Sports franchises, events, and city livability: an examination of spectator sports and crime rates.

Baumann, Robert ; Ciavarra, Taylor ; Englehardt, Bryan 等

Introduction

It is a common refrain among sports boosters, city officials, and professional teams and leagues that sports teams and major athletic events bring significant economic windfalls to host cities. For example, estimates for the annual economic impact of a major league professional sports franchise in the United States often exceed $100 million (Oregon Baseball Campaign 2002) while organisers of sporting events claim impacts ranging from the tens of millions for league all-star games (Selig, Harrington, and Healey 1999) to the hundreds of millions for major championships like the Super Bowl of American football or European soccer's Champions League Final (National Football League 1999) and even into the billions for the largest of the so-called 'mega-events' such as the Olympics or World Cups in sports such as soccer or rugby (Humphreys and Plummer 1995).

Most academic examinations of the direct economic impact of sports teams, stadiums, and events on observable economic variables such as employment (Baade and Matheson 2002; Hagn and Maennig 2008), personal income or personal income per capita (Coates and Humphreys 2002), taxable sales (Baade, Baumann and Matheson 2008) and tourist arrivals (Baumann, Matheson, and Muroi 2009) have found that spectator sports have little to no measurable effect on the economy.

However, even if sporting events have minimal direct economic impact, it is conceivable that professional sports could have large indirect impacts on other measures of quality of life that are not captured by traditional economic statistics. In this paper we investigate whether sporting events have any effect on citywide violent and property crime rates.

As most studies of the direct economic impact of sports have been unable to identify significant positive benefits that would justify public subsidies, researchers have increasingly turned to the examination of the potential indirect economic benefits of sports, and it is here in which our investigation of crime finds its niche. It is reasonable to assume that sporting events or franchises may be an important source of civic pride, serve as a cultural amenity, or increase social capital in important ways. As noted by former Minnesota governor Rudy Perpich, 'Without professional sports, Minneapolis would just be a cold Omaha'. On the opposite end of the temperature spectrum, the Hawaii Tourism Authority sounds a similar note by suggesting that subsidising the Pro Bowl and local Professional Golfers Association (PGA) events improves the quality of life of the Island's residents by allowing them opportunities to watch or participate in major sporting events. (HTA 2008).

Carlino and Coulsen (2004) address whether sporting events indirectly impact the local community by examining rental housing prices in NFL cities. They find them to be 8 per cent higher than in non-NFL cities. While their methodology has been questioned, the basic finding would support the hypothesis that professional sports make cities more attractive places to live because renters are willing to pay a premium to live in NFL cities. Numerous other studies have also studied the connection between housing prices and sports (Tu 2005; Dehring, Depken and Ward 2007; Coates and Matheson 2011) with distinctly mixed results.

Others have used contingent valuation to assess the value of sporting teams and events in the absence of observable economic data. Here, too, the data is mixed. While most studies of new stadiums and arenas (Groothius, Johnson, and Whitehead 2004), professional franchises (Johnson, Groothius, and Whitehead 2001), and mega-events (Atkinson et al. 2008) find that citizens are willing to pay for sports teams and events beyond just purchasing tickets, several of the studies also demonstrate that this willingness to pay is often far less than the subsidy granted to the sports entity.

The connection between direct and indirect economic benefits is perhaps best summed up by Maennig (2007) who concludes in his ex post analysis of the 2006 World Cup in Germany that claims of 'increased turnover in the retail trade, overnight accommodation, receipts from tourism and effects on employment [are] mostly of little value and may even be incorrect. Of more significance, however, are other (measurable) effects such as the novelty effect of the stadiums, the improved image for Germany and the feel good effect for the population' (Maennig 2007: 1).

Examining the relationship between crime and sports is of considerable interest. Our own rough estimate of the cost of violent and property crime is $150 million per year for an average large metropolitan area in the United States, a conservative estimate that only includes direct losses and the pain and suffering of victims. (1) As such, if the diversion of sporting events reduces crime rates by even a small fraction, then they could be providing large unaccounted for gains to the cities that sponsor them. For instance, if a professional sports franchise reduced crime by just 5 per cent, then the estimated benefits are $7.5 million greater per year than previously estimated. On the other hand, if they increased crime by 5 per cent, then the estimates would be overstating any claimed positive impact.

