Friday, May 30, 2014

Volient Videogames decrease crime


ABSTRACT: Psychological studies predominately find a positive relationship between
violent video game play and aggression. However, these studies cannot account for either
aggressive effects of alternative activities video game playing substitutes for or the possible
selection of relatively violent people into playing violent video games. That is, they lack
external validity. We investigate the relationship between the prevalence of violent video
games and violent crimes. Our results are consistent with two opposing effects. First, they
support the behavioral effects as in the psychological studies. Second, they suggest a larger
voluntary incapacitation effect in which playing either violent or non-violent games decrease
crimes. Overall, violent video games lead to decreases in violent crime.


1. Introduction
From the sensational crime stories of the 19th century (Comstock and Buckly 1883), to
the garish comic books of the early 20 th century, (Hadju 2009), to today’s violent video
games, Americans have made efforts to reduce children’s access to violent media because of
concerns over their social costs. These concerns may not be unfounded as numerous studies
purport to find that violent media of all sorts, including games, can cause increases in
measured aggression. Aided in part by mounting evidence that violent video game play cause
aggression, states have passed legislation criminalizing the distribution of violent video games
to minors. 1
The research is not clear on how large the increase in aggression caused by these
games. Craig Anderson, a long-time researcher in the effect of violent media on aggression
has contended that "one possible contributing factor [to the Columbine High School killings
was the shooters’ habits of playing] violent video games. [The shooters] enjoyed playing the
bloody shoot-`em-up video game Doom, a game licensed by the U.S. Army to train soldiers to
effectively kill" (quoted in Kutner and Olson 2009). 2
1 In 2010, California passed a law making it a punishable offense for a distributor to sell a banned violent video to a minor. The US Supreme Court struck down this law in June, 2011.
2 In the opening paragraph of his literature review, Anderson (2004) suggested violent video games were
responsible for the recent wave of school shootings since the late 1990s.
3
If violent video games can be shown to cause violence, then laws aimed at reducing
access may benefit society at large. Yet to date, though there is ample evidence that violent
video games cause aggression in a laboratory setting, laboratory stings cannot address issues
of selection or incapacitation. Ward (2010) shows that adolescents who are otherwise
predisposed to violence tend to select into video game play. Dahl and Dellavegna (2009)
suggest that violent movies incapacitate violent crime offenders. Likewise, since the hours it
takes to "beat the game" substitute for some other activity, a complete analysis of video game
effects must consider the opportunity cost of this time. Violence may fall because violent
people are attracted to violent games and because gamers engaged in virtual violence are not
simultaneously engaged in actual violence.
To date, there is no evidence that violent video games cause violence or crime. In fact,
two recently published studies analyzed the effect of violent media (movies and video game
stores) on crime, and found increased exposure may have caused crime rates to decrease
(Dahl and Dellavegna 2009; Ward 2011). These studies, unlike the laboratory studies, were
conducted with observational data, which poses unique scientific challenge to establishing
causality. However, since laboratory studies have never shown that video game violence
causes crime or violence, despite researchers out-of-sample predictions (Anderson 2004),
observational studies may be the only ethical and practical way to test for such a causal effect.
To many in this field, it is logical to assume that if exposure to violent media causes
aggression in the lab, it will therefore cause aggression when exposure occurs non-randomly
outside the laboratory. Psychologists have adapted the general aggression model, or GAM, to
the video game setting (Bushman and Anderson, 2002 and Anderson and Bushman, 2002).
GAM hypothesizes that violent media, including violent video games, increases a person’s
aggressive tendencies through a process of social learning that occurs simultaneous to the
exposure itself. Violent media causes the person to mistakenly develop certain scripts, or rules
4
of thumb, that are used to interpret social situations both before they occur, as well as
afterwards. GAM posits, in other words, that violent video games cause aggression by biasing
individuals towards forming incorrect beliefs about relative danger that they are in. Perception
biases towards hostility, therefore, can in turn cause the person to respond in either a “fight or
flight” fashion. It may also permanently alter a person’s point of view, creating an aggressive
personality as an outcome (Bushman and Anderson 2002). A variant of the “rational
addiction” model (Becker and Murphy 1988) may be a fair representation of GAM. The key
insight for GAM is that consumption of a good in one period not only affects current utility
directly but, through a capital stock accumulation mechanism, it also affects future utility
indirectly.
The opportunity cost of playing a video game is not just pecuniary but also includes
lost time. In fact, for many gamers, the value of the time spent playing a game may be worth
much more than the pecuniary cost of the game. This time spent gaming cannot be spent on
other activities, legitimate or otherwise, if time use is rival in consumption. The substitution
patterns from video games may derive more from time use effects than from pecuniary costs
(Becker, 1965). Evidence for video games having a time use component can be found in
Stinebrickner and Stinebrickner (2008). The authors identified a causal effect of studying on
academic performance by utilizing the random assignment of college students to roommates
with a video game console, relative to the counterfactual, which caused students to study less
often, and in turn, to perform worse in school
In this paper, we argue that since laboratory experiments have not examined the time
use effects of video games, which incapacitate violent activity by drawing individual gamers
into extended gameplay, laboratory studies may be poor predictors of the net effects of violent
video games in society. Consequently, they overstate the importance of video game induced
aggression as a social cost. We argue that since both aggression and time use are a
5
consequence of playing violent video games, then the policy relevance of violent video game
regulation depends critically on the degree to which the one outweighs the other. If, as we
find in our study, the time use effect of violent video games reduces crime by more than the
aggression effects increase it, then the case for regulatory intervention becomes weaker.
While some early work has been done on the long-term effects of video game play, nearly all
the laboratory evidence that currently exists has only uncovered very short-term effects,
which is when time use effects could be the most important. 3
As with Dahl and Dellavegna (2009) and Ward (2011), we use a proxy for individuals’
exposure to violent video games – the volume of sales of violent video games in a week
among the top 50 best-selling video games from 2005-2008 – and relate it to a marker for
violent behaviors – weekly aggregate violent crime incidents from the National Incident
Based Reporting System (NIBRS). Using time series modeling, as well as an instrumental
variables approach, we estimate the effect of an increased weekly volume of violent video
game sales on the number of criminal incidents recorded to law enforcement over the
subsequent weeks and find that increased violent video games are associated with decreases
in crime rates, similar to Dahl and Dellavegna (2009) and Ward (2011).
One advantage of our approach is that we can attempt to disentangle the separate
effects of both a behavioral change toward more aggression and incapacitation due to time
use. Our results provide some support for the psychological finding that, absent
incapacitation, violent video games lead to more aggression as measured by violent crimes.
