Caplan-Kennedy-Miller style risk terrain: kernel-density surfaces
for each risk-factor point layer on a common grid, then a Poisson
regression of gridded incident counts on the standardized layer
densities. Relative risk scores per cell come from the fitted
surface.
Usage
morie_crim_risk_terrain(incidents, layers, n_grid = 25L, bandwidth = NULL)
Arguments
- incidents
Two-column matrix/data.frame of incident x,y.
- layers
Named list of two-column matrices (risk-factor point
layers, e.g. bars, transit stops).
- n_grid
Grid cells per axis. Default 25.
- bandwidth
Kernel sd. Default Silverman per layer.
Value
List of class "morie_rtm": coefficients (one per
layer, log relative risk), risk_surface (matrix), grid_x, grid_y,
deviance_ratio, n, call.
References
Caplan, Kennedy & Miller (2011) Justice Quarterly 28(2).
Examples
set.seed(4)
inc <- cbind(runif(80), runif(80))
lay <- list(bars = cbind(runif(15), runif(15)))
morie_crim_risk_terrain(inc, lay, n_grid = 10L)
#> Risk terrain model, 80 incidents
#> log relative risk per layer (standardized densities):
#> bars
#> 0.0467
#> deviance explained: 0.1%
