Alcohol Availabilty and Crime in California Cities

By

Erin Riches

PPA 207 Quantitative Methods, Spring 1999
Professor Robert Wassmer
California State University, Sacramento

May 26, 1999

I. INTRODUCTION

This paper uses regression analysis to examine the relationship between alcohol availability and the level of crime in California cities, controlling for factors such as social structure, economic structure, demographic characteristics, law enforcement presence, and retail activity. While many researchers have shown the link between alcohol and violence, and the link between drugs and crime, much less data exists regarding the link between alcohol and crime. There is a growing perception in American cities, however, that reducing alcohol availability may help reduce urban problems such as crime and blight. With the fiscal cost of crime being so high ($2,331 per household in 1983), not to mention the larger costs to society in general and to victims in particular, there is clearly an interest for policymakers to find ways to reduce the problem.

Costs of Alcohol Availability. High levels of alcohol consumption carry significant fiscal, social, and human costs. According to the State Department of Alcohol and Other Drug Programs (1997), problems associated with alcohol and other drugs cost the State of California almost $20 billion annually. This cost includes premature deaths, crimes, incarcerations, diseases, lost productivity, motor vehicle crashes, social welfare programs, AIDS, and alcohol- and other-drug-related infant care. The Pacific Center for Violence Prevention (1996) argues that "approximately one dollar of societal costs is created for every retail dollar spent on alcoholic beverages." They point out that over 13,000 Californians died of alcohol-related causes in 1989. A significant portion of these (11%) of these were homicides; an almost equally significant portion (8%) were suicides.

The Alcohol-Crime Link. The correlation between alcohol consumption and criminal behavior has received a great deal of attention. A recent report by the National Center on Addiction and Substance Abuse (1997) found that 80% of individuals in prisons and jails in this country are or have been major substance abusers. According to the study, 21% of state prison inmates and 11% of federal inmates incarcerated for violent crimes were under the influence of alcohol (and no other substance) when they committed their crime. A United States Department of Justice Bureau of Justice Statistics survey (1992) noted that 59% of victims of violent crimes reported that the offender was under the influence of alcohol and/or drugs. A 1987 report by the Secretary of Health and Human Services cited that "alcohol is a key factor in up to 68% of manslaughters, 62% of assaults, 54% of murders/attempted murders, 48% of robberies, and 44% of burglaries" (Minnesota Institute of Public Health 1995). In addition to violent crimes, the problems of "DUI" or DWI" (driving under the influence or driving while intoxicated) crimes clearly pose dangers to law-abiding citizens.

Targeting Alcohol as an Urban Problem. More and more urban communities are stepping up responses to public inebriation and related problems, such as low-level crime, panhandling, and general blight. Colorado, New Jersey, New Mexico, and California all have state laws enforcing some form of liquor license-to-population ratio (Dunstan 1997). Cities such as Portland, New York, Los Angeles, Roseville, and Sacramento have all launched comprehensive efforts to combat public inebriation and its associated problems. Less successful efforts include those such as Oakland’s, which has targeted liquor store owners as the main source of the problem; efforts to limit liquor licenses and increase fines on problem establishments have been slowed by multiple lawsuits (Williams 1997).

The Relevance of Alcohol Availability. This paper will not argue that alcohol availability is the sole causal factor of crime problems in California cities. On the contrary, my model will control for numerous other causes such as the poverty level in the city, the racial makeup of the city, and the population density of the city. This paper is intended to explore whether there is a connection between alcohol and crime at all, and, if so, how strong that connection may be. As noted above, anti-crime efforts which target alcohol availability alone have not been overly successful; cities which have launched more comprehensive efforts, working with local businesses, police, and community leaders, have been much stronger. It is worthwhile from a policy standpoint, however, to know whether comprehensive crime reduction efforts should include measures to reduce alcohol availability. If regression analysis turns up no relationship whatsoever between alcohol availability and crime, or if much stronger relationships are found between other independent variables and crime, policymakers might be better served by moving their focus away from alcohol availability.

How This Paper Will be Presented. The next section of this paper provides a brief literature review, presenting several findings related to my hypothesis. Section III presents the model which will be used as the basis for my regression analysis. The data used in the model is discussed in Section IV. Section V presents the actual regression analysis. Section VI wraps up the paper and considers possible future research.