However, our results also suggest that this is dominated by possible incapacitation and
selection effects leading to a net reduction in violent crimes. This approach can help guide
3 In Anderson (2004), the author notes the glaring omission of longitudinal studies of effects of violent video
games on aggression in his conclusions on the state of the research, calling for more studies aimed at
investigating the long-term effects. If nothing else, though, this makes our point that the abundance of evidence
that we know does exist only speaks to short-term effects of violent video games on aggression, which is the
purpose of this study here.
6
investigators to develop more holistic research designs, such as field experimentation and
other quasi-experimental methodologies, to determine whether the net social costs of violent
games are non-trivial. The main shortcoming of our approach is due to the limitations of our
data on game sales. Unfortunately, the industry does not report cross-sectional variation in
game sales – only the national weekly sales of the top 50 highest grossing games are
available. As a result, our paper follows a methodology similar to Dahl and Dellavegna
(2009), who estimated the impact of violent movies, proxied by daily ticket sales, on crime
using only time series methods.
The paper is structured as follows: the second section our data and methodology. The
third presents and discusses our results. We conclude with a brief discussion of the
implications for public policy.
III. Data and Methodology
Randomized assignment of a treatment with comparison groups used to make
comparative counterfactuals is widely considered the “gold standard” in the social sciences
(Fisher 1935; Campbell and Stanley 1963; Rosenbaum 2002). Yet, it is widely known that
experimentalism may fail to identify true causal effects for a variety of reasons (Berk 2005;
Deaton, 2010; Heckman and Urzua, 2010; Imbens 2010). While others have noted the failure
of researchers in this literature to satisfy the rigorous conditions for establishing causality
(Ferguson and Kilburn 2008; Olson and Kuttner 2009) our study will focus on a separate
statistical challenge not mentioned in these earlier studies: the challenge of internal versus
external validity.
Finding of a positive effect of violent games on aggression does not therefore mean
that violent video games playe will cause crime if the incapacitation effects from time use
7
swamp the marginal increase in aggression in the person. By design, laboratory studies – both
by ignoring alternative time use and by treating both treatment and control groups with this
separate effect – cannot be used to guide researchers as to what expect outside the lab. In this
sense, the studies have internal validity, but may not have external validity on the incidence
of socially costly aggression from violent video game play (Campbell and Stanley 1963).
Quasi-experimental methods, such as panel econometric methods, regression discontinuity
and instrumental variables, as well as field experimentation (Harrison and List 2004; Angrist
2006) may be more suitable estimating the social costs of violent video games since they
allow for the estimation of all known and unknown theoretical mechanisms. In this section,
we explain our research design and the data used to overcome some of the limitations of a
purely experimental methodology.
A. Empirical Methodology
The models of video game violence suggest that the effect of violent video game play on
crime will depend on whether a sizable stock of aggressive tendencies accumulates and on the
games’ time use intensities.
Since the theoretical predictions are ambiguous and the policy relevance of the
laboratory studies is unclear, empirical work outside of a laboratory context is warranted.
However, without experimental data, causal inference is problematic. Correlations between
video game play and crime may or may not reflect a causal relationship if the unobserved
determinants of crime are correlated with the determinants of video game play. For instance,
bad weather such as rain or heavy snow which causes individuals to remain at home would
both increase the likelihood of playing video games and decrease the returns to crime through
higher chances of finding a resident at home. Hence, negative correlations between crime and
violent video game play could purely be a consequence of omitted variable bias. Similarly,
video game publishers could strategically release violent video games during periods of time
8
when gamers have a lower value of time. But a low opportunity cost of time would affect both
video game sales and crime. For example, both video game sales and the crime rate increase
during summer when most teenagers are out of school..
One solution to omitted variable bias when there is time-variant heterogeneity is to
employ instrumental variables (IVs). The researcher must have instruments that are strongly
correlated with individual game play but uncorrelated with the determinants of crime. This
approach exploits exogenous variation in video game play that is not due merely to changes in
the determinants of crime providing greater assurance that the estimated effect is causal. We
use the ratings of video games by a video games rating agency as IVs. Our IV strategy
exploits the variation in game sales correlated only with the variation in quality, and thus is
mostly free of variation due to factors related to crime.
Zhu and Zhang (2010) show that consumer reviews of video games are positively
related to game sales. Ratings are valuable pieces of information for video games because
games are complex experience goods for which gamers cannot know their preferences
without playing. Our data on professional ratings contain rich information that communicates
the kinds of information that gamers value in forecasting their beliefs about the game, and as
beliefs and anticipation are drivers of the game sales, we would expect these rating
institutions to play important roles in forming consumer prior beliefs about the game and
therefore their purchases. But we also have some evidence from other industries that would
suggest scores would independently cause purchases to rise, independent of the unobserved
factors that cause expert opinion and purchases to be highly correlated. Reinstein and Snyder
(2005) used exogenous variation in Siskel and Ebert movie ratings due to disruptions in their
pair’s reviewing to determine a causal effect on movie demand. More recently, Hilger Rafert
and Villas-Boas (2010) found that randomly assigned expert scores on bottles of wine in a
retail grocery store caused an increase in sales for the higher rated, but less expensive, wines.
9
While these studies do not confirm that there are exogenous forces in video game ratings that
drive consumer purchases, they are suggestive.
We begin by estimating a standard multivariate regression model of the incidence of
various crimes as functions of, among other controls, the prevalence of non-violent and
violent video games. Our outcome variables of interest, C t , are the total number of reported
criminal incidents in week t as well as the number of such incidents that are classified as
violent. While one might interpret any criminal incident as reflecting some level of
aggression, we interpret violent crimes as reflecting more aggression. While the dataset we
use documents criminal offenses on a daily basis, since the video game sales data are
available only on a weekly basis, we aggregate crimes into weekly measures to focus on
same-week exposure. Accordingly, we employ a simple least squares estimator so as to more
easily instrument for video game exposure. 6
Our main explanatory variables are aggregated current and lagged values of weekly
sales volumes for both non-violent and violent video games. Video games appear to
depreciate quickly with use. This may be because new games are played intensively for a few
weeks after purchase and are not replaced with a new game until after some diminishing
returns have been reached, or it may suggest that firms typically stagger the release dates of
games. We measure the cumulative effect of games with the sales volume of the current
week’s sales, along with the various lags of previous weeks’ sales, so as to capture the effect
of higher volume of gameplay with an unknown time lag to trigger crime.
Our benchmark specification is:
( ) ∑
[ (
)]
[ (
)]
where L  is the lag operator of length  . The number of crime incidents depends on the
exposure to violent video games
and non-violent games . The sum over  of can be
6 Our empirical methodology is in large part based on Dellavegna and Dahl’s (2008) study of the effect of movie
violence on crime.