There are numerous reasons why sporting events could increase or decrease crime. First, if the events decrease local unemployment or increase wages or income, then the opportunity costs of committing crime will rise and thus crime rates will fall. Given the preponderance of the evidence suggesting insignificant direct economic effects from sports, this is unlikely to be the avenue by which sporting events might alter crime rates. However, even in the absence of observed citywide economic effects, most economists acknowledge the potential for stadiums and events to cause localized changes in economic activity (Tu 2005). Thus, sports may serve as an anchor for neighbourhood or city revitalisation, and as such, the resulting reduction in crime in a particularly blighted area could have a measurable impact on citywide crime rates.

Alternatively, sporting events' primary effect on crime could be as a distraction from illegal activities. Indeed, the idea of using sports to pacify and distract the masses is centuries old. In circa 100 AD, the Roman poet Juvenal coined the phrase 'bread and circuses' to describe the use of food subsidies and cheap entertainment to limit public dissent during the height of the Roman Empire. Nearly 2,000 years later, star American football player Ray Lewis, noted more for his crunching tackles rather than his epic poetry, suggested a similar effect if labour strife were to cause the cancellation of the 2011 National Football League (NFL) season. Said Lewis, 'Do this research if we don't have a season--watch how much evil, which we call crime, watch how much crime picks up, if you take away our game'. When asked why he thought crime would increase if the NFL lost a season, Lewis said, 'There's nothing else to do' (ESPN 2011).

On the other hand, spectator sports could increase crime rates as large influxes of visitors may enlarge the pool of potential criminals and victims. In addition, excess alcohol consumption and unruly crowds are associated with sporting events, and these situations are known to be a major factor in perpetrating crime. Furthermore, sports may heighten emotions leading to impulsive criminal behaviour. Rees and Schnepel (2009) and Card and Dahl (2009) examine college and professional football, respectively, and find that arrests for assault, disorderly conduct and domestic violence rise during (and after) football games and, in particular, reach a peak when teams experience unexpected wins or losses, suggesting crime is not solely a function of the number of people attending a sporting contest but also a result of the emotional state of fans. Finally, sports may serve to exacerbate existing partisan divides resulting in fan violence and increasing crime as seen in the epidemics of hooliganism in English soccer in the 1980s and early 90s. Of course, sports could also serve to reduce tensions by providing a platform for conflicts to be played out without the need to resort to violence.

To reiterate, we add to the discussion on whether sporting events have an indirect effect on local economies by examining their effects on the city-level incidence of crime. The results we will present in the next sections generally indicate no positive or negative benefit along the crime dimension associated with major sporting events.

Model and Data

After the first attempts of theoretically modelling criminal behaviour (e.g. Becker, 1968 and Ehrlich 1973), several empirical analyses followed. Most empirical approaches test whether crime is influenced by some measure of wealth, such as unemployment (Gould, Weinberg, and Mustard 2002), wages (Grogger 1998), and education (Lochner 2004) to name only a few. In these cases, crime is modelled as a substitute for working. Each individual compares the expected return to criminal activity against expected punishment and foregone wages from legitimate employment. This produces a reduced-form equation where crime is a function of wealth and punishment. Other studies include demographic controls such as racial/ethnic, gender, and age distribution since these tend to influence the amount of crime.

We add controls for the presence and success of franchises in the four American major sports leagues--National Football League (NFL), Major League Baseball (MLB), National Basketball Association (NBA), and National Hockey League (NHL)--to determine whether these franchises affect crime. If a connection between the presence of a sports team and a reduction in crime can be identified, sports franchises may provide indirect economic benefits to their host cities that would not necessarily be captured in observable economic data such as income, employment or taxable sales.

We use the Federal Bureau of Investigation's Uniform Crime Reports (UCR) to measure crime. UCR data are available annually from 1981 to 2006 at the county-level. It is only fair to acknowledge two weaknesses in the UCR data. First, the data are available only annually and at the county level. This limits the ability to match individual crimes with specific sporting events at the neighbourhood level or daily time frame; however, these data should allow for an examination of whether professional sports alter the general crime climate in a city. The second problem is that the UCR data are compiled from local police reports meaning they only include reported crime. This creates two problems. First, the total amount of crime is underestimated since unreported crime is not measured. Second, Levitt (1998) notes reporting and classification tendencies differ across police stations. However, UCR data are by far the most common aggregate data set in the literature. The other alternative is victimisation data, and the most common is the National Crime Victimization Survey (NCVS). But the only geographic information in the NCVS is four broad regions of the U.S., which makes it impossible to merge the NCVS with franchise location data.

