In November of 2015, the Los Angeles Geographical Society (LAGS), continued their lecture series with a presentation by Dr. Chris Carter, a professor at Long Beach City College. His lecture was titled: “Spatial Patterns of Crime in Long Beach Using ArcGIS Spatial Statistics Tools”.
He started off by discussing Long Beach demographics for 2015. The population at that time was 471,210 with a diversity index of 87.8%. The per capita income was $25,807, which matches 86% of California’s per capita income. His crime data comes from the Long Beach Police Department (LBPD) and ranges from 2010-2014. This data includes crime type, crime location by address, arrestee in combination with city by address, and victim by address.
He examined patterns of assault and battery (not including domestic/family violence), robbery (not including commercial), burglary (not including commercial), and murder. With this data he performed geocoding, which is the act of transforming spatial, locationally-descriptive text into a valid spatial representation. For this he used ESRI software, specifically ArcGIS Desktop, which was one of his first tips. Another tip he gave was that with crimes and addresses of people you need to rename/re-code points from police station and police substations.
Carter then decided to do a hotspot analysis (HSA), examining by block, neighborhood, city, and regional patterns. For this you must be sure to determine your units of analysis. With the HSA, there is the Effect of Distance Threshold Robbery Rate equally a 3,000 ft distance threshold vs. a 12,00 ft distance. Both were considered statistically valid. The data was also examined in terms of Counts vs Rates. Crime counts are useful for patrol officers and commanders when answering questions such as, “Where are the crimes and where do I need more officers?” Crime rates on the other hand are useful for researchers and policy makers. They can be used as a control for population size/housing units. Another difference of counts is that it’s interested in local scale to create a fishnet tool. This creates a 500 ft grid based on location.
In terms of residential burglaries, they showed localized hotspots where patrols are most needed. These areas were higher density areas, such as downtown, Belmont Shores, and North Long Beach. When finding the Directional Distribution of HSA for residential burglary, Carter used the Directional Distribution Tool to visualize changing spatial patterns. When he ran the hotspots for each year from 2010-2014, he found one standard deviation (68%) of weighted values (GT_Bin), which showed a very consistent pattern except for 2011. The results for Directional Distribution of street assault/battery and street robbery showed that assault/battery had a shifting focus on downtown while robbery had an increasing focus on downtown. This raises the question of within each patrol district where should officers guard? Burglary was found to be more concentrated around downtown with new hotspots showing up in East Long Beach along Los Coyotes Ave. Robbery had narrow hotspots around downtown limited to a few blocks.
HSA Rates data is based on data from the US Census Block Groups via the American Community Service. Residential burglaries rates show the hotspots to disappear around downtown and Belmont Shores but the lower income area of north Long Beach becomes a visible hotspot. The higher income areas of Eastern Long Beach become a cold spot. Assault, battery and robbery (by rate) are clearly around downtown and where there is commercial and pedestrian activity, such as along Pine St., Anaheim St. and Pacific Coast Highway.
Carter also performed a Regression Analysis to test the Social Disorganization Theory by Shaw and Mckay (1940s-1960s). This theory states that crime is place specific and that there is a lack of social controls to guide personal behavior due to neighborhood instability. Part of this involves a measured arrest rate vs. poverty, ethnic, heterogeneity, population turn-over and so on. Carter then completed his own Dependent Analysis looking at arrest and citation rate. His independent variables came from the American Community Survey totaling five years and on the block group level.
Another tool he used was the Explanatory Regression Tool, but he couldn’t get passing models. Despite this the tool is great for ESRI documentation. When interpreting explanatory regression results he also used the ordinary least squares (OLS) regression tool, which included dependent variables and the log for arrests and citations. His interpretation of the results was that arrests and citations rates increase as block groups have higher rates of poverty, larger percentage of rental housing, lower rates of English speaking residents, and lower per capita income. This then supports the Social Disorganization Theory. Looking at the adjusted R2 of 0.55, it shows that about 50% of explanatory power is still missing. Reasons for this might be because there are missing variables that can influence arrest rates, such as the gang membership rate, racial profiling, police deployment patterns, and other reasons.
The policy implications for this issue then are to reduce social disorganization. Some ways to do this would be to create stability within neighborhoods to reduce population turn-over. This would mean encouraging people to get to know their neighbors better. There is also the option of providing assistance for first time homeowners. In terms ethnic heterogeneity, the issues of integration and communication with neighbors and authorities would need to be addressed.
Going back to the issue of distance, in this case it must be measured in relation to aresstee to crime and victim to crime. This allows for testing of the crime patterns theory, which states that crimes occur in the activity space of individuals with certain overlap in other areas. When working within ArcGIS, one needs to add XY fields to geocoded points. The results showed that 33% of arrest/battery cases occurred at the same address with 43% occurring within 500 ft. For robberies, 23% occurred within 1,500 ft and 37% was within 300 ft while 52% was within one mile. For burglary, 35% occurred within 3,000 ft with 47% occurring within one mile. Looking at these results, the next question is, “Given the ‘distance decay‘ patterns, where should detectives focus their searches for suspects?” Focusing on the distance of victim to crime, 37% of arrest/battery cases occurred at the same address with 50% within 500 ft. This then is similar to arrestees. For robbery, 43% occurred within 1,500 ft. Thus this shows classic distance decay patterns, meaning that people move around more within place closer to home. This also supports the crime pattern theory, showing overlapping activity spaces in local areas.
In conclusion, one should take care when geocoding. Think about your scale of analysis when working with hotspots, use theory to guide your regression analysis, and use the Exploratory Regression Tool to find valid models while being careful with non-normal distributions. One should also run a OSL regression with the final model, use the XY to line tool to see spatial relationships between arrestees/victims/crime location and remember that finding good crime data can be a challenge. For Carter’s study, he had to find public sources of crime data, such as from the LBPD, though they restrict the use of the full dataset. Another dataset he used can be found online at Crime Reports. A final dataset mentioned was retrieved from the LA County Sheriff, such as data by crime and address.