spider-Man

May 18 2002Star Wars 2

May 25 2002Star Wars 2

July 24/31 2004Bourne Supremacy

June 5 2004

Harry Potter 3

FIGURE IbWeekend Theater Audience of Mildly Violent Movies

Plot of weekend (Friday through Sunday) box-office audience in millions ofpeople for movies rated as strongly violent and mildly violent. The ten week-ends with the highest audience for strongly violent (mildly violent) movies arelabeled. Movies are rated as strongly violent (mildly violent) if they have a kids-in-mind.com rating 8–10 (5–7). The audience data are from box-office sales (fromthe-numbers.com) deflated by the average price of a ticket.


-0.1

-0.05

0

0.05

0.1

0.15

Jan-

95

Jul-9

5

Jan-

96

Jul-9

6

Jan-

97

Jul-9

7

Jan-

98

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Jan-

03

Jul-0

3

Jan-

04

Jul-0

4

Weekend

Lo

g a

ssau

lt r

esid

ual

s

Log assault residuals

Top 10 strongly violent (8-10) movies

Top 10 mildly violent (5-7) movies

FIGURE IcLog Assaults and the Top Ten Violent Movies (Controlling for Seasonality)Plot of average (Friday through Sunday) residuals of weekend log assaults

after controlling for seasonality, holidays, and weather controls (see text for list ofall the controls). The assault data are from NIBRS. The figures highlight the tenweekends with the largest strongly violent movie audience (see Figure I(a)) andthe ten weekends with the largest mildly violent movie audience (see Figure I(b)).

We find that, on days with a high audience for violent movies,violent crime is lower, even after controlling flexibly for season-ality. To rule out unobserved factors that contemporaneouslyincrease movie attendance and decrease violence, such as rainyweather, we use two strategies. First, we add controls for weatherand days with high TV viewership. Second, we instrument formovie audience using the predicted movie audience based onthe following weekend’s audience. This instrumental variablestrategy exploits the predictability of the weekly decrease inattendance. Adding in controls and instrumenting, the correlationbetween movie violence and violent crime becomes more negativeand remains statistically significant.

The estimated effect of exposure to violent movies is small inthe morning or afternoon hours (6 A.M.–6 P.M.), when movie atten-dance is minimal. In the evening hours (6 P.M.–12 A.M.), instead,we detect a significant negative effect on crime. For each millionpeople watching a strongly or mildly violent movie, respectively,violent crimes decrease by 1.3% and 1.1%. The effect is smallerand statistically insignificant for nonviolent movies. In the


nighttime hours following the movie showing (12 A.M.–6 A.M.), thedelayed effect of exposure to movie violence is even more nega-tive. For each million people watching a strongly or mildly violentmovie, respectively, violent crime decreases by 1.9% and 2.1%.Nonviolent movies have no statistically significant impact. Unlikein the psychology experiments, therefore, media violence appearsto decrease violent behavior in the immediate aftermath of expo-sure, with large aggregate effects. The total net effect of violentmovies is to decrease assaults by roughly 1,000 occurrences perweekend, for an annual total of about 52,000 weekend assaultsprevented. This translates into an estimated yearly social gain ofapproximately $695 million in avoided victimization losses (directmonetary costs plus intangible quality-of-life costs). The resultsare robust to a variety of alternative specifications, measures ofmovie violence, instrument sets, and placebo tests. Additional es-timates using variation in violent DVD and VHS video rentals areconsistent with our main findings.

We also examine the delayed impact of exposure to movie vi-olence on violent crime. Although our research design (like thelaboratory designs) cannot test for a long-run impact, we can ex-amine the medium-run impact in the days and weeks followingexposure. We find no impact on violent crime on Monday andTuesday following weekend movie exposure. We also find no im-pact one, two, and three weeks after initial exposure, controllingfor current exposure. Hence, the same-day decrease in crime isunlikely to be due to intertemporal substitution of crime from thefollowing days.

To interpret the results, we develop a simple model whereutility-maximizing consumers choose between violent movies,nonviolent movies, and an alternative activity. These options gen-erate violent crime at different rates. The model provides threemain insights. First, in the reduced form implied by the model,the estimates of exposure to violent movies capture the impact forthe self-selected population that chooses to attend violent movies,and not the population at large. In particular, the violent sub-population self-selects into more violent movies, magnifying anyeffects of exposure. Second, the reduced-form estimates capturethe net effect of watching a violent movie and not participatingin the next-best alternative activity. A blockbuster violent moviehas a direct effect on crime as more individuals are exposed toscreen violence, but also an indirect effect as people are drawnaway from an alternative activity (such as drinking at a bar) and


its associated level of violence. Third, it is possible to identify thedirect effect of violent movies if one can account for self-selection.

We interpret the first empirical result, that exposure toviolent movies lowers same-day violent crime in the evening(6 P.M. to 12 A.M.), as voluntary incapacitation. On evenings withhigh attendance at violent movies, potential criminals chooseto be in the movie theater and hence are incapacitated fromcommitting crimes. The incapacitation effect is larger for violentmovies because potential criminals self-select into violent, ratherthan nonviolent, movies. Indeed, using data from the ConsumerExpenditure Survey time diaries, we document substantialself-selection. Demographic groups with higher crime rates, suchas young men, select disproportionately into watching violentmovies.

The second result is that violent movies lower violent crimein the night after exposure (12 A.M. to 6 A.M.). These estimates re-flect the difference between the direct effect of movie violence andthe violence level associated with an alternative activity. Hence,the reduction in crime associated with violent movies is best un-derstood as movie attendance displacing more volatile alternativeactivities both during and after movie attendance. Because alco-hol is a prominent factor that has been linked to violent crime(Carpenter and Dobkin 2009), and alcohol is not served in movietheaters, one potential mechanism is a reduction in alcohol con-sumption associated with movie attendance. Consistent with thismechanism, we find larger decreases for assaults involving alco-hol or drugs and for assaults committed by offenders just over(versus just under) the legal drinking age.

A common theme to the findings above is the importance ofself-selection of potential criminals into violent movies. We pro-vide additional evidence on selection using ratings data from theInternet Movie Database (IMDb). We categorize movies based onhow frequently they are rated by young males. We find that, evenafter controlling for the level of violence, movies that dispropor-tionately attract young males significantly lower violent crime.

Our second result appears to contradict the evidence fromlaboratory experiments, which find that violent movies increaseaggression through an arousal effect. However, the field and lab-oratory results are not necessarily contradictory. The laboratoryexperiments estimate the impact of violent movies in partial equi-librium, holding the alternative activities constant. Our naturalexperiment instead allows individuals to decide in equilibriumbetween a movie and an alternative activity. Exposure to movie


violence can lower violent behavior relative to the foregone al-ternative activity (the field findings), even if it increases violentbehavior relative to exposure to nonviolent movies (the labora-tory findings). Under assumptions that allow us to estimate theamount of selection, our field estimates can be used to infer theeffect of exposure, holding the alternative activities constant (asin the laboratory).

Using this methodology, we find evidence of an arousal effectconsistent with the laboratory experiments; violent movies inducemore violent crime relative to nonviolent movies. However, thisestimated arousal effect is smaller than the time-use effect—onnet, violent movies still induce substantially less violent behaviorthan the alternative activity. Hence, the field evidence provides abound for the size of the arousal effect identified in the laboratory.This example also suggests that other apparent discrepanciesbetween laboratory and field studies (see Levitt and List [2007])might be reconciled if differences in treatment and setup aretaken into account.

Our research is related to a growing literature in economicson the effect of the media. Among others, Besley and Burgess(2002), Stromberg (2004), Gentzkow (2006), and DellaVigna andKaplan (2007) provide evidence that media exposure affects po-litical outcomes. Card and Dahl (2009) show that emotional cuesprovided by local NFL football games (in the form of unexpectedupset losses) cause a spike in family violence. Relative to thismedia literature that emphasizes the effect of content, our paperstresses the impact of time use. In our context, the substitutionin activities induced by violent movies dominates the effect ofcontent. This mechanism also operates in Gentzkow and Shapiro(2008), who show the introduction of television during preschoolhad positive effects on test scores for children of immigrants,who otherwise would have had less exposure to the Englishlanguage.

Our paper also complements the evidence on incapacitation,from the effect of school attendance (Jacob and Lefgren 2003)to the effect of imprisonment (Levitt 1996). Our paper differsfrom this literature because the incapacitation is optimallychosen by the consumers, rather than being imposed. Not allleisure activities have an incapacitation effect, however. Rees andSchnepel (2009) document an increase in crimes by spectators ofcollege football games in the host community. The prevalence ofalcohol consumption at football games, but not in movie theaters,plausibly explains the difference.


Finally, this paper is related to the literature on the impact ofemotions such as arousal (Loewenstein and Lerner 2003; Arielyand Loewenstein 2005) on economic decisions.

The remainder of the paper is structured as follows. Section IIpresents a simple model of movie attendance choice and its effecton violence. Section III describes the data, and Section IV presentsthe main empirical results. Section V provides interpretations andadditional evidence. Section VI concludes.

II. FRAMEWORK

II.A. Model

In this section we model the choice to view a violent movie andthe resulting impact on the level of violence following exposure.Our setup is meant to illustrate (i) the importance of self-selection,(ii) the effect of time use versus content for violent movies, and(iii) how estimates in the laboratory and field differ.

Individuals choose the utility-maximizing activity among fourmutually exclusive options: watching a strongly violent movie av,watching a mildly violent movie am, watching a nonviolent moviean, or participating in an alternative social activity as. Althoughwe could assume a standard multinomial choice model, any choicemodel implies probabilistic demand functions for movies P (av),P (am), P (an), and for the alternative activity 1 − P(av) − P(am) −P(an). For each type of movie, demand P

(aj

)varies based on the

quality and overall appeal of the movie (which we do not observe).We allow for heterogeneity in the taste for movies. We label

the group with high demand for violent movies as young y andthe other group as old o. Within each group, the fraction choosingactivity j is denoted as P(aj

i ) for i = y, o and j = v, m, n, s. Theaggregate demand functions for the young and old are simplythese probabilities multiplied by group size Ni.

Violence, which does not enter individuals’ utility functions,depends on the type of movies viewed, as well as on participationin the alternative social activity. We model the production functionfor aggregate log violence as linear in the demand for movies andthe alternative social activity, aggregated over young and old:

ln V =∑i=y,o

∑

j=v,m,n

αji Ni P

(aj

i

) + σi Ni(1 − P

(av

i

) − P(am

i

)−P(an

i

)) .

(1)


The parameters αvi , αm

i , αni , and σi, all (weakly) positive, capture

the effects on violence from the four alternative activities. Giventhe log specification (motivated by the similarity to a Poissonmodel), increasing the young audience size of violent movies by 1,ceteris paribus, results in roughly a αv

y percent increase in violence.Because individual-level data on movie attendance are not

available, we rewrite (1) in terms of aggregate movie attendancefor the young and old combined. (In Section IV, we discuss waysto identify consumer types using auxiliary data.) The effect oftotal audience size Aj = Ny P(aj

y) + No P(ajo ) on log violence is a

weighted average of the effects for the young and old subgroups:

ln V = (σyNy + σo No)

+∑

j=v,m,n

[x j(α j

y − σy) + (1 − x j)

(α j

o − σo)]

Aj,(2)

where x j = Ny P(ajy)/(Ny P(aj

y) + No P(ajo )) denotes the young audi-

ence share for movie j.The estimating equation we use in Section IV follows directly

from (2):

ln V = β0 + βv Av + βmAm + βnAn + ε,(3)

where ε is an additively separable error term. Comparing (3) and(2), we can write the coefficients as

β j = x j(α jy − σy

) + (1 − x j)(α j

o − σo)

for j = v, m, n.(4)

Notice the parameter β j is constant only if the young audienceshare x j is constant in response to changes in movie quality. Inwhat follows, we assume that this is approximately the case, thatis, that when movie quality changes, demand by the young and oldroughly rises and falls proportionately with each other (as wouldbe true for a multinomial logit model).

II.B. Interpretation

Expression (4) illustrates several points. First, the impactof a violent movie βv on violence is the sum of two effects: adirect effect, captured by αv

i , and an indirect effect, capturedby σi. The direct effect is the impact of violent movies, holdingeverything else constant. There are two broad theories about thedirect impact of violent movies immediately after exposure. Thefirst theory is that exposure to media violence triggers additional


aggression, whether through arousal or the imitation of violentacts (Anderson et al. 2003). The second, opposite theory is thatexposure to movie violence leads to a decrease in aggressionbecause of a cathartic effect of viewing violence on screen. Thistheory, which parallels Aristotle’s theory about the effect of theGreek tragedy, was a leading theory among psychologists until1960. Since the 1960s, a series of laboratory experiments, fromBandura, Ross, and Ross (1963) to Buschman (1995), have foundsubstantial support for arousal and imitation and little supportfor catharsis. In our model, αv

y is large if movie violence triggersviolence through arousal or imitation, and small if movie violencehas a cathartic effect.

In addition to the direct effect, there is an indirect effect due tothe displacement of alternative social activities that occurs whenan individual chooses to watch a violent movie. A first possibilityis that these displaced activities trigger crime at a lower rate thanmovie attendance. This can be the case, for example, if movies pro-vide a meeting point for potential criminals who would otherwisestay home. In this case, movie attendance, on net, increases crime(positive βv) after exposure. A second possibility is that the after-math of movie attendance is more dangerous that the alternativeactivity. This can occur, for example, if movie attendance leads toearlier bedtimes and lower alcohol consumption, compared to, say,bar attendance. In this case, movie attendance, on net, decreasescrime (negative βv).

We note that the effect of movies during exposure (the con-temporaneous effect) differs from the effect after exposure (thedelayed effect). During the movie showing, the direct effect ofmovie exposure α j approximately equals 0 for all types of moviesbecause very few crimes are committed while physically in themovie theater. In this sense, movie attendance can be viewed as aform of voluntary incapacitation: movies take individuals “off thestreets” and place them into relatively safe environments.

A second insight from (4) is that heavy moviegoers contributemost to the identification of βv. This parameter is a weightedaverage of the net effects for old and young people. To the extentthe young like violent movies more than the old, they will beoverrepresented in the audience for violent movies, and hencethe weight representing their audience share will be larger thantheir share in the population. Because the young and old havevery different crime patterns, this type of sorting can have a largeimpact on the aggregate estimate.


To illustrate the importance of selection, suppose that thedirect effect of movie exposure is the same for all movie types(αn

i = αmi = αv

i = α for i = y, o), but that the violent subpopulationengages in more dangerous alternative activities (σy > σo). In thiscase β j = α − σo − x j(σy − σo). Even in the absence of a differen-tial direct effect for violent movies, the level of violence in a moviecan affect crime. If violent movies are more likely to attract theviolent subpopulation (i.e., xv > xm > xn), as we document empir-ically below, then the effect of exposure becomes more negativewith the violence level of the movie: βv < βm < βn. Exposure toviolent movies can lower crime relative to nonviolent movies sim-ply because violent movies induce more substitution away fromdangerous activities for the violent subgroup.

In addition to this selection effect, there can be a direct effectof movie violence, as suggested by the arousal and catharsistheories. To capture this possibility, modify the example in thepreceding paragraph so that strongly violent movies have adirect effect αv (with nonviolent and mildly violent movies stillhaving impact α). Then the impact of exposure to a violentmovie is βv = (αv − α) + (α − σo) − xv(σy − σo). If we could observethe selection of criminals x j into the different types of movies,we could estimate the differential direct effect of violent movies(the parameter captured in the laboratory experiments) as

αv − α = βv −[βn + xv − xn

xm − xn (βm − βn)]

.(5)

The solution for αv − α is the difference between the actual impactof strongly violent movies (βv) and the predicted impact basedon selection (the term in square brackets). If strongly violentmovies trigger additional aggression due to arousal or imitation(αv − α > 0), the impact of strongly violent movies βv can be lessnegative than mildly violent movies βm. In Section IV.A, we pro-vide an estimate of αv − α under the assumptions outlined above.

Finally, although we have emphasized the impact of movieson potential criminals, we note that exposure to movies can alsohave a parallel effect on potential victims. During the durationof the movie, potential victims are likely to be protected fromcrime. After the movie, they may be more or less susceptible toassaults depending on whether their alternative activity wouldhave placed them in a more or less volatile situation (account-ing for any arousal or catharsis effects). Therefore, although we


cannot distinguish between effects on the supply side and on thedemand side of criminal activity, the interpretations of the resultsand the policy implications remain essentially unchanged. In fact,it is likely that any effect of movie attendance, such as a reduc-tion of alcohol consumption, would operate symmetrically on bothoffenders and victims.

