Department of Epidemiology and Biostatistics, Texas A & M University, System Health Science Center, School of Rural Public Health, Bryan, TX, USA
* Author to whom correspondence should be addressed at: Department of Epidemiology and Biostatistics, School of Rural Public Health, Suite 310, 3000 Briarcrest Drive, Bryan, Texas 77802, USA. Tel.: +979 458 2236; Fax: +979 862 8371; E-mail: gorman{at}srph.tamushsc.edu
(Received 10 February 2004; first review notified 2 March 2004; in revised form 6 March 2004; accepted 7 March 2004)
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
One aspect of the built environment that has received increased attention in recent years has been the location and concentration of alcohol outlets, especially in urban neighbourhoods (Lipton et al., 2003). Much of the early research in this area was limited by statistical weaknesses and a reliance on aggregated datasets pertaining to large units of analysis such as states and counties (Gruenewald, 1993
; Stockwell and Gruenewald, 2001
). The statistical weaknesses arose from the failure of ordinary least squares analyses to account for spatial autocorrelations that can bias statistical estimates of effects and lead to either Type I (in the case of positive spatial autocorrelation) or Type II (in the case of negative spatial autocorrelation) errors (Gruenewald et al., 2000
). Generalized least squares regression models that estimate and correct for residual spatial autocorrelation between geographic units can be used to guard against such failures of unit independence (Griffith, 1988
) and have been employed in some recent studies (Gruenewald et al., 2000
; Gorman et al., 2001
; Lipton and Gruenewald, 2002
). In addition, it is now recognized that the relationship between alcohol availability and violence is best studied at a relatively small level of analysis, and most recent studies use cities or sub-divisions of these (e.g. census tract) as the geographic unit of analysis (Stockwell and Gruenewald, 2001
; Lipton et al., 2003
).
The present study is intended to build on previous research, and specifically to address the following three issues. First, does the observed relationship between alcohol outlet density and violent crime exist within cities that are both larger than, and in a different part of the US, from those that have previously been studied? The foci of our study are the cities of Austin and San Antonio in the State of Texas, which had populations of 656 562 and 1 144 646, respectively, in 2000. Previous city-level studies have focused on places with populations (at the time that the studies were conducted) under 600 000, such as Camden, Cleveland and Newark (Roncek and Maier, 1991; Speer et al., 1998
; Gorman et al., 2001
). Only Scribner et al. (1999)
and Costanza et al. (2001)
have assessed the relationship between alcohol outlet density and violent crime in cities in the southern US. Second, what are the implications of using generalized least-squares analysis in assessing the relationship between alcohol outlet density and violent crime? Gruenewald and colleagues (1996
, 2000
) have described in detail the potential problems that emanate from the use of ordinary least square analysis in geospatial studies of the effects of alcohol availability. However, Scribner et al. (1999)
have argued that the extent of the problems emanating from spatial autocorrelation have yet to be empirically demonstrated in this field of research. Our study will add to the small body of research that has addressed this issue. Third, is the relationship between alcohol availability and violence contingent upon specific neighbourhood context? For example, might the effect of alcohol outlet density be limited to relatively small geographic areas in some cities but be more diffuse in others, depending on systemic features such as sociodemographic composition and transportation networks (Stockwell and Grunewald, 2001
; Gorman et al., 2004
). In the present study we assess this primarily through an examination of the effects of alcohol outlet density and other covariates across geographic units.
![]() |
MATERIALS AND METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Data
Three datasets were employed in the study: one pertaining to alcohol availability, one to violent crime, and one to neighbourhood sociostructural characteristics. For the alcohol availability variables a list of active alcohol outlets in Austin and San Antonio was obtained from the website of the Texas Alcoholic Beverage Commission, 2000. Each record in the dataset included the name, geographic location and type of permit or license of the outlet. There were 1486 alcohol outlets in the City of Austin in 2000, and 2690 in the City of San Antonio. Each alcohol outlet record was geocoded by street address. Geocoding rates were very high in each of the cities 100% for Austin and 99.5% for San Antonio.
