Graphs

Statistics Canada census tract (CT) spreadsheet data was extracted from the University of Toronto's CHASS data centre. The CT data, which included income, was extracted for the years of 1971, 1981, 1991, 2001 and 2006. This data was used to construct the graphs which compared income demographics for these years.

In order to compare different income demographics for the graphs there had to be some sort of defined classification system. For our purposes there were five classes that were used which showed the variation that there was from the average income of CTs. These categories were Very Low Income (40% below average), Low Income (40% to 20% below average), Middle Income (20% below to 20% above average), High Income (20% above average) and Very High Income (40% above average).

GIS

Tables collected from Statscan and CHASS were joined to Census Tract shapefiles obtained from Statscan (2006) and ABACUS (1971) using ArcMap 10.  These layers were then symbolized to show average household  income for 1971 and 2006 (Figure 5a - .png or .pdf and Figure 5b - .png or .pdf), unemployment rate for 1971 and 2006 (Figure 6a - .png or .pdf and Figure 6b - .png or .pdf) and the average number of children per household per census tract for 1971 and 2006 (Figure 7a - .png or .pdf and Figure 7b - .png or .pdf).

A number of tools from the Spatial Statistics toolbox (ArcGIS description of the toolbox can be found here) were then run on the income data from the 1971 and 2006 censuses.

From the Analyzing Patterns toolset, the High/Low Clustering (Getis-Ord General G) tool was run to measure whether or not there was any clustering of high or low values across the area for either date (Table 1a - .txt and Table1b - .txt). Second, the Spatial Autocorrelation (Global Moran’s I) tool, also from the Analyzing Patterns toolset, was run to determine whether the spatial pattern of the data is clustered, dispersed or random (Table 2a - .txt and Table 2b - .txt).   Both of these tools output a .txt file containing z-score and p-scores regarding statistical significance.

From the Mapping Clusters toolset, the Hot Spot Analysis (Getis-Ord G*) tool was run; this tool identifies spatially significant hot spots (high values) and cold spots (low spots) and creates  a feature class which displays how hot and cold the locations are as standard deviations from the mean (Figure 8a - .png or .pdf and Figure 8b - .png or .pdf).

Lastly, from the Modelling Spatial Relationships toolset, an Ordinary Least Squares analysis (Figure 9a - .png or .pdf and Figure 9b - .png or .pdf) tool as well as a Geographically Weighted Regression (Figure 10a - .png or .pdf and Figure 10b - .png or .pdf) tool were run to determine how well average household income could be explained by the average number of children and the unemployment rate of census tracts in 1971 and 2006 . These two variables were chosen because they were the only matching variables between the two census years that could be considered as contributing to income levels. All of the other variables had been changed in the 35 years between census years.