II.C. Comparison of Lab to Field

Before continuing, a brief comparison to the psychology exper-iments is in order. There are three factors that differ between thelaboratory and the field. The first and most important is the com-parison group. In the experiments, exposure to violent and nonvi-olent movies is manipulated as part of the treatment, whereas inthe field, subjects optimally choose relative to a comparison activ-ity as. Hence, in the laboratory, the treatment effects are estimatedas the difference between the effect of violent versus nonviolentmovies. In contrast, the effect of exposure in the field is measuredas the difference between the effect of movie violence and the effectof the foregone alternative activity. The second factor is selection.Subjects in the laboratory are a representative sample of the (stu-dent) population, while moviegoers in the field are a self-selectedsample. The sorting of violent individuals into violent movies,which could result in large displacement effects in the field, is notpresent in the lab. Finally, the third factor is the type of violence.The clips used in the experiments typically consist of five to tenminutes of selected sequences of extreme violence. In the field,instead, media violence also includes meaningful acts of reconcil-iation, apprehension of criminals, and nonviolent sequences. Theexposure to random acts of violence may induce different effectsfrom the exposure to acts of violence viewed in a broader context.

Within our empirical specification, an estimate of βv in thelaboratory experiment yields

βvlab = Ny

Ny + Noαv

y +(

1 − Ny

Ny + No

)αv

o .

Comparing this estimate to the estimate obtained from field datain (4) makes apparent the first two differences discussed above.First, the impact of media violence in the lab does not includethe indirect effect of σ, which operates through the alternativeactivity. By virtue of experimental control, the indirect effect isshut down. Second, the weights on the young and old coefficients


are different (compare Ny/(Ny + No

)to xv). The laboratory exper-

iments capture the reaction to media violence of a representativesample, whereas the field evidence assigns more weight to theparameter of the individuals that sort into the violent movies(the “young”). Hence, the laboratory setting is not representativeof exposure to movie violence in most field settings, where con-sumers choose what media to watch. However, it is representativeof instances of unexpected exposure, as in the case of a violentadvertisement or a trailer placed within family programming.

Recognizing these differences is important not only to betterunderstand the effect of media on violence, but also more generallyto understand the relationship between experimental and fieldevidence (Levitt and List 2007). In our setting, the field findingsare important to evaluate policies that would restrict access toviolent movies, as such policies would lead to substitution towardalternative activities in the short run. The results of the laboratoryexperiments, however, are useful to evaluate different policies,such as the short-run impact of unexpected exposure to mediaviolence. In addition, some of the differences between laboratoryand field can be altered by changes in the laboratory design. Forinstance, the laboratory experiments can incorporate sorting intoa violent movie (Lazear, Malmendier, and Weber 2006) to replicatethe selection in the field, or can change the exposure to a full-length movie.

One important limitation of both the laboratory and field de-signs is that neither provides evidence on the long-term effectsof repeated exposure to violent media. These cumulative effectscould be substantial, yet they are difficult to estimate causally.

III. DATA

In this section we introduce our various data sets, providesummary statistics, and describe general patterns of movie atten-dance and violent crime.

III.A. Movie Data

Data on box-office revenue are from the-numbers.com, whichuses the studios and Exhibitor Relations as data sources. Infor-mation on total weekend box-office sales is available for the topfifty movies consistently from January 1995 on. Daily revenueis available for the top ten movies beginning mid-August 1997.We focus on daily data for Friday, Saturday, and Sunday because


movie attendance, and therefore the identifying variation for ouranalysis, is concentrated on weekends (see Table I). To estimatemovie theater attendance, we deflate both the weekend and thedaily box-office sales by the average price of a ticket. For the pe-riod January 1995 to mid-August 1997 and for all movies that donot make the daily top-ten list, we impute daily box-office revenue(see Appendix I).

We match the box-office data to violence ratings fromkids-in-mind.com, a site recognized by Time Magazine in 2006 asone of the “Fifty Coolest Websites.” Since 1992, this nonprofit or-ganization has assigned a 0- to 10-point violence rating to almostall movies with substantial sales. The ratings are performed bytrained volunteers who, after watching a movie, follow guidelinesto assign a violence rating. In Table A.1, we illustrate the ratingsystem by listing the three movies with the highest weekend au-diences within each rating category. For most of the analysis, wegroup movies into three categories: strongly violent, mildly vio-lent, and nonviolent. Movies with ratings between 0 and 4 suchas Toy Story and Runaway Bride have very little violence; theirMPAA ratings range from G to R (for sexual content or profanity).Movies with ratings between 5 and 7 contain a fair amount ofviolence, with some variability across titles (Spider-Man versusMummy Returns). These movies are typically rated PG-13 or R.Movies with a rating of 8 and above are violent and almost uni-formly rated R, and are disproportionately more likely to be in the“Action/Adventure” and “Horror” genres. Examples are Hannibaland Saving Private Ryan. For a very small number of movies,typically with limited audiences, a rating is not available.

We define the number of people (in millions) exposed to moviesof violence level k on day t as Ak

t = ∑j∈J djεkaj,t, where aj,t is

the audience of movie j on day t, djεk is an indicator for film jbelonging to violence level k, and J is the set of all movies. Theviolence level varies between 0 and 10.1 We define three summarymeasures for movies with differing levels of violence. The measureof exposure to strongly violent movies on day t is the audiencefor movies with violence levels between 8 and 10, Av

t = ∑10k=8 Ak

t .

1. The rereleases of Star Wars V and VI in 1997 were not rated because theoriginal movie predates kids-in-mind.com. We assigned them the violence rating 5,the same rating as for the other Star Wars movies. To deal with the small numberof remaining movies with missing violence ratings, we assume ratings are missingat random with respect to the level of violence in a movie, and inflate each day’sexposure variables Ak

t accordingly. The average share of missing ratings is 4.1%across days.


TA

BL

EI

SU

MM

AR

YS

TA

TIS

TIC

S

Ass

ault

s(p

erda

y)

En

tire

day

6A.M

.to

6P.

M.

6P.

M.t

o12

A.M

.12

A.M

.to

6A.M

.(1

)(2

)(3

)(4

)

Ass

ault

data

for

alld

ays

Wee

ken

d(F

rida

y–S

un

day)

1,45

456

953

135

4F

rida

y1,

589

614

543

432

Sat

urd

ay1,

564

557

558

449

Su

nda

y1,

209

536

491

182

Wee

kday

(Mon

day–

Th

urs

day)

1,29

360

848

020

5

Sh

are

ofw

eeke

nd

assa

ult

sin

each

cate

gory

Ass

ault

data

for

wee

ken

ds(F

rida

y–S

un

day)

By

gen

der

ofof

fen

der

Sh

are

wit

hm

ale

offe

nde

r0.

779

0.75

50.

784

0.81

1B

yag

eof

offe

nde

rS

har

ew

ith

offe

nde

rof

ages

18to

290.

378

0.34

00.

359

0.46

7A

lcoh

ol-r

elat

edas

sau

lts

Sh

are

wit

hof

fen

der

susp

ecte

dof

usi

ng

alc.

ordr

ugs

0.17

00.

082

0.18

50.

290

Sh

are

wit

hof

fen

der

ofag

es17

to20

(un

dera

ge)

0.13

30.

125

0.13

90.

138

Sh

are

wit

hof

fen

der

ofag

es21

to24

(ove

r-ag

e)0.

135

0.11

80.

123

0.18

2N

um

ber

ofob

serv

atio

ns

N=

1,56

3da

ys,2

,272

,999

assa

ult

s,1,

781

agen

cies


TA

BL

EI

( CO

NT

INU

ED

) Mov

ieau

dien

ce(m

illi

ons

ofti

cket

sor

ren

tals

per

day)

Th

eate

rau

dien

ceV

HS

/DV

Dre

nta

ls(5

)(6

)

Mov

ieau

dien

ceda

tafo

ral

lday

sW

eeke

nd

(Fri

day–

Su

nda

y)6.

293.

92F

rida

y5.

744.

13S

atu

rday

7.90

4.82

Su

nda

y5.

242.

82W

eekd

ay(M

onda

y–T

hu

rsda

y)2.

002.

09M

ovie

audi

ence

data

for

wee

ken

ds(F

rida

y–S

un

day)

By

kids

-in

-min

d.co

mra

tin

gS

tron

gly

viol

ent

mov

ies

0.87

0.64

Mil

dly

viol

ent

mov

ies

2.43

1.56

Non

viol

ent

mov

ies

2.99

1.72

Not

es.A

nob

serv

atio

nis

ada

yov

erth

eye

ars

1995

–200

4.A

ssau

ltda

taco

me

from

the

Nat

ion

alIn

cide

nt

Bas

edR

epor

tin

gS

yste

m(N

IBR

S),

and

the

sam

ple

incl

ude

sag

enci

esth

atdo

not

hav

em

issi

ng

data

onan

ycr

ime

(not

just

assa

ult

s)fo

rm

ore

than

seve

nco

nse

cuti

veda

ysfo

rth

atye

ar.T

he

mov

ieau

dien

cen

um

bers

are

obta

ined

from

the-

nu

mbe

rs.c

oman

dar

eda

ily

box-

offi

cere

ven

ue

divi

ded

byth

eav

erag

epr

ice

per

tick

et.T

he

rati

ngs

ofvi

olen

tm

ovie

sar

efr

omki

ds-i

n-m

ind.

com

.Th

eau

dien

ceof

mil

dly

viol

ent

mov

ies

isth

eau

dien

ceof

allm

ovie

sw

ith

avi

olen

cera

tin

g5–

7.T

he

audi

ence

ofst

ron

gly

viol

ent

mov

ies

isth

eau

dien

ceof

allm

ovie

sw

ith

avi

olen

cera

tin

g8–

10.V

HS

/DV

Dre

nta

ldat

aco

me

from

Vid

eoS

tore

Mag

azin

e.


Similarly, exposure to mildly violent Amt and nonviolent An

t movieson day t are defined as the aggregated audiences for movies witha violence level between 5–7 and 0–4, respectively.

Figure Ia plots the measure of strong movie violence, Avt , over

the sample period 1995 to 2004. To improve readability, we plotthe weekend audience (the sum from Friday to Sunday) insteadof the daily audience. In the graph, we label the top ten weekendswith the name of the movie responsible for the spike. The seriesexhibits sharp fluctuations. Several weekends have close to zeroviolent movie audience. On other weekends, over twelve millionpeople watch violent movies. The spikes in the violent movie seriesare distributed fairly uniformly across the years, and decay withintwo to three weeks of the release of a violent blockbuster.

Figure Ib plots the corresponding information for the measureof mild movie violence, Am

t . Because more movies are included inthis category, the average weekend audience for mildly violentmovies is higher than for strongly violent movies, with peaks ofup to 25 million people. There is some seasonality in the releaseof violent movies, with generally lower exposure to movie violencebetween February and May. This seasonality is less pronouncedfor the strongly violent movies compared to the mildly violentmovies.

To put audience size into perspective, note that blockbustermovies are viewed by a sizable fraction of the U.S. population.Over a weekend, strongly violent and mildly violent blockbustersattract up to 4% and 8%, respectively, of the U.S. population(roughly 300 million). This extensive exposure provides the iden-tifying variation in our setup.

III.B. Violent Crime Data

Our source for violent crime data is the NIBRS, chosen for twoimportant features. First, it reports violent acts known to police,such as verbal intimidation or fistfights, which do not necessarilyresult in an arrest. Second, it reports the date and time of thecrime, allowing us to match movie attendance and violent crime atthe daily level. Alternative large-scale data sets on crime, such asthe Uniform Crime Report and the National Crime VictimizationSurvey, do not contain this same type of detailed information atthe daily level.

The NIBRS data collection effort is a part of the UniformCrime Reporting Program. Submission of NIBRS data is still


voluntary, and over time the number of reporting agencies hasincreased substantially. In 1995 (the first year of NIBRS data),only 4% of the U.S. population was covered, but by August 2005,there were 29 states certified to report NIBRS data to the FBI,for a coverage rate of 22% of the U.S. population (reporting is notalways 100% within a state). This 22% coverage represents 17%of the nation’s reported crime, which reflects the fact that NIBRScoverage is more heavily weighted toward smaller cities and coun-ties (where crime rates are lower). One limitation of NIBRS is thatit does not cover crime in the nation’s largest cities, although itdoes include medium-size cities such as Memphis and Cincinnati.

We use data from 1995 to 2004 for NIBRS city and countyreporting agencies, which include local police forces and countysheriff offices. Because not all agencies report consistently, in eachyear we exclude agencies that have missing data on crime (not justassaults) for more than seven consecutive days, where a report ofzero counts as nonmissing data. This filter eliminates 12.5% ofreported assaults. If no crime is reported on a given day afterthis filter, we set that day’s assault count to zero. Our main vio-lence measure is the total daily number of assaults, Vt, defined asthe sum of aggravated assault, simple assault, and intimidation,2

across all agencies on day t. In some specifications, we separateassaults into four time blocks: 6 A.M.–12 P.M., 12 P.M.–6 P.M., 6 P.M.–12 A.M., and 12 A.M.–6 A.M. We assign assaults occurring between12 A.M. and 6 A.M. to the previous calendar day to match them tomovies played the previous evening.

To provide graphical evidence on this series, we construct theresidual of log daily assaults, after controlling for an extensive setof indicator variables for year, month, day-of-week, day-of-year,and holidays as well as weather and TV audience measures (thesame set of variables used in our main specification and describedin Appendix I). Figure Ic plots the average of the Friday to Sundayresiduals (the days with highest movie audience) over time. Theresiduals behave approximately like white noise. Only 44 week-ends differ from the mean by more than 0.05 log points, and justone differs by more than 0.10 log points.

2. Aggravated assault is an unlawful attack by one person upon anotherwherein the offender uses a weapon or displays it in a threatening manner, orthe victim suffers obvious severe or aggravated injury. Simple assault is also anunlawful attack but does not involve a weapon or obvious severe or aggravatedbodily injury. Intimidation is placing a person in reasonable fear of bodily harmwithout a weapon or physical attack.


The figure also labels the top ten weekends for the audienceof strongly violent (see Figure Ia) and mildly violent movies (seeFigure Ib). Interestingly, Figure Ic offers an indication of a neg-ative relationship between violent movies and crime. For bothmildly violent and strongly violent movies, seven of the top tenweekends have residuals below the median. (One of the positiveresiduals is for Passion of the Christ, an atypical violent movie,both for its target audience and its potential effect on crime.) Inaddition, out of twenty weekends with a residual more negativethan −0.05 log points, two are among the top ten weekends forstrongly violent movies, and two are among the top ten weekendsfor mildly violent movies. We examine the relationship betweenviolent movies and violent crime in detail in the next section.

III.C. Summary Statistics

After matching the movie and crime data, the resulting dataset includes 1,563 weekend (Friday through Sunday) observa-tions, covering the time period from January 1995 to December2004. The data set contains a total of 2,272,999 assaults and 1,781reporting agencies. Table I reports summary statistics. The aver-age number of assaults on any given weekend day is 1,454. Theassaults occur mostly in the evening (6 P.M.–12 A.M.), but are alsocommon in the afternoon (12 P.M.–6 P.M.) and in the night (12 A.M.–6 A.M.). Assaults are highest on Friday and Saturday, and loweron Sundays and other weekdays. Assaults are three times largerfor males than for females, and are decreasing in the age of theoffender (for ages above 18). The share of assaults where the of-fender is suspected of using alcohol or drugs is 17.0% over thewhole day, with a much larger incidence in the night hours.

Table I also reports summary statistics for movie attendance.The average daily movie audience on a weekend day is 6.29 mil-lion people, with a peak on Saturday. The audience for stronglyand mildly violent movies is respectively 0.87 million and 2.43million. The table also presents information on VHS and DVDmovie rentals.

IV. EMPIRICAL RESULTS

IV.A. Theater Audience—Daily

To test for the short-run effects of exposure to violent movies,we focus on same-day exposure, a short time horizon similar tothe one considered in the psychology experiments. The outcome


variable of interest is Vt, the number of assaults on day t. Althoughthe number of assaults is a count variable, specifying explicitly thecount process (as in a Poisson regression) is not key because thenumber of daily assaults is sufficiently large. Hence, we adopt anOLS specification, which allows us to more easily instrument formovie exposure later in the paper. The benchmark specificationthat follows from the model developed in Section II is

ln Vt = βv Avt + βmAm

t + βnAnt + �Xt + εt.(6)

The number of assaults depends on the exposure to stronglyviolent movies Av

t , mildly violent movies Amt , and nonviolent

movies Ant . The coefficient βv can be interpreted as the percent

increase in assaults for each million people watching stronglyviolent movies on day t, with a similar interpretation for the co-efficients βm and βn. Identification of the parameters relies ontime-series variation in the violence content of movies at the the-ater (see Figures Ia and Ib). By comparing the estimates of βv andβm to the estimate of βn, one can obtain a difference-in-differenceestimate of the effect of violent movies versus nonviolent movies.

The variables Xt are a set of seasonal control variables: indica-tors for year, month, day-of-week, day-of-year, holidays, weather,and TV audience. Because new movie releases and movie atten-dance are concentrated on weekends, we restrict the sample toFriday, Saturday, and Sunday. All standard errors are robust andclustered by week, to allow for arbitrary correlation of errorsacross the three observations on the same weekend.