II. LITERATURE REVIEW

While volumes of literature exist regarding the relationship between alcohol and violence, much less can be found about the connection between alcohol availability and crime. This brief literature review surveys several findings that have been made on this issue and indicates some of the variables relating to crime.

Overview: An Economic Analysis of Crime. According to O’Sullivan (1996), an economic perspective of crime dictates that criminals base their decisions on how they perceive benefits and costs; i.e., the expected take versus the probability of imprisonment. In examining the demographics of crime, O’Sullivan argues that the poor are more likely to be victims of violent crime, while higher-income households are more likely to suffer from larceny crimes. Crime rates are higher in metropolitan areas and central cities than in suburban and nonmetropolitan areas, due to the high urban concentration of poverty, wealth, and population. Finally, Blacks tend to be victimized more frequently than whites. Police force increases have proven effective in reducing crime; according to one study, a 10 percent increase in the arrest ratio decreases the level of crime by about five percent. O’Sullivan concludes that policymakers must weigh the costs of crime prevention against the costs of crime.

Alcohol Availability, "Strangers," and Crime. Stephen Jarrell and Roy M. Howsen (1990) studied the relationship in Kentucky between the number of "strangers" in a given area and the levels of crime in that area. They measured "strangers" with variables representing tourists, college students, highway travelers, and shoppers at retail and liquor establishments. The model included a dummy variable to designate "dry" areas (counties which do not serve or sell alcohol) and "wet" areas (counties where alcohol is readily available). Significantly, Jarrell and Howsen found higher levels of each type of crime in "wet" counties than in "dry" counties, which they explain as follows:

The consumption of alcoholic beverages can alter the perpetrator’s benefit-cost ratio for the criminal act (increasing it), so that he/she is more likely to engage in the illegal activity, especially for crimes such as murder and assault. Also, the presence of liquor stores and bars increases the incidence of property crime, because liquor stores and bars are frequently targets of such criminal activity.

The authors also found that an increased number of strangers in an area positively affected nonviolent crimes, but had very little effect on violent crimes. They argue that "as the number of strangers increases in a particular area, certain crimes are expected to increase because apprehension becomes more difficult and the supply of potential victims increases." Finally, they found that unemployment rates have a positive impact on all types of crime. The authors conclude that anti-crime measures should be targeted toward areas associated with alcohol sales.

The Model for This Paper. Scribner, MacKinnon, and Dwyer (1995) used least squares regression analysis to examine the relationship between the rate of assaultive violence in 74 cities in Los Angeles County and alcohol-outlet density, economic structure, ethnicity, age structure, "urbanicity," and social structure. Their finding, which provided the inspiration for this paper, argued that

...higher levels of alcohol-outlet density are geographically associated with higher rates of assaultive violence. This association is independent of measured confounders, including city-level measures of unemployment, ethnic/racial makeup, income, age structure, city size, household size, and female-headed households.

While the authors admitted that no conclusions could be drawn regarding the direction of the relationship, and that 70% of the variance in the rate of assaultive violence was accounted for by sociodemographic variables, "adding the variable for alcohol-outlet density to the model yielded a significantly positive slope." According to their results, each outlet was associated with 3.4 additional assaultive violence offenses for the year studied (1990). The authors conclude that this evidence supports local efforts to reduce alcohol availability in an attempt to reduce alcohol-related problems.

Alcohol Availability Through Restaurants. An analysis by Gruenewald at al (1996) found no significant relationship between alcohol-related motor vehicle crashes and alcohol-outlet density. Interestingly, however, they did find significant relationships between restaurant densities and traffic crashes, implying that restaurants lead to more drunk driving than do alcohol retail outlets. Overall, they found, a 10% greater restaurant density was related to 1.7% higher crash rates. Moreover, they argued,

...rates of alcohol-related crashes within each study area were affected by restaurant density both within those areas and in surrounding areas, demonstrating the ‘spreading’ effects of alcohol availability in communities.

The authors concluded that communities should accordingly consider widening the scope of their prevention efforts.

