error/uncertainty

Modifiable aerial unit problem and ecological fallacy are both problems that we face within this research. Because data tend to be recorded around artificially defined units of measurement such as dissemination area, census tract, city boundary or larger administrative or political boundaries, patterns in crime vary quite substantially at different scales of aggregation. Also, the aggregation of the various units and boundaries from different sources of data always pose problems because the separation and aggregation of various polygons are often subjective to the users. Our crime analysis begins with a small spatial unit, dissemination area, and builds larger unit that reflects natural neighbourhoods. Each neighbourhood consists of a cluster of DAs. In order to interpret the result in one single map, we need to manually classify and aggregate the DAs into neighbourhoods. Some DAs do not fall into one single neighbourhood but instead lay across two or more neighbourhoods. Thus these DAs are reclassified into the appropriate neighbourhoods according to our perceptual estimation. In addition, because the results presented in our study are based on rather large neighbourhood boundaries, the values presented for each neighbourhood should only be taken as an average within the boundaries.

Further, as is common in much criminological research, there is some necessary mixing of time ranges when gathering the data from different sources. According to our research, our crime data are available for 2002 to 2006; census data are from 2001 and 2006; local area boundary data is from 1999. Due to data availability, this inconsistency of time ranges creates uncertainty about some of the data and should be taken into account when interpreting the results, unless one can assume neighbourhood stability in all characteristics used in the analysis. However, over time it should become possible to eliminate such temporal mixing by the ongoing data refinement.

suggestion on future research

Further research in crime patterns should begin by using a fine unit of analysis, and aggregating up to street blocks, neighbourhoods, major activity spaces, or even larger units when the theoretical orientation calls for it. Finer crime classifications may resolve the problem surrounding boundary reclassification. Also, point feature data of crime counts are essential in providing a more detailed and accurate analysis such that hotspot analysis or spatial regression procedure can then be able to perform. Hotspot analysis is good at showing the areas with a higher than expected level of criminal activity; whereas spatial regression model accounts for the spatial autocorrelation between crime rates and different types of socio-economic characteristics. With these analytical techniques, it is easier to predict the crime rate by controlling for the population at risk.