In column (1) of Table II we begin by estimating equation (6)with only year controls included. The year controls are necessarybecause the cities and counties in the sample vary year-to-year. Inthis specification, exposure to media violence appears to increasecrime. However, we also obtain the puzzling result that exposureto nonviolent movies increases crime significantly, suggesting thatat least part of this correlation is due to omitted variables. Einav(2007) documents seasonality in movie release dates and under-lying demand, with the biggest ticket sales in the beginning ofthe summer and during holidays. Because assaults are also ele-vated during summers and holidays, it is important to control forseasonal factors. In columns (2) and (3), we include indicators formonth-of-year and for day-of-week. Although introducing thesecoarse seasonal variables increases the R2 substantially, from.9344 to .9846, these variables do not control for additional effects

DOES MOVIE VIOLENCE INCREASE VIOLENT CRIME? 697T

AB

LE

IIE

FF

EC

TO

FM

OV

IEV

IOL

EN

CE

ON

SA

ME-D

AY

AS

SA

ULT

S

Spe

cifi

cati

on:

OL

Sre

gres

sion

sIV

regr

essi

ons

Dep

.var

.:L

og(n

um

ber

ofas

sau

lts

inda

yt)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

Au

dien

ceof

stro

ngl

yvi

olen

tm

ovie

s0.

0324

0.00

05−0

.006

1−0

.005

1−0

.007

2−0

.009

1−0

.010

6(m

illi

ons

ofpe

ople

inda

yt)

(0.0

053)

∗∗∗

(0.0

053)

(0.0

033)

∗(0

.003

3)(0

.003

3)∗∗

(0.0

026)

∗∗∗

(0.0

031)

∗∗∗

Au

dien

ceof

mil

dly

viol

ent

mov

ies

0.02

460.

0017

−0.0

084

−0.0

042

−0.0

056

−0.0

079

−0.0

102

(mil

lion

sof

peop

lein

day

t)(0

.003

0)∗∗

∗(0

.002

9)(0

.002

0)∗∗

∗(0

.002

6)(0

.002

7)∗∗

(0.0

022)

∗∗∗

(0.0

028)

∗∗∗

Au

dien

ceof

non

viol

ent

mov

ies

0.00

82−0

.016

4−0

.006

2−0

.002

3−0

.002

9−0

.003

5−0

.005

0(m

illi

ons

ofpe

ople

inda

yt)

(0.0

029)

∗∗∗

(0.0

030)

∗∗∗

(0.0

021)

∗∗∗

(0.0

024)

(0.0

026)

(0.0

024)

(0.0

029)

∗

Con

trol

vari

able

sYe

arin

dica

tors

XX

XX

XX

XD

ay-o

f-w

eek

indi

cato

rsX

XX

XX

XM

onth

indi

cato

rsX

XX

XX

Day

-of-

year

indi

cato

rsX

XX

XH

olid

ayin

dica

tors

XX

XW

eath

eran

dT

Vau

dien

ceco

ntr

ols

XX

F-t

est

onad

diti

onal

con

trol

s1,

934.

021,

334.

3188

.56

13.3

715

.05

18.5

8A

udi

ence

inst

rum

ente

dw

ith

pred

icte

dau

dien

ceu

sin

gn

ext

wee

ken

d’s

audi

ence

XR

20.

9344

0.97

110.

9846

0.99

040.

9912

0.99

31N

1,56

31,

563

1,56

31,

563

1,56

31,

563

1,56

3

Not

es.

An

obse

rvat

ion

isa

Fri

day,

Sat

urd

ay,

orS

un

day

over

the

year

s19

95–2

004.

Ass

ault

data

com

efr

omth

eN

atio

nal

Inci

den

tB

ased

Rep

orti

ng

Sys

tem

(NIB

RS

),w

her

eth

esa

mpl

ein

clu

des

agen

cies

that

don

oth

ave

mis

sin

gda

taon

any

crim

e(n

otju

stas

sau

lts)

for

mor

eth

anse

ven

con

secu

tive

days

for

that

year

.Th

em

ovie

audi

ence

nu

mbe

rsar

eob

tain

edfr

omth

e-n

um

bers

.com

and

are

dail

ybo

x-of

fice

reve

nu

edi

vide

dby

the

aver

age

pric

epe

rti

cket

.T

he

rati

ngs

ofvi

olen

tm

ovie

sar

efr

omki

ds-i

n-m

ind.

com

.T

he

audi

ence

ofst

ron

gly

viol

ent

mov

ies

isth

eau

dien

ceof

allm

ovie

sw

ith

avi

olen

cera

tin

g8–

10.T

he

audi

ence

ofm

ildl

yvi

olen

tm

ovie

sis

the

audi

ence

ofal

lmov

ies

wit

ha

viol

ence

rati

ng

5–7.

Th

esp

ecifi

cati

ons

inco

lum

ns

(1)t

hro

ugh

(6)a

reO

LS

regr

essi

ons

wit

hth

elo

g(n

um

ber

ofas

sau

lts

occu

rrin

gin

day

t)as

the

depe

nde

nt

vari

able

.Th

esp

ecifi

cati

onin

colu

mn

(7)i

nst

rum

ents

the

audi

ence

nu

mbe

rsw

ith

the

pred

icte

dau

dien

cen

um

bers

base

don

nex

tw

eeke

nd’

sau

dien

ce.

Det

ails

onth

eco

nst

ruct

ion

ofth

epr

edic

ted

audi

ence

nu

mbe

rsar

ein

the

text

.R

obu

stst

anda

rder

rors

clu

ster

edby

wee

kar

ein

pare

nth

eses

.*

Sig

nifi

can

tat

10%

;**

sign

ifica

nt

at5%

;***

sign

ifica

nt

at1%

.


such as the Christmas season in the second half of December orfor holidays such as Independence Day. In columns (4) and (5), wetherefore add 365 day-of-year indicators (dropping February 29 inleap years) and holiday indicators (see Appendix I), raising the R2

further to .9912. As we add these variables, the coefficients βv andβm on the violent movie measures flip sign and become negative,significantly so in column (5). This suggests that the seasonalityin movie releases and in crime biases the estimates upward.

This negative correlation, however, may still be due to anunobserved variable that contemporaneously increases violentmovie attendance and decreases violence εt. For example, on rainydays assaults are lower, but movie attendance is higher. To ad-dress this possibility, we use two strategies. First, we add a set ofweather controls to account for hot and cold temperatures, humid-ity, high winds, snow, and rain. We also control for distractors thatcould affect both crime and movie attendance by controlling forthe day of the Super Bowl and for the other days with TV showshaving an audience in excess of fifteen million households ac-cording to Nielsen Media Research. (These controls are describedin Appendix I.) Adding these controls makes the estimates morenegative (column (6)).

Second, we instrument for movie audience on day t usinginformation on the following weekend’s audience for the samemovie. This instrumental variable strategy exploits the pre-dictability of the weekly decrease in attendance. At the sametime, it removes the effect of any shocks that affect violence andattendance in week w(t), but are not present in week w (t) + 1.Examples include one-time TV events or transient weathershocks that are not already captured in our TV and weathercontrols. This procedure, detailed in Appendix II, generatespredictors for the audience of strongly violent, mildly violent,and nonviolent movies on day t. Panel B in Table III shows thatthese predictors are strongly correlated with the actual audiencenumbers they are instrumenting for. In the first stage for theaudience of strongly violent movies (column (1)), the coefficienton the predicted audience for strongly violent movies is highlysignificant and close to 1 (.9145), as predicted. The other two co-efficients in this regression are close to 0, though also significant.We obtain similar first stages for the audience of mildly violentmovies (column (2)) and nonviolent movies (column (3)).

Column (7) in Table II presents the IV estimates, where wehave instrumented for the movie audience variables with their


TA

BL

EII

IE

FF

EC

TO

FM

OV

IEV

IOL

EN

CE

ON

SA

ME-D

AY

AS

SA

ULT

SB

YT

IME

OF

DA

Y

A.B

ench

mar

kre

sult

sS

peci

fica

tion

:In

stru

men

talv

aria

ble

regr

essi

ons

Dep

.var

.:L

og(n

um

ber

ofas

sau

lts

inda

yt

inti

me

win

dow

)(1

)(2

)(3

)(4

)

Au

dien

ceof

stro

ngl

yvi

olen

tm

ovie

s−0

.005

0−0

.003

0−0

.013

0−0

.019

2(m

illi

ons

ofpe

ople

inda

yt)

(0.0

066)

(0.0

050)

(0.0

049)

∗∗∗

(0.0

060)

∗∗∗

Au

dien

ceof

mil

dly

viol

ent

mov

ies

−0.0

106

−0.0

001

−0.0

109

−0.0

205

(mil

lion

sof

peop

lein

day

t)(0

.006

0)∗

(0.0

045)

(0.0

040)

∗∗∗

(0.0

052)

∗∗∗

Au

dien

ceof

non

viol

ent

mov

ies

−0.0

033

0.00

16−0

.006

3−0

.006

0(m

illi

ons

ofpe

ople

inda

yt)

(0.0

060)

(0.0

046)

(0.0

043)

(0.0

054)

Tim

eof

day

6A.M

.–12

P.M

.12

P.M

.–6

P.M

.6

P.M

.–12

A.M

.12

A.M

.–6

A.M

.n

ext

day

Con

trol

vari

able

sF

ull

set

ofco

ntr

ols

XX

XX

Au

dien

cein

stru

men

ted

wit

hpr

edic

ted

audi

ence

usi

ng

nex

tw

eek’

sau

dien

ceX

XX

XN

1,56

31,

563

1,56

31,

562


TA

BL

EII

I( C

ON

TIN

UE

D)

B.F

irst

stag

eS

peci

fica

tion

:IV

regr

essi

on,fi

rst

stag

e

Dep

.var

.:A

udi

ence

ofst

ron

gly

Au

dien

ceof

mil

dly

Au

dien

ceof

viol

ent

mov

ies

viol

ent

mov

ies

non

viol

ent

mov

ies

(1)

(2)

(3)

Pre

d.au

dien

ceof

stro

ngl

yvi

olen

tm

ovie

s0.

9145

−0.1

431

−0.1

694

(mil

lion

sof

peop

lein

day

t)(0

.019

6)∗∗

∗(0

.021

0)∗∗

∗(0

.028

1)∗∗

∗P

red.

audi

ence

ofm

ildl

yvi

olen

tm

ovie

s−0

.039

90.

8532

−0.1

817

(mil

lion

sof

peop

lein

day

t)(0

.010

1)∗∗

∗(0

.025

5)∗∗

∗(0

.029

6)∗∗

∗P

red.

audi

ence

ofn

onvi

olen

tm

ovie

s−0

.048

0−0

.136

30.

8138

(mil

lion

sof

peop

lein

day

t)(0

.009

7)∗∗

∗(0

.019

5)∗∗

∗(0

.030

9)∗∗

∗C

ontr

olva

riab

les

Fu

llse

tof

con

trol

sX

XX

F-t

est

onin

stru

men

ts1,

050.

8988

9.02

730.

85N

1,56

31,

563

1,56

3

Not

es.

See

not

esto

Tab

leII

.T

he

nu

mbe

rof

obse

rvat

ion

sin

colu

mn

(4)

ofP

anel

Ais

one

few

erth

anin

colu

mn

s(1

)–(3

)of

Pan

elA

beca

use

we

are

mis

sin

gth

eas

sau

ltda

tafo

rJa

nu

ary

1,20

06,f

orth

eh

ours

betw

een

12A.M

.an

d6

A.M

.*

Sig

nifi

can

tat

10%

;**

sign

ifica

nt

at5%

;***

sign

ifica

nt

at1%

.


predicted values. Instrumenting makes the correlation betweenmovie violence and violent crime become more negative. An in-crease of one million in the audience for violent movies decreasesviolent crime by 1.06% (strongly violent movies) and 1.02% (mildlyviolent movies), substantial effects on violence. Nonviolent movieshave a smaller (marginally significant) negative effect on assaults.The IV estimates do not noticeably change if the weather controlsare excluded (not reported), suggesting that the instruments aretaking care of temporary shocks, such as those due to weather.

IV.B. Theater Audience—Time of Day

Table II implies that exposure to violent movies diminishescrime in the short run. To clarify this potentially puzzling result(relative to the findings in the laboratory experiments), we sep-arately examine the effect of violent movies on violent crime bytime of day. In these and all subsequent specifications, we includethe full set of controls Xt and instrument for the actual audiencesAv

t , Amt , and An

t using the predicted audiences.In Table III, we present our baseline estimates by time of

day: assaults committed in the morning (6 A.M.–12 P.M.), afternoon(12 P.M.–6 P.M.), evening (6 P.M.–12 A.M.), and nighttime (12 A.M.–6 A.M.). Because movie audiences are unlikely to watch movies inthe morning and in the afternoon, and especially so for violentmovies, we expect to find little or no effect of exposure to vio-lent movies in the first two time blocks. There are small negativeeffects for assaults in the morning hours which are not very signif-icant. This appears to be due to a spillover from the previous day’smovie exposure (which is highly correlated with today’s movie ex-posure). Exposure to violent movies has no differential impact onassaults in the afternoon (column (2)). Because we consistentlyfind similar effects for these two time periods (small negative ef-fects in the early morning and no effect in the afternoon), we poolthem in subsequent tables to save space.

During the evening hours (column (3)), we find, instead, a sig-nificant negative effect of exposure to violent movies. An increasein the audience of mildly violent movies of one million decreasesviolent crime by 1.09%. Exposure to strongly violent movies hasa slightly larger effect. Exposure of one million additional peo-ple reduces assaults by 1.30%. Exposure to nonviolent movies isnegatively correlated with violent crime, but the point estimateis smaller than for violent movies, and not significant. Over thenight hours following exposure to a movie (column (4)), violent


movies have an even stronger negative impact on violent crime.Exposure to mildly and strongly violent movies for one millionpeople decreases violent crimes by, respectively, 2.05% and 1.92%.The impact of nonviolent movies is also negative but substantiallysmaller and not significantly different from 0.

To put these estimates into perspective, on an unseasonablycold day (20–32 degrees Fahrenheit) assaults go down by 11%in the evening hours and 8% in the night hours.3 In compari-son, the blockbuster strongly violent movie Hannibal (with anaudience size of 10.1 million on opening weekend) is predicted toaccount for a 4.4% reduction in assaults in the evening hours anda 6.5% reduction in the night hours (see footnote 14 for details onthis calculation). In Section V, we provide interpretations of thesefindings.

IV.C. Theater Audience—Timing of Effects

So far, we have estimated the impact of exposure to movieviolence on same-day violent crimes. We now estimate whetherthere is a delayed impact at various time intervals. If violentmovies increase violent crime in the medium run, or if they leadto intertemporal substitution of crime (as in the case of weathershocks in Jacob, Lefgren, and Moretti [2007]), violent crime islikely to be higher in the period following movie exposure.

Monday and Tuesday. In columns (1) and (2) of Table IV,we estimate the impact of average weekend movie audience onviolent crime for the Monday and Tuesday following the weekend.Because the movie audience on these weekdays is limited, to afirst approximation this specification captures the delayed effectof movie exposure one to three days later. We find no evidence ofan increase in violent crime due to either imitation or intertempo-ral substitution. Most coefficients are close to zero, and the onlymarginally significant coefficient indicates a delayed negative im-pact of mildly violent movies.

One Week, Two Weeks, and Three Weeks Later. In the fol-lowing specifications, we estimate the impact one, two, and threeweeks after the original exposure, controlling for contemporane-ous exposure. Separate identification is made possible by newreleases occurring after the initial exposure. Lagged movie at-tendance is instrumented using a similar methodology as for the

3. These are coefficients from the baseline IV regression, with 33–79 degreesFahrenheit as the omitted category.


AB

LE

IVM

ED

IUM

-RU

NE

FF

EC

TO

FM

OV

IEV

IOL

EN

CE

Spe

cifi

cati

on:

OL

Sre

gres

sion

s

Tim

ing:

Nex

tM

onda

yan

dT

ues

day

Nex

tw

eek

Tw

ow

eeks

late

rT

hre

ew

eeks

late

r

Dep

.var

.:L

og(n

um

ber

ofas

sau

lts

onM

onda

yan

dT

ues

day

inti

me

win

dow

)L

og(n

um

ber

ofas

sau

lts

inda

yt

inti

me

win

dow

)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Au

dien

ceof

stro

ngl

y−0

.012

7−0

.008

1−0

.014

2−0

.020

9−0

.013

6−0

.019

9vi

olen

tm

ovie

s(m

illi

ons

(0.0

045)

∗∗∗

(0.0

060)

(0.0

051)

∗∗∗

(0.0

067)

∗∗∗

(0.0

051)

∗∗∗

(0.0

063)

∗∗∗

ofpe

ople

inda

yt)

Au

dien

ceof

mil

dly

−0.0

061

−0.0

087

−0.0

096

−0.0

194

−0.0

114

−0.0

199

viol

ent

mov

ies

(mil

lion

s(0

.003

1)∗∗

(0.0

043)

∗∗(0

.004

2)∗∗

(0.0

056)

∗∗∗

(0.0

041)

∗∗∗

(0.0

052)

∗∗∗

ofpe

ople

inda

yt)

Au

dien

ceof

non

viol

ent

−0.0

027

0.00

30−0

.005

0−0

.007

9−0

.007

0−0

.007

6m

ovie

s(m

illi

ons

(0.0

033)

(0.0

050)

(0.0

046)

(0.0

061)

(0.0

044)

(0.0

056)

ofpe

ople

inda

yt)

Lag

ged

audi

ence

ofst

ron

gly

0.00

19−0

.000

40.

