![Sample our Social Sciences journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/110822/TOC-Subject-Sample-2021-Social-Sciences.jpg"})
ABSTRACT
The current study adds the context of the immediate microgeographic environment (measured as the street segment) to the study of individual victimization. Using residential survey and physical observation data collected on 449 street segments nested within 53 communities in Baltimore, MD, we employ multilevel logistic regression models to examine how individual risky lifestyles, the microgeographic context of the street, and community-level measures influence self-reported property and violent crime victimization. Results confirm prior studies that show that risky lifestyles play a key role in understanding both property and violent crime victimization, and community indicators of disadvantage play a role in explaining violent crime victimization. At the same time, our models show that the street segment (micro-geographic) level adds significant explanation to our understanding of victimization, suggesting that three-level models should be used in explaining individual victimization. The impact of the street segment is particularly salient for property crime.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1. A detailed description of the sampling approach and methodology for the project is available online: https://cebcp.org/wp-content/uploads/2020/07/NIDA-Methodology.pdf.
2. Crime call data for 2012 were obtained from the Baltimore City Police Department and geocoded to the street centerline. They were then spatially joined to obtain counts of crime for every street segment in Baltimore City. The geocoding match rate for the crime call data was 98.8%.
3. Violent crime calls for service included rape (force), robbery (armed or unarmed), bank hold up, aggravated assault (cutting, hands, or gun), common assault, carjacking, abduction, and sniper. Drug crime calls included narcotics, narcotics outside, and narcotics on-view.
4. We used data obtained from the Baltimore City Mayor’s Office for year 2010 to identify occupied households on city streets.
5. Due to our sampling criteria for residential streets (20+ dwelling units), streets located in the inner harbor, which are largely businesses and commercial, were not eligible for sampling.
6. Streets were selected through a random sampling procedure developed in Model Builder (in ArcGIS) that prevented any two sample streets from being within a one block buffer area. Once residential street segments were selected based on these data, occupancy of dwelling units was verified through a physical census conducted by field researchers using a series of vacancy indicators. Streets were replaced in this process as needed to reach our sample goals.
7. The contact rate was calculated by dividing those households with contact/eligible households, and the cooperation rate was calculated by dividing households with a completed survey/households with contact. The average number of visits to households was 4, but we visited as many as 25 times in order achieve a high contact and cooperation rate.
8. Four guidelines were followed when constructing CSAs; 1) the boundaries had to align with Census Tracts, 2) consist of 1–8 tracts with populations ranging from 5,000 to 20,000, 3) define relatively homogenous areas, and 4) reflect the boundaries of communities recognized by city planners, institutions, and residents (Baltimore Neighborhood Indicators Alliance, 2018).
9. One distinct methodological problem with using census tracts, is that 50 of the sampled streets are located on census tract borders. Furthermore, of the 200 census tracts in Baltimore, 50 were not represented in our data because they had no sampled street segments, 49 census tracts had one street segment, and 36 census tracts had two sampled street segments located in the census tract, making it problematic to aggregate to this unit of analysis. We note that census tracts were not defined to create community areas, but for administrative purposes (e.g., see, Weisburd et al., Citation2009).
10. At the beginning of the survey, the field researcher defined what was meant by “block” in the survey – “When I talk about your block, I only mean [ADDRESS STREET NAME ONLY] between STREET A and STREET B, including both sides of your street.” This definition aligns with our use of the street segment as the unit of analysis.
11. Offenses included driving a vehicle under the influence of alcohol, damaged someone else’s property on purpose, taken something that didn’t belong to you, used someone else’s credit card or a personal check to steal something, owned or carried a gun without a license, broken into a home or business to steal something, sold illegal drugs, stolen a car or some other type of motor vehicle, used violence against someone-like in a fist fight or assault, and taken something from someone using violence or the threat of violence.
12. The sample of individuals included in the analysis departs from the full sample due to missing data at the individual level. The amount of missing data for each variable did not exceed 2% missing data, and total number of missing cases was 6.8%. Given minimal missing data, listwise deletion was used to deal with the missing data in the current analysis (see, Bennett, Citation2001; Schafer, Citation1999). There were no missing data at the street level in part due to aggregating survey responses.
13. Building counts and purposes were obtained during the physical observations. Mixed-use buildings that included a commercial establishment were included in the calculation. We also obtained aggregated business and employee counts for our sample of street segments from Baltimore Neighborhood Indicators Alliance (BNIA) retrieved from InfoUSA. The number of businesses on the street was highly correlated with our measure of percent commercial (r >0.75), and there were no substantive changes in the regression models, we opted to use our direct measure consistent with the other data collected. Furthermore, we considered the role of different types of commercial establishments such as bars and liquor stores, and included a measure of whether a bar or liquor store was located on the street in different versions of the models. The direction of the variable was negative but not significant, and there were no other substantive changes to the models, including the effect of % commercial. For parsimony, we only include the % commercial in the final models.
14. Bivariate correlations between the independent variables (based on Pearson’s r) were calculated to assess potential problems relating to multicollinearity. None of the correlations exceeded 0.5, which is below the 0.70 threshold traditionally used to identify collinearity problems (Licht, Citation1995). Further collinearity diagnostics were conducted to rule out the presence of harmful collinearity. Variation inflation factors among variables included in the models are below the threshold of 4 and the tolerance levels are well above 0.4, thus there is little concern for multicollinearity (see, Tabachnick & Fridell, Citation2001).
15. The results from the three-level model did not vary substantively, nor did a penalized logistic regression without accounting for nesting; therefore, we report the 2-level multilevel model most appropriate for the nested structure of the data. Covariates at the street-level are still included as they can have an impact on the individual-level outcome, but the variables explain individual-level variance, not street.