10
interpreted as the cumulative percentage increase over the  weeks in criminal incidents for
each percent increase in violent video games sold in week t while the similar sum for
can
be similarly interpreted for non-violent video games. The trend and month dummies attempt
to account for secular increases and seasonality in video game purchases. The identification of
the parameters is based on the time-series variation in the style of violence in the video
games. Again, we instrument for both types of games using average quality ratings of the
games on the market that week.
The measured effect from this specification can represent a confluence of multiple
effects. It is possible for there to be a positive behavioral effect, as found in the laboratory,
and a negative voluntary incapacitation effect. This specification will typically only measure
the net effect. However, it may be possible to disentangle the behavioral effects from the
incapacitation effect from the estimated cumulative effects from non-violent and violent
games. Both should incorporate incapacitation effects but only the former will include a
behavioral effect toward aggression. The difference between the two provides an estimate of a
pure aggression effect.
Besides the benchmark specification we employ two additional specifications as
robustness checks. These specifications identify specific segments of the population and
locations where we expect a differential a gaming-to-violence link, e.g. crimes committed by
teens and young adults and those committed at high school and college campuses. For each
crime incident, NIRBS provides information on the age of the offender and on the location of
the incident. In the first robustness check, we select a sun-sample of offenders aged between
15 and 30 years and compare these results to the results obtained from a sub-sample of
offenders who are 35 to 50 years old. In our second check, we extend our estimation
procedure to compare the effects on the number of incidents reported on school campuses to
the number committed at other locations.
11
B. Video Game Sales Data
Our treatment variables for video game play are derived from the volume of video
game unit sales data from VGChartz 7 . Beginning consistently in 2005, this site has provided
unit sales volume information for each of the top 50 selling video console based games each
week. Among other information, volumes are reported worldwide as well as for several
geographical areas including USA, Japan, Europe, Middle East, Africa or Asia. In our sample
period 2005 to 2008 the VGChartz dataset contains 1,091 different titles over the 208 weeks
for the US with some of these titles being the same game for different gaming consoles. In
sum, the games are provided from 47 different publishers and designed for nine different
gaming consoles. While VGChartz includes the top 50 selling games each week, it only
covers a portion of all sales in the US video game market. A game’s week of release is almost
always its top selling week. Figure 1 indicates that most games stay in the top 50 for only a
few weeks. Moreover, as Figure 2 indicates, the top selling games sell much more than even
the lower ranked top 50 games. These features suggest that there is considerable week-to-
week variation in the games, and the types of games, being played. According to the
Entertainment Software Association (ESA) 8 VGChartz account for about one-quarter of all
units in 2005 (ESA Annual Report, 2010). The ESA also includes sales of non-console based
games such as computer and smartphone games. Still, this fraction rises to almost one-half in
2008.
Our measure of violent videogame content stems from the Entertainment Software
Rating Board (ESRB). 9 This non-profit body independently assigns a technical rating (E, E10,
T, M, and A) which defines the audience the game is appropriate for where E classifies games
for everybody, E10 for everyone aged 10 and up, T for teens, M games for a mature audience,
and A for adult content. In addition, ESRB provides detailed description of the content in each
7 http://www.vgchartz.com/
8 http://www.theesa.com – The reported numbers from ESA also include games for personal computers which
amount to about 10 percent of the market each year and are intentionally not included in VGChartz.
9 http://www.esrb.org
12
game on which the rating was made, including the style of violence, e. g. language, violence,
or adult themes. For all of the 1,091 titles in our sample we collected the appropriate ESRB-
rating and all content descriptors. Based on this content information we identify 762 non-
violent and 329 violent games, of which 105 titles are described as intensely violent. Almost
all violent games are rated T or M. All intensely violent games are rated M. Since most of the
policy concern stems from these intensely violent games, these are the games we concentrate
on. 10 Merging both data sources together we can construct measures of the aggregate unit
sales of non-violent and intensely violent video games for each week. The weekly sales are
pictured in Figure 3 for all games and for intensely violent games. Overall, the two graphs
follow a similar pattern with a peak around the Christmas gift purchasing period. In the mid of
2008, however, the intense violent games seem to account for almost all sales of the violent
games.
As argued above the prevalence of video games in a week is not randomly distributed
over the sample and therefore may be endogenous. For instance, if changing economic
conditions that caused a rise in unemployment, and in turn crime rates, may also have caused
leisure activities like video games to rise, then we might observe positive correlations
between video game play and crime that is driven purely by these changing economic factors
(Raphael and Winter-Ebmer 2001; Gould, Weinberg and Mustard 2002). We address the
potential endogeneity of video games with instrumental variables using expert review of each
title as an instrument for purchases.
Our expert review data comes from the GameSpot website. 11 GameSpot provides
news, reviews, previews, downloads and other information for video games. Launched in
May 1996 GameSpot’s main page has links to the latest news, reviews, previews and portals
10 We have also performed our analysis for the broader “violent” and “intensely violent” definition of a violent
game. Qualitatively, all of our general results described below hold however parameter estimates are smaller (in
absolute value terms) and are less precisely estimated.
11 http://www.gamespot.com
13
for all current platforms. It also includes a list of the most popular games on the site and a
search engine for users to track down games of interest. The GameSpot staff reviewed all but
a handful of the games in our sample and rated the quality of the titles on a scale from 1 to 10
with 10 being the best possible rank. These so-called GameSpot-scores assigned to each game
are intended to provide an at-a-glance sense of the overall quality of the game. The overall
rating is based on evaluations of graphics, sound, gameplay, replay value and reviewer’s tilt.
A possible issue with this measure is that GameSpot changed the rating system in mid of 2007
to employ guidelines and a philosophy focusing more on a prospective customer rather than a
hardcore-fan that the reviewers had focused on before. Nevertheless, the five mentioned
aspects are essential parts of a game that are still reviewed in detail by a GameSpot reviewer
but will not get an aspect-specific rating score anymore. We do not consider this change in the
GameSpot focus to noticeably affect the overall GameSpot-score.
We expect the quality rating of the games to be positively correlated with their sales as
better-rated games usually are more highly demanded. It is possible that some games have the
opposite relationship if they are based on a popular tie-in from a movie, e. g. Harry Potter, or
sequels, e. g. the Final Fantasy series. Developers know that these games will sell well due to
their popular tie-in which may lower the returns to investment in game quality. However, in
table 2 we show that, a game title’s weekly sales are positively related to the Game Spot score
for games of different violence profiles.
