Mapping Crime in Savannah
Social Disadvantage, Land Use, and
Violent Crimes Reported to the Police
Social Science Computer Review
Volume 25 Number 2
Summer 2007 194–209
© 2007 Sage Publications
10.1177/0894439307298935
http://ssc.sagepub.com
hosted at
http://online.sagepub.com
Daniel Lockwood
Savannah State University, Georgia
This article is an area study on the hypothesis that violent crime is linked to a subculture of
violence, social disadvantage, or land uses such as rental, retail/office/commercial, or public/institutional property. Rates and types of reported violent crimes are related to neighborhood characteristics available through U.S. Census data and the county planning commission’s
geographical information office. The research design called for the geocoding of 26,467 violent reported crimes, which were aggregated by census block group. Variables in the analysis
of these spatial units include frequency and characteristics of reported violent crimes, rates of
violent crime, indicators of social disadvantage, housing types, housing values, and land use.
Analysis employed multiple regression, with simple assault, aggravated assault, homicide, and
robbery as the dependent variables. Variables helpful in the prediction of violent crime rates
were also placed as map layers on data maps. Results indicate that assault in neighborhoods is
associated with social disadvantage and the land uses of retail/office/commercial and
public/institutional. Robbery is associated with recreational, retail/commercial/office, renter,
and public/institutional land uses. Both percentage of renters and percentage of African
Americans fail to predict violence when controlling for social disadvantage. Implications concern issues of theoretical criminology related to routine activities theory, the southern regional
subculture of violence, and social disorganization theory. Land use planning and crime are also
discussed.
Keywords: homicide; assault; robbery; social disadvantage; crime mapping; Savannah;
violence
Space is the stage on which man’s behavior unfolds. Space provides the occasions for motives—the opportunities, temptations and pressures. Space conditions human relationships, brings people together and separates
them. Space undergirds social climate, sets limits, inspires, beckons, frustrates, isolates, crowds, intrudes, liberates. It pretends vistas and forecloses them, yields privacy and violates it.
Toch (1980, p. xi)
Authors’ Note: The Savannah Police Department generously provided complete electronic data files, with
addresses, for violent reported crimes. I am especially grateful to Richard Strait, the analyst with the
Department’s Intelligence Office, for his assistance with geocoding the events. I am also grateful to David
Donald Charles of the Savannah Morning News for his help in extracting and interpreting the U.S. Census data.
I particularly wish to thank Ron Wilson, program manager of the Mapping and Analysis for Public Safety
Program and Data Resources of the National Institute of Justice, who ably helped me with the spatial regression using GeoDa.
194
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
Lockwood / Mapping Crime in Savannah
195
T
he characteristics of neighborhoods can help explain why violent crime varies across
neighborhoods in a city. Neighborhood rates of violent crime can also aid in understanding why certain neighborhoods have the features that they do. Neighborhoods affect
crime, and crime affects neighborhoods. Analyses of community characteristics and neighborhood crime rates may give us information useful for developing violence prevention
strategies matched to specific types of social environments. Spatial analyses of crime may
also make contributions to criminological theory. Indeed, some of the earliest theoretical
thinking on the causes of crime, the Chicago School of Sociology research of Shaw and
McKay (1942), for example, developed ecological explanations for delinquency that linked
crime rates in urban areas to the characteristics of those areas.
Theories of Neighborhood and Crime
Christopher S. Dunn, in a literature review entitled “Crime Area Research,” summarized
various research perspectives on neighborhoods and crime (Dunn, 1980). According to
Dunn, up to the early 1980s, the main environmental attributes criminologists related to
crime rates were land use and social structure. Some of the earliest theoretical criminology
in this country connected land use to crime. White, for example, found offenses associated
with areas near businesses, industry, and factories (White, 1932). Shaw and McKay found
that those areas adjacent to the central business district and industrial centers had more
crime than other areas (Shaw & McKay, 1942).
Research connecting crimes to social structure has looked at such variables as socioeconomic status, family stability, and ethnicity. Often, indicators from the U.S. Census,
separately or combined as an index, are used in the analysis. Criminologists of the
Chicago School, such as Shaw and McKay, viewed poverty, ethnic diversity, and residential instability as leading to high crime rates because these factors cause social disorganization that prevents residents from coming together to solve neighborhood crime
problems (Shaw & McKay, 1942). Researchers have tested this theory with survey data,
finding that scant friendship networks and low levels of participation in neighborhood
organizations characterize communities in Great Britain with high crime rates (Sampson
& Groves, 1989). The hypothesis that social cohesion and a willingness to act to improve
the community (a concept given the term social efficacy) are linked to reduced violence
rates was further tested with survey data from Chicago. This study found that “collective
efficacy” can overcome the influence of community disadvantage and instability on
crime rates (Sampson, Raudenbush, & Earls, 1997). In the field of environmental psychology, research techniques are being developed to directly measure the social attributes of
neighborhoods through survey research and field observations. Such work tries to characterize “neighboring,” examining “social ties,” “neighborhood attachment,” “neighborhood
identity,” “involvement in the neighborhood,” and “neighbor annoyance” (Skjoeveland,
2001).