Like most of its counterparts in other states, the Texas Alcoholic Beverage Commission employs a permit classification system that allows outlets to be distinguished by types of beverages sold and type of consumption allowed (specifically, off-premise versus on-premise). Indeed, the State of Texas employs a rather complex system of primary (e.g. mixed beverage) and secondary (e.g. food and beverage) licences; an outlet must have the former in order to obtain the latter. In addition, outlets can have more than one primary licence for example, both a mixed beverage permit and a beer/wine retailer permit. Moreover, in some cases one of these permits may be for on-sale and the other for off-sale. In line with previous studies, analyses were conducted separately for on-sale only outlets, off-sale only outlets and combined on-/off-sale outlets, as well as for total outlet densities. Of the 1486 outlets in Austin, 486 (32.7%) were on-sale, 612 (41.2%) were off-sale, and 388 (26.1%) were combined on-/off-sale. In San Antonio, 610 (22.7%) of the total 2690 outlets were on-sale, 1113 (41.4%) were off-sale, and 967 (35.9%) were combined on-/off-sale. In the analysis, census tract densities were entered as outlets per 100 persons.
The violent crime data available for each city differed in respect to type of crime reported. Data pertaining to reports of violent crime (murder, rape, robbery and aggregated assault i.e. crimes involving an unlawful attack by one person on another for the purpose of inflicting severe bodily harm, usually through the use of a weapon) for the City of Austin were extracted from the website of the city police department. The Austin police department posts data pertaining to the occurrence of violent crime on a monthly basis. The data contained on the website are aggregated up to the census tract level, and are based on first reports of offences (that is, before investigation and final classification of crimes). Such call for assistance data have been used in previous studies of the relationship between alcohol availability and violent crime, and have strengths as well as limitations relative to official crime records (such as the state-level Uniform Crime Reports). Specifically, such data are subject to less refinement than are official records and are therefore considered to capture events that better reflect the criminogenic nature of geographic locations (Sherman et al., 1989; Nelson et al., 2001
). However, given that official data have been used in most previous studies, a comparison of the total reports of violent crime contained on the police department website for the year 2000 with the official total contained in the 2000 Uniform Crime Reports (Texas Department of Public Safety, 2001
) was conducted. Interestingly, this showed that the former had 295 (9.6%) fewer crimes than the latter (2779 versus 3074). Most of this discrepancy was accounted for by differences in the aggregated assault category, in which there were 283 fewer cases in the website reports than the official Uniform Crime Reports (1683 vs 1400; 16.8%). There were just 24 fewer robberies contained in the website reports than in the Uniform Crime Reports (328 vs 352; 6.8%), while the former contained slightly more robberies than the latter (1018 vs 1006; 1.2%). The number of murders (33) contained in the website reports and Uniform Crime Reports were identical. Given the size of the discrepancy between the two data sources in the aggregated assault category, this was excluded from the analyses.
For San Antonio, violent crime arrest data for 2002 were obtained directly from the city police department. There were a total of 10 465 violent crimes for the year, of which 99 (0.9%) were murders, 527 (5.0%) rapes, 2221 (21.2%) robberies, and 7618 (72.8%) assaults. In the analysis, the census tract violent crime rate for both Austin and San Antonio was calculated as crimes per 100 persons.