Further Examination of the Alcohol-Outlet Density Theory. Gorman et al (1998) basically replicated the Scribner, et al study (see above), applying the model to 223 municipalities in New Jersey. They found no geographic association between assaultive violence rates and alcohol outlet density. They caution, however, that New Jersey and Los Angeles differ significantly in several of the variables used; the New Jersey cities have a lower rate of assaultive violence, lower population sizes, lower percentages of ethnics, higher median incomes, etc. They theorize that

…alcohol outlet density may relate to assaultive violence only when certain conditions prevail, for example, when average population size is large, alcohol outlet density crosses a certain threshold, and/or alcohol is sold through certain types of ‘easy access’ retail outlets such as mini-markets.

The authors further point out that the relationship between alcohol availability and alcohol-related behavior is not necessarily "fixed and unidirectional," and argue for more comprehensive research.

How I Plan to Use These Findings. These studies demonstrate some of the major variables which must be considered in analyzing causal factors of crime. As O’Sullivan pointed out, race, poverty, and population density are all connected to crime levels; accordingly, I will include measures for these factors. O’Sullivan further argues that police force sizes can affect crime rates; I will include a measure for police for each city. Jarrell and Howsen, Scribner et al, and Grunewald et al all found that alcohol availability was somehow related to crime; to account for this, I will use a measure of total alcohol availability (i.e., total liquor licenses issued) for each city. Scribner et al controlled for factors such as income and female-headed households; these variables will also be included in my model. Finally, Jarrell and Howsen, among others, have pointed out that the amount of retail which exists in a given area is often connected to the level of crime in that area; accordingly, I will include in my model a variable to measure retail activity.

III. MODEL

Dependent Variable and Causal Factors. The dependent variable for my regression model is the total number of crimes per California city with a 1990 population of 25,000 or more. This gave me a sample size of 215, a fairly large sample (the larger the sample size, the stronger the statistical analysis). Broad causal factors which are generally associated with crime include social structure, economic structure, public safety structure, and demographic characteristics. This paper will also examine two additional factors which some researchers believe are linked to crime; specifically, alcohol availability and retail activity.

The Model. The model is given below as the following function:

Crime in California Cities = f (Alcohol Availability, Economic Characteristics of Citizens, Demographic Characteristics of Citizens, Social Characteristics of Citizens, Size of City, Level of Retail Activity, Level of Law Enforcement, Regional Area).

Specific variables were chosen to proxy for these causal factors as follows:

Alcohol Availability = f (Liquor Licenses per City),

Economic Characteristics = f (Median Income per City, Percentage of Persons per City with Income Below Poverty Level),

Demographic Characteristics = f (Percentage of Black Persons per City, Percentage of Hispanic Persons per City, Percentage of Persons per City with Bachelor’s Degree),

Social Characteristics = f (Percentage of Female-Headed Households per City),

Size of City = f (Persons per Square Kilometer per City),

Retail Activity = f (Paid Employees per City in Retail Trade),

Law Enforcement = f (Total Police Employees per City),

Regional Area = f (Regional Dummy Variables Representing North, Bay Area, Valley, Coast, and South Areas of California).

Expected Signs

Hypothesis. Predicting the signs of the independent variables in this model is relatively straightforward, as much research has been done on causal factors of crime. First, as indicated above, my hypothesis argues that alcohol availability is positively related to the level of crime in a city; therefore, the sign is positive. This agrees with the Scribner et al and Jarrell and Howsen studies, which found some degree of positive correlation between alcohol availability and crime.

Economic Characteristics. The signs of the variables representing economic characteristics are expected to vary. Scribner et al argue that economic theory dictates that "poverty and lack of opportunity predispose individuals to violent behavior." O’Sullivan, on the other hand, contends that the poor are more likely to be victims of violent crime (see Literature Review). Accordingly, I expect median income to exert a negative effect upon crime; as the median income in a city rises, the crime level should drop. Conversely, the poverty level is expected to exert a positive effect; as the number of persons with income below the poverty level increases, the level of crime should increase.

Demographic Characteristics. Demographic characteristics are expected to exert varying effects. As the nonwhite population of a city grows (i.e., the "Black" and "Hispanic" variables), the crime level is likely to increase. (While this may sound like a racist argument, Scribner et al point out that "minority status is strongly associated with being a victim of violence"). Conversely, a higher overall level of education should exert a negative effect; as the number of people with bachelor’s degrees increases, crime levels should drop. This is in keeping with the expected drop in crime when median income rises or poverty falls, as discussed above; college education should lead to higher incomes and less poverty.