UCR provide data on eight types of crime, which we combine into two larger groups. Violent crime, which is committed with force, consists of murder/manslaughter, rape, robbery, and assaults. Property crime, which is not done with force and typically when the victim is not present, consists of burglaries, larceny, arson, and motor vehicle theft. Both types of crime are scaled so that each is per 100,000 people to control for differences in population.

Our measure of wealth is per capita income, which is available at the MSA level from the Bureau of Economic Analysis (BEA). We use a sample of 56 metropolitan standardized areas (MSAs) between 1981 and 2006. With a few exceptions (2), these MSAs represent the largest cities in the United States and include all MSAs that host an NFL, MLB, NHL, or NBA franchise. This list also includes cities without a franchise in any of the four major sports leagues to serve as part of our control group, e.g. Austin, Las Vegas, and Riverside. While cities without a franchise tend to be smaller, the other portion of our control group includes cities whose franchise status changes. The largest MSA in this group is Los Angeles, which once had two NFL teams but lost them both to relocation by 1995. In addition, Washington, D.C. did not have a MLB team until 2005. In addition, there are several MSAs with franchises in some but not all of the four major sports, e.g. Houston (no NHL), St. Louis (no NBA), and Portland, Oregon (no MLB, NFL, or NHL).

Since UCR data is county-level, we aggregate the UCR data to the MSA level using the county compositions of the MSAs provided by the BEA. This creates a sample of 56 MSAs over the time period 1981 to 2006. Table 1 provides summary statistics for the data.

The following is our baseline model:

[CR.sub.it] = [[beta].sub.0] + [[beta].sub.1][INC.sub.it] + [[beta].sub.2][F.sub.it] + [[beta].sub.3][NF.sub.it] + [[beta].sub.4][S.sub.it] [[beta].sub.5][H.sub.it] + [[alpha].sub.i] + [[gamma].sub.i] + [[epsilon].sub.it] (1)

Because the motivations for property and violent crime are different, we present separate estimations for property and violent crime, [CR.sub.it]. [INC.sub.it] is the per capita income level. [F.sub.it] is a vector of four dummy variables that indicates whether the MSA has a franchise in each of the four major sports leagues. [NF.sub.it] is a vector of four dummy variables that equal one the first year a franchise is in the MSA. This variable captures any novelty effect that a new franchise has on crime. [S.sub.it] is a vector of four dummy variables that indicates whether the MSA has a franchise that made the championship game or series, i.e. the Stanley Cup Finals, NBA Finals, World Series, and Super Bowl. Although there are many ways to measure success, the grand finals are the pinnacle of each league and should have a larger effect than, say, winning percentage or making the playoffs/finals. (3) Championship games have also been known to spark violent fan reaction in participating cities such as the rioting that followed the NBA championships won by the Detroit Pistons in the early 1990s, while such incidents are almost unheard of during mere playoff games. In addition, changing the specification of [S.sub.it] to winning percentage or making the playoffs has no substantial impact on the results. [H.sub.it] is a vector of dummy variables that equal one if the MSA hosted the Super Bowl, Olympics, or World Cup soccer match. Finally, controls for each year ([[gamma].sub.t]) and MSA ([[alpha].sub.i]) are included to capture any MSA-specific or year-specific effects on crime.

The MSA controls are particularly important since they account for time-invariant reporting and classification tendencies specific to the MSA (see Levitt 1998). In addition, these controls also absorb some of the other factors of crime, such as police expenditures and punishment, not included in our model. Omitting these controls is not likely to influence our key results for two reasons. First, these controls have little variation over time within an MSA, which means most of their impact is in the fixed effect. For example, if one observed an increase in crime in 1984 in Los Angeles, the year the city hosted the Summer Olympics, it is extremely unlikely such a result could be explained by changes in the standard demographic or punishment related variables such as male share or racial mix of the population or the median sentence length that unpin the theoretical Becker (1968) and Ehrlich (1973) models of crime and punishment, since these variables exhibit almost no movement within an MSA from year to year. While these variables may be useful in explaining differences in crime rates between cities in any given year, the inclusion of MSA fixed effects serves largely the same purpose. Second, leaving out demographic and punishment data is also unlikely to influence the result because the omitted controls likely have little to no correlation with our sports variables of interest.