Other researchers, extending social disorganization theory, have tested the hypothesis
that the presence of institutions in neighborhoods serves to lower crime rates. In this theory, organizations are understood as agents promoting social efficacy. To test this theory,
for example, a study examined the link between the presence of social institutions such as
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
196
Social Science Computer Review
major retail chain grocery stores, banks, public libraries, and recreation centers and violent
crime. However, evidence failed to find any positive effect of organizations reducing crime
rates in the neighborhoods in which they are found (Peterson, Krivo, & Harris, 2000).
Other researchers have looked at theories using data about neighborhood sociodemographic conditions from the U.S. Census and crime. For example, McNulty and Holloway
(2000) tested the hypothesis that the strength of the association between race and crime
rates diminishes as the distance from public housing projects increases. Analyzing census
and crime data from Atlanta, they found this theory supported for violent crimes except for
robbery (McNulty & Holloway, 2000).
Theories also concern the effect of high crime rates on urban transition. National data on
the United States’ large cities for the period 1970–1980 show that crime rates can help
explain migration out of cities (Sampson & Wooldredge, 1986). Also, theories have been
tested about the link between violent crime and population decline in neighborhoods, and
the role of sociodemographic disadvantage and racial segregation in the process of neighborhood population change (Bursik, 1986; Morenoff & Sampson, 1997).
The geography of crime also has been related to the question of what is more important
to look to in explaining violent crime—the subculture-of-violence thesis or relative deprivation theory. Is it culture or social structures of inequality that best explain differences in
the rates of violent crime in areas? Researchers, using ethnicity as the indicator of culture,
have come to different conclusions on this question. Rosenfeld analyzed Uniform Crime
Reports (UCR) and U.S. Census data for Standard Metropolitan Areas. He concluded, “The
investigation finds strong support for the cultural model of crime” (Rosenfeld, 1986,
p. 127). Sampson, on the other hand, who analyzed National Crime Victimization Survey
data, concluded, “Neighborhood family structure has significant and substantively important influences on personal criminal victimization” (Sampson, 1986, p. 45). This is an issue
we shall take up in our study of violence.
Hypotheses and Purposes of the Study
Our research has exploratory and descriptive aims as well as the intent to test hypotheses. We designed the project to create a database of violent reported crimes for Savannah,
Georgia, for the period 1993–1997. Then we geocoded these events to points in a geographical information system (GIS). This permitted us to summarize the crimes to small areas
within the city (census block groups) about which we have other information. Based on the
availability of this secondary data and the literature on the social ecological approach, questions such as these drove the work:
1.
2.
3.
4.
In what census block groups are the number and rates of violent crimes the greatest?
Do census block groups with different numbers of retail/office/commercial, recreational, and public/institutional spaces differ from residential census block groups in
their number and rates of violent crime? Do types of crime differ by land uses?
Is ethnicity a factor in differential rates of violent crime within communities?
Do census block groups with a high percentage of rental housing differ on violent crime
from predominantly owner-occupied neighborhoods?
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
Lockwood / Mapping Crime in Savannah
197
The literature reviewed suggested several major hypotheses that we could test with the
secondary data available to us for the City of Savannah. They are as follows:
Hypothesis 1: The violent crime rate within a neighborhood is related to social disadvantage. The greater the
percentage of socially disadvantaged households in a neighborhood, the higher the crime rate is likely to be.
Hypothesis 2: The violent crime rate of a neighborhood is related to the type of housing unit. The greater the
percentage of rental housing units, the higher the crime rate.
Hypothesis 3: The violent crime rate of a neighborhood is related to ethnicity. The higher the percentage of
African Americans, the higher the violent crime rate.
Hypothesis 4: The violent crime rate of a neighborhood is related to land use. The greater the amount of
retail/office/commercial or public/institutional land use, the greater the crime rate.
Data and Method
Description of Research Setting
Savannah, Georgia, offers an excellent setting for an ecological study of violence in
neighborhoods. The areas of the city are diverse in regard to demographic characteristics
and land uses. With a population in 1990 of about 135,000, the city contained at that time
145 census block groups, a sufficient number to allow for statistical comparisons. Above
all, Savannah has many violent crimes. There has been one self-reported victimization survey in the history of the city, in 1997. According to these data, the violent victimization rate
in Savannah was 81 per 1,000, in comparison to 51 per 1,000 for the National Crime
Victimization Survey urban crime rate for 1997. Thus, Savannah had a higher rate of selfreported violent victimization than such cities as Chicago, Kansas City, or Los Angeles
(Smith, Stedman, Minton, & Townsend, 1999).