Social Characteristics. Social characteristics are expected to exert a positive effect. Scribner et al indicate that

As the emerging Black middle class has moved out of inner-city areas, the remaining residents have ever-declining resources. These communities can be identified by a high rate of female-headed households.

If the number of female-headed households is a good indicator of inner-city, low-income areas, both of which are generally expected to be associated with crime, it follows that this variable will come out with a positive sign.

City Size. The size of a city is also expected to exert a positive effect. As the density of an area’s population grows, more potential exists for crime to increase; as discussed in the literature review, urban centers tend to have a high concentration of poverty, wealth, and population, leading to higher levels of crime.

Retail Activity. I expect retail activity to come out with a positive sign. As discussed in the literature review, higher levels of retail activity in an area tend to be connected with higher levels of crime due to increased levels of "strangers," wealth, etc.

Law Enforcement. Law enforcement is expected to exert a negative effect. As the number of police officers per city grows, the level of crime should decrease; in O’Sullivan’s economist terms, a greater police forces increases the "costs" of committing a crime (i.e., increased probability of arrest). If the size of the police force has no effect whatsoever upon the level of crime, the policy rationale for funding police forces would appear to be in question.

Regional Area. Finally, the regional dummy variables are expected to exert varying directional effects. I expect crime levels to be higher in the Bay Area and in Southern California (and in Los Angeles, the omitted variable), since they are highly urbanized regions; as discussed above, density is expected to exert a positive effect upon crime. I expect crime in the Northern California region to be lower, since it is made up of cities with lower populations. I am uncertain as to the direction of the effects for the Coastal and Valley regions.

IV. DATA

The variables used to proxy for the broad causal factors in this model were selected based on readings done for the literature review. As noted above, so much research and analysis has been done on the issue of crime that it is fairly easy to find proxies with which to measure broad causal factors.

Proxies. Most of the specific proxies have already been discussed above; this section will provide a summary. While some debate exists as to whether causal factors differ for violent versus nonviolent crime, my model will take a broad approach by using total crime as the dependent variable. Similarly, alcohol availability will be measured by total liquor licenses per city, rather than breaking down availability into on-premise (i.e., restaurant) and off-premise (i.e., liquor stores) licenses. Median household income and the percentage of city residents below the poverty level will proxy for each city’s economic structure. The demographic breakdown for each city will be represented by measures of Black and Hispanic populations, as well as a measure for the percentage of city residents who hold a Bachelor’s degree. Social structure will be measured by the percentage of female-headed households per city. City size will be measured by population density; specifically, the number of persons per square kilometer per city. Retail activity will be measured via the number of paid employees in the retail trade per city, while law enforcement will be represented by the number of total police employees per city. (These last two are the most easily available statistics and appear to be logical measures). Finally, regional dummy variables (based on a breakdown performed by other researchers) will be included to measure the effects of urban and rural communities upon crime in California cities.

Tables. The three tables on the following pages provide information on the variables in my model. Table I lists the variable name, description, and data source. Table II lists the descriptive statistics for each variable. Table III presents a correlation matrix for the independent variables.

TABLE 1: LIST OF VARIABLES WITH SOURCES

TABLE II: DESCRIPTIVE STATISTICS

TABLE III: CORRELATION MATRIX (PART ONE)

* Denotes significance at the 95% level (2-tailed).

**Denotes significance at the 99% level (2-tailed).

TABLE III: CORRELATION MATRIX (PART TWO)

TABLE III: CORRELATION MATRIX (PART THREE)

In all correlation matrix tables, * denotes significance at the 95% level (2-tailed) and ** denotes significance at the 99% level (2-tailed).

TABLE IV: REGRESSION RESULTS

* Denotes coefficients that are significant with greater than 80% and less than 90% confidence. ** Denotes coefficients that are significant with greater than 90% and less than 99% confidence. *** Denotes coefficients that are significant with greater than 99% or greater confidence.

V. REGRESSION ANALYSIS

This section will present the results of the regression analysis which are outlined in Table 4 above. Different functional forms will be compared, multicollinearity problems will be explored, and corrections to refine the model will be discussed.