We use a variety of tests to check for unit roots in property crime, violent crime, and per capita income. First, we perform Dickey-Fuller and Phillips-Perron tests on each MSA. These tests do not reject the existence of a unit root in nearly every MSA for all three variables. Second, we test for unit roots using panel data tests from Levin, Lin, and Chu (2002) and Pesaran (2007). These tests allow the entire data to be tested at once, and allow each MSA to have their own time trend and autoregressive path. Further, Pesaran (2007) suggests a unit root test for panel data that allows for dependence across the panels.

Table 2 presents the results from the panel unit root tests. These tests suggest the crime variables are free of a unit root in levels, but per capita income is not free of a unit root. The same tests reject the existence of a unit root for the first difference of each variable. For this reason, we take a conservative approach, and the first difference of property crime, violent crime, and per capita income is used in all estimations. Using the levels of the crime variables produces no substantial changes in the results.

Autocorrelation is a concern in this model since it is likely the unexplained portion of crime in a given period is correlated with the unexplained portion of crime in the previous period. In the presence of autocorrelation, the least squares estimates will be consistent but the standard errors will be wrong. Wooldridge (2002) suggests testing for autocorrelation using two steps. First, estimate the baseline model at (1). Second, generate the residuals and estimate [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. If there is no autocorrelation, then [rho] = -0.5. For both the property crime and violent crime models, we reject the hypothesis that [rho] = -0.5 which suggests the model has autocorrelation.

We correct for autocorrelation by including an autoregressive term to (1):

[DELTA][CR.sub.it] = [[beta].sub.0] + [[beta].sub.1][DELTA][CR.sub.it-1] + [[beta].sub.2][DELTA][INC.sub.it] + [[beta].sub.3][F.sub.it] + [[beta].sub.4][NF.sub.it] + [[beta].sub.5][S.sub.it] + [[beta].sub.6][H.sub.it] + [[alpha].sub.i] + [[gamma].sub.t] + [[epsilon].sub.it] (2)

The first differences of crime and per capita income are included to ensure unit roots do not produce a spurious correlation. In the presence of a lagged dependent variable, least squares estimates are likely to be biased because of the correlation between [DELTA][CR.sub.it-1] and [[epsilon].sub.it]. Instead, we use the Arellano and Bond (1991) technique which produces consistent estimates. Other descriptions of this technique can be found in Bond (2002) and Roodman (2006). The Arellano and Bond (1991) technique differences the entire model, which eliminates the MSA fixed effect [[alpha].sub.i]. Next, higher-order lags of the dependent variable are used to instrument for the endogenous [DELTA][CR.sub.it-1]. This technique also allows any other endogenous or predetermined independent variables (i.e., variables independent to the current error but not previous errors) to be instrumented. Since it is plausible that per capita income is also endogenous (or at least predetermined), we instrument for [DELTA][INC.sub.it].

Our original sample frame ranges from 1981 to 2006. However, we use the first difference of the data to guard against unit roots, and the lag of the already first-differenced dependent variable is included to account for autocorrelation. This changes the sample frame to 1983 to 2006. Since T = 24, there are 22 higher-order lags of the dependent variable that could serve as instruments. These higher-order lags create missing values, e.g. if t = 1985 then the third lag and higher of [DELTA][CR.sub.it] are not defined since the sample frame begins in 1983. Nevertheless, Holtz-Eakin, Newey, and Rosen (1988) point out that each higher-order lag is a useful moment condition. In this scenario, the moment condition is E[[Z'.sub.it] [DELTA][[epsilon].sub.it]] = 0, where [Z'.sub.it] is a vector that contains the higher-order lags of the dependent variable. For the second order lag, [summation over (i)][y.sub.i,t-2] [DELTA][[epsilon].sub.it] = 0 if t [greater than or equal to] 3; for the third-order lag, [summation over (i)][y.sub.i,t-3] [DELTA][[epsilon].sub.it] = 0 if t [greater than or equal to] 4; and so on.

These moment conditions require the error term to be independently and identically distributed. This is unlikely in panel data because the error variance probably differs across MSAs. For this reason, a weighting matrix W is included in the moment condition that asymptotically corrects this problem:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [[??].sub.i] and [DELTA][[??].sub.i], are MSA-specific vectors with (T - 2) elements. The weighting matrix is:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Since the weighting matrix includes [DELTA][[??].sub.i], the model must be estimated in two steps. First, a second weighting matrix:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

is used to produce [DELTA][[??].sub.i], where H is a (T - 2) square matrix with 2 on the diagonal, -1 on all of the immediate off-diagonals, and zero elsewhere. Once [DELTA][[??].sub.i], is estimated, the second step minimises:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

to produce the estimates.

