IS6833 Homicide Prediction 2011 Michelle Bergesch Jeff Stahlhuth
Issue: Determine Next Homicide Homicide: Includes murder and non-negligent manslaughter which is the willful killing of one human being by another Key Considerations: – Generally independent non-related events – Too easy to jump to conclusion – Data granularity (Region, District, Ward, Zip, Neighborhood) – Victim to Offender relationship
Causation Vs. Correlation Homicide Contributing Factor Assumptions: 1.Economy 2.Abandoned property 3.Census demographics Sex, Race, Education, Income 4.Homicide as a result of another crime 5.Previous conviction type and frequency 6.Felon location
Causation Vs. Correlation 1.Economy: – Presumption is bad economy = more crime – Most research is inclusive on this relationship – Available census data was from Abandoned Property – Presumption increased number of abandoned properties would be used for drugs and crime – Property data organized by address not neighborhood, required extensive
Causation Vs. Correlation 3.Census Demographics: – Needed data formatted by neighborhood – Available census data was from Homicide as a result of another crime – FBI expanded homicide data sort data by relationship, sex, weapon, related crime, etc. – Related crime categories do not completely match UCR categories
Causation Vs. Correlation 5.Previous conviction type and frequency – Strongest predictive factor to predict murders is previous offender history (Berk) – Court case data provides frequency and timing of previous offenses 6.Felon location – Need a mechanism to identify previous offender location (i.e. sex offender registry) – No research to support proximity of murder to offender dwelling
What Data Was Considered? Missouri Census Data – St. Louis Police Dept. UCR Data – St. Louis Circuit Court (Cases & Protective Orders) – – Neighborhood Background Data – Missouri Highway Patrol UCR – FBI UCR Crime Statistics – Berk Crime Prediction Tool – Univ. of Pennsylvania – prediction-and-probability-in-crime-patterns/3598/ prediction-and-probability-in-crime-patterns/3598/
What Data Was Used? St. Louis Police Dept. UCR Data Neighborhood Background Data FBI UCR Crime Statistics
Methodology FBI Expanded data breaks down annual homicides by contributing circumstances (Rape, Burglary, Robbery, etc.) Data granularity does not match standard UCR categories, however 5 crimes do match
Methodology FBI Expanded Data
Methodology Develop a holistic scoring number to predict homicides and related crime trend for each neighborhood Predicted Homicides (PH) – linear trend analysis of homicide rates by neighborhood. Predicted Crime Index (PCI)- predicted sum of projected related crimes based on reported UCR data for each neighborhood Report then outlines where the next homicide will be AND a related crime index. Patrol deployment recommendation based on sum of PH + PCI
Conclusion Projected Homicides by Neighborhood NEIGHBORHOOD TOTAL MurderTOTAL Rape TOTAL ROBBERY Burglary TOTAL Larceny TOTALAUTO Theft ANCILLARY HOMICIDEPHPCI Jeff-Vanderlou Average Mark-Twain Average Kingsway-West Average Wells-Goodfellow Average Baden Average Penrose Average Fairground Average North-Point Average O'Fallon Average Hyde-Park Average Dutchtown Average College-Hill Average Hamilton-Heights Average Walnut-Park-West Average Tower-Grove-South Average Gravois-Park Average Downtown-West Average Academy 2011 prediction Benton-Park-West Average Old-North-St.-Louis Average
Conclusion Resource Deployment by Neighborhood NEIGHBORHOOD TOTAL MurderTOTAL Rape TOTAL ROBBERY Burglary TOTAL Larceny TOTALAUTO Theft ANCILLARY HOMICIDEPHPCI Jeff-Vanderlou Average Dutchtown Average Wells-Goodfellow Average Tower-Grove-South Average Baden Average Mark-Twain Average Penrose Average Kingsway-West Average O'Fallon Average The-Greater-Ville Average Downtown-West Average Gravois-Park Average North-Point Average Downtown Average Carondelet Average Walnut-Park-West Average Benton-Park-West Average Fairground Average Central-West-End Average Hamilton-Heights Average