Linear Regression Results. The results from the initial linear regression are presented in the "Standard OLS Results" column of Table 4. Nine of the fifteen variables (ALCPOP, INCOME, POVERTY, BLACK, DENSITY, RTAILPOP, POLCEPOP, AND SOUTH) were significant at a confidence level of 85% or above. Three of the variables (HISPANIC, POLCEPOP, AND BAYAREA) did not exhibit the signs predicted in the model.

Log-Log Regression Results. Next, I ran a log-log regression, the results of which are presented in the "Log-Log Results" column of Table 4. Eight of the fifteen variables (LNALCPOP, LNPOVRTY, LNBLACK, LNFHHPOP, LNDNSTY, LNRTLPOP, LNPOLPOP, AND SOUTH) were significant at a confidence level of 85% or higher. Four of the variables (LNINCOME, LNPOLPOP, NORTH, and BAYAREA) came out with unexpected signs.

Multicollinearity. The initial model included population as an explanatory variable, instead of controlling for it within each variable. This resulted in a very high degree of multicollinearity; a large number of variables came out with an r of 0.9. I then dropped the population variable and instead divided the relevant variables by population, which significantly reduced multicollinearity. The correlation matrix run for the linear regression, however, still exhibited a high degree of collinearity (seven variables came out with a correlation of 0.8 or above). In an effort to reduce this problem, I ran a correlation on the logged variables instead. In this case, only one instance of collinearity arose (LNINCOME and LNPOVERTY). Since I had already dropped a variable representing unemployment, and the income and poverty measures were expected to exhibit opposite directional signs, I decided to keep both the income and poverty variables as good measures of economic structure.

F-Test. I next ran an F test to determine whether the regional dummy variables exerted a large enough collective effect to be included in the final model. The F test compared the overall fit of the model equation including the regional dummies, to a revised equation which dropped the dummies. This is accomplished by calculating the following equation:

(RSSm – RSS)/ M
F = RSS/ (n-K-1)

where,

RSSm = the residual sum of squares from the original regression (including regional dummies) = 9.279
RSS =the residual sum of squares from the revised regression (not including regional dummies) = 10.241
M = the number of regional dummies in the original regression = 4
K = the number of coefficients in the original regression = 15
n-K-1 = the number in the sample minus the number of coefficients minus one = 215-15-1 = 199

The equation yields a calculated F value of 0.389. The critical F value significantly exceeds this at 1.83 (assuming a 5% level of confidence, 12 degrees of freedom in the numerator, and 120 degrees of freedom in the denominator). Because the calculated F value did not exceed the critical F value, the null hypothesis, which states that the regional dummies exert zero effect, cannot be rejected. Therefore, I concluded that the regional dummies could conceivably be removed from the equation, since they do not appear to exert any significant effect.

Testing for Heteroskedacity. I was hopeful that heteroskedacity would not be a problem in my regression, since I had converted most of the variables to percentages by dividing them through by population and I was using a log functional form (percentages and log forms are both methods which theoretically minimize heteroskedacity) . However, when I ran a Park Test, I found a 93.9% level of confidence that heteroskedacity did indeed exist. The Park Test was performed by first recovering the residuals, then squaring them. Then the log of the squared residuals was computed and regressed against the log of the "Z" proportionality factor. I chose population as my Z, since a good Z measures the size of the observation relative to the dependent variable.

Correcting for Heteroskedacity. Once I discovered the heteroskedacity problem, I performed the Weighted Least Squares (WLS) procedure to correct for it. WLS removes the scale effect by dividing through all the logged variables by (unlogged) population; it effectively "weights" each variable by population. The WLS results are reported in Table 4 ("Log-Log Corrected for Heteroskedacity"). In the corrected results, eight of the fifteen variables were significant at a level of 85% or above; only two variables (LNPOLPOP and BAYAREA) exhibited signs that were not predicted by the model. Endogeneity. It is quite likely that my model contains endogeneity problems. Endogenous variables are variables which are simultaneously determined; this dual causality effect can make it difficult to determine the impact of one variable (sort of a chicken-and-egg relationship). For example, the income and crime variables are probably endogenous: on the one hand, low-income areas tend to be crime-ridden because people are poor and perhaps steal to help survive; on the other hand, crime-ridden areas tend to attract lower-income people because the property values (and therefore the rents or housing prices) are lower because of the high crime rates. In order to correct for endogeneity, I would need to run a Two-Stage Least Squares regression (2SLS). 2SLS is a method of creating instrumental variables to replace endogenous variables in simultaneous equations systems. An instrumental variable is a proxy for one of the endogenous variables; the instrumental variable must be both a good substitute for the removed variable, and totally unrelated to the other endogenous variable. Time constraints prevented me from undertaking this difficult task.

