The program inputs incident locations (e.g., robbery locations) in 'dbf', 'shp', ASCII or ODBC-compliant formats using either spherical or projected coordinates. It calculates various spatial statistics and writes graphical objects to ArcView/ArcGIS, MapInfo, Atlas*GIS, Surfer for Windows, and Spatial Analyst.

CrimeStat is organized into seven sections:

Data Setup

• Primary file - this is a file of incident or point locations with X and Y coordinates. The coordinate system can be either spherical (lat/lon) or projected. Intensity and weight values are allowed. Each incident can have an associated time value.

• Secondary file - this is an associated file of incident or point locations with X and Y coordinates. The coordinate system has to be the same as the primary file. Intensity and weight values are allowed. The secondary file is used for comparison with the primary file in the risk-adjusted nearest neighbor clustering routine and the duel kernel interpolation.

• Reference file - this is a grid file that overlays the study area. Normally, it is a regular grid though irregular ones can be imported. CrimeStat can generate the grid if given the X and Y coordinates for the lower-left and upper-right corners.

• Measurement parameters - This page identifies the type of distance measurement (direct, indirect or network) to be used and specifies parameters for the area of the study region and the length of the street network. New in CrimeStat IV is the ability to utilize a network for linking points. Each segment can be weighted by travel time, travel speed, travel cost or simple distance. This allows the interaction between points to be estimated more realistically.

Spatial Description

• Spatial distribution - statistics for describing the spatial distribution of incidents, such as the mean center, center of minimum distance, standard deviational ellipse, directional mean, convex hull, Moran's I spatial autocorrelation index, Geary's C spatial autocorrelation index, Getis-Ord G spatial autocorrelation index, Moran correlogram, Geary correlogram, and Getis-Ord correlogram routines.

• Spatial autocorrelation - statistics for describing the amount of spatial concentration for zone data including Moran's I spatial autocorrelation index, Geary's C spatial autocorrelation index, Getis-Ord G spatial autocorrelation index, Moran correlogram, Geary correlogram, and Getis-Ord correlogram routines.

• Distance analysis I - statistics for describing properties of distances between incidents including nearest neighbor analysis, linear nearest neighbor analysis, and Ripley's K statistic. New in CrimeStat IV is a routine that assigns the primary points to the secondary points, either on the basis of nearest neighbor or point-in-polygon, and then sums the results by the secondary point values.

• Distance analysis II - calculates matrices representing the distance between points for the primary file, for the distance between the primary and secondary points, and for the distance between either the primary or secondary file and the grid.

Hot Spot Analysis

• 'Hot spot' analysis I - routines for conducting 'hot spot' analysis on point data including the mode, the fuzzy mode, hierarchical nearest neighbor clustering, and risk-adjusted nearest neighbor hierarchical clustering. The hierarchical nearest neighbor hot spots can be output as ellipses or convex hulls.

• 'Hot spot' analysis II - more routines for conducting hot spot analysis on point data including the Spatial and Temporal Analysis of Crime (STAC) and K-means clustering. The STAC and K-means hot spots can be output as ellipses or convex hulls.

• 'Hot spot' analysis of zones - routines for conducting hot spot analysis on zone data including Anselin's local Moran, the Getis-Ord local G statistics, and Zonal nearest neighbor hierarchical spatial clustering.

Spatial Modeling I

• Interpolation I - a single-variable kernel density estimation routine for producing a surface or contour estimate of the density of incidents (e.g., burglaries) and a dual-variable kernel density estimation routine for comparing the density of incidents to the density of an underlying baseline (e.g., burglaries relative to the number of households).

• Interpolation II - includes the Head Bang statistic which smooths zonal data to ensure reliable estimates. There is a standard Head Bang and an interpolated Head Bang for further smoothing over a study region.

• Space-time analysis - a set of tools for analyzing clustering in time and in space. These include the Knox and Mantel indices, which look for the relationship between time and space, a spatial-temporal moving average, and the Correlated Walk Analysis module, which analyzes and predicts the behavior of a serial offender.