According to Uniform Crime Reports, the average annual homicide rate in the city for
the period 1980–2005 is 21 per 100,000, among the highest of U.S. cities. The average
annual robbery rate is 489. Moreover, although many other places experienced sharp
declines in serious violent crime, the rates of reported murders and robberies from 1980 to
2005 in Savannah have remained about the same. In 2005, the murder and robbery rates in
Savannah were much higher than those of the other cities of the region such as Jacksonville,
Charleston, Macon, and Columbia. Among the cities of Georgia, only Atlanta has a higher
rate of serious violent crime.
The violence problem is particularly acute in certain residential neighborhoods of
Savannah, a few of which have extremely high homicide and aggravated assault rates.
Not far from these high-crime neighborhoods is the Savannah Historic District, which
attracts over a million and a half tourists a year to the city. The Historic District contains a college and also employs a large number of retail, commercial, and office workers, for it is a center for business, government, and education. Their spending holds up
the economic well-being of many of the surrounding poor residential neighborhoods.
The CAP Index and APBnews.com, which ranks colleges on the murder, rape, and robbery rates in the neighborhoods around them, rated the college in the Savannah Historic
District as 9 on a 10-point scale of dangerousness. Ten on the scale is “extremely high”
(Few, 2001).
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
198
Social Science Computer Review
The prevalence of handguns also distinguishes Savannah from other places. Survey data
comparing 12 cities of the United States tell us that 34% of the households in Savannah in
1997 kept a self-defense weapon in the home, a higher percentage than most of the other
12 cities. Of these 12 cities, Savannah had the highest percentage of violent crimes in which
a gun was used, 46% (Smith et al., 1999).
Methods
Operational Definitions
Table 1 shows the variables analyzed in this study. These analyses looked at the connection between various independent variables available from the U.S. Census and the
Metropolitan Planning Commission and violent crime rates for block groups in the City of
Savannah for the years 1993–1997. We selected census block groups as our unit of analysis, rather than the city-designated neighborhoods, so that we could include variables available through the census in our analysis. The choice of the 145 block groups over the 42
census tracts in the city gave us a sufficient number of area units for statistical analysis.
Although the City of Savannah officially is divided into “neighborhoods,” in fact, in many
Table 1
Variables in the Analysis
Variable
Description
HOM (homicide)
ROB (robbery)
AGG (aggravated assault)
SIM (simple assault)
A-A (percentage African American)
DIS (disadvantage index)
Five-year average annual rate of homicide, 1993–1997
Five-year average annual rate of robbery, 1993–1997
Five-year average annual rate of aggravated assault, 1993–1997
Five-year average annual rate of simple assault, 1993–1997
Percentage of the total census block group that is African American
Index made by adding together the z-scores of the following factors
from the 1990 U.S. Census:
1. Percentage of block group residents below the poverty level
2. Percentage of female-headed households with children in the
block group
3. Percentage of block group population in the workforce
unemployed according to the 1990 Census
4. Percentage of block group households receiving public
assistance according to the 1990 Census
RENT (percentage renter)
RTCOM (retail/commercial)
PUBINS (public/institutional
REC (recreational)
Percentage of population of block group occupying rental housing
according to the 1990 Census
Number of polygons in block group designated retail/commercial
by zoning map
Number of polygons in block group designated public/institutional
by zoning map
Number of polygons in block group designated recreational by
zoning map
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
Lockwood / Mapping Crime in Savannah
199
areas of the city, residents have a very weak sense of belonging to a certain neighborhood,
and many residents do not even know the official neighborhood name. Geographers concerned with crime argue that census block groups are better to use as units of analysis than
census tracts, because they give better spatial definition (Harries & Kovandzic, 1999). Also,
other recent social ecological approaches to the study of crime have successfully used the census block group as the operational definition of neighborhood (McNulty & Holloway, 2000).
Because our crime data came from the Savannah Police Department (SPD), and we
wanted to construct rates, we chose only those census block groups that are entirely within
the boundary of the city. This gives us 145 block groups in our data set, made up of 128,498
people. This is some 10,000 fewer people than the 1990 population of the city, because
people living in census block groups that extend beyond the city boundary are excluded.
Crimes
This article examines data from the digital files of crimes reported to the SPD. Rates for
the violent crimes of homicide, robbery, aggravated assault, and simple assault give us the
dependent variables. We deliberately left out forcible rape, because our data files contain
the exact address of the event and we preferred to protect the anonymity of rape victims.
Although most other studies of this type use the FBI’s violent index crimes of homicide,
forcible rape, robbery, and aggravated assault as the dependent variables, we also included
simple assault in our operational definition of violence. Reported simple assaults were
available and, being numerous, provided a measure of violent crime with fine gradations.