C. Crime Data
For our measure of weekly crime, we used the National Incident Based Reporting
System (NIBRS). NIBRS is a federal data collection program begun by the Bureau of Justice
Statistics in 1991 for gathering and distributing detailed information on criminal incidents for
participating jurisdictions and agencies. Participating agencies and states submit detailed
information about criminal incidents not contained in other data sets, such as the Uniform
Crime Reports. For instance, whereas the Uniform Crime Reports contain information on all
14
arrests and cleared offenses for the eight Index crimes, NIBRS consists of individual incident
records for all eight index crimes and the 38 other offenses (Part II offenses) at the calendar
date and hourly level (Rantala and Edwards 2001). A potential drawback of NIBRS is that
many law enforcement agencies do not participate. [Need a sentence or two here Scott] We
aggregate across only the jurisdictions that participated during each of our sample years.
Because of the detailed information about the incident, including the precise time and date of
the incident, economists such as Dahl and Dellavegna (2009), Card and Dahl (2009), Jacob
and Moretti (2003) and Lefgren, Jacobs and Moretti (2007) have used it for event studies. In
our case, we exploit detailed information about the age of offenders and the crime’s location –
on school campuses or not – for our robustness checks.
Crimes follow a seasonal pattern. Figure 4 indicates a consistent pattern of gradual
increases in both violent and non-violent crimes from winter to summer. Our method was
developed to account for seasonality in both of our main variables of interest crime and
games. Much of the seasonality in crimes is believed to be due to weather while seasonality in
games is likely due to holiday gift giving (Lefgren, Jacobs and Moretti 2007). Failure to
address this will likely lead to spurious correlations. As indicated above, we accommodate
this in two ways. First, month dummy variables should capture much of the seasonality.
Second, using Game Spot scores as IVs should isolate the variation in game sales due to game
quality.
Our final sample includes 208 weekly observations on video games sales and crimes
from early 2005 through 2008. However, eight observations are excluded from final
regressions because of the use of lagged video game sales. Table 3 reports basic descriptive
statistics for our sample.
IV. Results
A. Basic Results
15
Our basic regression results are presented in Tables 4 and 5. Table 4 reports estimates
of specifications for various lags of the effect video games sales on all crimes. Video games
are separated between those that the ESRB rated as “intensely violent” and those that are not.
Recall that the lesser rating of merely “violent” does not warrant an ESRB rating of “M.” 12
Control variables include month dummies to capture seasonality and a time trend to capture
any secular trend. The columns from left to right add more lags of video games to the
specification so as to measure possible inter-temporal effects of game purchases in one week
affecting crime in subsequent weeks through continued play. Finally, each regression employs
a 2SLS estimator with the same set of current and eight lags of Game Spot scores averaged
over intensely violent games and over games that are not intensely violent. Since the
specifications are over-identified, we test for possible endogeneity of the instrument set. As
expected, in all cases, we fail to reject the exogeneity of Game Spot scores with respect to the
level of crime. 13
The estimated effect of video games sales in any single week is small. Most individual
coefficient estimates are negative but few are significantly different from zero. It appears that
lags of up to five weeks of video game sales may be associated with current crime. It is not
clear from this table whether violent games have a different effect from those that are not
violent. For ease of comparison, we report the sum of the coefficients for various lags for both
in the top panel of Table 6 to calculate the cumulative effect of a change in video games over
time. Here it becomes clearer that video games are estimated to have an overall negative
effect on crime for specifications that include from two to six lags. That is, both violent and
non-violent games are associated with reductions in crimes. However, the effect is small.
Since our specification is double log, these estimates can be interpreted as elasticities with
12 Unreported regressions comparing games that are either “intensely violent” or “violent” versus all other games
generally yield much less precisely estimated parameters.
13 Estimates assuming that game sales are exogenously determined typically generated smaller (in absolute value
terms) and much less precisely estimated coefficients.
16
values of up to -0.025 for non-violent games and -0.010 for violent games. These estimates
suggest that, over all the mechanisms through which videogame play can affect crime, the net
effect is to reduce crime.
As mentioned above, these estimates may also allow us to make some inferences that
distinguish between potential mechanisms. While both violent and non-violent games are
hypothesized to have incapacitation effects, only violent games are hypothesized to alter
behaviors. Indeed, the top panel of Table 6 indicates that the difference in effects between
violent and non-violent games is for violent games to reduce crime by a smaller amount and
that this difference is statistically significant for specifications that include between one and
five lags. Moreover, it is possible that the incapacitation effect for violent games is greater
than for non-violent games, though we cannot test this hypothesis. If so, the difference of
these estimates may represent a downwardly biased estimate of a behavioral effect. This
provides some support for the laboratory findings of a reinforcing behavioral effect that
partially counterbalances the incapacitation effect.
Table 5 repeats these specifications where the dependent variable is now the log of
violent crimes. By doing so, we focus on criminal acts that clearly entail an element of
aggression. Again, we include various lags for the effects of video games and, again, more
individual estimates are negative than positive but few are significantly different from zero.
The bottom panel of Table 6 reports the aggregation of the lagged video game coefficients to
calculate the cumulative effects. From this panel we usually find an overall negative effect of
video games on the number of violent criminal incidents. These estimates are quite similar to
those for all crimes in upper panel of this table. If anything, these parameter estimates are
slightly larger (in absolute value terms) and aggregations with more specifications yield
results significantly different from zero. These estimates indicate that both violent and non-
17
violent video game play is generally associated with reductions in the number of violent
crimes.
The test for a difference in the effects for violent and non-violent games may be more
informative. There are no known previously hypothesized mechanisms through which non-
violent games would affect violent crimes. We propose that the appropriate test for violent
video games affecting violent behavior is the difference in these effects by game type. In this
case, the marginal effect of violent video games, relative to non-violent games, is to increase
violent crimes. Decomposing the two effects suggests that a one hundred percent increase in
violent video game sales implies an incapacitation effect reducing violent crime by as much
as 2.6% and an aggression effect increasing violent crimes by as much as 1.5%.
B. Age of Offender Results
A potential robustness check is to examine the effects of video games on criminal
offenders by age of offender. While the age profile of video game players is increasing, video
games are still primarily played by children, teens and younger adults. For most offenses, the
NIBRS data records information on the age of the offenders for an incident. We separately
examine the effects of video game sales on offenders aged 15-30, the prime video game
playing population, versus those 35-50, a population for which video game play is not as
popular. If our basic results were spurious and did not reflect any direct link between video
game play and criminal acts, we would have no reason to expect a differential effect by age
group. In contrast, under our hypotheses, we would expect larger effects for the younger
group.
Table 7 reports cumulative estimates for both these younger and older groups. The
specifications are otherwise identical to those reported in Table 4. However, rather than report
the individual estimates as in Tables 4 and 5, we report the estimated sums over all lags as in
18
Table 6. As before, specifications with lags from between two and five achieve some level of
statistical significance for both the young and the old. The estimated effects of both violent
and non-violent video games are both negative, as before. And, as before, violent video games
decrease crime by less than do non-violent video games. That is, there are few, if any,
qualitative differences across the two groups.