0046

−0.0

017

−0.0

028

0.00

200.

0017

−0.0

065

viol

ent

mov

ies

(mil

lion

s(0

.005

8)(0

.008

7)(0

.004

2)(0

.005

4)(0

.004

7)(0

.006

2)(0

.004

4)(0

.005

6)of

peop

lein

day

t)L

agge

dau

dien

ceof

mil

dly

−0.0

07−0

.014

6−0

.001

80.

0001

−0.0

061

−0.0

056

0.00

02−0

.010

5vi

olen

tm

ovie

s(m

illi

ons

(0.0

050)

(0.0

076)

∗(0

.002

6)(0

.003

7)(0

.003

7)(0

.004

9)(0

.003

1)(0

.004

5)∗∗

ofpe

ople

inda

yt)

Lag

ged

audi

ence

of0.

0012

−0.0

065

−0.0

007

0.00

31−0

.006

00.

0012

0.00

11−0

.004

9n

onvi

olen

tm

ovie

s(m

illi

ons

(0.0

054)

(0.0

074)

(0.0

028)

(0.0

041)

(0.0

042)

(0.0

055)

(0.0

036)

(0.0

048)

ofpe

ople

inda

yt)


TA

BL

EIV

( CO

NT

INU

ED

)

Spe

cifi

cati

on:

OL

Sre

gres

sion

s

Tim

ing:

Nex

tM

onda

yan

dT

ues

day

Nex

tw

eek

Tw

ow

eeks

late

rT

hre

ew

eeks

late

r

Dep

.var

.:L

og(n

um

ber

ofas

sau

lts

onM

onda

yan

dT

ues

day

inti

me

win

dow

)L

og(n

um

ber

ofas

sau

lts

inda

yt

inti

me

win

dow

)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Lag

spec

ifica

tion

Lag

:wee

ken

dbe

fore

Lag

:7da

ysbe

fore

Lag

:14

days

befo

reL

ag:2

1da

ysbe

fore

Tim

eof

day

6P.

M.–

12A.M

.–6

P.M

.–12

A.M

.–6

P.M

.–12

A.M

.–6

P.M

.–12

A.M

.–12

A.M

.6

A.M

.12

A.M

.6

A.M

.12

A.M

.6

A.M

.12

A.M

.6

A.M

.n

ext

day

nex

tda

yn

ext

day

nex

tda

yC

ontr

olva

riab

les

Fu

llse

tof

con

trol

sX

XX

XX

XX

XA

udi

ence

inst

rum

ente

dw

ith

XX

No

No

XX

XX

pred

icte

dau

dien

ceu

sin

gfo

llow

ing

wee

k’s

audi

ence

N1,

041

1,04

11,

559

1,55

81,

556

1,55

51,

553

1,55

2

Not

es.S

een

otes

toT

able

II.T

he

spec

ifica

tion

sar

eIV

regr

essi

ons

wit

hth

elo

g(n

um

ber

ofas

sau

lts

occu

rrin

gin

day

t)as

the

depe

nde

nt

vari

able

.Th

esp

ecifi

cati

ons

inco

lum

ns

(3)

and

(4)

are

not

inst

rum

ente

d,be

cau

seth

epr

edic

tors

for

the

audi

ence

ofth

epr

evio

us

wee

kar

eh

igh

lyco

llin

ear

wit

hth

eco

nte

mpo

ran

eou

sau

dien

ce.

*Sig

nifi

can

tat

10%

;**s

ign

ifica

nt

at5%

;***

sign

ifica

nt

at1%

.


other movie attendance variables, except for the one-week lag(columns (3) and (4)). In this specification, we report the OLS re-sults, because the instrument for lagged exposure would be essen-tially collinear with contemporaneous exposure. Across the threespecifications (columns (3)–(8)), we find no evidence of a delayedeffect of movie exposure. Of eighteen coefficients for lagged ex-posure, only one is significant (negative) at the 5% level. At thesame time, we find strong evidence of a negative impact of con-temporaneous exposure to violent movies, as in our benchmarkspecifications. These results suggest that there is no medium-runeffect of exposure to movie violence due to either imitation or in-tertemporal substitution.

IV.D. Theater Audience—Robustness

Before discussing how to interpret the results, in Table V weassess the robustness of the benchmark estimates of Table III,reproduced in column (1).

In column (2), we use a different set of instruments for movieattendance—information on the production budget and the num-ber of theaters in which a movie is playing in week w (t) (seeAppendix II for details). Production budgets are decided far inadvance, whereas the number of screens is finalized one or twoweeks in advance (Moretti 2008). These instruments, like ourbaseline instruments, should purge the estimates of short-termshocks affecting both attendance and crime. We supplement theseinstruments with an additional instrument for total movie au-dience size, based on our standard procedure.4 The results areremarkably similar to the benchmark IV results.

Column (3) uses the standard instrument but includes allseven days of the week instead of just the weekend (column (3)).Many of the point estimates for the effect of movie violence in theevening and night (Panels B and C) become more negative, includ-ing the estimate for nonviolent movies, which is now significant.The latter finding may reflect an impact of nonviolent movies forthe same reasons as for violent movies (with smaller magnitudes),

4. We supplement with total movie audience size because the new instrumentsdo not predict overall movie audience well. This is because total number of theatersis essentially fixed in any given week, and production budgets do not provide muchidentifying variation. The joint F-tests for the first stages of this instrument setrange from 280 to 378, with most of the power coming from the variables for thenumber of theaters.

706 QUARTERLY JOURNAL OF ECONOMICST

AB

LE

VR

OB

US

TN

ES

S

Spe

cifi

cati

on:

Inst

rum

enta

lvar

iabl

esre

gres

sion

sO

LS

reg.

Poi

sson

reg.

Dep

.var

.:L

og(n

um

ber

ofvi

olen

tcr

imes

inda

yt

inti

me

win

dow

)N

o.of

assa

ult

s

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

A.E

ffec

tsin

mor

nin

gan

daf

tern

oon

(6A.M

.–6

P.M

.)A

udi

ence

ofst

ron

gly

−0.0

037

−0.0

046

0.00

050.

0005

−0.0

075

−0.0

047

−0.0

096

−0.0

081

viol

ent

mov

ies

(mil

lion

s(0

.004

6)(0

.004

5)(0

.003

9)(0

.003

7)(0

.005

6)(0

.004

4)(0

.003

5)∗∗

∗(0

.002

9)∗∗

∗

ofpe

ople

inda

yt)

Au

dien

ceof

mil

dly

−0.0

03−0

.004

6−0

.000

6−0

.000

6−0

.002

8−0

.003

−0.0

088

−0.0

102

viol

ent

mov

ies

(mil

lion

s(0

.004

1)(0

.004

2)(0

.003

3)(0

.003

3)(0

.003

9)(0

.004

0)(0

.002

7)∗∗

∗(0

.002

3)∗∗

∗

ofpe

ople

inda

yt)

Au

dien

ceof

0.00

03−0

.001

2−0

.001

2−0

.001

2−0

.001

30

−0.0

079

−0.0

098

non

viol

ent

mov

ies

(mil

lion

s(0

.004

1)(0

.004

2)(0

.003

5)(0

.003

4)(0

.004

4)(0

.003

9)(0

.002

8)∗∗

∗(0

.002

3)∗∗

∗

ofpe

ople

inda

yt)

B.E

ffec

tsin

the

even

ing

(6P.

M.–

12A.M

.)A

udi

ence

ofst

ron

gly

−0.0

13−0

.015

8−0

.014

4−0

.014

4−0

.013

9−0

.015

3−0

.009

9−0

.008

1vi

olen

tm

ovie

s(m

illi

ons

(0.0

049)

∗∗∗

(0.0

048)

∗∗∗

(0.0

046)

∗∗∗

(0.0

044)

∗∗∗

(0.0

063)

∗∗(0

.004

4)∗∗

∗(0

.003

7)∗∗

∗(0

.003

0)∗∗

∗

ofpe

ople

inda

yt)

Au

dien

ceof

mil

dly

−0.0

109

−0.0

107

−0.0

165

−0.0

165

−0.0

109

−0.0

119

−0.0

065

−0.0

075

viol

ent

mov

ies

(mil

lion

s(0

.004

0)∗∗

∗(0

.004

2)∗∗

(0.0

035)

∗∗∗

(0.0

032)

∗∗∗

(0.0

039)

∗∗∗

(0.0

038)

∗∗∗

(0.0

029)

∗∗(0

.002

3)∗∗

∗

ofpe

ople

inda

yt)

Au

dien

ceof

−0.0

063

−0.0

062

−0.0

098

−0.0

098

−0.0

08−0

.006

9−0

.002

6−0

.003

non

viol

ent

mov

ies

(mil

lion

s(0

.004

3)(0

.004

4)(0

.004

0)∗∗

(0.0

036)

∗∗∗

(0.0

042)

∗(0

.004

0)∗

(0.0

030)

(0.0

024)

ofpe

ople

inda

yt)

C.E

ffec

tsin

the

nig

ht

(12

A.M

.–6

A.M

.)A

udi

ence

ofst

ron

gly

−0.0

192

−0.0

202

−0.0

206

−0.0

206

−0.0

252

−0.0

211

−0.0

098

−0.0

133

viol

ent

mov

ies

(mil

lion

s(0

.006

0)∗∗

∗(0

.005

9)∗∗

∗(0

.005

4)∗∗

∗(0

.005

5)∗∗

∗(0

.006

8)∗∗

∗(0

.006

6)∗∗

∗(0

.005

2)∗

(0.0

035)

∗∗∗

ofpe

ople

inda

yt)


AB

LE

V(C

ON

TIN

UE

D)

Spe

cifi

cati

on:

Inst

rum

enta

lvar

iabl

esre

gres

sion

sO

LS

reg.

Poi

sson

reg.

Dep

.var

.:L

og(n

um

ber

ofvi

olen

tcr

imes

inda

yt

inti

me

win

dow

)N

o.of

assa

ult

s

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Au

dien

ceof

mil

dly

−0.0

205

−0.0

202

−0.0

245

−0.0

245

−0.0

187

−0.0

205

−0.0

089

−0.0

106

viol

ent

mov

ies

(mil

lion

s(0

.005

2)∗∗

∗(0

.005

4)∗∗

∗(0

.004

0)∗∗

∗(0

.003

9)∗∗

∗(0

.005

0)∗∗

∗(0

.005

2)∗∗

∗(0

.004

1)∗∗

(0.0

029)

∗∗∗

ofpe

ople

inda

yt)

Au

dien

ceof

non

viol

ent

−0.0

06−0

.004

7−0

.010

3−0

.010

3−0

.010

4−0

.007

50.

0045

0.00

05m

ovie

s(m

illi

ons

(0.0

054)

(0.0

056)

(0.0

042)

∗∗(0

.004

1)∗∗

(0.0

053)

∗(0

.005

3)(0

.004

3)(0

.002

9)of

peop

lein

day

t)R

obu

stn

ess

spec

ifica

tion

Ben

chm

ark

IV:I

nst

rum

ents

Ben

chm

ark

+B

ench

mar

k+

Ben

chm

ark

+B

ench

mar

k+

OL

Sre

gres

s.P

oiss

onIV

spec

ifica

tion

budg

etan

din

clu

deM

o–T

hin

clu

deM

oTh

use

MP

AA

dep.

vari

able

(no

inst

rum

ents

)re

gres

sion

no.

thea

ters

+N

ewey

-Wes

tm

easu

reof

isal

lcri

mes

(no

inst

rum

ents

)28

-day

corr

.m

ovie

viol

ence

agai

nst

pers

onC

ontr

olva

riab

les

Fu

llse

tof

con

trol

sX

XX

XX

XX

XA

udi

ence

inst

rum

ente

dX

XX

XX

wit

hpr

edic

ted

audi

ence

usi

ng

foll

owin

gw

eek’

sau

dien

ceN

1,56

31,

563

3,64

53,

645

1,53

91,

563

1,56

31,

563

Not

es.

Th

ista

ble

pres

ents

ase

ries

ofro

bust

nes

sch

ecks

toth

ere

sult

sin

Tab

leII

I,re

prod

uce

din

colu

mn

(1).

Col

um

n(2

)u

ses

inst

rum

ents

con

stru

cted

asin

the

ben

chm

ark

inst

rum

ents

,bu

tu

sin

gth

en

um

ber

ofth

eate

rssh

owin

gth

em

ovie

inw

eek

w(t

)an

dth

epr

odu

ctio

nbu

dget

(wh

enav

aila

ble)

aspr

edic

tors

.T

his

spec

ifica

tion

also

incl

ude

sth

ein

stru

men

tfo

rov

eral

lmov

ieau

dien

ceco

nst

ruct

edw

ith

the

bech

mar

kin

stru

men

ts.(

See

text

for

addi

tion

alde

tail

s)C

olu

mn

(3)u

ses

data

also

from

Mon

day–

Th

urs

day,

inad

diti

onto

Fri

day–

Su

nda

y.C

olu

mn

(4)

use

sth

esa

me

sam

ple

asco

lum

n(3

)bu

tw

ith

New

ey-W

est

stan

dard

erro

rsw

ith

a21

-day

lag.

Col

um

n(5

)pr

esen

tsth

ere

sult

sfo

ran

alte

rati

vem

easu

reof

mov

ievi

olen

ceba

sed

onth

eM

PA

Ara

tin

gs.T

he

nu

mbe

rof

obse

rvat

ion

sis

smal

ler

beca

use

inth

efi

rst

wee

ksof

1995

,th

eM

PA

Ara

tin

gis

mis

sin

gfo

ra

nu

mbe

rof

mov

ies;

we

set

the

MP

AA

viol

ence

mea

sure

mis

sin

gfo

rth

ete

nw

eeks

inw

hic

hth

era

tin

gis

avai

labl

efo

rle

ssth

an70

%of

the

mov

ieau

dien

ce.

Inco

lum

n(6

)th

ede

fin

itio

nof

crim

esag

ain

sta

pers

on,

inad

diti

onto

assa

ult

san

din

tim

idat

ion

,in

clu

des

robb

ery,

hom

icid

e,an

dse

xof

fen

ses.

Col

um

n(7

)pr

esen

tsan

OL

Ssp

ecifi

cati

on,a

nd

colu

mn

(8)p

rese

nts

aP

oiss

onre

gres

sion

(als

on

otin

stru

men

ted)

.Th

en

um

ber

ofob

serv

atio

ns

inP

anel

Cis

one

few

erth

anin

Pan

els

Aan

dB

beca

use

we

are

mis

sin

gth

eas

sau

ltda

tafo

rJa

nu

ary

1,20

06,f

orth

eh

ours

betw

een

12A.M

.an

d6

A.M

.See

also

not

esto

Tab

leII

.*S

ign

ifica

nt

at10

%;*

*sig

nifi

can

tat

5%;*

**si

gnifi

can

tat

1%.


for example by incapacitating potential criminals. An alternativepossibility is that the instrument, which is based on next week-end’s audience, does not completely remove the impact of short-term shocks, especially for Wednesdays and Thursdays, which fallimmediately before the next weekend.

Column (4) assesses the robustness of the standard errors toautocorrelation. One may worry that violent crime is positivelycorrelated across weeks, even after controlling flexibly for sea-sonality. In this case, clustering by week (which assumes inde-pendence across weeks) may lead to standard errors that are toosmall. To address this concern, we replicate the specification ofcolumn (3) using Newey-West standard errors with a 28-day win-dow.5 The Newey-West standard errors are on average 5% lowerthan the clustered standard errors, suggesting that autocorrela-tion is a minor issue.

Next, we use an alternative measure of movie violence. Inaddition to rating movies (R, PG, etc.), the MPAA summarizes inone sentence the reason for their rating. We characterize as mildlyviolent those movies whose MPAA rating contains the word “Vio-lence” or “Violent,” with two exceptions. If the reference to violenceis qualified with “Brief,” “Mild,” or “Some,” we classify the movieas nonviolent. If qualified with either “Bloody,” “Brutal,” “Dis-turbing,” “Graphic,” “Grisly,” “Gruesome,” or “Strong,” we classifythe movie as strongly violent. The kids-in-mind.com and MPAA-based measures have correlations of .68 (mild violence) and .66(strong violence).6 The correlation is also apparent in Table A.1,which lists the violence ratings for blockbuster movies. Using thisMPAA-based measure of movie violence yields similar results (col-umn (5)). When we include both measures of violence (not shown),however, the effects on assaults load almost exclusively onto thekids-in-mind.com measures.

We also consider an alternative definition of violent crimes,including any type of crime against a person (column (6)). Inaddition to assaults and intimidation, this definition includes alsorobbery, homicide, and sex offenses. The results are very similar

5. We use data for the seven-weekday data rather than the benchmark three-day weekend data because Newey-West standard errors imply a decay that is afunction of the temporal distance between observations.