The simple assault rate correlated highly with the rates for the other, more serious index
crimes and could serve as a general indicator of the relative level of violence within an area
as small as a census block group. Five-year (1993–1997) annual average crimes per 1,000
were constructed to reduce the effect of annual fluctuations. Because Savannah is a city
with high rates of violent crime, and we included simple assaults, we had sufficient incidents to construct reliable rates even for small areas.
A self-reported victimization survey conducted in Savannah in 1997 by the U.S. Bureau
of Justice Statistics, fortuitously, gives us a basis to discuss the validity of these official violent crime data. According to this survey, 40% of the violent crimes of murder, rape, aggravated assault, robbery, and simple assault against individuals 12 years or older were
reported to Savannah police in 1997. The number of robberies reported by the victimization survey was very close to the number of robberies reported by the official UCR data.
Aggravated assaults were also similar in self-report and official data. There is, however,
according to this survey, a large underreporting of simple assault. The numbers of simple
assault reported, for example, may be only half of the actual number.
Land Uses
Land uses were obtained from an ArcView GIS data file created in 1992 by the Chatham
County Metropolitan Planning Commission (MPC). Using the Union procedure in ArcView,
the large polygons in the land use data file were broken down into census block group
boundaries and then added up to give a value for each block group. The operational definition of land use, thus, is the total number of polygons for each land use within each census
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
200
Social Science Computer Review
block group. The validity of this measure was tested and found to be adequate by examining visually the overlay of the original MPC land use map layer on top of the graduated
color data maps of the new land use variables that were created.
Method of Analysis
The SPD provided us with a database of reported violent crimes (N ⫽29,221) that included
address as one of the fields. Using ArcView GIS, we geocoded the crimes, locating each one
as a point in a computerized GIS. We successfully geocoded 95% of the events. Then, the
crimes were aggregated to the census block groups entirely within the city boundaries.
Sociodemographic data at the U.S. Census block group level were selected out of two
census data systems. Population, race, and housing type (whether owner or renter occupied) were selected from the census short-form data that are asked of every household.
Percentage below the poverty level, percentage unemployed, and percentage of households on public assistance came from the long form, which is asked of one out of six
households, chosen at random.
Following similar research, we created a disadvantage index from U.S. Census variables
(McNulty & Holloway, 2000). This is based on the z score. A z score is a statistical measure that tells how any individual score (such as the percentage unemployed in a census
block group) differs from the average of all scores (the mean unemployment rate for all
block groups). It is expressed in standard deviation units so that measures with very different ranges can be added together. We added together the z scores of the following: percentage of persons below the poverty level, percentage of households receiving public
assistance, percentage of female-headed households with children, and percentage of unemployed. This gave us a database made up of sociodemographic, land use, and criminal event
variables, with the census block group as the unit of analysis. These data were then analyzed using ArcView GIS and SPSS.
The causal link between income inequality and violent crime has been demonstrated
using the Gini index and homicide and robbery rates within and among countries. Crime
rates and inequality are positively correlated, even after controlling for other crime determinants (Fajnzylber, Lederman, & Loayza, 2002). The Gini coefficient has also been used
in a large number of studies to show the relationship between homicide and income inequality in Canada and the United States (Wilson, Wilson, & Vasdev, 2001). Our disadvantage index not only incorporates this notion of income inequality but also includes
measures of structural disadvantage.
Statistical Analyses
Our model sees violent crime as a function of land use, social disadvantage, and control
variables. We used SPSS multiple linear regression to estimate models of simple assault,
aggravated assault, robbery, and homicide. Using the index of social disadvantage avoided
multicollinearity among the social disadvantage variables, for this was made up of four
highly correlated indicators (percentage below the poverty level, percentage receiving public assistance, percentage unemployed, and percentage of female-headed households with
children). Interpretation of collinearity diagnostics in SPSS indicated no difficulties with
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
Lockwood / Mapping Crime in Savannah
201
the data set (SPSS, 1998). We first conducted ordinary least squares (OLS) regression in
SPSS. Next, we analyzed the data in GeoDa for spatial effects (Anselin, 2004).
We also created multivariate data maps in ArcView GIS using the variables revealed by
the regression analysis to be important. This gave us an independent analytical procedure
against which we could visually compare the findings of our regression analysis. The visual
presentations of the quantitative data from the GIS analysis also allow us to present the
findings of our multivariate analysis in visual displays that are understandable to the generally educated person unfamiliar with regression statistics.
Results
Descriptive and Bivariate Results
Figure 1 is a series of maps showing the relationship among homicide, simple assault,
social disadvantage, and percentage of African Americans. As we see in Figure 1, homicides
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
202
Social Science Computer Review
cluster with simple assaults. In fact, for analytical purposes, simple assault rates predict
homicide rates and serve quite well as a general indicator of violence. Both assault and homicide are highest in block groups of high social disadvantage. In block groups with high percentages of disadvantaged African Americans, the problem of assault and homicide is the
greatest. However, as we see in Figure 1, the northwestern part of the city has block groups
of moderate social disadvantage that are almost entirely African American. In these areas,
assault and homicide rates are also moderate. There appears to be a violent subculture
engaging in high levels of dispute-related violence in the city that is associated with
disadvantage—not just race.