Table 8 reports cumulative estimates where the dependent variable is violent crimes,
for both these younger and older groups. The specifications are otherwise identical to those
reported in Table 5 and again we report the estimated sum of effects over all lags as in Table
6. Now, there are noticeable differences across the two groups. None of the estimates for the
older group approach traditional levels of statistical significance. In contrast, the estimates for
the younger group are generally larger (in absolute value) and many are statistically
significant. In addition, the differences in estimates between violent and nonviolent games are
often statistically significant. We again find that, for the younger group, non-violent games, as
well as violent games, reduce the number of violent crimes. In these specifications, the
measured by the difference between the coefficients in the two rows which is about 0.06.
Thus, this is evidence that the behavioral effect of violent video games on violent behavior is
found only within the younger population that tends to play video games more intensively.
C. On Campus Results
Another potential robustness check is to distinguish between crimes committed at
schools and colleges and those committed elsewhere. Schools and colleges tend to aggregate
people who are of video game playing age. The NIBRS data record the location of each
incident as a categorical variable where one possible choice out of eleven is “school or college
campus.” One advantage of this variable over the age of offender variable is that it is recorded
for all incidents while the age of offender can be missing if no one witnessed the incident in
progress. One disadvantage is that crimes committed at schools and colleges need not be
19
committed by a member of the younger video game playing demographic, though most are.
Perhaps a bigger problem is that many of the younger video game playing population commit
crimes away from schools. Finally, since such a small number of crimes are committed on
campus, we may lose statistical power for that sub-sample while the off-campus sub-sample
will be quite similar to the overall sample.
Table 9 reports cumulative estimates for both crimes committed on campuses and
those committed off-campus. The specifications are otherwise identical to those reported in
Table 4 but we report the estimated cumulative effect over all lags as in Table 6. As before,
specifications with lags from between two and five achieve some level of statistical
significance for both the young and the old. The pattern of estimated effects for both violent
and non-violent video games is similar to before except that they are much larger for the on-
campus sample than off-campus sample. In the lower panel, the estimates are qualitatively
similar to the base results in Table 6. However, the upper panel estimates are about five times
larger. Other than the difference in magnitudes, the pattern of effects on-campus is
unchanged. There is still a negative effect for non-violent video games in columns 2-5 that we
interpret as an incapacitation effect. The estimated effect for violent video games is
statistically significantly smaller (in absolute value) and we interpret the difference as a
possible estimate of a behavioral effect of violent video games on crime for this sub-sample.
Table 10 reports cumulative estimates where the dependent variable is the number of
violent crimes, for both crimes on and off campus. The specifications are otherwise identical
to those reported in Table 2 and again we report the estimated sum of effects over all lags as
in Table 6. In this case, fewer effects are estimated to be significantly different from zero.
However, the pattern is similar to those for all crimes in table 9. The magnitudes are about
five times larger for the on-campus sub-sample relative to the off-campus sub-sample. As
20
expected, the off-campus results are more similar our basic results reported in the bottom
panel of Table 6.
V. Conclusion
Content regulation of the video game industry is usually predicated on the notion that
the industry has large and negative social costs through games’ effect on aggression. Many
researchers have argued that these games may also have caused extreme violence, such as
school shootings, because laboratory evidence has found an abundance of evidence linking
gameplay to aggression. Yet few studies before this one had examined the impact of these
games on crime, with the exception of Ward (2011) and Dahl and Dellavegna (2009).
Consistent with these studies, we find that the social costs of violent video games may be
considerably lower, or even non-existent, once one incorporates the time use effect into
analysis.
These analyses are suggestive of the hypothesis that violent video games, like all video
games, paradoxically may reduce violence while increasing the aggressiveness of individuals
by simply shifting these individuals out of alternative activities where crime is more likely to
occur. Insofar as our findings suggest that the operating mechanism by which violent
gameplay causes crime to fall is the gameplay itself, and not the violence, then regulations
should be carefully designed so as to avoid inadvertently reducing the time intensity, or the
appeal, of video games.
Our findings also suggest unique challenges to game regulations. Because GAM
proposes that the individual playing violent video games is developing, accidentally, a biased
hermeneutic towards people wherein they believe they are in danger, then the decrease in
violent outcomes that we observe in our study – the incapacitation effect from time use – may
21
be masking the long-run harm to society if these violent behaviors are developing within
gamers. This suggests that regulation aimed at reducing violent imagery and content in games
could in the long-run reduce the aggression capital stock among gamers, but potentially also
cause crime to increase in the short-run if the marginal player is being drawn out of violent
activities. This may be too costly a tradeoff, and may not pass any cost-benefit test. But
another possibility is that individuals who play games could be regularly taught to recognize
these errors in their framing of situations, which theoretically would reduce the aggressive
capital and thus reduce any negative outcome that is determined by the amount of aggression
the person has built up, without losing the short-run gains from crime reduction.
22
References
Anderson, Craig A. (2004). “An update on the effects of playing violent video games,”
Journal of Adolescence 27, 113–122.
Anderson, Craig A, Douglas A. Gentile, Katherine E. Buckley. (2007) Violent video game
effects on children and adolescents: theory, research and public policy. Oxford
University Press, 1st edition.
Anderson, C. A. and B. J. Bushman. 2002. “Human aggression.” Annual Review of
Psychology, 53:27–51.
Angrist, J. D. (2006). “Instrumental variables methods in experimental criminological
research: what, why and how.” Journal of Experimental Criminology 2: 23–44.
Becker, Gary S. (1965). “A theory of the allocation of time.” Economic Journal 75 (299), pp.
493-517.
Becker, Gary and Kevin M. Murphy (1988). “A theory of rational addiction.” The Journal of
Political Economy 96: p. 675-700.
Berk, R (2005). “Randomized experiments as the bronze standard.” Journal of Experimental
Criminology. 1:417-433.
B. J. Bushman and C. A. Anderson (2002). “Violent video games and hostile expectations: a
test of the general aggression model.” Personality and Social Psychology Bulletin 28
(12): 1679-1686.
D. T. Campbell and J. C. Stanley (1963). Experimental and quasi-experimental designs for
research. Chicago: Rand McNally.
Card, David and Gordon B. Dahl (2011). “Family violence and football: the effect of
unexpected emotional cues on violent behaviour.” Quarterly Journal of Economics
(forthcoming).
Comstock, Anthony and J. M. Buckley (1883) Traps for the young. Republished in 1967 by
Beknap Press, First Edition edition.
Dahl, Gordon and Stefano DellaVigna (2009). “Does movie violence increase violent crime?”
Quarterly Journal of Economics, 124(2) 637–675.