Finally, Arellano and Bond (1991) note the two-step estimation process causes the standard errors to be downward biased. Windmeijer (2005) offers a finite-sample correction which we use here.

Results

Table 3 presents the estimation results for the property crime model. The Arellano-Bond tests for autoregressive errors suggest only a first-order autoregressive term is necessary. We also present the result from a Hansen (1982) test to determine whether the model is over-identified. We use Hansen tests to determine the ideal number of higher-order lags to use as instruments. In the property crime model, the Hansen test suggests the second- and third-order lags do not over-identify the model. We also suppress the results for the MSA and the year dummies for brevity, but these are available upon request.

The only sports variable that is statistically significant is Olympics location. Hosting the Olympics raises property crime by about 445 per 100,000 people or an increase of about 10 per cent. The other sports estimates suggest there is no effect of a franchise or its success on property crime rates. While the Olympics result may simply be a spurious correlation that is the result of the inclusion of a large number of sports-related variables, it is noteworthy that the Olympics are far and away the largest sports mega-event drawing a far larger number of visitors than any other sporting event. An increase in reported crime fits the hypothesis that a rise in visitors raises the crime rate by increasing the number of potential victims and criminals partially echoing the conclusions of Rees and Schnepel (2009). Alternatively, the vastly increased police presence during the Olympics could create a perception that law enforcement may be better equipped to solve any given crime leading to higher reporting rates than during non-Olympic periods. Other specifications of success, e.g. winning percentage or making the playoffs, do not substantially change the results. Per capita income has a negative and statistically significant effect on property crime, meaning higher wealth is correlated with lower property crime.

Table 4 presents the estimation results for the violent crime model. The Arellano-Bond tests for autoregressive errors again suggest a first order autoregressive term is appropriate, and the Hansen (1982) test allows for the second- through fifth-order lags to serve as instruments. The only sports variable that is statistically significant is the Super Bowl location, which decreases violent crime by about 17.5 per 100,000 people, a decrease of about 2.5 per cent. Similar to property crime, the other sports estimates suggest there is no effect of a franchise or its success on violent crime. One difference between the property and violent crime models is the effect of per capita income. For violent crime, the effect is positive, suggesting higher wealth correlates with more violent crime. There are several possible explanations for this result. Since the UCR data only collect reported crime, it is possible that an increase wealth also increases reporting habits. In addition, the motivations of violent crime tend to be psychological rather than pecuniary, which means there is no ex ante expectation of the relationship between wealth and violent crime.

Again, the one significant sports variable is noteworthy. The Super Bowl, along with the Olympics, is among the few mega-events for which cities can plan in advance. For example, the World Series or NBA finals are played in the cities of the teams involved, so their locations are only known as teams advance in the playoffs. The Super Bowl, however, is held at a neutral site designated well in advance. Knowing that the eyes of the world will be on the host city, the local law enforcement agencies may take steps to 'clean up the town' in advance of the big game, and these crime eradication efforts may carry through for some time after the event. In the language of the Becker (1968) and Ehrlich (1973) models, the fall in violent crime around the time of the Super Bowl can potentially be explained by a reduced return to criminal activity caused by higher expected punishment in the form of increase policing.

Conclusion

The results of this paper overall suggest no significant link between crime and the presence of professional sports teams or events at the metropolitan-area wide level with two notable exceptions. The Olympics Games are associated with roughly a 10 per cent increase in property crime while the Super Bowl is associated with a 2.5 per cent decrease in violent crime. In the whole, however, spectator sports do not seem to automatically carry with them any reductions in criminal behaviour.

Further research is required to examine nuisance crimes, arrests versus reports of crime, the geographic distribution of crime within a city, the effect of new stadiums, and the changes in crime rates in years leading up to planned events such as the Olympics.

References

Arellano, M. and Bond, S. (1991) 'Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations, Review of Economic Studies, 58(2), pp. 277-297.

Atkinson, G., Mourato, S., Szymanski, S. and Ozdemiroglu, E. (2008) 'Are we willing to pay enough to "back the bid"?: Valuing the intangible impacts of London's bid to host the 2012 Summer Olympic Games, Urban Studies, 45(2), pp. 419-444.