Figure 2 is a multimap showing the relationship between robbery and land use. We see in
this figure that robberies occur most often in block groups that contain spaces given to public
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
Lockwood / Mapping Crime in Savannah
203
Table 2
Correlations, Means, and Standard Deviations of Variables: Savannah, Georgia,
Census Block Groups Entirely Within City Boundaries (N ⫽145)
HOM
HOM
ROB
AGG
SIM
A-A
DIS
RENT
RTCOM
PUBINS
REC
ROB
AGG
SIM
A-A
DIS
RENT
RTCOM
PUBINS
REC
.709
.776
.142
.287
.346
.660
.617
.459
1.00
.953
.469
.615
.436
.601
.536
.079
1
0.41
0.58
0.46
0.62
0.59
.147
1
0.771
0.314
0.316
0.119
⫺.205
1
0.54
0.34
0.23
⫺0.1
1
0.303
0.283
⫺0.13
1
0.695
0.282
1
0.245
47.32
3.92
3.92
.94
27
3.45
2.85
1.51
1
.361
.615
.536
.356
.412
.232
.348
.266
⫺.086
1
Mean
.206
7.044
3.775
36.50
53.30
.03
Standard
Deviation
.3817
8.531
4.838
36.57
38.91
3.18
1
Note: The indicators making up the disadvantage scale had the following means and standard deviations: percentage below the poverty level, 24.20 (20.85); percentage of households receiving public assistance, 14.12
(14.25); unemployment, 4.12 (3.07); female-headed household with children, 8.68 (7.29).
or semipublic land uses such as retail, office, commercial, institutional, and recreational. The
robbery rate is also high in most disadvantaged block groups that contain mixed-use spaces.
Looking at these two figures, we see that in Savannah, commercial areas with high robbery
rates are generally close to residential block groups with high rates of homicide and assault.
Table 2 contains the correlation matrix, means, and standard deviation of the variables
in the analysis. The crime means of the 145 census block groups in Savannah present average rates per 1,000 for the years 1993–1997. Even though some 10,000 of the city’s residents are left out of the study because they reside in census block groups that are only partly
in the city boundaries, and even though the rates come from block groups of different population sizes, these means are very close to the actual average UCR rates for the city during that period. The mean per 1,000 for homicide is .206; robbery, 7.04; and aggravated
assault, 3.785. The 20,528 simple assaults make up the greatest number of the 26,467 violent crimes in this analysis, with a mean rate per 1,000 of 36.5. The mean percentage of residents below the poverty level in the block groups is 24%, indicating that Savannah has
high levels of social disadvantage.
As for the land use variables, the mean of 47% of block group residents who occupy
rental property gives us a picture of a city where about half of the residents live in rental
units. This rate is confirmed by the actual number of persons living in rental property
according to the 1990 U.S. census, 60,000 of the study population of 128,000. The mean
of land use polygons within the block groups designated retail/office/commercial is 3.92;
public/institutional, 3.92; and recreational, .94.
The correlations between crime rates and land uses give us some ability to predict where
violent crimes are most likely to occur. Homicide, aggravated assault, and simple assault
have strong positive correlations with social disadvantage. Most of these events involve
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
204
Social Science Computer Review
disputes, and most occur in poor parts of the city where members of a subculture of violence live who readily use violence to resolve conflicts. Moreover, in accordance with the
theoretical expectations of the Chicago School, homicide, aggravated assault, and simple
assault also have strong positive correlations with retail/office/commercial and public/institutional land uses. Residents of disadvantaged block groups with high rates of violence
share their space with stores, businesses, offices, other small businesses, and institutions.
The number of these mixed land uses within block groups in Savannah is strongly associated with the number of homicides, aggravated assaults, and simple assaults.
The correlations of the independent variables with robbery in Table 2 present a different
portrait. There is only a weak positive association between race and robbery. Although
there is still an association with disadvantage, this association is weaker than with the other
violent crimes. Disadvantaged block groups have high rates of robbery; so do nearby block
groups with retail/office/commercial and public/institutional land uses. There is also a
strong association between robbery and the number of recreational land uses in block
groups. Parks appear to provide space for robberies to take place. Although recreational
land use has a strong positive association with robbery, recreational land use has no association with homicide, aggravated assault, or simple assault. This may be because there are
few recreational spaces in the disadvantaged areas of the city where most of these disputerelated events take place. In conclusion, it seems that most dispute-related violence occurs
within the African American disadvantaged neighborhoods of the city (which also have a
robbery problem), but a large percentage of robberies occur outside of these areas.