Deaton, A. (2010). “Instruments, randomization, and learning about development.” Journal of
Economic Literature 48: 424–455.
Ferguson, C. J. and J. Kilburn (2008). “The public health risks of media violence: a meta-
analytic review.” The Journal of Pediatrics 154(4): 759−763.
Fisher, R. A. (1935). The design of experiments. Oliver and Boyd, Edinburgh.
Gould, E. D., B. A. Weinberg and D. B. Mustard (2002). “Crime rates and local labor market
opportunities in the united states: 1979–1997.” The Review of Economics and
Statistics 84(1): 45–61
Hadju David (2009), The ten-cent plague: the great comic-book scare and how it changed
america. New York: Farrar, Strauss and Giroux.
23
Harrison, G. W. and J. A. List (2004). “Field experiments.” Journal of Economic Literature
42(4): 1013-1059.
Heckman, J. and S. Urzua (2010) “Comparing IV with structural models: what simple IV can
and cannot identify.” Journal of Econometrics 156 (1): 27-37.
Imbens, G. W. (2010). Better LATE Than Nothing: Some Comments on Deaton (2009) and
Heckman and Urzua (2009). Journal of Economic Literature 48(2): 399-423.
Hilger, James, Greg Rafert and Sofia Villas-Boas (2010). “Expert opinion and the demand for
experience goods: an experimental approach in the retail wine market.” Review of Economics
& Statistics, forthcoming.
Jacob, B. A., and E. Moretti (2003). “Are idle hands the devil's workshop? incapacitation,
concentration, and juvenile crime.” The American Economic Review 93 (5): 1560-
1577.
Jacob, B. A., L. Lefgren and E Moretti (2007). “The dynamics of criminal behavior: evidence
from weather shocks.” Journal of Human Resources 42 (3): 489–527.
Kutner, Lawrence and Cheryl Olson (2008). Grand theft childhood: the surprising truth about
violent video games and what parents can do. Simon & Schuster.
Rantala, Ramona R and Thomas J. Edwards (2000). “Effects of NIBRS on crime statistics”.
NCJ Publication 178890. US Department of Justice.
Raphael, S. and R. Winter-Ebmer (2001). “Identifying the effect of unemployment on crime.”
Journal of Law and Economics 44: 259-283.
Reinstein, D. and C. Snyder (2005). “The influence of expert reviews on consumer demand
for experience goods: A case study of movie critics.” Journal of Industrial Economics
53(1): 27–51.
Rosenbaum, P. R. (2002) Observational studies. New York: Springer.
Rosenbaum, P. R. (2002). “Covariance adjustment in randomized experiments and
observational studies.” Statistical Science 17(3): 286–304.
Stinebrickner, Ralph Todd R. Stinebrickner (2008). “The causal effect of studying on
academic performance.” The B.E. Journal of Economic Analysis & Policy: Frontiers
8(1) Article 14.
Ward, Michael R. (2010). “Video games and adolescent fighting.” Journal of Law and
Economics, 53(3) 611-628.
Ward, Michael R. (2011). “Video games and crime.” Contemporary Economic Policy. 29(2)
261-273.
Zhu, Feng and Xiaoquan (Michael) Zhang (2010). “Impact of online consumer reviews on
sales: the moderating role of product and consumer characteristics.” Journal of
Marketing, Vol. 74 (March 2010), 133–148.
24
Figure 1
25
Figure 2
26
Figure 3
27
Figure 4
28
Table 1
Unit Sales of Video Games (millions) from VGChartz and ESA
Year VGChartz ESA Pct
2005 56.7 240.7 23.6%
2006 76.2 267.8 28.5%
2007 107.0 298.2 35.9%
2008 141.3 273.5 51.7%
VGChartz from authors’ calculations and ESA from
http://www.theesa.com/facts/pdfs/VideoGames21stCentury_2010.pdf.
29
Table 2
The Effect of Game Quality (Game Spot Score) on Log Sales
All Intensely Violent Not Intensely Violent
Games Games Games
GameSpot
Score
0.0803** 0.1221** 0.0769**
(0.0060) (0.0181) (0.0065)
Week of
Release
-0.0039** -0.0081** -0.0036**
(0.0002) (0.0008) (0.0003)
Trend 0.0058** 0.0040** 0.0060**
(0.0001) (0.0003) (0.0001)
February -0.0902* -0.2169* -0.0663+
(0.0361) (0.1020) (0.0385)
March -0.0212 -0.0576 -0.0081
(0.0348) (0.0967) (0.0371)
April -0.1770** -0.3466** -0.1361**
(0.0344) (0.0945) (0.0369)
May -0.2838** -0.4069** -0.2485**
(0.0355) (0.1004) (0.0378)
June -0.1663** -0.3593** -0.1217**
(0.0363) (0.1036) (0.0386)
July -0.2251** -0.5266** -0.1732**
(0.0358) (0.1059) (0.0378)
August -0.3607** -0.6881** -0.3126**
(0.0364) (0.1151) (0.0381)
September -0.2700** -0.4117** -0.2422**
(0.0358) (0.1200) (0.0374)
October -0.1326** 0.0065 -0.1333**
(0.0365) (0.1159) (0.0383)
November 0.6122** 0.6812** 0.6051**
(0.0361) (0.1052) (0.0382)
December 1.2038** 1.1363** 1.2153**
(0.0349) (0.1073) (0.0367)
Constant -4.8503** -0.5994 -5.3472**
(0.2957) (0.8309) (0.3189)
Observations 10,648 1,345 9,303
R-squared 0.38 0.40 0.38
Standard errors in parentheses
+ significant at 10%; * significant at 5%; ** significant at 1%
30
Table 3
Summary Statistics
Variable Mean Std. Dev.
Ln All Video Game Sales 0.407 0.632
Ln Intensely Violent Video Game Sales -1.900 1.037
Ln Not Intensely Violent Video Game Sales 0.781 0.340
Average GameSpot Score 7.634 0.435
Average Intensely Violent GameSpot Score 8.546 0.646
Average Not Intensely Violent GameSpot Score 7.506 0.468
Ln All Crimes 10.889 0.085
Ln Violent Crimes 9.967 0.083
Ln All Crimes on Campuses 7.463 0.421
Ln Violent Crimes on Campuses 6.663 0.506
Ln All Crimes Not on Campuses 10.852 0.091
Ln Violent Crimes Not on Campuses 9.925 0.091
Ln All Crimes Offender Aged 15-30 9.854 0.068
Ln Violent All Crimes Offender Aged 15-30 9.360 0.084
Ln All Crimes Offender Aged 35-50 9.040 0.082
Ln Violent All Crimes Offender Aged 35-50 8.603 0.095
Descriptive statistics of the 200 observations used in later tables.