6. These are the correlations of the residuals from OLS regressions on thestandard set of control variables appearing in column (6) of Table II, excluding themovie violence measures.


to the benchmark ones.7 We find qualitatively similar results forthe three component categories of our assault measure (intimida-tion, simple assault, and aggravated assault), for assaults withand without injury, for assaults occurring at home and away fromhome, and for crimes involving a weapon (see Online AppendixTables 1 and 4). We find larger effects for assaults against aknown person, as opposed to against a stranger. We find smallnegative but statistically insignificant effects for property crimes(burglary, theft, motor vehicle theft, and vandalism).8

Finally, we estimate two specifications that do not instrumentfor movie audience: OLS (column (7)) and Poisson MLE (column(8)). In these specifications, the effect in the evening and nighthours is qualitatively similar to the benchmark estimates, withsomewhat smaller effects. Exposure to all types of movies in themorning and afternoon has a negative (significant) effect on vio-lent crime. These small differences are likely due to omitted vari-ables that are correlated with overall movie audience and crime.Indeed, if one considers the differential impact of violent versusnonviolent movies, the results mirror the IV results: no differen-tial effect in the morning and afternoon, and large negative effectsin the evening and night.

An Online Appendix presents additional robustness checks,including (i) the use of 52 week-of-year indicators instead of 365day-of-year indicators, (ii) estimates using only the audience forthe first week of release, (iii) estimates for the set of agencies thatreport consistently for the entire sample, (iv) separate estimatesfor violence levels 0 through 10, and (v) estimates in two-hourblocks. The pattern of findings is similar in these specifications.

In addition, the Online Appendix includes two placebo tests:one that reassigns movie attendance to the other date in the sam-ple that falls on the same day of year and same day of week,and another that examines whether future exposure, controllingfor current exposure, affects violent crime. We find no system-atic impact for either set of placebo variables, suggesting that ourfindings are not due to unobserved seasonal factors.

7. Homicide and sex offenses are relatively infrequent, and not significantindividually. Regressions for robbery by itself yield negative estimates that aresignificant in the evening hours but not in the nighttime hours.

8. Insofar as alcohol plays an important role (Section V.B), the smaller findingsfor property crimes are consistent with Carpenter and Dobkin (forthcoming) whofind a smaller spike around the legal drinking age in property crimes, comparedto violent crimes. It is also possible that movie attendance creates additionalopportunities for property crimes because property owners may be in the theater.


IV.E. DVD and VHS Rental Audience

While this paper focuses on the effect of movies shown intheaters, a similar design exploits the releases of movie rentalson VHS and DVD. These releases occur several months after thetheatrical release, and rentals of newly released VHSs and DVDspeak in the first week of release, with the top one to two moviescapturing a substantial share of total rental revenue.

We use data on weekly DVD and VHS rental revenue fromVideo Store Magazine covering the top 25 movies over the periodJanuary 1995–December 2004.9 The average number of rentalson a weekend day is 3.92 million (Table I). Weekend rentals ofstrongly violent (mildly violent) movies total 0.64 (1.56) million.While rentals are 30% to 40% smaller than the theater attendance,these numbers underestimate the audience reached because mul-tiple people often view a single rented movie. The violent audi-ence size for DVD and VHS rentals is positively correlated to thebox-office measure in the corresponding week: the conditional cor-relation between the two measures of strong (mild) violence is .15(.39) (see footnote 6).

In columns (1)–(3) of Table VI, we estimate equation (6) usingDVD and VHS rentals instead of box-office audience. We includethe full set of controls and instrument using a predictor based onnext week’s rentals. We find, as might be expected, no effect ofexposure to violent movies in the morning and afternoon hours(column (1)). In the evening hours (column (2)), we find a largenegative impact of exposure to mildly violent movies (a 1.48%decrease in assaults per million rentals), and a smaller, insignifi-cant impact of strongly violent movies. In the night hours (column(3)), we find large negative effects of exposure to rentals of violentmovies, but also a significant negative effect of the rental audi-ence of nonviolent movies. These estimates are less precise thanthe estimates for box-office releases, with standard errors about30% larger. When we also control for box-office movie audiencein the regressions, the results are similar, although with largerstandard errors (columns (4)–(6)).

9. To convert revenue data into an estimated number of rentals, we deflaterental revenue by the average price of a rental estimated using the ConsumerExpenditure Survey. We impute daily rentals using the within-week distribution ofrentals in the Consumer Expenditure Survey. As with the box-office data, we focuson weekend rentals. Data are missing for twenty weeks in which the magazine didnot publish the relevant numbers.


AB

LE

VI

EF

FE

CT

OF

DV

D/V

HS

MO

VIE

VIO

LE

NC

EO

NS

AM

E-D

AY

AS

SA

ULT

S

Spe

cifi

cati

on:

Inst

rum

enta

lvar

iabl

ere

gres

sion

s

Dep

.var

.:L

og(n

um

ber

ofas

sau

lts

inda

yt

inti

me

win

dow

)(1

)(2

)(3

)(4

)(5

)(6

)

DV

D/V

HS

ren

tals

ofst

ron

gly

viol

ent

mov

ies

−0.0

042

−0.0

078

−0.0

148

−0.0

051

−0.0

044

−0.0

107

(mil

lion

sof

peop

lein

day

t)(0

.005

8)(0

.006

3)(0

.007

8)∗

(0.0

101)

(0.0

104)

(0.0

120)

DV

D/V

HS

ren

tals

ofm

ildl

yvi

olen

tm

ovie

s−0

.004

1−0

.014

8−0

.031

1−0

.003

4−0

.022

7−0

.019

3(m

illi

ons

ofpe

ople

inda

yt)

(0.0

059)

(0.0

052)

∗∗∗

(0.0

071)

∗∗∗

(0.0

103)

(0.0

092)

∗∗(0

.010

2)∗

DV

D/V

HS

ren

tals

ofn

onvi

olen

tm

ovie

s−0

.002

9−0

.004

3−0

.022

5−0

.005

4−0

.004

1−0

.019

9(m

illi

ons

ofpe

ople

inda

yt)

(0.0

066)

(0.0

060)

(0.0

076)

∗∗∗

(0.0

115)

(0.0

106)

(0.0

114)

∗

Th

eate

rau

dien

ceof

stro

ngl

yvi

olen

tm

ovie

s0.

0017

−0.0

098

−0.0

192

(mil

lion

sof

peop

lein

day

t)(0

.008

2)(0

.007

7)(0

.008

9)∗∗

Th

eate

rau

dien

ceof

mil

dly

viol

ent

mov

ies

0.00

34−0

.011

9−0

.020

2(m

illi

ons

ofpe

ople

inda

yt)

(0.0

076)

(0.0

070)

∗(0

.008

0)∗∗

Th

eate

rau

dien

ceof

non

viol

ent

mov

ies

0.00

42−0

.004

9−0

.007

1(m

illi

ons

ofpe

ople

inda

yt)

(0.0

078)

(0.0

070)

(0.0

079)

Tim

eof

day

6A.M

.–6

P.M

.6

P.M

.–12

A.M

.12

A.M

.–6

A.M

.6

A.M

.–6

P.M

.6

P.M

.–12

A.M

.12

A.M

.–6

A.M

.n

ext

day

nex

tda

yC

ontr

olva

riab

les

Fu

llse

tof

con

trol

sX

XX

XX

XR

enta

lan

dth

eate

rau

dien

ces

inst

rum

ente

dX

XX

XX

Xw

ith

pred

icte

dau

dien

ces

usi

ng

nex

tw

eek’

sau

dien

ces

N1,

475

1,47

51,

475

1,47

51,

475

1,47

5

Not

es.T

he

dail

yau

dien

cen

um

bers

are

com

pute

dfr

omw

eekl

yda

taon

DV

Dan

dV

HS

ren

tal

reve

nu

efr

omV

ideo

Sto

reM

agaz

ine.

Th

ew

eekl

yre

ven

ue

isdi

vide

dby

the

aver

age

pric

eof

are

nta

lan

dpr

opor

tion

atel

yat

trib

ute

dto

the

Fri

day,

Sat

urd

ay,a

nd

Su

nda

yw

indo

wu

sin

gth

eav

erag

ew

ith

in-w

eek

dist

ribu

tion

ofre

nta

lsin

the

CE

Xdi

arie

s.T

he

spec

ifica

tion

sar

eIV

regr

essi

ons

wit

hth

elo

g(n

um

ber

ofas

sau

lts

occu

rrin

gin

day

t)as

the

depe

nde

nt

vari

able

.See

also

not

esto

Tab

leII

.∗ S

ign

ifica

nt

at10

%;∗

∗ sig

nifi

can

tat

5%;∗

∗∗si

gnifi

can

tat

1%.


The results on DVD and VHS releases are consistent with anegative impact of violent movies on violent crime, especially overthe evening hours. The similarity with the results from theaterreleases is interesting in light of the differences in setting (e.g.,alcohol consumption is possible at home but not at the theater).

V. INTERPRETATION AND ADDITIONAL EVIDENCE

We summarize the findings so far as follows: (i) exposure toviolent movies lowers same-day violent crime in the evening; (ii)this exposure also lowers violent crime in the night after exposure;(iii) in the night, strongly violent movies have a somewhat smallereffect on crime compared to mildly violent movies; (iv) nighttimehours have larger negative effects compared to evening hours;(v) there is no lagged effect of exposure in the weeks followingmovie attendance. We now provide interpretations and additionalevidence for the first four of these findings (the fifth finding isstraightforward to interpret).

We stress that, because of data limitations, the interpreta-tions in this section are based on ecological inference and notindividual-level analysis. As such, alternative explanations forthe findings are also possible. For example, whereas the decreasein crime in the evening hours has a natural interpretation asincapacitation of criminals, an alternative, complementary inter-pretation is protection of potential victims.

V.A. Lower Crime in the Evening—Voluntary Incapacitationand Sorting

We interpret the first finding, that violent movies lower crimein the evening hours, as voluntary incapacitation. Because it isvirtually impossible to commit an assault while in the theater, asmovie attendance rises, violent acts fall relative to the counterfac-tual. Interestingly, as simple as this explanation is, incapacitationhas largely been ignored in discussions on the effect of movie vi-olence. This voluntary incapacitation differs from the standardincapacitation in the literature because it is optimally chosen bythe consumers, rather than being imposed, as in the case of schoolclosings (Jacob and Lefgren 2003) or incarceration (Levitt 1996).

Although the qualitative findings are consistent withincapacitation, are the magnitudes also consistent with thisinterpretation? Suppose watching a movie (including time spentbuying tickets, waiting in the lobby, and traveling to and from the


theater) occupies roughly one-half of the 6 P.M.–12 A.M. time periodand fully incapacitates individuals. For the rest of the time block,assume that crime rates are the same as for the alternativeactivity. Using the framework of Section II, denoting criminalswith a y subscript, and assuming no crime is committed by nonvi-olent individuals (σo = 0) yields β j = −0.5x jσy. If criminals wereequally represented in the audience of a movie with one millionviewers, about 1/300th (i.e., 1 million out of a total populationof 300 million) of the criminals would be incapacitated, leadingto βv

equal = −0.5 ∗ (1/300) ≈ −0.0017, compared to the observedvalues βv = −0.0130 and βm = −0.0109. This implies violentindividuals are overrepresented by about 0.0130/0.0017 = 7.6times in strongly violent movies and 0.0109/0.0017 = 6.4 timesin mildly violent movies.

Although this is a substantial amount of selection, it is notimplausibly large. To provide evidence on the sorting of more vi-olent individuals into more violent movies, we turn to data fromthe Consumer Expenditure Survey (CEX). We take advantage ofthe fact that the CEX diaries record all expenditures of surveyedhouseholds day by day for a period of one or two weeks, includ-ing demographic information about the households that purchasemovie tickets.

For each day t in the years 1995–2004, we compute the shareof interviewed households that watch a movie at the theater,shareCEX

t . We regress this share on shares of the population at-tending movies of different violence levels according to our pri-mary movie attendance data10:

shareCEXt = α + βv Av

t

Popt+ βm Am

t

Popt+ βn An

t

Popt+ �Xt + εt,(7)

where Popt is the U.S. population in year t (Table VII). BecauseshareCEX

t and Ajt /Popt are both measures of the share of the pop-

ulation attending a movie on day t, we expect, and indeed find,that the estimated regression coefficients β j are statistically in-distinguishable from 1 when we include all demographic groups(column (1)).

10. The regressions include Friday, Saturday, and Sunday and are weightedby the number of households reporting consumption expenditures for day t, whichaverages 157.88. We include the standard set of controls Xt. We obtain similarresults when using an imputed individual-level measure of movie attendance, andsimilar, but less precisely estimated, results if we instrument for movie attendance.

714 QUARTERLY JOURNAL OF ECONOMICST

AB

LE

VII

PA

TT

ER

NS

OF

MO

VIE

AT

TE

ND

AN

CE

BY

DE

MO

GR

AP

HIC

S(C

EX

DA

TA)

Spe

cifi

cati

on:

OL

Sre

gres

sion

s

Dep

.var

.:S

har

eof

hou

seh

olds

inte

rvie

wed

wat

chin

ga

mov

ieat

the

thea

ter

inda

yt

(1)

(2)

(3)

(4)

(5)

Sh

are

ofau

dien

ceof

stro

ngl

yvi

olen

tm

ovie

s0.

9469

2.09

41.

146

0.43

232.

7751

(in

shar

eof

U.S

.pop

ula

tion

inda

yt)

(0.1

883)

∗∗∗

(0.5

602)

∗∗∗

(0.3

328)

∗∗∗

(0.2

580)

∗(1

.455

0)∗

Sh

are

ofau

dien

ceof

mil

dly

viol

ent

mov

ies

0.77

361.

4642

1.44

990.

1259

2.78

25(i

nsh

are

ofU

.S.p

opu

lati

onin

day

t)(0

.141

9)∗∗

∗(0

.440

7)∗∗

∗(0

.262

3)∗∗

∗(0

.171

1)(1

.311

0)∗∗

Sh

are

ofau

dien

ceof

non

viol

ent

mov

ies

0.76

141.

0786

1.15

550.

392

0.40

31(i

nsh

are

ofU

.S.p

opu

lati

onin

day

t)(0

.144

0)∗∗

∗(0

.465

2)∗∗

(0.2

491)

∗∗∗

(0.1

741)

∗∗(1

.292

6)D

emog

raph

icgr

oups

(by

hea

dof

hou

seh

old)

All

Age

s18

–29

Age

s30

–44

Age

s45

+S

ingl

em

ales

age

18–2

9

Fu

llse

tof

con

trol

sX

XX

XX

Reg

ress

ion

sw

eigh

ted

byn

um

ber

ofX

XX

XX

hou

seh

olds

inte

rvie

wed

inda

yt

Ave

rage

nu

mbe

rof

hou

seh

olds

in15

7.88

22.6

153

.94

81.2

93.

96de

mog

raph

icgr

oup

inte

rvie

wed

onda

yt

N1,

563

1,55

81,

560

1,56

31,

474

Not

es.A

nob

serv

atio

nis

aF

rida

y,S

atu

rday

,or

Su

nda

yov

erth

eye

ars

1995

–200

4.T

he

depe

nde

nt

vari

able

isth

esh

are

ofth

eh

ouse

hol

dsin

the

diar

yC

EX

sam

ple

that

repo

rted

atte

ndi

ng

am

ovie

onda

yt.

Th

eau

dien

cesh

ares

are

obta

ined

from

dail

ybo

x-of

fice

reve

nu

esdi

vide

dby

the

aver

age

pric

epe

rti

cket

and

then

divi

ded

agai

nby

the

U.S

.pop

ula

tion

.B

ecau

sebo

thth

ede

pen

den

tva

riab

lean

dth

ein

depe

nde

nt

vari

able

sar

em

easu

res

ofat

ten

dan

ceto

the

thea

ter

insh

ares

,th

eco

effi

cien

tsin

colu

mn

(1)

shou

ldbe

clos

eto

1.T

he

coef

fici

ents

inco

lum

ns

(2)–

(4)

indi

cate

the

degr

eeof

self

-sel

ecti

onof

diff

eren

tde

mog

raph

icca

tego

ries

into

mov

ies

ofdi

ffer

ent

viol

ence

leve

ls.S

eeal

son

otes

toT

able

II.

∗ Sig

nifi

can

tat

10%

;∗∗ s

ign

ifica

nt

at5%

;∗∗∗

sign

ifica

nt

at1%

.


Although different types of movies should have the same im-pact on overall attendance, we expect differential sorting when wesplit the data by demographics (columns (2)–(5)). Indeed, youngerhouseholds (heads ages 18 to 29, column (2)) have larger estimatedcoefficients, indicating that they attend the movies more oftenthan older people. Younger households also select disproportion-ately into violent movies: they are 2.094/0.9469 = 2.2 times over-sampled in strongly violent movies and 1.4642/0.7736 = 1.9 timesoversampled in mildly violent movies, but only 1.0786/0.7614 =1.4 times oversampled in nonviolent movies. Middle-age house-holds (heads ages 30 to 44, column (3)) and especially olderhouseholds (heads over 45 years, column (4)) attend the movietheater less and display a flatter attendance pattern with respectto the violence content of movies. The age groups with highercrime rates (Table I), therefore, select into violent movies, a resultconsistent with selective incapacitation.