The percentage of residents living in rental units is also positively associated with violent crime. The positive association between renters and violence is particularly strong for
aggravated assault and simple assault. It is moderately strong for homicide and robbery.
These correlation coefficients are significant at the .01 level. However, our multivariate
analysis, which we turn to next, shows an interesting interpretation of this correlation.
Multivariate Results
Table 3 shows the result of the multiple regression analysis on rates of simple assault,
examining the effects of ethnicity, social disadvantage, and land use. Overall, this model
Table 3
Regression of Simple Assault Rates on Social Disadvantage,
Land Use, and Control Variablesa
Model
(Constant)
Percent African American
Disadvantage Index
Percent renter
Retail/commercial
Public/institutional
B
Standard Error
Beta
t
Sig.
8.271
⫺6.304E-02
4.838
.122
2.972
3.583
7.258
.083
1.131
.090
.845
.990
⫺.067
.421
.090
.280
.279
1.140
⫺.755
4.277
1.352
3.516
3.619
0.256
0.451
0
0.179
0.001
0
Note: Total R2 ⫽.580. F(5,139) ⫽40.73. p ⬍.000.
a. Dependent variable: simple assault rate per 1,000.
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
Lockwood / Mapping Crime in Savannah
205
Table 4
Regression of Aggravated Assault Rates on Social Disadvantage,
Land Use, and Control Variablesa
Model
(Constant)
Percent African American
Percent renter
Disadvantage Index
Public/institutional
Retail/commercial
B
Standard Error
Beta
t
Sig.
.391
⫺1.111E–03
7.797E–03
.688
.385
.394
.973
.011
.012
.152
.133
.113
⫺.009
.044
.452
.227
.281
.402
⫺.099
.646
4.534
2.902
3.481
.688
.921
.520
.000
.004
.001
Note: Adjusted R2.568. F(5, 139) ⫽38.90. P ⬍.000.
a. Dependent variable: aggravated assault rate per 1,000.
is helpful in predicting rates of violence in neighborhoods. Knowing social disadvantage,
retail/office/commercial, public/institutional, percent renter, and percent African American
explains 58% of the variability of simple assault. Through this model, we can judge
whether land use has independent effects on violence and if these effects help explain the
influence of social disadvantage. The t statistic in the second-to-last column in Table 3
gives us an idea of the relative importance of each variable in the model. Helpful predictors have t values below ⫺2 or above ⫹2 (SPSS, 1998). Accordingly, social disadvantage,
retail/office/commercial, and public/institutional remain as predictors of simple assault.
As expected by theory, social disadvantage is the strongest explanatory variable. Percent
African American and percent renter fail to be useful predictors and make little useful contribution to the model.
Table 4 shows the multiple regression analysis for aggravated assault. Race and renter
again fail to be useful predictors of this violent crime. However, social disadvantage,
retail/office/commercial, and public/institutional create a model that explains 57% of the
variance in aggravated assault. Social disadvantage is again the strongest predictor of
aggravated assault.
We first conducted OLS regression in SPSS. Next, we analyzed the data in GeoDa for
spatial effects (Anselin, 2004). Lagrange multiplier statistics gave us our measure for
spatial error. This procedure indicated no spatial error for the models using the dependent
variables of homicide, aggravated assault, and simple assault. What this means is that the
explanatory variables were not suffering from any unknown spatial effects. However, these
tests did indicate that there is significant spatial lag in one dependent variable, robbery.
Then, using the software program for spatial statistical analysis, GeoDa (Anselin, 2004),
we estimated our model again for robbery, including proper spatial structure. The results let
us compare the OLS regression to the spatial regression to examine for spatial dependence
when considering the predictors of robbery. These findings are presented in Table 5, which
we discuss below.
If we use the log likelihood as our measure of fit, as recommended by Anselin (2005)
for spatial regression analyses, we see this measure changing little, from ⫺449 in the OLS
to ⫺446 for the spatial regression. As for the coefficients of the dependent variables, as
Table 5 shows, the only important change by applying spatial regression in GeoDa is in the
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
206
Social Science Computer Review
Table 5
Ordinary Least Squares and Spatial Lag Regressions of Robbery
for Savannah, 1993–1997 (N ⫽145)
Ordinary Least Squares
Coefficient
A-A
DIS
RENT
RTCOM
PUBINS
REC
Constant
Rho
Log likelihood
⫺.016
.325
.056**
.771**
.705**
1.865**
⫺2.626
0.218
⫺449.131
Standard Error
.020
.267
.022
.222
.231
.342
⫺3.109
Spatial Lag
Coefficient
Standard Error
⫺.014
.197
.048*
.688**
.604**
1.687**
.019
.258
.0207
.213
.224
.335
⫺446.832
*p ⬍.05. **p ⬍. 0 1
coefficient for social disadvantage. This went from .324 in the OLS to .196 in the spatial
regression. It was diminished because social disadvantage was not as concentrated as the
OLS indicated. The spatial regression adjusted the influence of this variable based on its
scale of influence. That is, it is not as concentrated but is more widespread.
