The Possibilities and Constraints of Crime Rate Forecasts

• In Event: Lethal Inequalities: Understanding the Structural Predictors, Gendered Patterns, and Life Lost to Violence

Abstract

This paper builds on Richard Rosenfeld’s pathbreaking work on forecasting crime rates. It concentrates its discussion on forecasting for large units (cities, nations) and relatively long temporal intervals (months, years). Within this framework it considers both univariate and multivariate approaches. Although both methods can provide enlightening forecasts, they also have limitations. Absent the existence of nonlinear generating mechanisms, forecasts using univariate methods (for example ARIMA models) amount to linear projections of the recent past into the future. The future then looks much like the present, with few surprises. Multivariate methods (for example, vector autoregressive models) can produce more intricate patterns, but they often require forecasting all of the model variables. The models are also most useful where some components strongly lead or lag the others. Acknowledging these constraints, the paper undertakes a variety of univariate and multivariate forecasts of local and national data.

Author

• David McDowall, University at Albany