Conclusion
The overall spatial pattern of weapon violations in Chicago was shown to be relatively scattered throughout the city with a number of identified “hot spots” including the West, the far Southwest and the far Southeast regions. The results from the three types of crime analyses performed are mostly consistent with each other. The location of land features such as water bodies and major road networks was found to have significant impacts on determining the locations with high crime rates. In terms of the usefulness of the tools used, unlike the typical hot spot analyses that produces statistical summaries of incidents, kernel density interpolation provides information on the precise locations, spatial extent and intensity of crime hotspots for the entire area of interest, which allows for sophisticated interpretation of crime data.
Using the exploratory regression analysis, three socio-economic factors were found to have the strongest correlation with the number of weapon violations in Chicago, Illinois: percentage of crowded housing, percentage of unemployed population over the age of 16, and percentage of population over the age of 25 without a high school diploma. The results from another global regression model, OLS, confirmed the relationships between these three variables with the occurrence of weapon violations. Based on the results from the local model, GWR, unemployment generally have a strong positive correlation with weapon violation incidents. Therefore, it can be concluded that policies should be implemented to help combat crime through the improvement of employment support of workers. Percent of crowded housing, however, showed spatially diverging behaviour in terms of its relationship with weapon offences across the city. Surprisingly, the level of education was shown to be positively related to this type of crime with some variation observed locally. In terms of the usefulness of the models, compared to other non-spatial or global regression models, GWR was found to be more effective in honoring the importance of spatial non-stationarity when performing spatial analyses.
Limitations and Sources of Error
Firstly, we acknowledge that a change in input parameters for regression analyses in ArcGIS and CrimeStat can generally pose a major impact on the outputs. For example, choosing different options with bandwidth values for GWR and parameter values for kernel density estimation may have produced different results.
Different crime types were aggregated into the “weapon violations” category, which may have neglected the difference in the nature of the individual crime types. This can potentially lead to over generalization of the unique features related to the sub-classes of weapon violations.
In this study, zoning districts were used in place of a land use layer due to data availability. Because of the issue of non-conformal (“Grandfathering”) land use within zoning districts, the land use classes derived from it may not reflect the current state of land use in the city. For the purpose of this study, it was assumed to be up-to-date.
Lastly, since the modifiable areal unit problem (MAUP) may have been present during the spatial analyses performed, errors or artifacts may have resulted when aggregating spatial data into units. As this issue cannot be easily solved, information should be carefully grouped during analyses, and should only be done when necessary.
Potential future research on similar topics should address a number of issues. Crime analyses should be performed on multiple spatial scales for a more sophisticated interpretation. The majority of the spatial analyses performed in this study are limited to community level in Chicago due to the limited availability of socio-economic data at a finer resolution. Therefore, the resulting relationships discovered can be an artifact of the spatial scale used. Performing the same set of analyses at a different scale may confirm the findings with greater insights with higher level of confidence and possibly allow more in-depth interpretation of the results.