Collect
The best available crime data is continuously pulled from key sources including federal, regional, and local law enforcement, and all cross-referenced to establish a more realistic portrait of crime.
Clean
Each source of crime data is closely inspected to flag and correct missing information and data-entry errors as well as account for periodic revisions and irregular fluctuations in reported crime. The potential of underreporting across communities is also accounted for, with the addition of crime victimization and perception of security surveys, local demographics, and economic conditions.
Contextualize
Crime types are assigned a severity weight based on prison sentence length and victim expenses (property damage, healthcare, judiciary fees), with more serious crimes contributing more to the crime risk score.
Compete
Forecasting algorithms are tested against one another in a double-elimination tournament, assessing the predictive accuracy of each algorithm against known benchmark data. The final crime index is constructed using an ensemble of winning algorithms unique to each crime type and community.
Unique Challenges. Specialized Solutions.
What makes the Pinkerton Crime Index different from other crime forecasting solutions is in the way we approach the data – building a customized solution tailored to the nuances and unique challenges each country faces in crime data collection.
STANDARD SCORING
National and local crime data differs in terms of quality, frequency, and included crime subtypes across countries.
The Pinkerton Crime Index takes a country-specific approach to analysis, with scores defined as a multiple of the country’s median district score; a 1.0x score indicating equal risk to the country median, 2.0x twice the risk, 0.5x half the risk, and so on.
Forgoing a one-size-fits-all scoring system for every country enables us to deliver a higher level of accuracy, using country-specific blended data sources, forecasting strategies, and crime classifications.
UNDERREPORTING
Though present in all countries, the extent of underreporting differs by country. For example, according to the 2020 victimization survey (produced by Instituto Nacional de Estadística y Geografía), 89% of crimes in Mexico went unreported to the police.
Utilizing victimization surveys and other sources of crime data, we overcome this data gap by constructing scalars that can accurately inflate underreported crimes.
POPULATION COUNTS
To accurately compare areas of different population size, it is necessary to construct a crime rate (crime per capita) normalizing crime counts by the number of people living in the area.
While most crime rates are calculated using residential population, factors such as heavy tourism and commuting will result in an over representation of the amount of crime in the space.
Relying on residential population alone does not adequately capture the amount of people in a space. A significant challenge in Mexico and the United Kingdom, we leverage information on commuting behavior, number of hotels, and other metrics of tourism to more accurately account for the daily number of people in the space.