Finally, data pertaining to 12 neighbourhood characteristics were extracted from Summary File 1 and Summary File 3 of the 2000 US Census (US Census Bureau, 2003). These variables, that were grouped under three broad headings, were chosen as they had been used in previous ecologic studies of alcohol availability and violent crime (Gorman et al., 2001
; Lipton and Gruenewald, 2002
), as well as studies of violent crime in urban neighbourhoods (Peterson et al., 2000
). Of the 12 neighbourhood sociostructural variables, six were measures of concentrated disadvantage [percent families below poverty line, per cent families receiving public assistance, per cent unemployed individuals (16+) in civilian workforce, per cent female-headed households with children, per cent black, and per cent Latino], three were measures of residential instability (per cent residents over the age 5 years who have lived in the same house for 5 or more years, per cent homes that are owner-occupied, and per cent vacancy rate), and three were sociodemographic measures of the resident population (adult : child ratio : ratio of adults (18 or older) to children under age 18, population density: number of persons per square mile, and percent population that is male and 1524 years).
Data analyses
First, basic descriptive analyses (means and standard deviations) were calculated and all variables were then transformed to their natural logarithm to adjust for skew. Next, bivariate and multivariate regression analyses were conducted to examine the relationship between the neighbourhood sociostructural characteristics, alcohol outlet densities and violent crime. Multivariate regression analysis with stepwise selection (using P = 0.05 for retention) was used to produce the most parsimonious sociostructural models for each city. As the stepwise procedure can lead to biased estimates for observational data, special cares should be taken to select a candidate pool of explanatory variables. Elimination of key covariates can seriously damage the explanatory power, while inclusion of too many independent variables often results in smaller statistical power. The candidate covariates in this study cover 16 factors pertaining to neighbourhood sociostructural characteristics and alcohol availability. Before performing the regression analysis, the distribution of variables was checked to make sure they were not omitted in the selection process due to a narrow range of values. Residual plots after the regression analysis were also checked for nonrandom pattern which indicates that some important independent variable was not incorporated in the model.
As noted above, analyses of small area data are subject to a variety of biases due to unobserved correlations between geographic units (i.e. spatial autocorrelations) and spill-over of the effects of some measures (e.g. densities of alcohol outlets) on outcomes observed in adjacent areas (e.g. violent crime rates). In order to detect and correct for spatial autocorrelated errors in the current analysis as well as to assess the potential dynamic effects between geographic units, three specialized spatial analyses were conducted. The first extended the bivariate and ordinary least squares regression analysis by including tests for spatial autocorrelationsspecifically the raw Moran coefficient in the bivariate analysis and Moran's I in the multivariate analysis. The second analysis further addressed the issue of potential unit dependence by taking into account the correlated error between adjacent census tracts (Gruenewald et al., 2000). Here one is testing whether the effects observed in the ordinary least squares (OLS) model can be replicated under more rigorous analytic conditions using generalized least squares (GLS) analysis (Griffith, 1988
). Specifically, for each model the spatial autocorrelation coefficient,
, is presented along with a test of its significance. The value of
ranges from 1.00 to +1.00; Type I errors in analysis occur when the value of
is less than 0, and Type II errors when it is greater than 0. Finally, the third form of spatial analysis presented examined the effects of independent variables measured in adjacent census tracts on rates of violence in target census tractsso-called spatial effects or spatial lags (Gruenewald et al., 2000
). Specialized geostatistical software developed by Gruenewald and colleagues was used in the spatial analyses. Details of the software and the computations used in the analyses are contained elsewhere (Gruenewald et al., 1996
; 2000
; Ponicki and Gruenewald, 1998
).
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
|
Bivariate associations among variables are also presented in Tables 1 and 2 for Austin and San Antonio, respectively. In the former city, violent crime rate was positively associated with alcohol outlet density and all of the neighbourhood sociostructural variables except vacant housing. Total alcohol outlet density was positively correlated with three of the five sociostructural variables. In San Antonio, the violent crime rate was positively associated with five of the six neighbourhood variables, and negatively associated with the other. Total alcohol outlet density was not significantly associated with either of the sociodemographic variables, but was negatively associated with one of the measures of residential instability and positively correlated with the remaining three neighbourhood sociostructural variables.