VI. CONCLUSION

This concluding section will discuss the final corrected regression results (as shown in the ""Log-Log Corrected for Heteroskedacity" column of Table 4). As discussed above, endogeneity was not corrected for; in a longer-term project, such as a thesis, it would need to be addressed.

Elasticities. Since all of the relevant variables in the model were logged, there is no need to convert them to elasticities; elasticity can be interpreted via the coefficients of the variables. The elasticities are thus exhibited in Table 5 below:

TABLE V: COEFFICIENTS AND ELASTICITIES

**Denotes significance at a greater than 80% but less than 90% level.

***Denotes significance at a 99% or greater level.

Elasticity measures the responsiveness of the dependent variable to a given change in an independent variable (holding all other independent variables constant). For example, the elasticity for the alcohol variable is 0.26; this implies that as alcohol availability rises by 1%, crime rises by 0.26%. All of the elasticities in my model are less than one, meaning that that all of these variables are quite inelastic; a change in any of these independent variables will lead to only a very small change in the dependent variable. This is not an overly surprising result. There are so many different causal factors of crime that it would probably be more unexpected to find a single causal variable exerting a very strong impact. The rather discouraging conclusion this leads to, however, is that the policy impact of changing any one of these variables will be minimal; any successful effort to reduce crime levels clearly needs to be quite broad-based.

Significant Coefficients. Five variables had coefficients that were significant at the 99% level or above; these were the variables representing alcohol availability, poverty, Black population, police force, and the regional dummy variable representing Southern California. Four other variables were significant between the 90-99% level: density, retail, and the Bay Area and Valley regional dummies. The Northern California and Coastal California regional dummies were not significant. Neither, surprisingly, were the variables representing income, Hispanic population, persons with Bachelor’s degrees, and female-headed households. Perhaps, however, they were simply redundant variables; after all, income and poverty did show collinearity. It is possible that the poverty and Black population variables were sufficient measures of socioeconomic status.

Expected Signs. Most variables came out with signs predicted in the original model; many of these variables, however, were not significant. Of the significant variables, alcohol availability, poverty, Black population, population density, and retail activity all came out positive as expected. Surprisingly, however, the police variable came out positive rather than negative, implying that as the police force increases, so does crime. This may be due to endogeneity problems, which were not corrected for (as discussed above). The regional dummy results were mixed; while North and South came out as expected, the Bay Area variable came out with the opposite sign (the Valley and Coast variables were question marks in the original prediction). As discussed in the F-Test section above, however, the regional dummies could be dropped from the model.

Significance v. Elasticity. Of the significant non-dummy variables – alcohol availability, poverty, Black population, population density, retail activity, and police force – poverty was the most elastic at 0.31. The least elastic was the density variable at 0.05. Both were significant at the 99% level or above. As discussed above, all of the significant variables came out quite inelastic.

R-Squared Interpretation. The R Squared, or coefficient of determination, represents how well the estimated regression equation fits the data sample. Specifically, it is defined as the percentage of the variation of the dependent variable around the mean that is explained by the equation. The closer the R Squared is to one, the better the fit; an R Squared close to zero shows that the equation does not fit the data well and that the mean would be a better estimator. The R Squared for the linear regression was actually fairly high (0.681); the R Squared for the log-log regression was slightly higher (0.696). The R Squared for the final model (adjusted for heteroskedacity) was the highest of the three (0.798). While this statistic is a good measure for overall fit, however, it is more important to first look at the significance of the individual coefficients. The R Squared is more useful in comparing models with different functional forms, data sets, or combinations of independent variables.

