Forecasting Knife Homicide Risk from Prior Knife Assaults in 4835 Local Areas of London, 2016–2018

How accurately can all recorded locations of 97 knife homicides in one year be forecast across all 4835 Lower Layer Super Output Areas (LSOAs) of London, based only upon all 3506 known locations of nonfatal knife injury assaults in the preceding year? All recorded “knife crimes” in the Metropolitan Police Service (MPS) area of London in the financial year 2016–2017 (year 1) were manually reviewed to identify all 3506 reported locations of knife-enabled (KE) but nonfatal injuries, as distinct from other events digitally coded as “knife crimes”, such as displaying, threatening with, or carrying knives. All KE homicides in 2017–2018 (year 2) were then added to the database. Each KE injury assault in year 1 was classified for occurrence in one of London’s 4835 LSOAs. The total N of such crimes within each LSOA was summed across all records to divide all LSOAs into seven categories of frequency of KE injury assaults in 2016/2017: 0, 1, 2, 3, 4, 5, and 6 or more. We then divided the N of LSOAs in each category in 2016/2017 (year 1) into the total N of knife homicides in 2017/2018 (year 2). False positives, false negatives, and percentage of homicides targeted were calculated for six different targeting plans. Over two thirds (69%) of KE homicides in 2017/2018 (year 2) occurred in just 67 (1.4%) of all 4835 LSOAs, comprising 3.3% of the 2048 LSOAs that had had one or more of the 3506 KE injury assaults mapped in 2016/2017 (year 1). The proportion of LSOAS with a KE homicide in year 2 was higher where there were higher numbers of KE injury assaults per LSOA in year 1. Among LSOAs with zero assaults in year 1, only 1% had a KE homicide in year 2. Among LSOAs with six or more KE injury assaults in year 1, 15% had a KE homicide in year 2. While the risk of homicide was 1400% higher in the hottest spots of knife assaults than in the coolest spots, the absolute number of year two KE homicides in those hottest areas was only 6% of the total. All LSOAs with one or more year 1 KE assaults had three times as much KE homicide risk than LSOAs with no year 1 KE assaults. Targeting all 2048 areas that had one or more year 1 KE assaults would have yielded a true positive rate of over 3% and covered 69% of actual homicide locations. The true negative rate for 2781 areas that would not have been targeted based on having no year 1 KE injuries would have been 99%. Predicting which local areas are most likely to suffer knife-enabled homicides, based only on recent nonfatal knife injuries, can pinpoint risks of homicide in local areas that are up to 1400% higher than in most local areas, offering a range of strategies for resource allocation.