The Raw Moran coefficient measures presented at the bottom of Tables 1 and 2 assess the extent to which the observations in one geographic unit resemble those in geographic units. Both the Austin and San Antonio analyses show that for each variable included in the model there was significant positive spatial autocorrelation. This means that data pertaining to each variable (e.g. alcohol outlet density) collected from any single census tract tended to resemble the data pertaining to that variable collected from adjacent census tracts. Such a pattern of spatial autocorrelation suggests that OLS analysis of these data would exhibit considerable Type I errors.
Table 3 presents the results of the multivariate and spatial analyses for the City of Austin. Model 1 considers only the neighbourhood sociostructural covariates in the ordinary least square analysis. Six covariates remained after stepwise selection procedure and they explained close to 59% of the variability in violent crime rates in the city. Adding alcohol outlet densities (total, on-sale, off-sale, and combined on-/off-sale) to the neighbourhood sociostructural covariates and going through stepwise selection again, Model 2 eliminated the significance of adult : child ratio and diminished the effect of per cent Latino by 41%. All of the coefficients in the model were positively associated with violent crime. This model explained 64% of the variability in violence rates. Model 3 keeps all the covariates in Model 2 and adds the first order spatial lag for each of them. None of the lags for the sociostructural variables was statistically significant, so only the result for the spatial lag for total alcohol outlet density (which was significant) is presented. The introduction of this variable reduces the effect of alcohol outlet density in the target census tract by 31%. It also reduces the magnitude of four of the five neighbourhood sociostructural coefficients, but the effects are smaller than that found for outlet density in the target tract (e.g. a 19% reduction for per cent living in poverty and a 15% reduction for population density).
|
The results of the multivariate and spatial analyses for San Antonio are shown in Table 4. Model 1 (the model of neighbourhood sociostructural variables based on OLS analysis) explained close to 39% of the variability in violent crime rates in the city. Total alcohol outlet density was added in Model 2. This eliminated the significance of per cent vacant housing and population density while the coefficient for per cent female-headed households was now significant. This variable and per cent Latino were both positively related to violent crime, while per cent owner-occupier housing and per cent men aged 1524 years were both negatively related to violence rates in Model 2. This model explained an additional 4% of the variance in violent crime. The first order spatial lags for covariates were added in Model 3. The lag for total alcohol outlet density and per cent female-headed household were both statistically significant. The addition of these variables to the model eliminated the effect of per cent female-headed households in the target census tract. Model 3 explained 48% of the variability in violent crime rates.
|
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
As noted above, the spatial analysis presented enabled identification and correction for spatially-autocorrelated errors along with assessment of the potential dynamics effects between geographic units. The assessment of these first order lagged effects using OLS analysis indicated that alcohol outlet density in adjacent areas influenced rates of violence in target areas in both Austin and San Antonio. In San Antonio, however, the effect was eliminated once the autocorrelated error was taken into account. This was in line with our previous analysis in Camden, New Jersey, which also found the effects of alcohol outlet densities on violent crime rates to be spatially limited (Gorman et al., 2001). Such differences suggest that the spatial effects of alcohol availability on violence may vary from location to location, thereby indicating the importance of understanding this relationship within a specific community context (Gorman et al., 2004
; Stockwell and Gruenewald, 2001
).
The spatial analyses also demonstrate the importance of controlling for autocorrelated error in small area analyses of the association between alcohol availability and violence. Scribner et al. (1999) recently observed that the extent of the potential problems emanating from spatial autocorrelation had yet to be determined in this field of research. And, indeed, in our previous study in Camden, spatial autocorrelation was not found to be a significant source of bias (Gorman et al., 2001
). However, in the present study considerable Type I error was present in the models estimated using OLS analysis. Specifically, in San Antonio, the first order lag of alcohol outlet density was no longer significant once the effects of autocorrelated error were taken into account. In Austin, a number of sociostructural variables appeared significant in the OLS model but were not found to be in the more rigorous GLS analysis. Thus, the model for Austin was quite different from that found in our earlier analysis of Camden, New Jersey or in other ecologic studies of crime in large urban centres such as Chicago, Illinois and Columbus, Ohio (Sampson et al., 1997
; Peterson et al., 2000
). One possible reason for this is that aggregated assault was excluded from the Austin analysis, whereas it was included in previous studies. In addition, it may be that the underlying social mechanisms related to crime in older urban centres in the northeast and mid-west US are different from those in a southwestern city such as Austin (Ousey, 2000
). Again, this points to the importance of understanding the alcohol-violence relationship within a specific neighbourhood context.
