All projects have some problems, whether they are methodological or operational errors. While I was working on this project I did encounter some shortcomings.
Methodological Errors:
1.) The Canadian Census data is available for each Dissemination Area, which is the smallest geographic unit available for census data. Since I compared crime rates for each neighbourhood, the DA's had to be assigned to their corresponding neighbourhoods. After displaying the neighbourhood shapefile and the DA shapefile for Vancouver, I discovered that some DA's crossed over multiple neighbourhood boundaries. To solve this problem, I assigned a centroid to each DA. The 'feature to point' tool in ArcToolBox came in handy for this calculation. The DA's were then grouped into their appropriate neighbourhoods. A few points were located on a neighbourhood boundary which made it difficult to decide which neighbourhood the DA would belong to. With such a large area, this leads to calculating the averages of the DA's, which may result in a loss of characteristics, thus leading to the scale problem. The scale problem is defined as "the variation which can occur when data from one scale of areal units is aggregated into more or less areal units" (Ratcliffe, 1999). The DA data was aggregated to the neighbourhood level which results in a loss of variation. Discrepancies exist in average income and population density for each neighbourhood.
2.) Modifiable areal unit problem (MAUP) occurs when a region is subdivided, affecting the model results. Geographical data is aggregated to present a study showing a spatial phenomena. Furthermore, areal units for geographic data are arbitrary in nature (Ratcliffe, 1999). The census data is available at the DA level and crime rates is available at the neighbourhood level. The data at the neighbourhood level may have been assigned arbitrarily on the basis that public may be more familiar with neighbourhoods. Aggregating data to an areal unit is an inaccurate presentation of the spatial distribution of crime (Muscat & Losoncz, 2000). There is influence and bias caused through the subdivision of the sample area or region. In this case, Vancouver is divided into 24 neighbourhoods with crime rates for each neighbourhood. But depending on how you divide the neighbourhoods, crime data will vary for each neighbourhood. For example, if boundaries are changed spatially, crime data will vary for each neighbourhood. Different combinations of areal units available to aggregate data is astounding (Ratcliffe, 1999). Working with neighbourhoods, it is difficult to examine spatial patterns with such a large unit.
3.) Ecological fallacy is another problem I encountered. Ecological fallacy is the assumption a researcher makes about an individual that is based on aggregate data for a group. In other words, by using highly aggregated data, an attempt is made to infer those results to lower levels of aggregation (McGrew & Monroe, 2000). Since the crime rate for Downtown was quite high and the average income was lower than other neighbourhoods, this does not imply that all persons with lower incomes are criminals. The correlation shows that the rate of crime is higher in areas that have a greater number of people with low incomes.
4.) The crime statistics that is available to the public through the Vancouver Police Department is not too detailed because of privacy and legal issues. I wanted to know where crime actually occurred so I could create a map showing the hotspots in Vancouver. Instead, the data is only available at the neighbourhood level, which does not handle small amounts of data very efficiently. Also larger polygons affects the data by skewing the data. Although point features of crime locations are more precise than area based methods, digitizing all the hotspots in Vancouver would have been a tedious task.
5.) The shapefile for the neighbourhoods did not contain the Musqueam and Stanley Park area. To create these polygons, I digitized both areas in ArcMap based on the neighbourhood map which is available through the Vancouver Police Department.
7.) For the MCE, the constraints and factors I chose was purely based on personal assumptions. Other factors such as employment rates, home ownership, government assistance and education levels could have been taken into account. The constraints and factors that were used for the project assumes influence on the locational analysis of a new police station. In the pairwise comparison matrix table, assumptions were made in assigning the weights to the factors. By changing the values in the pairwise comparison matrix tables, factors can be given more weight than other factors. Depending on the weights placed on each factors, emphasis could be placed on a specific factor which would result in different suitable locations. As a result, different weights placed on different factors yield varying suitable locations.
8.) Since the crime data is in Adobe Portable Document Format, I had to enter it in Microsoft Excel. The crime rate totals from January 2004 to June 2004 were summed for each category and grouped. Calculation errors are possible in the process of data manipulation.
I did not encounter any operational problems in ArcGIS or IDRISI Kilimanjaro.
Although my project looks at occurrences of crime from January 2004 to June 2004, the temporal timeframe is relatively short. In reality, more research and planning is needed in order to decide where the best location for a new police station would be in the City of Vancouver.