31
Table 4
The Effects of Video Game Sales on the Log of both Violent and Non-Violent Crime
(1) (2) (3) (4) (5) (6) (7)
Ln Video Game Sales
Not Intensely Violent
-0.028 0.029 0.030 0.041 0.042 0.032 0.044
(0.60) (0.50) (0.44) (0.54) (0.52) (0.40) (0.55)
Ln VG Sales Not
Intensely Violent lag 1
-0.130+ -0.110 -0.090 -0.089 -0.099 -0.088
(1.92) (1.35) (1.03) (1.03) (1.15) (1.02)
Ln VG Sales Not
Intensely Violent lag 2
-0.131+ -0.098 -0.095 -0.044 -0.040
(1.71) (1.16) (1.13) (0.50) (0.46)
Ln VG Sales Not
Intensely Violent lag 3
-0.068 -0.067 -0.064 -0.075
(0.91) (0.88) (0.86) (0.90)
Ln VG Sales Not
Intensely Violent lag 4
0.010 0.042 0.029
(0.12) (0.53) (0.35)
Ln VG Sales Not
Intensely Violent lag 5
-0.125+ -0.126+
(1.73) (1.72)
Ln VG Sales Not
Intensely Violent lag 6
0.026
(0.30)
Ln Intensely Violent
Video Game Sales
-0.009 0.014 0.019 0.030 0.031 0.023 0.026
(0.44) (0.56) (0.64) (0.94) (0.94) (0.71) (0.81)
Ln Intensely Violent
VG Sales lag 1
-0.055+ -0.043 -0.029 -0.029 -0.034 -0.027
(1.77) (1.11) (0.70) (0.69) (0.83) (0.65)
Ln Intensely Violent
VG Sales lag 2
-0.063+ -0.044 -0.042 -0.021 -0.017
(1.72) (1.06) (1.02) (0.49) (0.39)
Ln Intensely Violent
VG Sales lag 3
-0.048 -0.047 -0.047 -0.051
(1.41) (1.27) (1.29) (1.22)
Ln Intensely Violent
VG Sales lag 4
0.001 0.011 0.006
(0.04) (0.29) (0.15)
Ln Intensely Violent
VG Sales lag 5
-0.036 -0.032
(1.10) (0.91)
Ln Intensely Violent
VG Sales lag 6
-0.000
(0.01)
Sample includes 200 weekly observations from 2004-2008. Month dummy variables and a time trend were also
included but are not reported. Average GameSpot scores for intensely violent and not and for the current period
and eight lags are used as IVs. The Sargon statistic for over-identification always fails to reject the exogeneity of
the instrument set. Absolute value of z-statistics in parentheses. + significant at 10%; * significant at 5%; **
significant at 1%.
32
Table 5
The Effects of Video Game Sales on the Log of Violent Crime
(1) (2) (3) (4) (5) (6) (7)
Ln Video Game Sales
Not Intensely Violent
-0.061 0.006 0.012 -0.001 0.001 -0.015 -0.006
(1.21) (0.09) (0.16) (0.02) (0.01) (0.17) (0.07)
Ln VG Sales Not
Intensely Violent lag 1
-0.154* -0.138 -0.109 -0.107 -0.114 -0.108
(2.04) (1.49) (1.14) (1.12) (1.22) (1.16)
Ln VG Sales Not
Intensely Violent lag 2
-0.147+ -0.138 -0.134 -0.099 -0.093
(1.70) (1.50) (1.46) (1.04) (0.98)
Ln VG Sales Not
Intensely Violent lag 3
0.001 0.003 0.009 -0.009
(0.01) (0.04) (0.11) (0.10)
Ln VG Sales Not
Intensely Violent lag 4
0.009 0.031 0.029
(0.11) (0.35) (0.31)
Ln VG Sales Not
Intensely Violent lag 5
-0.072 -0.078
(0.93) (0.99)
Ln VG Sales Not
Intensely Violent lag 6
0.041
(0.44)
Ln Intensely Violent
Video Game Sales
-0.024 0.001 0.003 0.007 0.009 0.000 0.003
(1.15) (0.04) (0.10) (0.20) (0.24) (0.01) (0.08)
Ln Intensely Violent
VG Sales lag 1
-0.060+ -0.050 -0.031 -0.030 -0.036 -0.033
(1.73) (1.14) (0.68) (0.66) (0.80) (0.72)
Ln Intensely Violent
VG Sales lag 2
-0.066 -0.062 -0.060 -0.044 -0.040
(1.61) (1.37) (1.32) (0.97) (0.87)
Ln Intensely Violent
VG Sales lag 3
-0.020 -0.017 -0.016 -0.025
(0.54) (0.42) (0.41) (0.54)
Ln Intensely Violent
VG Sales lag 4
-0.001 0.000 0.000
(0.02) (0.00) (0.00)
Ln Intensely Violent
VG Sales lag 5
0.011 -0.012
(0.30) (0.30)
Ln Intensely Violent
VG Sales lag 6
0.011
(0.24)
Sample includes 200 weekly observations from 2004-2008. Month dummy variables and a time trend were also
included but are not reported. Average GameSpot scores for intensely violent and not and for the current period
and eight lags are used as IVs. The Sargon statistic for over-identification always fails to reject the exogeneity of
the instrument set. Absolute value of z-statistics in parentheses. + significant at 10%; * significant at 5%; **
significant at 1%.
33
Table 6
The Cumulative Effect of Video Games on Crimes
Aggregate Effect on All Crimes (from Table 4)
Number of Lags Included
0 1 2 3 4 5 6
Not Intensely
Violent Coefs.
-0.028 -0.102 -0.213* -0.216* -0.203+ -0.257* -0.230+
(0.046) (0.062) (0.096) (0.105) (0.122) (0.124) (0.139)
Intensely
Violent Coefs.
-0.009 -0.041 -0.088* -0.093* -0.088+ -0.104* -0.096+
(0.020) (0.027) (0.041) (0.044) (0.050) (0.050) (0.056)
Chi-Sq test of
difference
0.42 2.70+ 4.75* 3.83* 2.40 4.07* 2.50
Aggregate Effect on Violent Crimes (from Table 5)
Violent/
All Crimes
Number of Lags Included
0 1 2 3 4 5 6
Not Intensely
Violent Coefs.
-0.061 -0.148* -0.272* -0.247* -0.227+ -0.260* -0.224
(0.050) (0.069) (0.109) (0.116) (0.134) (0.135) (0.148)
Intensely
Violent Coefs.