The latter points to one of the main limitations of the present study, namely that it only took into account fairly limited aspects of neighbourhood ecology. The study focuses on the socioeconomic and sociodemographic composition of census tracts along with alcohol outlet density, but fails to account for other aspects of neighbourhood life that influence the occurrence of acts of violence. These include other potential crime attractor locations such as convenience stores and major street intersections (Block and Block, 1995; Nelson and Bromley, 2001
), signs of physical and social disorder (Skogan, 1990), as well as cultural aspects of neighbourhood life such as social integration and cohesion (Sampson et al., 1997
). A second limitation of the study is that it is unable to address the question of exactly what it is about alcohol outlets that is important in explaining violent crime. For example, is it the density of outlets per se that matters or is it that attractor bars tend to be found in areas of high outlet density (Block and Block, 1995
)? More detailed micro-spatial and temporal analysis of the type undertaken in recent British studies (Nelson et al., 2001
; Bromley and Nelson, 2002
) will be needed to answer such questions.
![]() |
ACKNOWLEDGEMENTS |
---|
![]() |
REFERENCES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Bottoms, A. E. and Wiles, P. (1997) Environmental criminology. In The Oxford Handbook of Criminology, Maguire, M., Morgan, R. and Reiner, R., eds, pp. 305359. Oxford University Press, Oxford.
Brantingham, P. L. and Brantingham, P. J. (1993) Nodes, paths and edges: considerations on the complexity of crime and the physical environment. Journal of Environmental Psychology 13, 328.[ISI]
Bromley, D. F. B. and Nelson, A. L. (2002) Alcohol-related crime and disorder across urban space and time: evidence from a British city. Geoforum 33, 239254.[CrossRef][ISI]
Costanza, S. E., Bankston, W. B. and Shihadeh, E. (2001). Alcohol availability and violent crime rates: a spatial analysis. Journal of Crime and Justice 24, 7183
Farrington, D. P. and Loeber, R. (2000) Epidemiology of juvenile violence. Juvenile Violence 9, 733748.
Gorman, D. M., Speer, P. W., Gruenewald, P. G. and Labouvie, E. W. (2001) Spatial dynamics of alcohol availability, neighborhood structure and violent crime. Journal of Studies on Alcohol 62, 628636.[ISI][Medline]
Gorman, D. M., Gruenewald, P. J., Hanlon, P. J., Mezic, I., Waller, L., Castillo-Chavez, C., Bradley, E. and Mezic, J. (2004). Implications of systems dynamic models and control theory for environmental approaches to the prevention of alcohol- and drug-related problems. Substance Use and Misuse.
Griffith, D. A. (1988) Advanced Spatial Statistics. Kluwer, Dordrecht, The Netherlands.
Gruenewald, P. J. (1993) Alcohol problems and the control of availability: theoretical and empirical issues. In Economics and the Prevention of Alcohol-Related Problems, NIAAA Research Monograph No. 25, Hilton, M. E. and Bloss G, eds, pp. 5990. National Institute on Alcohol Abuse and Alcoholism, Rockville, MD.
Gruenewald, P. J., Millar, A. B., Treno, A. J., Yang, Z., Ponicki, W. R. and Roeper P. (1996) The geography of availability and driving after drinking. Addiction 91, 967983.[CrossRef][ISI][Medline]
Gruenewald, P. J., Millar, A. B., Ponicki, W. R. and Brinkley G. (2000) Physical and economic access to alcohol: the application of geostatistical methods to small area analysis in community settings. In The Epidemiology of Alcohol Problems in Small Geographic Areas, Wilson, R. A. and Dufour, M. C., eds, pp. 163212. National Institute on Alcohol Abuse and Alcoholism, Bethesda, MD.