Baade, R. and Matheson, V. (2002) 'Bidding for the Olympics: Fool's gold?' in C. Pestana Barros, M. Ibrahimo, and S. Szymanski (eds) Transatlantic Sport: The Comparative Economics of North American and European Sports, Edward Elgar Publishing, London, pp. 127-151.

Baade, R., Baumann, R., and Matheson, V. (2008) 'Selling the game: Estimating the economic impact of professional sports through taxable sales, Southern Economic Journal, 74(3), pp. 794-810.

Baumann, R., Matheson, V., and Muroi, C. (2009) 'Bowling in Hawaii: Examining the effectiveness of sports-based tourism strategies, Journal of Sports Economics, 10(1), pp. 107-123.

Becker, G. (1968) 'Crime and punishment: An economic approach, Journal of Political Economy, 76, pp. 169-217.

Bond, S. (2002) 'Dynamic panel data models: A guide to micro data methods and practice, Portuguese Economic Journal, 1, pp. 141-162.

Carlino, G. and Coulson., E. (2004) 'Compensating differentials and the social benefits of the NFL, Journal of Urban Economics, 56(1), pp. 25-50.

Card, D. and Dahl, G. (2009) 'Family violence and football: The effect of unexpected emotional cues on violent behavior, NBER Working Paper No. 15497.

Coates, D. and Humphreys, B. (2002) 'The economic impact of post-season play in professional sports', Journal of Sports Economics, 3(3), pp. 291-299.

Coates, D. and Matheson, V. (2011) 'Mega-events and housing costs: Raising the rent while raising the roof?', Annals of Regional Science, 46(1), pp. 119-137.

Cohen, M. (1988) 'Pain, suffering, and jury awards: A study of the cost of crime to victims. Law and Society Review', 22(3), pp. 537-556.

Dehring, C., Depken, C. and Ward, M. (2007) 'The impact of stadium announcements on residential property values: Evidence from a natural experiment in Dallas-Ft. Worth, Contemporary Economic Policy, 25(4), pp. 627-638.

Ehrlich, I. (1973) 'Participation in illegitimate activities: A theoretical and empirical investigation', The Journal of Political Economy, 81, pp. 521-565.

ESPN (2011) 'Ray Lewis discusses lockout issues, 25 May, available: http://sports. espn.go.com/nfl/news/story?id=6575570 [accessed 3 June 2011].

Gould, E., Weinberg, B., and Mustard, D. (2002) 'Crime rates and local labor market opportunities in the United States: 1979-1997, The Review of Economics and Statistics, 84(1), pp. 45-61.

Grogger, J. (1998) 'Market wages and youth crime', Journal of Labor Economics, 16(4), pp. 76-91.

Groothuis, P., Johnson, B., and Whitehead, J. (2004) 'Public funding of professional sports stadiums: Public choice or civic pride?', Eastern Economic Journal, 30(4), pp. 515-526.

Hagn, F. and Maennig, W. (2008) 'Employment effects of the football World Cup 1974 in Germany, Labour Economics, 15(5), pp. 1062-1075.

Hansen, L. (1982) 'Large sample properties of generalized method of moments estimators', Econometrica, 50(4), pp. 1029-1054.

HTA (Hawaii Tourism Authority) (2007) HTA 2007 Annual Report to the Hawaii State Legislature, available: http://www.hawaiitourismauthority.org [accessed 31 December 2011].

Holtz-Eakin, D., Newey, W., and Rosen, H. (1988) 'Estimating vector autoregressions with panel data, Econometrica, 56(6), pp. 1371-1395.

Humphreys, J. and Plummer, M. (1995) The economic impact on the State of Georgia of hosting the 1996 Summer Olympic Games, Mimeograph, Selig Center for Economic Growth, The University of Georgia.

Im, K., Pesaran, H., and Shin, Y. (2003) 'Testing for unit roots in heterogeneous panels, Journal of Econometrics, 115(1), pp. 53-74.

Johnson, B., Groothuis, P., and Whitehead, J. (2001) 'The value of public goods generated by a major league sports team: The CVM approach, Journal of Sports Economics, 2(1), pp. 6-21.

Levin, A., Lin, C., and Chu, C. (2002) 'Unit root tests in panel data: Asymptotic and finite sample properties, Journal of Econometrics, 108(1), pp. 1-24.