Because men also have higher assault rates compared towomen (Table I), it would be useful to differentiate by gen-der. Although this is generally problematic in the CEX data(which only report purchases at the household level), we canconsider single men ages 18–29. In this group (column (5)), wefind even greater evidence of selection. Single young males are2.7751/0.9469 = 2.9 times oversampled in strongly violent moviesand 2.7825/0.7736 = 3.6 times oversampled in mildly violentmovies. Although the estimates for this small group should betaken with caution given the large standard errors, they indicatesubstantial sorting into violent movies.11

We find substantial sorting even using relatively poor cor-relates of criminal behavior—age and gender. In addition tobetween-group sorting, we expect substantial within-group sort-ing. The combination of between- and within-group sorting canplausibly generate overrepresentation of potential criminals by afactor of 6 or 7, as implied by the effect on assaults.

V.B. Lower Crime after Exposure—Sobriety

The second result is that exposure to movie violence also low-ers violent crime in the night. We interpret this to mean that anevening spent at the movies leads to less dangerous activities in

11. When we split households by income (results not shown), we find strongevidence of selection into more violent movies by lower-income households, a se-lection pattern consistent with research that documents that the poor are morelikely to be victims of aggravated assaults.


the night hours following exposure (i.e., αi < σ in expression (4)).This could be because a visit to the movie theater involves lessalcohol consumption, disrupts and alters an evening’s activities,or places potential criminals in relatively safer environments oncethe movie is over. This is not a trivial finding, because attendanceat movie theaters could have provided a meeting point for poten-tial criminals, leading to an increase in crime.

Alcohol is a prominent factor that has been linked to violentcrimes, and assaults in particular (Carpenter and Dobkin forth-coming). Alcohol is banned in almost all movie theaters in theUnited States, so a mechanism for reduced crime in the night-time could well be sobriety. To test this explanation, we examinewhether the displacement is larger for assaults involving alcoholor drugs (columns (1) and (2) of Table VIII) than for assaults notinvolving such substances (columns (3) and (4)). Indeed, althoughthe negative impact of movie violence on assaults is present inboth samples, the estimates are on average 1.5 times larger forassaults involving alcohol or drugs. We also find large displace-ment effects in the night hours for assaults in bars and nightclubsand for arrests for drunkenness, although these estimates areimprecise (Online Appendix Table 3).

To further test the impact of alcohol, in columns (5)–(8) weseparately estimate the effect for offenders just under the legaldrinking age (ages 17–20) and offenders just over the legal drink-ing age (ages 21–24). If the effect is due to alcohol consumption, itshould be larger for the latter group, because the younger group isless likely to drink as part of their displaced alternative activity.Indeed, the effect of violent movies is two to three times larger forthe over-age group.

Finally, to provide direct evidence that movie attendance low-ers alcohol consumption, we use data from the CEX time diaries.We examine whether exposure to violent movies reduces the shareof respondents consuming alcohol away from home (column (9)).We find suggestive evidence that violent movies may have reducedalcohol consumption, though the estimates are not significantlydifferent from zero.

V.C. Nonmonotonicity in Violent Content—Arousal

The third finding is that the negative effect in the night hoursis not monotonic: strongly violent movies have a slightly smallereffect than mildly violent movies (−.0192 versus −.0205). This at


AB

LE

VII

IT

ES

TO

FS

OB

RIE

TY:E

FF

EC

TO

FA

LC

OH

OL

CO

NS

UM

PT

ION

Spe

cifi

cati

on:

Inst

rum

enta

lvar

iabl

ere

gres

sion

sR

eg.(

CE

Xda

ta)

Dep

.var

.:S

har

eco

nsu

min

gal

coh

olaw

ayL

og(n

um

ber

ofas

sau

lts

ofa

type

inda

yt

inti

me

win

dow

)fr

omh

ome

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Au

dien

ceof

stro

ngl

y−0

.012

−0.0

287

−0.0

137

−0.0

164

−0.0

239

−0.0

376

−0.0

125

−0.0

058

−0.3

303

viol

ent

mov

ies

(mil

lion

s(0

.008

0)(0

.010

9)∗∗

∗(0

.005

6)∗∗

(0.0

070)

∗∗(0

.010

3)∗∗

(0.0

115)

∗∗∗

(0.0

114)

(0.0

149)

(0.2

696)

ofpe

ople

inda

yt)

Au

dien

ceof

mil

dly

−0.0

183

−0.0

25−0

.008

8−0

.019

7−0

.022

9−0

.033

8−0

.011

2−0

.017

1−0

.192

1vi

olen

tm

ovie

s(m

illi

ons

(0.0

071)

∗∗(0

.010

7)∗∗

(0.0

046)

∗(0

.005

9)∗∗

∗(0

.008

4)∗∗

∗(0

.010

7)∗∗

∗(0

.010

0)(0

.013

3)(0

.207

7)of

peop

lein

day

t)A

udi

ence

of−0

.006

8−0

.010

2−0

.005

7−0

.003

9−0

.02

−0.0

213

0.00

650.

0011

−0.0

271

non

viol

ent

mov

ies

(mil

lion

s(0

.007

6)(0

.011

4)(0

.004

8)(0

.006

0)(0

.008

9)∗∗

(0.0

110)

∗(0

.010

6)(0

.013

9)(0

.199

3)of

peop

lein

day

t)A

ssau

ltby

offe

nde

rA

ssau

ltby

offe

nde

rT

ype

ofcr

ime

Ass

ault

sin

volv

ing

Ass

ault

sn

otin

volv

ing

ages

21–2

4ag

es17

–20

Alc

ohol

con

sum

ptio

nal

coh

olor

dru

gsal

coh

olor

dru

gs(o

ver

drin

kin

gag

e)(u

nde

rdr

inki

ng

age)

away

from

hom

eT

ime

ofda

y6

P.M

.–12

A.M

.–6

P.M

.–12

A.M

.–6

P.M

.–12

A.M

.–6

A.M

.–12

A.M

.–S

ame

12A.M

.6

A.M

.12

A.M

.6

A.M

.12

A.M

.6

A.M

.12

A.M

.6

A.M

.da

yn

ext

day

nex

tda

yn

ext

day

nex

tda

yC

ontr

olva

riab

les

Fu

llse

tof

con

trol

sX

XX

XX

XX

XX

Au

dien

cein

stru

men

ted

XX

XX

XX

XX

wit

hpr

edic

ted

audi

ence

usi

ng

nex

tw

eek’

sau

dien

ceN

1,56

31,

560

1,56

31,

562

1,56

31,

562

1,56

31,

561

1,56

3

Not

es.T

he

spec

ifica

tion

sar

ein

IVre

gres

sion

sfo

rsp

ecifi

cty

pes

ofas

sau

lts

usi

ng

NIB

RS

data

inco

lum

ns

(1)–

(8).

Col

um

n(9

)use

sth

eC

EX

data

use

din

Tab

leV

III;

the

depe

nde

nt

vari

able

isth

esh

are

ofth

eh

ouse

hol

dsin

the

diar

yC

EX

sam

ple

that

repo

rted

con

sum

ing

alco

hol

away

from

hom

e.In

colu

mn

(9),

the

mov

ieex

posu

reva

riab

les

are

insh

are

ofth

eto

talp

opu

lati

on.S

eeal

son

otes

toT

able

II.

∗ Sig

nifi

can

tat

10%

;∗∗ s

ign

ifica

nt

at5%

;∗∗∗

sign

ifica

nt

at1%

.


first is puzzling, because strongly violent movies attract more po-tential criminals, and the additional selection should render theeffect more negative. As discussed in Section II, however, this puz-zle can be explained if strongly violent movies have a differentialdirect impact.

We estimate the differential impact of strongly violent movies,αv − α, under the assumptions used to derive expression (5). Es-timation of αv − α requires information about the selection of po-tential criminals x j into different movies. Although this selectionis unobservable, we do observe selection along dimensions thatcorrelate with criminal behavior, age, and gender. As Table I in-dicates, crimes are committed disproportionately by young males.We make the assumption that the selection of potential criminalsinto movie theaters, x j, is an affine transformation of the selectionof young males, yi; that is, x j = λ0 + λ1yi. We can then estimateexpression (5) by substituting the term (yv − yn) / (ym − yn) for theunobserved (xv − xn) / (xm − xn) .

To estimate the sorting of young males, we turn to an auxil-iary source of data, the IMDb.12 IMDb maintains a popular web-site for movie-goers, which invites its users to rate movies. A typ-ical blockbuster movie is rated by tens of thousands of viewers.IMDb displays, for each movie, statistics on the rating for eachcombination of gender (male, female) and four age groups (under18, 18 to 29, 30 to 44, and over 45). As a measure of the attrac-tiveness of a movie to potential criminals, we use the share ofraters that are male and are ages 18 to 29, a group disproportion-ately likely to commit crimes (see Table I). Figure II shows thatthe share of young male reviewers is fairly linear in the 0 to 10violence ratings for movies from kids-in-mind.com. The extent ofselection is substantial: while the fraction of raters of nonviolentmovies that are young males, yn, is 0.459, the corresponding frac-tion for strongly violent movies, yv, is 0.546. These data allow usto estimate (yv − yn) / (ym − yn) as 1.718.

Figure III displays both the actual impact of movie violence β j

(solid lines) and the predicted impact purely due to sorting (dottedlines). The two estimates are very close for crime in the eveninghours, and one cannot reject the hypothesis that they are the same.This is to be expected, because a large share of the evening is

12. The CEX data used in Table VIII also indicate substantial selection: younghouseholds (with heads ages 18–29) select into strongly violent movies at a ratethat is 43% higher compared to mildly violent movies. We use the IMDb databecause they provide a substantially more precise estimate.


0.35

0.4

0.45

0.5

0.55

0.6

0 1 2 3 4 5 6 7 8 9 10

Rating of violent content of movie (from kids-in-mind.com)

Sh

are

of

you

ng

mal

es a

mo

ng

IMD

b r

evie

wer

s o

f m

ovi

e

Young males in audience as functionof violence of movieConfidence interval (lower bound)

Confidence Interval (upper bound)

Average for nonviolent

Average for strongly violent movies

Average for mildly violent movies

FIGURE II

Share of Young Males in Audience as Function of Movie Violence (Internet MovieDatabase Data)

This plot employs IMDb rating data to provide a measure of the attractivenessto young males of movies of varying degrees of violence (0 is least violent, 10 ismost violent). The measure of attractiveness to young males is the share of ratersof a movie that report being male and ages 18 to 29. The plotted variable is theaverage share across all movies of a given violence level, weighted by the numberof raters for the movie. The violence rating of movies is from kids-in-mind.com.The dotted lines are pointwise 95% confidence intervals.

spent inside the movie theaters, which mechanically implies αv ≈α ≈ 0. In the night hours, instead, the observed impact of movieviolence is substantially larger than the predicted impact becauseof selection, and the difference is marginally significant (p-value of.08).13 The estimated differential impact of movie violence αv − α

is sizable (.011) and equal to about one-third of the predictedimpact of strongly violent movies because of sorting.

We therefore detect some evidence that, after accounting forselection, violent movies induce more violent crime relative to non-violent movies, consistent with an arousal effect. This may occurfor the same reasons as in the laboratory—an emotional effect ofarousal, or short-term imitation of violent acts. As in the labora-tory, we find no evidence of a cathartic effect, which would havemade the effect of strongly violent movies even more negative. Ourfield evidence, hence, provides a natural comparison of the size of

13. Bootstrap standard errors take into account the sampling variabilityassociated with (yv − yn) / (ym − yn).


–0.035

–0.03

–0.025

–0.02

–0.015

–0.01

–0.005

0

Nonviolent movies Mildly violent movies Strongly violent movies

Eff

ect

of

exp

osu

re t

o m

ovie

vio

len

ce o

n a

ssau

lts

Actual estimates (evening, 6 P.M.–12 A.M.)

Predicted estimates with selection (no arousal) (evening, 6 P.M.–12 A.M.)

Actual estimates (night, 12 A.M.– 6 A.M.)

Predicted estimates with selection (no arousal) (night, 12 A.M.–6 A.M.)

Difference between

observed and predicted

estimates = .0117

(arousal effect )p-value=.08

Differencebetween

observed and predicted

estimates = .0012

p-value=.85

–.0063

–.0109

–.0205

–.0130

–.0142

–.0192

–.0309

–.0060

FIGURE III

Effect of Movie Violence on Assaults: Selection and Arousal EffectsThis figure displays both the actual impact of movie exposure on violent crime

(solid lines) and the predicted impact with linear selection (dotted lines) by typeof movie (nonviolent/mildly violent/strongly violent) and by time block (evening6 P.M.–12 A.M./night 12 A.M.–6 A.M.). The estimates of the actual impact (solidlines) are reproduced from columns (3) and (4) of Table III, Panel A, and can beinterpreted as the percent change in violent crime due to the exposure of onemillion people to movies of type j in time period t. For example, an increase inone million of the audience of mildly violent movies lowers violent crime by 1.09%in the evening time block and by 2.05% in the nighttime block. The estimatesof the predicted impact with linear selection (dotted lines) are computed usingthe estimates for nonviolent and mildly violent movies, taking into account theincreased selection of criminals into strongly violent movies and assuming thatall types of movies have the same direct effect on violent crime. The (unobserved)selection of criminals into movies is assumed to be related linearly to the (observed)selection of young males into movies. The comparison between the predicted andthe actual effect of violent movies provides an estimate of the differential effectof strongly violent movies relative to mildly violent and nonviolent movies. Thefigure shows a marginally significant difference in the actual and predicted impactfor the nighttime block: compared to the predicted impact, strongly violent moviescause more crime, consistent with an arousal effect of strongly violent movies.Details on the calculations of the difference are in the text.

the arousal effect to the other main impact of movie violence, timeuse. Although the estimated arousal effect on violence is sizable,it is one-third as large as the foregone violence associated withthe alternative activity.

We also point out that this evidence should be consideredsuggestive, given the assumptions involved. Other explanationsfor this nonmonotonic pattern are also possible. For example,


a potential offender may attend a mildly violent movie with agirlfriend and a strongly violent movie with drinking buddies.This could have an independent effect on the level of violence.

V.D. Larger Nighttime Estimates—Compositional Effects

The fourth finding is that, in the night hours following movieexposure (12 A.M.–6 A.M.), the impact of movie violence on assaultsis higher than in the evening hours (6 P.M.–12 A.M.). This find-ing might seem puzzling, because the highest decrease in crimeshould occur when potential criminals are in the movie theater,when committing crimes is nearly impossible.

However, the composition of crimes in the two time periodsis different, making a direct comparison of the size of the effectsdifficult. For example, assaults involving alcohol or drugs and as-saults committed by offenders just over the legal drinking age aremuch more common in the night hours than in the evening hours(Table I). As previously noted, alcohol-related assaults respondmore to violent movie exposure (Table VIII). Hence, the decreasein alcohol consumption, a primary mechanism for the effects, islikely to prevent a higher fraction of violent crimes in the night(when inebriation would have the most impact) compared to theevening. The activities prevented by movie attendance in the nighthours are more dangerous (in the model, have a larger σ ) than theactivities prevented in the evening hours.

Broadly speaking, we obtain similar compositional differ-ences in the pattern of assaults by demographics (shown in OnlineAppendix Table 5). The impact of exposure to violent movies islarger (i.e., more negative) for male offenders than for female of-fenders, especially in the night hours, and male offenders commita higher share of the assaults at night than in the evening hours(Table I). We also find a relatively monotonic decrease of the effectsizes by age (with the exception of the 45–54 age group), whichcontributes to explaining the findings, because the younger agegroup also contributes disproportionately to nighttime assaults(Table I).

V.E. Additional Evidence on Selection

In both the evening and the night hours, violent movies lowercrime more than nonviolent movies. Our explanation for thesefacts is selection: violent movies are more likely to attract poten-tial criminals. We now test another implication of selection, that


movies that draw young men tend to decrease violent crime, evenif the movies are not violent.

We divide movies into thirds based on the fraction of youngmen rating a movie in the IMDb (see Figure II), and label thecategories as Not Liked, Liked, and Highly Liked by young males.Table IX reports information on the blockbusters within the threecategories, holding constant the kids-in-mind.com violence rating.Among nonviolent movies, Runaway Bride is not liked by youngmales, while Austin Powers in Goldmember is highly liked. Formildly violent movies, Save The Last Dance and Spider-Man arebest sellers in the Not Liked and Highly Liked categories, respec-tively. Among strongly violent movies, there are essentially noblockbusters that are not liked by young males, because movieviolence and liking by young males are highly correlated. How-ever, the IMDb information distinguishes between movies in themiddle group such as Passion of the Christ and movies in the topgroup such as Hannibal.