• Journey-to-crime analysis - a criminal justice method for estimating the likely location of a serial offender given the distribution of incidents and a model of travel distance. The routine allows the user to estimate a travel model with a calibration file and apply it to the serial events. It can be used to identify an optimal central location given the distribution of 'points' and assumptions about travel behavior. Also, there is a routine for drawing lines between origins and destinations (crime trips)

• Bayesian Journey-to-crime analysis - a routine for adding information on where other offenders lived who committed crimes in the same locations to the Journey-to-crime estimate. This method improves accuracy and precision of the Journey-to-crime method and has been tested on sets of serial offenders in Baltimore County, Chicago, the Hague (Netherlands), and Manchester (England)

Spatial Modeling II

• Regression I - a regression module for modeling crime and other rare events as a function of one or more independent variables. Version 4.02 includes Ordinary Least Squares and Poisson models with or without a spatial autocorrelation component (conditional autoregression - CAR, and simultaneous autoregression -SAR). The Poisson is generally a better method for modeling skewed distributions, such as crime. The current version includes "pure" Poisson, Poisson with linear correction, Poisson-Gamma, Poisson-Gamma-CAR/SAR, Poisson-lognormal, Poisson-lognormal-CAR/SAR, Binomial logit, Binomial-logit-CAR/SAR, and Binomial probit models. The models are estimated either with maximum likelihood or Markov Chain Monte Carlo (MCMC) methods.

• Regression II - applies coefficients estimated in the Regression I module to a new data set to make predictions.

• Discrete Choice I - a module for modeling discrete dependent variables and includes multinomial logit and conditional logit models. The former is used to estimate predictors associated with the cases while the latter is used to estimate predictors associated with the choices.

• Discrete Choice II - applies coefficients estimated in the Discrete Choice I module to a new data set to make predictions.

• Time Series Forecasting - uses time series data over different geographical areas to predict the number of events in the next time period.

Crime Travel Demand Modeling

Crime travel demand modeling is an application of travel demand modeling, widely used in transportation planning, to crime analysis. The analysis is done by zones. First, crime 'trips' are defined as a link between an offender residence/origin location and a crime location. The number of crimes originating in each zone is counted as is the number of crimes ending in each zone. Second, the model is run sequentially in four separate stages with multiple routine in each stage:

• Trip Generation - Separate models are produced that predict the number of crimes originating in each zone (origins) and the number of crimes ending in each zone (destinations). CrimeStat IV uses a multivariate Poisson regression model, with stepwise options, to create the prediction. Trips from outside the study area (external trips) can be added to the origin model to account for travel from outside the region. Once the models are created, a balancing procedure ensures that the number of origins equals the number of destinations.

• Trip Distribution - Using the predicted number of crime trips originating in each zone and the predicted number of trips occurring in each zone, the second stage distributes trips from each zone to every other zone using a gravity model. There are routines for calculating the actual (observed) distribution from individual data, for estimating the prediction coefficients, and for applying the predicted coefficients to the predicted origins and destinations. Another routine allows a comparison of the predicted trip distribution with the observed trip distribution.

• Mode Split - The predicted number of trips for each zone-to-zone pair can be split into likely travel modes using an accessibility function that approximates the utility of one mode relative to the others.

• Network Assignment - Finally, the predicted trips from each zone to every other zone by travel mode are assigned to a likely route based on the A* shortest path algorithm. The output includes the likely routes taken for each origin-destination zone pair and the total volume of trips on network links. This step requires a travel network, one for each travel mode. There are additional utilities for calcuating transit networks from station/stop locations and for testing for one-way streets.

Options

• Parameters can be saved and re-loaded.

• Tab colors can be changed.

• The Primary file can be converted to a KML file.

• An Excel file can be converted to a dbf files.

• Monte Carlo simulation data can be output.

CrimeStat is accompanied by sample datasets, a tutorial, a manual that gives the background behind the statistics and examples, and library files that can be embedded in third party applications. The manual also discusses applications of CrimeStat developed by other analysts and researchers. The program and sample data sets are in Windows-based zipped files that can be downloaded. The manual is a set of individual chapters in PDF files which can be viewed online or downloaded.

Here are some maps from Baltimore County (MD) that illustrate statistical calculations made by CrimeStat.

The above map shows the distribution of vehicle thefts in Precinct 11 between June and July 1996.

The above map shows vehicle theft risk (vehicle thefts relative to population).

The above map shows robbery 'Hot Spots'.

Finally, the above map shows crime trips on the Baltimore road system.