-0.024 -0.061* -0.113* -0.107* -0.099+ -0.108* -0.095
(0.021) (0.030) (0.046) (0.048) (0.054) (0.055) (0.056)
Chi-Sq test of
difference
1.36 4.44* 5.99* 4.10* 2.45 3.40* 1.95
For both the top and bottom panels, each column represents results from a separate instrumental variables
regression. Each row reports the sum of coefficients for a variable for different possible lag lengths. Not
reported are coefficients of month dummies and a time trend. Absolute value of z-statistics in parentheses. +
significant at 10%; * significant at 5%; ** significant at 1%
34
Table 7
The Effect of Video Games on All Crimes by Offender Age
Aged 15-30 Number of Lags Included
0 1 2 3 4 5 6
Not Intensely
Violent Games
-0.028 -0.098 -0.182* -0.178+ -0.167 -0.218+ -0.214
(0.046) (0.061) (0.092) (0.100) (0.115) (0.117) (0.134)
Intensely
Violent Games
-0.012 -0.043 -0.079* -0.081+ -0.077+ -0.093* -0.093+
(0.019) (0.026) (0.039) (0.042) (0.046) (0.047) (0.054)
Chi-Sq test of
difference
0.32 2.31 3.56+ 2.62 1.66 3.04+ 2.18
Aged 35-50 Number of Lags Included
0 1 2 3 4 5 6
Not Intensely
Violent Games
-0.020 -0.089 -0.236* -0.210+ -0.214 -0.243+ -0.235
(0.049) (0.068) (0.112) (0.117) (0.136) (0.138) (0.157)
Intensely
Violent Games
-0.014 -0.042 -0.103* -0.096* -0.098+ -0.105+ -0.102
(0.021) (0.029) (0.047) (0.049) (0.055) (0.056) (0.062)
Chi-Sq test of
difference
0.05 1.37 4.01* 2.63 1.99 2.69 1.91
For both the top and bottom panels, each column represents results from a separate instrumental variables
regression. Each row reports the sum of coefficients for a variable for different possible lag lengths. Not reported
are coefficients of month dummies and a time trend. Absolute value of z-statistics in parentheses. + significant at
10%; * significant at 5%; ** significant at 1%
35
Table 8
The Effect of Video Games on Violent Crimes by Offenders Age
Aged 15-30 Number of Lags Included
0 1 2 3 4 5 6
Not Intensely
Violent Games
-0.034+ -0.057* -0.090* -0.087* -0.100* -0.087+ -0.059
(0.019) (0.025) (0.038) (0.040) (0.048) (0.051) (0.057)
Intensely
Violent Games
-0.010 -0.018+ -0.031+ -0.029+ -0.034+ -0.028 -0.017
(0.008) (0.011) (0.016) (0.017) (0.019) (0.021) (0.022)
Chi-Sq test of
difference
4.32* 6.48** 6.84** 5.61* 5.11* 3.54+ 1.40
Aged 35-50 Number of Lags Included
0 1 2 3 4 5 6
Not Intensely
Violent Games
-0.023 -0.021 -0.028 -0.011 0.005 0.020 0.046
(0.016) (0.020) (0.028) (0.033) (0.040) (0.043) (0.048)
Intensely
Violent Games
-0.008 -0.006 -0.009 -0.002 0.003 0.009 0.019
(0.007) (0.009) (0.012) (0.014) (0.016) (0.017) (0.019)
Chi-Sq test of
difference
1.94 1.42 1.24 0.20 0.00 0.17 0.84
For both the top and bottom panels, each column represents results from a separate instrumental variables
regression. Each row reports the sum of coefficients for a variable for different possible lag lengths. Not reported
are coefficients of month dummies and a time trend. Absolute value of z-statistics in parentheses. + significant at
10%; * significant at 5%; ** significant at 1%
36
Table 9
The Aggregate Effect of Video Games on All Crimes by Campus Location
Crimes on
Campus
Number of Lags Included
0 1 2 3 4 5 6
Not Intensely
Violent Games
-0.041 -0.434 -0.837* -1.126* -1.099+ -1.381* -0.976
(0.266) (0.308) (0.417) (0.486) (0.567) (0.600) (0.724)
Intensely
Violent Games
0.018 -0.174 -0.340* -0.465* -0.458* -0.557* -0.399
(0.112) (0.133) (0.176) (0.203) (0.231) (0.242) (0.289)
Chi-Sq test of
difference
0.13 1.97 4.00* 5.14* 3.40+ 5.05* 1.70
Crimes off
Campus
Number of Lags Included
0 1 2 3 4 5 6
Not Intensely
Violent Games
-0.024 -0.088+ -0.192* -0.187+ -0.175 -0.222+ -0.208
(0.045) (0.061) (0.095) (0.103) (0.120) (0.121) (0.137)
Intensely
Violent Games
-0.008 -0.035 -0.080* -0.081+ -0.076 -0.091+ -0.087
(0.019) (0.026) (0.040) (0.043) (0.049) (0.049) (0.055)
Chi-Sq test of
difference
0.30 2.04 3.95* 2.93+ 1.80 3.15+ 2.09
For both the top and bottom panels, each column represents results from a separate instrumental variables
regression. Each row reports the sum of coefficients for a variable for different possible lag lengths. Not
reported are coefficients of month dummies and a time trend. Absolute value of z-statistics in parentheses. +
significant at 10%; * significant at 5%; ** significant at 1%
37
Table 10
The Effect of Video Games on Violent Crimes by Campus Location
Crimes on
Campus Number of Lags Included
0 1 2 3 4 5 6
Not Intensely
Violent Games
0.048 -0.376 -0.758 -1.052+ -0.953 -1.265+ -0.810
(0.302) (0.353) (0.456) (0.550) (0.649) (0.688) (0.832)
Intensely
Violent Games
0.039 -0.171 -0.327 -0.455+ -0.422 -0.529+ -0.349
(0.128) (0.152) (0.201) (0.230) (0.263) (0.278) (0.332)
Chi-Sq test of
difference
0.00 0.93 2.31 3.27+ 1.81 3.06+ 0.82
Crimes off
Campus Number of Lags Included
0 1 2 3 4 5 6
Not Intensely
Violent Games
-0.061 -0.136* -0.255* -0.219+ -0.200 -0.224 -0.202
(0.050) (0.069) (0.109) (0.116) (0.134) (0.137) (0.153)
Intensely
Violent Games
-0.025 -0.056+ -0.106* -0.095+ -0.088 -0.093+ -0.086
(0.021) (0.030) (0.046) (0.048) (0.055) (0.055) (0.061)
Chi-Sq test of
difference
1.32 3.74+ 5.02* 3.14+ 1.86 2.46 1.53
For both the top and bottom panels, each column represents results from a separate instrumental variables
regression. Each row reports the sum of coefficients for a variable for different possible lag lengths. Not
reported are coefficients of month dummies and a time trend. Absolute value of z-statistics in parentheses. +
significant at 10%; * significant at 5%; ** significant at 1%