Levitt, S. (1998) 'The Relationship between crime reporting and police: Implications for the use of uniform crime reports, Journal of Quantitative Criminology, 14(1), pp. 61-81.

Lochner, L. (2004) 'Education, work, and crime: A human capital approach, International Economic Review, 45(3), pp. 811-843.

Maennig, W. (2007) 'One year later: A re-appraisal of the economics of the 2006 Soccer World Cup, International Association of Sports Economists Working Paper Series, No. 07-25.

National Football League (1999) 'Super Bowl XXXIII generates $396 million for South Florida, NFL Report, 58(7).

Oregon Baseball Campaign (2002) 'Why MLB, why Oregon, why now?, available: http://www.oregonbaseballcompaign.com/fact.htm [accessed 5 September 2002].

Pesaran, H. (2007) 'A simple panel unit root test in the presence of cross-section dependence, Journal of Applied Econometrics, 22(2), pp. 265-312.

Rees, D. and Schnepel, K. (2009) 'College football games and crime', Journal of Sports Economics, 10(1), pp. 68-87.

Roodman, D. (2006). 'How to do xtabond2: An introduction to "difference" and "system" GMM in Stata, Working Paper 103, Center for Global Development, Washington, DC.

Selig, B., Harrington, J. and Healey, J. (1999) 'New ballpark press briefing: July 12, 1999, available: http://www.asapsports.com/baseball/1999allstar/071299BS. html [accessed 29 August 2000].

Tu, C. (2005) 'How does a new sports stadium affect housing values? The case of FedEx Field, Land Economics, 81(3), pp. 379-395.

Windmeijer, F. (2005) 'A finite sample correction for the variance of linear two-step GMM estimators, Journal of Econometrics, 126(1), pp. 25-51.

Wooldridge, J. (2002) Introductory Econometrics: A Modern Approach, Second edition, South-Western College Publishers, New York.

Robert Baumann *

Taylor Ciavarra *

Bryan Englehardt *

Victor A. Matheson *

* Department of Economics, College of the Holy Cross, Worcester, MA, USA

Notes

(1.) To be conservative, we have excluded unreported crime as well as other expenditures including police, judicial costs, and detention. We use the dollar value of direct costs and pain and suffering found in Cohen (1988) to calculate the weighted average of the total cost each city incurs per year. The dollar value is in 2008 dollars for the 56 largest metropolitan areas in the United States examined later in the paper.

(2.) Because of inconsistencies in the UCR data, we omit Akron, Ohio, Chicago and Champaign, Illinois MSAs from the data.

(3.) Note that in American terminology, 'playoffs' refer to any games played in the post-regular season, or the equivalent of the Australian term 'finals, while the term 'final' refers to the championship, or the equivalent to the Australian term 'grand final'.

Dr Robert Baumann is an Assistant Professor in the Department of Economics, College of the Holy Cross, Worcester, MA, USA. His research interests are labour economics, industrial organisation and econometrics. he can be contacted at [email protected].

Taylor Ciavarra is a product manager and analyst for Enservio Corporation of Needham, MA, USA. He is a graduate of the Department of Economics at the College of the Holy Cross where he engaged in research on crime and sports economics. He can be contacted at [email protected].

Dr Bryan Engelhardt is an Assistant Professor in the Department of Economics, College of the Holy Cross, Worcester, MA, USA. His research interests are labour economics, macroeconomics theory and mathematics for economists. He can be contacted at [email protected].

Dr Victor A. Matheson is an Associate Professor in the Department of Economics, College of the Holy Cross, Worcester, MA, USA. His research interests are public finance, sports economics, lotteries and gaming and teaching issues in economics. He can be contacted at [email protected].

Table 1: Summary statistics

Variable                                   Mean (Standard Deviation)

Property Crimes per 100,000 people               4,748.71 (1,503.01)
Violent Crimes per 100,000 people                    618.74 (263.11)
Per capita income                             $31,687.08 ($5,618.73)
MSA has NHL team                                               0.267
MSA has NBA team                                               0.413
MSA has NFL team                                               0.481
MSA has MLB team                                               0.372
NHL team appeared in Stanley Cup finals                        0.024
NBA team appeared in finals                                    0.031
NFL team appeared in Super Bowl                                0.035
MLB team appeared in World Series                              0.030
MSA hosted Olympics                                            0.002
MSA hosted World Cup                                           0.006

Note: (1) There are three observations that hosted the Olympics:
Los Angeles in 1984, Atlanta in 1996, and Salt Lake City in 2002.