Thesis Question. As predicted in the original model, the final regression results show that alcohol availability does exert a positive effect, at a high level of significance, upon the level of crime in a city. As previously discussed, however, all of the causal variables, including alcohol, exerted only small effects upon crime. This clearly implies that any policy effort to reduce crime in a given city needs to be broad-based, giving consideration to all of these factors, rather than targeting a single factor such as alcohol availability.

Master’s Thesis Potential. I plan to base my Master’s Thesis upon the hypothesis of this paper. I believe there is a great deal more work to be done with this model. The endogeneity problem discussed above clearly needs to be addressed in order to test the integrity of the model. I would also like to compare different regressions using violent crime and nonviolent crime and the dependent variable, as opposed to simply using total crime as in this paper. It might also be interesting to try using off-sale and on-sale alcohol licenses to represent alcohol availability to see whether any significant differences emerge. I plan to submit this paper to my community contact, John Peirce, Staff Counsel for the State Department of Alcoholic Beverage Control, to solicit his feedback and suggestions on other avenues to be explored. With so little literature to be found on the alcohol availability-crime link, and with crime being such a major social issue, I believe that attempts to add to this knowledge are worthwhile.

APPENDIX

Values for the following were entered as averages due to unavailable data. Values were calculated for each city by averaging the values for cities with population within 10,000 of that city. Encinitas, Laguna Niguel, Temecula and Yucaipa were removed from the sample, as data for three variables (crime, police, and alcohol) were not available for each of these cities.

REFERENCES

Department of Alcoholic Beverage Control. 1990. "Alcoholic Beverage Licenses as of June 30, 1990." Sacramento: State of California Department of Alcoholic Beverage Control.

Dunstan, Roger, California Research Bureau. June 17, 1997. Memo to Erin Keyes, c/o Assemblymember Michael Sweeney, re: Liquor Licenses.

Federal Bureau of Investigation. 1990. Uniform Crime Reports for the United States. Washington, DC: US Department of Justice.

Gorman, Dennis M., Speer, Paul W., Labouvie, Erich W., and Apana P. Subaiya. "Risk of Assaultive Violence and Alcohol Availability in New Jersey." American Journal of Public Health, January 1988, vol. 88, no.1.

Gruenewald, Paul J., Millar, Alexander B., and Peter Roeper. "Access to Alcohol: Geography and Prevention for Local Communities." Alcohol Health and Research World, Fall 1996, vol. 20, no. 4.

Jarrell, Stephen and Roy M. Howsen. "Transient Crowding and Crime: The More ‘Strangers’ in an Area, the More Crime Except for Murder, Assault, and Rape." Journal of Economics and Sociology, October 1990, vol. 49, no. 4.

National Center on Addiction and Substance Abuse. 1997. "Behind Bars: Substance Abuse and American’s Prison Population." New York: Columbia University.

O’Sullivan, Arthur. 1996. "Crime and Punishment." Chapter 22 of Urban Economics. 3^rd Ed. Chicago: Irwin, A Times Mirror Higher Education Group, Inc. Company.

Pacific Center for Violence Prevention. April 1996. "Alcohol-Related Violence." http://www.pcvp.org/

Pacific Center for Violence Prevention. April 1996. "Alcohol Related Violence: Cost." http://www.pcvp.org/

Scribner, Richard A., McKinnon, David P., and James H. Dwyer. "The Risk of Assaultive Violence and Alcohol Availability in Los Angeles County." American Journal of Public Health, March 1995, vol. 85, no.3.

Slater, Courtenay M. and George E. Hall, eds. 1993. 1993 County and City Extra: Annual Metro, City, and County Data Book. Lanham, MD: Bernan Press.

State of California. Health and Welfare Agency. Department of Alcohol and Drug Programs. June 1997. Alcohol and Other Drugs Databook. Sacramento: Department of Alcohol and Drug Programs.

US Bureau of the Census. 1994. County and City Data Book: 1994. Washington, DC: US Government Printing Office.

US Department of Justice. Office of Justice Programs. Bureau of Justice Statistics. September 1994. Fact Sheet: Drug-Related Crime. Washington: Department of Justice.

Williams, Diana A. "New Age of Enforcement: Oakland Launches Concerted Attack on Liquor-Store Loitering." New Frontiers in Alcoholic Beverage Nuisance Abatement Management: November 1997. League of California Cities: Sacramento, California.

Return to Course Sylabi and Information Page