In summary, for the violent crimes of simple assault, aggravated assault, and robbery,
land use contributes to creating helpful models for explaining crime in Savannah during
this period. Social disadvantage is a strong predictor in the models of the dispute-related
crimes of simple assault and aggravated assault, but fails to contribute to the model for robbery. Only in the case of robbery does the percentage of renters contribute to the model. For
all three of these violent crimes, the association between percentage of African Americans
and violent crime vanishes when we control for social disadvantage and land use.
Conclusions
Literature has discussed the role that nearby institutions, including businesses, might play
in supporting residents to exert collective efforts to control crime in neighborhoods. Crime
prevention specialists have called retail, commercial, and public institutions within neighborhoods “community assets.” In this view, some hypothesize that the greater the number of
such organizations in a neighborhood, the lower the violent crime rate. Our results show the
opposite: The greater the presence of retail/office/commercial or public/institutional organizations in an area, the greater the violent crime rate.
The crime maps of Savannah, Georgia, with overlays for land use and social disadvantage, match the theoretical framework of some of the first Chicago School criminologists
to analyze the spatial characteristics of crime. In Savannah, as in the Chicago of Shaw and
McKay’s era, high levels of social disadvantage in neighborhoods are often accompanied
by the presence of land use given to public and commercial purposes. In such neighborhoods of mixed land use, violent crime rates are high.
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
Lockwood / Mapping Crime in Savannah
207
A glance at the association between violent crime and race in Savannah may lead one to
hypothesize that an ethnic subculture of violence best explains crime in the city. However,
as the results of our analysis indicate, as the level of social disadvantage goes down, the
effects of race on violent crime tend to disappear. Although the southern subculture of violence thesis may help a great deal to explain the shape of violence in disadvantaged neighborhoods, the influence of those cultural values, norms, and patterns of response that facilitate
violence appears to diminish as social class increases.
Looking at the high correlation between renters and violent crime, we might also
hypothesize that the percentage of rental properties in a neighborhood influences crime
rates. Indeed, even in the literature on crime, percent renter is used as an indicator of social
disorganization. Programs encouraging home ownership are often justified as social control measures in a city, with people arguing openly that renters are bad, home owners are
good, and a shift toward more home ownership will alter behavior. In a city such as
Savannah, where about half the residents live in rental units, the basis for such aspersions
against renters should be questioned. The results of our analyses show that, like race, the
effects of renters on violent crime are dependent on social disadvantage. Renters are associated with violent crime in Savannah because the socially disadvantaged rent. As social
disadvantage goes down, so does violent crime, regardless of whether housing is rental or
owner occupied.
The data we present, which are primarily reports to police by victims of violent crime,
indicate the extreme degree to which residents of disadvantaged neighborhoods are disproportionately impacted by violent crime. For example, from 1993 to 1997, 14,792 violent
crimes (56% of the total) occurred in neighborhoods where a quarter or more of the residents
lived below the poverty level ($12,000 for a family of four). In our calculations of “disadvantage,” we might consider adding to the burdens of poverty, unemployment, and single
parenthood the costs of such violent crime: loss of income because of the incarceration of
family members and intimate partners, diminished job opportunities because of criminal records, legal fees, restricting lifestyles because of fear, forced relocation to avoid victimization, and the collapse of supportive relations when disputes are resolved by violence.
We need to also consider the impact on business enterprises and institutions when such
organizations occupy spaces of high social disadvantage and high crime. Our analysis
shows that retail/office/commercial and public/institutional land uses are strongly associated with both crime and social disadvantage in Savannah. This helps us to understand
crime patterns in the city, particularly robbery. The ecology of such urban geography in
Savannah places socially disadvantaged young males in close proximity to crime targets
attracted into their space by land uses such as offices, shops, restaurants, museums, and
schools. Propinquity of offender to target itself probably provides the best explanation for
such victimization. In other research, we have examined this “adjacency hypothesis,” analyzing the home addresses of persons arrested for robbery in 2004 in Savannah. The conclusions of this work affirm the hypothesis, for we found that most persons arrested for
robbery in Savannah’s Historic District lived in adjacent disadvantaged neighborhoods (see
Lockwood, 2005).
In conclusion, the situation we describe in this article poses particular challenges for
both proponents for the socially disadvantaged and proponents for the economic development of the city. Although most violent crimes are disputes among intimates in their
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
208
Social Science Computer Review
own neighborhoods, the clients and employees of nearby retail/office/commercial and
public/institutional organizations also become victims of robbery and, occasionally, felony
murder. If such events occur too often, such organizations relocate, and the residents of the
nearby socially disadvantaged neighborhoods lose places of employment. On the other
hand, the low rent that makes it possible for disadvantaged families to afford a place to live
is itself a function of high crime rates. Lowering crime rates in disadvantaged areas may
well raise rents to the point where the poor are forced to relocate. Research on the spatial
characteristics of violent crime such as we have seen in this article contributes to the understanding of these social forces. The findings of such research should be considered in discussions of zoning and urban land use planning as well as crime prevention.
