Quantifying Crime Waves

Using a longitudinal data set containing high resolution spatial and temporal data of crimes committed in the city of Philadelphia during 1999, we attempt to quantify correlations and relationships between crime events as well as locations throughout the city. We use a variety of techniques from autoregressive models, time-series analysis, and random matrix theory to deduce the correlation structure within the data. Results show that the largest signal in the data is most likely an artifact of police procedure and find the strongest correlation of crime rate to day of the week, with Sundays experiencing low crime levels, while Tuesday and Wednesday having the highest. We show this daily correlation appears stronger for certain types of crime (mainly drug related incidents) and that crime rates for all other offenses can’t be distinguished from noise. In addition, we find no significant pattern of causation where in the presence of one type crime at a location predicts the presence of another type later in time.