Limitations & Uncertainty
- Knox Index: While the Knox Index was constructed to pick up on any type of space-time interaction, it does not distinguish whether such interactions are due to shifting population distribution or other-related phenomena (Kulldorff & Hjalmars, 1999). In other words, the Knox Index can be biased when there are geographical population shifts, which can be a considerable problem since the statistic could lead to false interpretation as a result. Another problem with the Knox Index is due to the definition of ‘close’ distance as it is determined by the analyst. Depending on the values chosen for ‘close’ time and ‘close’ distance, the results of the test may vary and subsequently change the interpretation. Thus, without having absolute justification for defining ‘closeness’, the interpretation could potentially be unreliable. A possible solution to address the problems with the Knox Index is to perform the Mantel Index instead, which is also possible through CrimeStat; but again, it is not without its own statistical limitations.
- Single Kernel Density: Even though the single kernel density estimate was intentionally created to compare with the dual kernel density due to the knowledge that the City of Toronto is the most populous city in Canada, the stark contrast was still surprising. Overall, this portion of the analysis further emphasized the importance of understanding the underlying population distribution, similar to above but with different meaning, given the dramatic change when standardized by population for the dual kernel density estimation.
- Population Aged 15 and Above, by Dissemination Area: While we do not strongly believe changing the missing population values to ‘0’ for the 14 DAs would have biased our results given the underlying geography (i.e., the land use) of these areas, we still would have liked to compare the results if areal interpolation of population had worked during the data preparation process. However, this is with the understanding that areal interpolation of population would not have (necessarily) generated the correct population values for the 14 DAs either.
Concluding Thoughts
Considering the number of existing research that provides evidence for spatial and temporal patterning of crime events, the importance of GIS to study and examine crime seems blatantly obvious. However, environmental criminology is still a growing interest within criminology, and the integration of GIS has only occurred in the recent decades. Crime is inherently part of the way we live, where people are mobile; time passes; and change is constant. So, “it is tied to the physical distribution of people and objects, to the routine activity patterns of daily life, and to the ways in which people perceive and use information about the environment” (Brantingham & Brantingham, 1993, p. 20). Thus, it should be natural to examine crime patterns from the perspective of environmental criminology when considering our relationship with the environment.
This subsequently leads us to CrimeStat, a free software that includes more than 100 statistical routines for the spatial analysis of crime and other incidents. It is also designed to interface with most GIS programs which provides added opportunities to visualize or further examine the outputs. While the software is rather straightforward to maneuver, understanding each statistical routine is a bit convoluted when examining spatial analysis of crime for the first time. The number of available statistical routines actually led to the paradox of choice, where we chose to explore what we knew best rather than venture into new options. So, given more time to conduct our study, we wish to further explore the CrimeStat Manual to identify other applicable analysis to expand our study. Furthermore, the great benefit of CrimeStat is the manual and the software is free to download, and thus the spatial analysis of crime is more readily available.