Crime Science, Published Online 16 July 2015
By Yasemin Irvin-Erickson and Nancy La Vigne
“Transit stations are acknowledged as particularly criminogenic settings. Transit stations may serve as crime “generators,” breeding crime because they bring together large volumes of people at particular geographies and times. They may also serve as crime “attractors,” providing well-known opportunities for crimes. This paper explores the node and place characteristics that can transform Washington DC, Metro stations to generators and attractors of different crimes at different times of the day. The crime-generating and crime-attracting characteristics of stations are modeled with Negative Binomial Regression analysis. To reflect the temporal trends in crime, crime counts are stratified into three temporal groups: peak hours, off-peak day hours, and off-peak night hours.
Robbery density at peak, non-peak day, and non-peak night hours
“The findings from this study not only suggest that stations assume different nodal and place-based crime-generating and crime-attracting characteristics, but also these roles vary for different crimes and different times. The level of activity and accessibility of a station, the level of crime at a station, and the connectedness of a station to other stations are consistent indicators of high crime rate ratios. Different characteristics of a station—such as being a remote station or belonging to a high or low socioeconomic status block group—are significant correlates for particular crime outcomes such as disorderly conduct, robbery, and larceny. ”
GIS and Science 
I’m not familiar with Negative Binomial Regression but what seems rare to me is that the R^2 is too low.
With that outcome can we say that it is an acceptable model?