(2) Eight MSAs in the sample hosted World Cup games in 1994:
Boston, Dallas, Detroit, Los Angeles, New York City, Orlando, San
Jose, and Washington, D.C.

Table 2: Unit root test results for panel

                                           standardised
                                                   test
                                              statistic    p value

Property Crime per 100,000 people
  Pesaran                                        -2.180      0.001
  Levin-Lin-Chu                                 -11.123     0.0004
Monthly Difference of Property Crime
  Pesaran                                        -3.238     <0.001
  Levin-Lin-Chu                                 -23.244     <0.001
Violent Crime per 100,000 people
  Pesaran                                        -2.070      0.009
  Levin-Lin-Chu                                 -10.004     0.0408
Monthly Difference of Violent Crime
  Pesaran                                        -3.595     <0.001
  Levin-Lin-Chu                                 -24.997     <0.001
Per Capita Income
  Pesaran                                        -1.695      0.656
  Levin-Lin-Chu                                  -8.875      0.330
Monthly Difference of Per Capita Income
  Pesaran                                        -2.541     <0.001
  Levin-Lin-Chu                                 -19.181     <0.001

Note: The null hypothesis in the Im-Pesaran-Shin and Levin-Lin-Chu
tests is that all series are  non-stationary.

Table 3: Arellano-Bond results for property crime model

                                     Estimate
Variable                     (Standard Error)

per capita income                     -0.1220 *
                                     (0.0479)

NHL Franchise                         144.198
                                    (148.207)

NBA Franchise                          65.872
                                    (207.342)

NFL Franchise                         -72.159
                                    (111.676)

MLB Franchise                         -62.795
                                    (220.749)

New NHL Franchise                      11.688
                                    (168.393)

New NBA Franchise                     -33.464
                                    (133.225)

New NFL Franchise                     -20.874
                                     (75.113)

New MLB Franchise                      77.971
                                    (155.470)

Stanley Cup Finals                    139.156
                                    (105.910)

NBA Finals                             51.952
                                     (42.954)

Super Bowl Team                       -12.679
                                     (84.402)

World Series Team                      20.251
                                    (106.680)

Super Bowl Location                  -106.195
                                     (82.423)

Olympics Location                   445.489 *
                                    (185.293)

World Cup Location                   -108.053
                                    (151.294)

Arellano-Bond test                  Z = -2.84
for AR(1)                           p = 0.005

Arellano-Bond test                  Z = -0.65
for AR(2)                           p = 0.513

Instruments (lags of
differenced dep. var.)                    2,3

Hansen test for over-     [chi square] = 2.09
identification                      p = 0.553

Note: (1) * indicates the estimate is statistically significant
at a = 0.05.

(2) Year dummies are included in the model but not presented
here. These estimates are available upon request.

Table 4: Arellano-Bond results for violent crime model

                                     Estimate
Variable                     (Standard Error)

per capita                             0.0255 *
income                               (0.0054)

NHL Franchise                          17.008
                                     (19.568)

NBA Franchise                           6.669
                                     (16.018)

NFL Franchise                           6.559
                                     (16.119)

MLB Franchise                         -24.286
                                     (37.538)

New NHL                               -24.894
Franchise                            (22.984)

New NBA                               -12.821
Franchise                            (19.898)

New NFL                                -2.054
Franchise                            (15.492)

New MLB                                 4.979
Franchise                            (18.267)

Stanley Cup                             8.898
Finals                               (15.490)

NBA Finals                            -10.196
                                     (10.282)

Super Bowl                             -1.383
Team                                 (12.463)

World Series                           -1.264
Team                                 (12.463)

Super Bowl                          -17.567 *
Location                              (9.935)

Olympics                               0.1892
Location                             (14.631)

World Cup                             -20.763
Location                             (23.185)

Arellano-Bond test for              Z = -3.69
AR(1)                               p = 0.000

Arellano-Bond test for               Z = 0.84
AR(2)                               p = 0.400

Instruments (lags of
differenced dep. var.)                2,3,4,5

Hansen test for over-     [chi square] = 5.64
identification                      p = 0.228

Note: (1) * indicates the estimate is statistically significant
at a = 0.05.

(2) Year dummies are included in the model but not presented
here. These estimates are available upon request.