To estimate the impact of movie attendance on violence withineach of the nine cells, we estimate ln Vt = ∑9

j=1 β j Ajt + �Xt + εt,

where j = 1, . . . , 9 denotes the nine cells. We adopt the full set ofcontrols and use the baseline instrument. Table IX reports withineach cell the coefficients for the evening time block and for thenight time block. Moving down within a column shows that moreviolent movies are generally associated with lower crime, evenholding constant the liking by young males (except for movies notliked by young males, where the violent movie category is verysparse and hence the estimates very noisy). For example, amongthe movies highly liked by young males, the estimated parametersβ j are −.0090 (nonviolent), −.0111 (mild violence), and −.0140(strong violence) for the evening hours. These patterns are broadlyconsistent with the interpretations discussed in Sections V.A–V.D.

More important for a test of selection, moving along arow the coefficients also generally become more negative. Innine of twelve pairwise comparisons, the estimates become morenegative as the liking by males increases (seven of ten if weexclude the bottom-left group, which is very sparse). Moviesthat attract more young males, therefore, appear to lower theincidence of violent crimes more, even holding constant the levelof violence in a movie. These results underscore the importanceof selection. Exposure to movies that attract more violent groups(along observable lines) is associated with lower rates of violentcrime.


TA

BL

EIX

MO

VIE

BL

OC

KB

US

TE

RB

YIM

DB

RA

TIN

GA

ND

VIO

LE

NC

E

Vio

len

ceB

lock

bust

erm

ovie

sn

otli

ked

Blo

ckbu

ster

mov

ies

like

dB

lock

bust

erm

ovie

sh

igh

lyli

ked

rati

ng

byyo

un

gm

ales

(dat

e,m

illi

ons

byyo

un

gm

ales

(dat

e,m

illi

ons

byyo

un

gm

ales

(dat

e,m

illi

ons

ofpe

ople

)of

peop

le)

ofpe

ople

)(1

)(2

)(3

)(4

)

0–4

Non

viol

ent

Top

1H

arry

Pot

ter

and

the

Ch

ambe

rof

Sec

rets

(11/

16/0

2,15

.2)

Sh

rek

2(5

/22/

04,1

7.4)

Au

stin

Pow

ers

inG

old

mem

ber

(7/2

7/02

,12.

6)m

ovie

sT

op2

Har

ryP

otte

ran

dth

eC

ham

ber

ofS

ecre

ts(1

1/23

/02,

7.3)

Har

ryP

otte

ran

dth

eS

orce

rer’

sS

ton

e(1

1/17

/01,

15.9

)In

cred

ible

s(1

1/6/

04,1

1.3)

Top

3R

un

away

Bri

de

(7/3

1/99

,6.8

)S

hre

k2

(5/2

9/04

,11.

8)B

ruce

Alm

igh

ty(5

/24/

03,1

1.2)

Top

4–6

Sw

eet

Hom

eA

laba

ma,

Am

eric

a’s

Sw

eeth

eart

s,E

rin

Bro

ckov

ich

Fin

din

gN

emo,

Toy

Sto

ry2,

Mon

ster

sIn

c.A

ceV

entu

ra:W

hen

Nat

ure

Cal

ls,

Wat

erbo

y,B

igD

add

yE

ffec

ton

−0.0

041

(0.0

062)

(6P.

M.–

12A.M

.)−0

.003

5(0

.004

2)(6

P.M

.–12

A.M

.)−0

.009

0∗(0

.005

3)(6

P.M

.–12

A.M

.)cr

ime

0.00

49(0

.007

1)(1

2A.M

.–6

A.M

.)−0

.005

7(0

.005

5)(1

2A.M

.–6

A.M

.)−0

.007

9(0

.006

3)(1

2A.M

.–6

A.M

.)

5–7

Mil

dly

Top

1D

oubl

eJe

opar

dy

(9/2

5/99

,4.6

)H

arry

Pot

ter

and

the

Pri

son

erof

Azk

aban

(6/5

/04,

15.1

)S

pid

er-M

an(5

/4/0

2,19

.8)

viol

ent

Top

2S

ave

Th

eL

ast

Dan

ce(1

/13/

01,4

.1)

Mu

mm

yR

etu

rns

(5/5

/01,

12.4

)M

atri

xR

eloa

ded

(5/1

7/03

,15.

2)m

ovie

sT

op3

Dou

ble

Jeop

ard

y(1

0/2/

99,3

.3)

Pla

net

ofth

eA

pes

(7/2

8/01

,12.

3)L

ost

Wor

ld:J

ura

ssic

Par

k(5

/24/

97,

14.3

)T

op4–

6A

bsol

ute

Pow

er,R

and

omH

eart

s,U

nfa

ith

ful

Day

afte

rT

omor

row

,In

dep

end

ence

Day

,Pea

rlH

arbo

rS

pid

er-M

an2,

X2:

X-M

en,S

tar

War

s2

Eff

ect

on0.

0049

(0.0

111)

(6P.

M.–

12A.M

.)−0

.009

9∗∗

(0.0

047)

(6P.

M.–

12A.M

.)−0

.011

1∗∗∗

(0.0

039)

(6P.

M.–

12A.M

.)cr

ime

−0.0

268

(0.0

141∗

)(1

2A.M

.–6

A.M

.)−0

.017

7∗∗∗

(0.0

057)

(12

A.M

.–6

A.M

.)−0

.017

9∗∗∗

(0.0

052)

(12

A.M

.–6

A.M

.)


TA

BL

EIX

( CO

NT

INU

ED

)

Vio

len

ceB

lock

bust

erm

ovie

sn

otli

ked

Blo

ckbu

ster

mov

ies

like

dB

lock

bust

erm

ovie

sh

igh

lyli

ked

rati

ng

byyo

un

gm

ales

(dat

e,m

illi

ons

byyo

un

gm

ales

(dat

e,m

illi

ons

byyo

un

gm

ales

(dat

e,m

illi

ons

ofpe

ople

)of

peop

le)

ofpe

ople

)(1

)(2

)(3

)(4

)

8–10

Top

1M

issi

ng

(11/

29/0

3,1.

8)P

assi

onof

the

Ch

rist

(2/2

8/04

,13.

5)H

ann

ibal

(2/1

0/01

,10.

1)S

tron

gly

viol

ent

Top

2N

urs

eB

etty

(9/9

/00,

1.3)

Pas

sion

ofth

eC

hri

st(3

/6/0

4,8.

5)Ju

rass

icP

ark

3(7

/21/

01,9

.1)

mov

ies

Top

3C

opyc

at(1

1/4/

95,1

.2)

Air

For

ceO

ne

(7/2

6/97

,7.9

)S

cary

Mov

ie(7

/8/0

0,8.

2)T

op4–

6Ja

de,

InD

ream

s,A

Ric

hM

an’s

Wif

eR

anso

m,S

leep

yH

ollo

w,G

ener

al’s

Dau

ghte

rB

adB

oys

2,T

roy,

Ter

min

ator

3

Eff

ect

on0.

0625

(0.0

384)

(6P.

M.–

12A.M

.)−0

.008

4(0

.008

2)(6

P.M

.–11

2A.M

.)−0

.014

0∗∗∗

(0.0

047)

(6P.

M.–

12A.M

.)cr

ime

0.05

26(0

.054

9)(1

2A.M

.–6

A.M

.)−0

.025

2∗∗∗

(0.0

087)

(12

A.M

.–6

A.M

.)−0

.015

0∗∗

(0.0

061)

(12

A.M

.–6

A.M

.)

Not

es.W

edi

vide

mov

ies

into

thir

dsu

sin

gth

efr

acti

onof

IMD

bra

ters

ofa

mov

ieth

atar

em

ale

and

ofag

es18

–29.

Mov

ies

not

like

dby

you

ng

mal

esar

ede

fin

edby

mov

ies

inth

ebo

ttom

thir

dof

this

dist

ribu

tion

,mov

ies

like

dby

you

ng

mal

esar

ein

the

mid

dle

thir

d,an

dm

ovie

sst

ron

gly

like

dby

you

ng

mal

esar

ein

the

top

thir

d.T

he

rati

ngs

ofm

ovie

viol

ence

are

from

kids

-in

-min

d.co

m.T

he

tabl

edi

vide

sm

ovie

sin

ton

ine

cate

gori

esde

fin

edby

the

inte

ract

ion

ofh

owli

ked

the

mov

ieis

byyo

un

gm

ales

and

the

viol

ence

leve

l.T

he

top

thre

em

ovie

sw

ith

the

hig

hes

tw

eeke

nd

audi

ence

are

repo

rted

for

each

cate

gory

,alo

ng

wit

hth

en

ext

thre

ela

rges

tdi

stin

ctbl

ockb

ust

erm

ovie

s.T

he

"Eff

ect

oncr

ime"

row

sre

port

the

coef

fici

ents

onth

eau

dien

cesi

zes

for

each

ofth

en

ine

cate

gori

esfr

omtw

ose

para

tere

gres

sion

sfo

rth

eev

enin

g(6

P.M

.–12

A.M

.)an

dn

igh

ttim

eh

ours

(12

A.M

.–6

A.M

.),w

her

eth

ede

pen

den

tva

riab

leis

log(

nu

mbe

rof

assa

ult

soc

curr

ing

inda

yt

inth

esp

ecifi

edti

me

bloc

k)an

dth

ein

depe

nde

nt

vari

able

sar

eth

eau

dien

ces

inm

illi

ons

ofpe

ople

for

mov

ies

inea

chof

the

nin

eca

tego

ries

.S

eeal

son

otes

toT

able

II.

∗S

ign

ifica

nt

at10

%;∗

∗si

gnifi

can

tat

5%;∗

∗∗si

gnifi

can

tat

1%.


VI. CONCLUSION

We have provided causal evidence on the short-run effect ofexposure to media violence on violent crime. We exploit the natu-ral experiment induced by time-series variation in the violence ofmovies at the box office. We show that exposure to violent movieshas three main effects on violent crime: (i) it significantly reducesviolent crime in the evening on the day of exposure; (ii) by an evenlarger percent, it reduces violent crime during the night hoursfollowing exposure; (iii) it has no significant impact in the daysand weeks following the exposure.

We interpret the first finding as voluntary incapacitation: po-tential criminals that choose to attend the movie theater foregoother activities that have higher crime rates. As simple as thisfinding is, it has been neglected in the literature, despite its quan-titative importance. We interpret the second finding as substitu-tion away from a night of more volatile activities, in particular, areduction in alcohol consumption. The third finding implies thatthe same-day impact on crime is not offset by intertemporal sub-stitution of crime. An important component of these interpreta-tions is the sorting of more violent individuals into violent movieattendance.

These findings appear to contradict evidence from laboratoryexperiments that document an increase in violent behavior follow-ing exposure to movie violence. However, the field and laboratoryfindings are not contradictory. Exposure to movie violence canlower violent behavior relative to the foregone alternative activity(the field finding), even if it increases violent behavior relative toexposure to nonviolent movies (the laboratory finding). In fact, wedocument suggestive evidence that, after accounting for selection,violent movies induce more violent crime relative to nonviolentmovies, consistent with an arousal effect. This example suggeststhat other apparent discrepancies between laboratory and fieldstudies (see Levitt and List [2007]) might be reconciled if differ-ences in treatment and setup are taken into account. In addition,the field evidence provides a bound for the laboratory finding ofan arousal effect, which we estimate in the field to be one-third aslarge as the time-use effect.

Given that movie attendance occupies a significant portionof leisure time use, our findings imply first-order welfare effects.We can calculate the change in assaults that would occur if theaudience of violent movies did not go to the movies but instead


engaged in their next best alternative. The total number ofevening and nighttime assaults prevented is 997 assaults perweekend, adding up to almost 52,000 weekend assaults preventedyearly.14 With an estimated (in year 2007 dollars) direct monetarycost of $2,217 and an estimated intangible quality-of-life cost of$11,154 per assault (Miller, Cohen, and Wiersema 1996), thisimplies a benefit of roughly $695 million each year. Our estimatessuggest that a strongly violent blockbuster movie such as Han-nibal (with 10.1 million viewers on opening weekend) reducedassaults by 1,056 on its opening weekend, which amounts to a5.2% decrease in assaults, about half the impact of the reductionin crime due to a cold day. This substantial short-term impact ofviolent movies had been overlooked by the previous literature.

Of course, if strongly violent movies were banned as a mat-ter of public policy, our estimated short-term effects could be offsetpartly if studios respond by producing more mildly violent movies.The degree to which this would temper our findings depends onhow substitutable strongly and mildly violent movies are for eachother. This substitution, however, is likely to be imperfect; a re-gression of strongly violent movie attendance on mildly violentmovie attendance (including all the baseline controls of Table III)yields a coefficient of −.196 (s.e. .028). This implies that there willbe substantial substitution to other nonmovie activities as well,and our empirical results suggest that these nonmovie activitiesare more conducive to violent behavior.

In the paper, we find no impact of violent movies in the daysand weeks following exposure. Still, our design (like the labora-tory experiments) cannot address the important question aboutthe long-run effect of exposure to movie violence. As such, thispaper does not provide evidence on the long-term effects of a pol-icy limiting the level of violence allowed in the media. However,it does indicate that in the short run these policies will likelyincrease violent crime, because they induce substitution towardmore dangerous activities.

Finally, a central point of our paper is that the merits ofany particular activity must be viewed relative to the next bestactivity in utility terms. As such, our findings are relevant beyond

14. We assume: (i) no impact of media violence on assaults beyond the eveningand night of the media exposure, (ii) no substitution toward other movies, and (iii)effects for the whole population being the same as for the set of cities in theNIBRS sample. We calculate the effect separately for each time block (eveningand night) and level of violence (strong and mild). We multiply the estimatedbaseline coefficient by the assault rate in NIBRS data times the U.S. population(300 million), times average violent movie attendance.


the case of movies. For example, violent video games may wellincrease aggression, but they also incapacitate potential offendersfor a substantial period of time. More generally, we hypothesizethat other activities with a controlled, alcohol-free environmentthat attract young men, such as Midnight Basketball, should alsoreduce crime in the short run.

APPENDIX I: DATA APPENDIX

A. Imputation of Daily Box-Office Audience

The daily box-office movie revenue for the ten highest-sellingmovies is available starting in September 1997. To extend cov-erage to January 1995–August 1997 and to movies that do notmake the daily top-ten list, we make use of weekend revenue forthe fifty highest-selling movies, because this is available through-out the whole sample. We take advantage of the regularity in thewithin-week pattern of sales and impute the daily data, whenevermissing, using the weekend box-office data for the same movie inthe same week. Denote by aj,t the daily audience of movie j ondate t, and by aw

j,w(t) the weekend audience of movie j on weekendw(t) corresponding to date t. (Because most movies are released onFriday, the function w (t) assigns the days from Monday throughThursday to the previous weekend.) We assume that the daily au-dience is a share s of the weekend audience, where the share isallowed to depend on a set of controls Y , s (Y ): aj,t = s (Y ) aw

j,w(t).

In logs, the model can be written as ln(aj,t

) = ln(s (Y )) + ln(awj,w(t)).

The most important control for the share ln (s (Y )) is the set of day-of-week indicators dd

t , because different days of the week capturea different share of the overall revenue (Table I). In addition, weuse the following controls Xj,t for the weekday share: month indi-cators (in the summer the Monday–Thursday audience is larger),a linear time trend, indicators for the level of violence (nonvio-lent versus mildly violent versus strongly violent), indicators forrating type (G/PG/PG-13/R/NC-17/Unrated/Missing Rating), in-dicators for week of release (up to week 26), and indicators foraudience size in week w (t) (audience <0.5M, ≥0.5M and <1M,≥1M and <2M, ≥2M and <5M, ≥5M). This set of controls X is in-teracted with the day-of-week dummies, as well as being presentin levels. Finally, we control for a set of holidays Ht, describedbelow. We estimate

ln(aj,t) − ln(awj,w(t)) =

∑d∈D

βdddt +

∑d∈D

�d,Xddt Xj,t + �Xj,t + �Ht + ε j,t


and obtain the predicted daily audience aj,t using aj,t =exp [ln(aw

j,w(t)) + ln(aj,t) − ln(awj,w(t))]. The final daily box-office au-

dience is defined as the actual box-office data aj,t whenever avail-able, and the predicted value otherwise. In the subsample, whereboth the daily and the weekend data are available, a regressionof predicted daily revenue on actual daily revenue yields a slopecoefficient of .9559 and has an R2 of .9590.

B. Holiday Controls

The extensive set of holiday indicators takes into account that(i) holidays generally increase movie attendance; (ii) differentholidays have different impacts on attendance; (iii) attendanceincreases in the days preceding a holiday, and for major holidaysin the week surrounding. Hence, we include separate indicatorsfor Martin Luther King Day, President’s Day, Memorial Day,Labor Day, and Columbus Day; separate indicators for the Friday,Saturday, and Sunday preceding each of these holidays, and a sep-arate indicator for the Tuesday following these Monday holidays.We also include an indicator for Independence Day, Veteran’s Day,three Easter indicators (Friday, Saturday, and Sunday), threeThanksgiving indicators (Wednesday, Thursday, and Thanks-giving weekend), four Christmas indicators (December 20–23,December 24, December 25, and December 26–30), and threeNew Year’s indicators (December 31, January 1, and January2–3). In addition, we include an indicator for holidays if they fallon a weekend (Independence Day, Veteran’s Day, Christmas, NewYear’s, and Valentine’s Day). Finally, we include indicators for St.Patrick’s Day, Valentine’s Day, Halloween, Cinco de Mayo, andMother’s Day. (Notice that several holiday indicators drop out inthe benchmark sample that includes only Friday through Sunday.)