Koop, C. and Lundberg, G. D. (1992) Violence in America: a public health emergency. Journal of the American Medical Association 267, 30753076.[CrossRef][ISI][Medline]
Krivo, L. J. and Peterson, R. D. (1996) Extremely disadvantaged neighborhoods and urban crimes. Social Forces 75, 619650.[ISI]
LaScala, E. A., Johnson, F. W. and Gruenewald, P. J. (2001) Neighborhood characteristics of alcohol-related pedestrian injury collision: a geospatial analysis. Prevention Science 2, 123134.[CrossRef][Medline]
Lipton, R. and Gruenewald, P. J. (2002) The spatial dynamics of violence and alcohol outlets. Journal of Studies on Alcohol 63, 187195.[ISI][Medline]
Lipton, R., Gorman, D. M., Wieczorek, W. F. and Gruenewald, P. (2003) The application of spatial analysis to the public health understanding of alcohol and alcohol-related problems. In Geographic Information Systems and Health Applications, Khan, O. A., ed, pp. 5779. Idea Group Publishing, Hershey, PA.
Nelson, A. L., Bromley, R. D. F. and Thomas, C. J. (2001) Identifying micro and temporal patterns of violent crime and disorder in the British city center. Applied Geography 21, 249274.[CrossRef][ISI]
Ousey, G. C. (2000). Explaining regional and urban variation in crime: a review of research. In Criminal Justice 2000 Volume 1: The Nature of Crime: Continuity and Change, LaFree, G., ed, pp. 261308. National Institute of Justice, Washington, DC.
Peterson, R. D., Krivo, L. J. and Harris, M. A. (2000) Disadvantage and neighborhood violent crime: do local institutions matter? Journal of Research on Crime and Delinquency 37, 3163.
Ponicki, W. R. and Gruenewald, P. J. (1998) S3: The Spatial Statistical System Users' Guide, V4.0. Prevention Research Center, Pacific Institute for Research and Evaluation, Berkeley, CA.
Roncek, D. W. and Maier, P. A. (1991). Bars, blocks, and crimes revisited: linking the theory of routine activities to the empiricism of hot spots. Criminology 29, 725753.[ISI]
Sampson, R. J., Raudenbush, R. W. and Earls, F. (1997) Neighborhoods and violent crime: a multilevel study of collective efficacy. Science 277, 918924.
Scribner, R. A., Cohen, D. A., Kaplan, S. and Allen, S. H. (1999) Alcohol availability and homicide in New Orleans: conceptual considerations for small area analysis of the effect of alcohol outlet density. Journal of Studies on Alcohol 60, 310316.[ISI][Medline]
Sherman, L. W., Gartin, P. R. and Buerger, M. R. (1989) Hot spots of predatory crime: routine activities and the criminology of place. Criminology 27, 2755.[ISI]
Speer, P. W., Gorman, D. M., Labouvie, E. W. and Ontkush, M. J. (1998). Violent crime and alcohol availability: relationships in an urban community. Journal of Public Health Policy 19, 303318.[ISI][Medline]
Stockwell, T. and Gruenewald, P. (2001) Controls on the physical availability of alcohol. In International Handbook of Alcohol Dependence and Problems, Heather, N., Peters, T. J. and Stockwell, T., eds, pp. 699719. Wiley & Sons, New York.
Texas Department of Public Safety (2001) Crime in Texas 2000: The Texas Crime Report. Texas Department of Public Safety, Austin, TX.
US Census Bureau (2003) 2000 United States Decentennial Census (Summary File 1 and Summary File 3). US Department of Commerce, Washington, DC.