References
Anselin, L. (2004). GeoDa 9.5-I release notes. Urbana: Spatial Analysis Laboratory (SAL), Department of
Agricultural and Consumer Economics, University of Illinois, Urbana-Champaign. Retrieved, from https://
www.geoda.uiuc.edu/documentation/manuals.
Anselin, L. (2005). Exploring spatial data with GeoDa™: A workbook. Urbana: Spatial Analysis Laboratory
(SAL), Department of Geography, University of Illinois, Urbana-Champaign. Retrieved, from https://
www.geoda.uiuc.edu/documentation/manuals.
Bursik, R. J. (1986). Delinquency rates as sources of ecological change. In J. M. Byrne & R. J. Sampson (Eds.),
The social ecology of crime (pp. 25–46). New York: Springer-Verlag.
Daly, M., Wilson, M., & Vasdev, S. (2001, April). Income inequality and homicide rates in Canada and the
United States. Canadian Journal of Criminology, 219–236.
Fajnzylber, P., Lederman, D., & Loayza, N. (2002) Inequality and violent crime. The Journal of Law &
Economics, 45, 1–39.
Few, J. (2001, October 13). SCAD pumps money into local economy. Savannah Morning News, 7A.
Harries, K., & Kovandzic, E. (1999). Persistence, intensity, and areal extent of violence against women.
Violence Against Women, 5, 813–828.
Lockwood, D. (2005, September). Mapping violent crime in Savannah’s Historic District, 1993–2003. Paper presented at the Eighth Annual International Crime Mapping Research Conference, Savannah, GA. Retrieved,
from http://www.ojp.usdoj.gov/nij/maps/savannah2005/papers/Lockwood.pdf.
McNulty, T. L., & Holloway, S. (2000). Race, crime, and public housing in Atlanta: Testing a conditional effect
hypothesis. Social Forces, 79, 707–729.
Morenoff, J. D., & Sampson, R. (1997). Violent crime and the spatial dynamics of neighborhood transition.
Social Forces, 76(1), 31–64.
Peterson, R. D., Krivo, L. J., & Harris, M. A. (2000). Disadvantage and neighborhood violent crime: Do local
institutions matter. Journal of Research in Crime and Delinquency, 37(1), 31–63.
Rosenfeld, R. (1986). Urban crime rates: Effects of inequality, welfare dependency, region, and race. In J. M.
Byrne & R. J. Sampson (Eds.), The Social Ecology of Crime (pp. 25–46). New York: Springer-Verlag.
Sampson, R., & Groves, W. (1989). Community structures and crime: Testing social disorganization theory.
American Journal of Sociology, 94, 774–802.
Sampson, R. J. (1986). Neighborhood family structure and the risk of personal victimization. In J. M. Byrne &
R. J. Sampson (Eds.), The social ecology of crime (pp. 25–46). New York: Springer-Verlag.
Sampson, R. J., Raudenbush, S. W., & Earls, F. (1997). Neighborhoods and violent crime: A multilevel study
of collective efficacy. Science, 277, 918–924.
Sampson, R. J., & Wooldredge, J. D. (1986). Evidence that high crime rates encourage migration away from
central cities. Sociology and Social Research, 90, 310–14.
Shaw, C., & McKay, H. (1942). Juvenile delinquency and urban areas. Chicago: University of Chicago Press.
Skjoeveland, O. (2001). Effects of street parks on social interactions among neighbors: A place perspective.
Journal of Architectural and Planning Research, 18, 131–146.
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
Lockwood / Mapping Crime in Savannah
209
Smith, S. K., Stedman, G. W., Minton, T., & Townsend, M. (1999). Criminal victimization and perceptions
of community safety in 12 cities, 1998. Washington, DC: U.S. Department of Justice, Bureau of Justice
Statistics.
SPSS. (1998). SPSS Base 8.0 applications guide. Chicago: Author.
Toch, H. (1980). Foreword. In D. George-Abeyie & K. Harries (Eds.), Crime: A spatial perspective (pp. xi-xii).
New York: Columbia University Press.
Daniel Lockwood is an associate professor in the Department of Social and Behavioral Sciences, Savannah
State University, where he specializes in geographic information systems, quantitative and qualitative
methodologies, survey data analysis, ecology of disadvantaged neighborhoods, program evaluation, planned
change, violence, crime analysis, crime prevention, corrections, and criminal behavior. He may be contacted
at [email protected].
Downloaded from ssc.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016