C. TV Audience Controls

We include two controls for TV audience: (i) an indicator forthe date of the Super Bowl; (ii) the TV audience for TV programswith an audience above fifteen million viewers, and 0 otherwise.The latter variable was constructed using Nielsen data on topshows of the year listed in Time Almanac; the variable is zero forthe season 2000–2001, for which we could not locate the data.

D. Weather Controls

The source for the weather variables is the “Global SurfaceSummary of Day Data” produced by the National Climatic Data


AB

LE

A.1

MO

VIE

BL

OC

KB

US

TE

RS

BY

VIO

LE

NC

EL

EV

EL W

eeke

nd

MP

AA

Lik

ing

Vio

len

ceW

eeke

nd

thea

ter

viol

ence

byyo

un

gra

tin

gT

itle

ofbl

ockb

ust

erda

teau

dien

cera

tin

gm

ales

(1)

(2)

(3)

(4)

(5)

(6)

0B

ird

cage

3/9/

1996

4,02

6,08

3L

owL

owYo

u’v

eG

otM

ail

12/1

9/19

983,

925,

587

Low

Low

You

’ve

Got

Mai

l12

/26/

1998

3,85

5,01

1L

owL

ow1

Ru

naw

ayB

rid

e7/

31/1

999

6,77

1,65

4L

owL

owE

rin

Bro

ckov

ich

3/18

/200

05,

178,

850

Low

Low

Not

tin

gH

ill

5/29

/199

94,

355,

314

Low

Low

2L

iar

Lia

r3/

22/1

997

6,70

9,56

9L

owH

igh

Toy

Sto

ry11

/25/

1995

6,59

9,61

0L

owM

ediu

mS

wee

tH

ome

Ala

bam

a9/

28/2

002

6,13

5,75

5L

owL

ow3

Sh

rek

25/

22/2

004

17,3

97,4

04L

owM

ediu

mS

hre

k2

5/29

/200

411

,838

,217

Low

Med

ium

Fin

din

gN

emo

5/31

/200

311

,650

,366

Low

Med

ium

4H

arry

Pot

ter

and

the

Sor

cere

r’s

Sto

ne

11/1

7/20

0115

,953

,113

Low

Med

ium

Har

ryP

otte

ran

dth

eC

ham

ber

ofS

ecre

ts11

/16/

2002

15,2

07,8

29M

ediu

mL

owA

ust

inP

ower

sin

Gol

dm

embe

r7/

27/2

002

12,5

76,5

92L

owH

igh

5H

arry

Pot

ter

and

the

Pri

son

erof

Azk

aban

6/5/

2004

15,0

86,5

32M

ediu

mM

ediu

mX

2:X

-Men

Un

ited

5/3/

2003

14,1

88,8

45M

ediu

mH

igh

Sta

rW

ars

2:A

ttac

kof

the

Clo

nes

5/18

/200

213

,774

,151

Med

ium

Hig

h6

Spi

der

-Man

5/4/

2002

19,7

66,6

28M

ediu

mH

igh

Spi

der

-Man

27/

3/20

0414

,195

,850

Med

ium

Hig

hP

lan

etof

the

Ape

s7/

28/2

001

12,2

97,2

62M

ediu

mM

ediu

m7

Mat

rix

Rel

oad

ed5/

17/2

003

15,2

19,6

37M

ediu

mH

igh

Los

tW

orld

:Ju

rass

icP

ark

5/24

/199

714

,255

,579

Med

ium

Hig

hM

um

my

Ret

urn

s5/

5/20

0112

,467

,726

Med

ium

Med

ium


TA

BL

EA

.1( C

ON

TIN

UE

D)

Wee

ken

dM

PA

AL

ikin

gV

iole

nce

Wee

ken

dth

eate

rvi

olen

ceby

you

ng

rati

ng

Tit

leof

bloc

kbu

ster

date

audi

ence

rati

ng

mal

es(1

)(2

)(3

)(4

)(5

)(6

)

8Ju

rass

icP

ark

37/

21/2

001

9,10

4,50

5M

ediu

mH

igh

Sca

ryM

ovie

7/8/

2000

8,24

0,15

7M

ediu

mH

igh

Scr

eam

212

/13/

1997

8,18

8,45

4H

igh

Hig

h9

Bad

Boy

s2

7/19

/200

37,

715,

185

Hig

hH

igh

Sav

ing

Pri

vate

Rya

n7/

25/1

998

6,50

0,63

9M

ediu

mH

igh

Sle

epy

Hol

low

11/2

0/19

995,

751,

378

Hig

hM

ediu

m10

Pas

sion

ofth

eC

hri

st2/

28/2

004

13,4

84,4

02H

igh

Med

ium

Han

nib

al2/

10/2

001

10,1

14,1

35H

igh

Hig

hP

assi

onof

the

Ch

rist

3/6/

2004

8,53

1,67

3H

igh

Med

ium

Mis

sin

gA

Per

fect

Mu

rder

6/6/

1998

3,54

5,84

2M

ediu

mM

issi

ng

AP

erfe

ctM

urd

er6/

13/1

998

2,40

4,99

4M

ediu

mM

issi

ng

AC

ind

erel

laS

tory

7/17

/200

42,

207,

419

Low

Low

Not

es.T

he

audi

ence

nu

mbe

rsar

eob

tain

edfr

omda

ily

box-

offi

cere

ven

ue

divi

ded

byth

eav

erag

epr

ice

per

tick

et.T

he

rati

ngs

ofm

ovie

viol

ence

inco

lum

n(1

)ar

efr

omw

ww

.kid

s-in

-min

d.co

m.T

he

nex

tth

ree

colu

mn

sre

port

the

titl

e(c

olu

mn

(2))

,th

ew

eeke

nd

(col

um

n(3

)),a

nd

the

wee

ken

dau

dien

cesi

ze(c

olu

mn

(4))

for

the

thre

em

ovie

sw

ith

hig

hes

tw

eeke

nd

sale

sin

agi

ven

viol

ence

cate

gory

.Col

um

n(5

)re

port

san

alte

rnat

ive

viol

ence

rati

ng

usi

ng

MP

AA

desc

ript

ion

s,an

dco

lum

n(6

)re

port

sa

mea

sure

ofh

owli

ked

the

mov

ieis

byyo

un

gm

ales

usi

ng

IMD

bm

ovie

rati

ngs

.Th

em

easu

res

use

din

colu

mn

s(5

)an

d(6

)ar

ede

scri

bed

inde

tail

inth

ete

xt.


Center and available from ftp://ftp.ncdc.noaa.gov/pub/data/gsod.Weather information is collected for the capital of each state inour sample (except Kentucky, where Lexington is used because ofdata issues). We then average these variables, using as weightsthe state-year-specific NIBRS population. The variables used aremaximum and minimum daily temperature measured in Fahren-heit, heat index, wind speed measured in knots (in Beaufort scale),rainfall, and snow. Before averaging, the variables are categorizedas dummy variables for the maximum daily temperature fallingin one of three categories (>80 and ≤90, >90 and ≤100, >100),the minimum daily temperature falling in one of three categories(≤10, >10 and ≤20, >20 and ≤32), the heat index falling in oneof three categories (>100 and ≤115, >115 and ≤130, >130), thewind speed falling in one of two categories (>17 and ≤21, >21),any rain, and any snow.

APPENDIX II: INSTRUMENTS

A. Benchmark Instrument

Our set of instruments uses information on the followingweekend’s audience for the same movie to predict movie atten-dance, and then aggregates these predictors across all moviesof a given violence level. The procedure is similar to the impu-tation procedure described in Appendix I. We assume the dailyaudience of movie j on day t, aj,t, is a share of the weekendaudience in the same week w(t), where the share is allowed todepend on a set of controls. In addition, we assume that the week-end audience decays each week at a rate that is also a functionof the controls. This specification allows the decay rate to varyby weekday and differentially so for different types of movies.We use the same controls (including interactions with day ofweek) as for the imputation procedure described in AppendixI with three differences: (i) the indicators for audience size re-fer to week w(t) + 1 (as opposed to week w(t)); (ii) we add twoindicators for slow releases, that is, indicators for the cases inwhich the weekend audience for week w(t) is less than 3 and lessthan 5 times smaller than in week w(t) + 1; (iii) we add 365 day-of-year indicators ηd(t) (not interacted with day of week). As inAppendix I, we estimate a log model, with ln(aj,t) − ln(aw

j,w(t)+1)as the dependent variable. The regression uses observations withnonimputed movie audience and is weighted by next weekend’s


audience awj,w(t)+1. We obtain the predicted daily audience using

aj,t = exp[ln(awj,w(t)+1) + ln(aj,t) − ln(aw

j,w(t)+1)]. To generate the pre-dicted audiences An

t , Amt , and Av

t , we aggregate across movies inthe relevant violence category.

We note that a coarser, but simpler, approach is to use asinstruments the audience in week w(t) + 1 of all movies in a cate-gory (strongly violent, mildly violent, and nonviolent). The empir-ical results using this approach are similar, although somewhatnoisier (see Online Appendix Table 1).

B. Instrument for DVD/VHS Rentals

The instrument for DVD and VHS rentals is constructed sim-ilarly to the benchmark instrument, except that Video Store Mag-azine only publishes the DVD and VHS rental at the weekly level.Hence, we estimate the equivalent of the predictive specificationfor the benchmark instrument, but without day-of-week dummiesand day-of-week interaction variables. The regression is weightedby the next week’s rentals aw

j,w(t)+1. The set of controls, as for thestandard instrument, includes month indicators, a linear timetrend, indicators for the level of violence, indicators for ratingtype, and indicators for rentals in week w (t) + 1. The holiday con-trols are separate indicators for whether the week w (t) includesany of the holidays described in Appendix I, and whether the weekw (t) + 1 includes any of these holidays. The predicted values fromthe regressions are used to generate the predicted weekly rentalsaj,t. These predicted rentals are then apportioned to each day ofweek using the within-week shares of rentals from the CEX timediaries.

C. Theaters and Budget Instrument

The estimates in column (2) of Table V use instruments basedon the number of theater screens on which a movie plays andits production budget (Moretti 2008). We use data from the-numbers.com and renormalize the number of screens and bud-get by the corresponding 90th percentile of each variable for thatyear. We use the number of screens in levels and take the log ofproduction budget (setting it equal to zero for missing productionbudgets and adding an indicator variable for missing productionbudgets). Because the predictability of audience using number ofscreens and budget varies with both the weekday and the num-ber of weeks a movie has been out, we interact these screen and


budget variables with indicators for day of week as well as numberof weeks out (0 weeks, 1 week, 2–4 weeks, 5–9 weeks, 10–19 weeks,20–26 weeks, >26). We estimate a log model, with ln(aj,t) as thedependent variable, using observations with nonimputed movieaudience and weighting by the number of screens next week. Theset of controls is the same as for the standard instrument, exceptthat we do not use information on the audience next week.

UC SAN DIEGO AND NBERUC BERKELEY AND NBER

REFERENCES

American Academy of Pediatrics, American Academy of Child and AdolescentPsychiatry, American Medical Association, American Psychological Associ-ation, American Academy of Family Physicians, and American PsychiatricAssociation, Joint Statement on the Impact of Entertainment Violence onChildren, Congressional Public Health Summit, July 26, 2000. Available athttp://www.aap.org/advocacy/releases/jstmtevc.htm.

Anderson, Craig, Leonard Berkowitz, Edward Donnerstein, L. Rowell Huesmann,James D. Johnson, Daniele Linz, Neil M. Malamut, and Ellen Wartella, “TheInfluence of Media Violence on Youth,” Psychological Science in the PublicInterest, 4 (2003), 81–110.

Anderson, Craig A., and Brad J. Bushman, “Effects of Violent Video Games onAggressive Behavior, Aggressive Cognition, Aggressive Affect, PhysiologicalArousal, and Prosocial Behavior: A Meta-analytic Review of the ScientificLiterature,” Psychological Science, 12 (2001), 353–359.

Ariely, Dan, and George Loewenstein, “The Heat of the Moment: The Effect ofSexual Arousal on Sexual Decision Making,” Journal of Behavioral DecisionMaking, 18 (2005), 1–12.

Bandura, Albert, Dorothea Ross, and Sheila A. Ross, “Imitation of Film-MediatedAggressive Models,” Journal of Abnormal and Social Psychology, 66 (1963),3–11.

Besley, Timothy, and Robin Burgess, “The Political Economy of GovernmentResponsiveness: Theory and Evidence from India,” Quarterly Journal ofEconomics, 117 (2002), 1415–1451.

Buschman, Brad J., “Moderating Role of Trait Aggressiveness in the Effects ofViolent Media on Aggression,” Journal of Personality and Social Psychology,69 (1995), 950–960.

Card, David, and Gordon Dahl, “The Impact of Emotional Cues on Domestic Vio-lence,” Rochester Center for Economic Research Working Paper No. 546, 2009.

Carpenter, Christopher, and Dobkin, Carlos, “The Effect of Alcohol Consumptionon Mortality: Regression Discontinuity Evidence from the Minimum DrinkingAge,” American Economic Journal: Applied Economics, 1 (2009), 16, 482.

DellaVigna, Stefano, and Ethan Kaplan, “The Fox News Effect: Media Bias andVoting,” Quarterly Journal of Economics, 122 (2007), 1187–1234.

Einav, Liran, “Seasonality in the U.S. Motion Picture Industry,” RAND Journal ofEconomics, 38 (2007), 127–145.

Federal Trade Commission, Marketing Violent Entertainment to Children: AReview of Self-Regulation and Industry Practices in the Motion Picture,Music Recording, and Electronic Game Industries, 2000. Available online athttp://www.ftc.gov/reports/violence/vioreport.pdf.

Gentzkow, Matthew, “Television and Voter Turnout,” Quarterly Journal ofEconomics, 121 (2006), 931–972.

Gentzkow, Matthew, and Jesse Shapiro, “Preschool Television Viewing and Ado-lescent Test Scores: Historical Evidence from the Coleman Study,” QuarterlyJournal of Economics, 123 (2008), 279–323.


Jacob, Brian, and Lars Lefgren, “Are Idle Hands the Devil’s Workshop? Incapac-itation, Concentration and Juvenile Crime,” American Economic Review, 93(2003), 1560–1577.

Jacob, Brian, Lars Lefgren, and Enrico Moretti, “The Dynamics of CriminalBehavior: Evidence from Weather Shocks,” Journal of Human Resources, 42(2007), 489–527.

Johnson, Jeffrey G., Patricia Cohen, Elizabeth M. Smailes, Stephanie Kasen,and Judith S. Brook, “Television Viewing and Aggressive Behavior duringAdolescence and Adulthood,” Science, 295 (2002), 2468–2471.

Lazear, Edward, Ulrike Malmendier, and Roberto Weber, “Sorting in Experimentswith Application to Social Preferences,” NBER Working Paper No. 12041,2006.

Levitt, Steven, “The Effect of Prison Population Size on Crime Rates: Evidencefrom Prison Overcrowding Litigation,” Quarterly Journal of Economics, 111(1996), 319–352.

Levitt, Steven, and John List, “What Do Laboratory Experiments MeasuringSocial Preferences Tell Us about the Real World?” Journal of EconomicPerspectives, 21 (2007), 153–174.

Loewenstein, George, and Jennifer Lerner, “The Role of Affect in Decision Mak-ing,” in Handbook of Affective Sciences, Richard J. Dickson, Klaus R. Scherer,and H. Hill Goldsmith, eds. (Oxford, UK: Oxford University Press, 2003).

Miller, Ted, Mark Cohen, and Brian Wiersema, Victim Costs and Conse-quences: A New Look, Research report prepared for National Institute ofJustice, U.S. Department of Justice, 1996. Available at http://www.ncjrs.gov/pdffiles/victcost.pdf.

Moretti, Enrico, “Social Learning and Peer Effects in Consumption: Evidencefrom Movie Sales,” NBER Working Paper No. 13832, 2008.

Rees, Daniel, and Kevin Schnepel, “College Football Games and Crime,” Journalof Sports Economics, 10 (2009), 68–87.

Stromberg, David, “Radio Impact on Public Spending,” Quarterly Journal ofEconomics, 119 (2004), 189–221.

DOES MOVIE VIOLENCE INCREASE VIOLENT CRIME?gdahl/papers/movies-and-violence.pdf · DOES MOVIE...

Documents

Transcript of DOES MOVIE VIOLENCE INCREASE VIOLENT CRIME?gdahl/papers/movies-and-violence.pdf · DOES MOVIE...