Methodology
The Ten Cities project aims to think about urban landscapes as a
collection of different neighbourhood typologies. Where most of
our maps analyze cities through one filter such as average
income, this approach asks how we can distill a number of
indicators at once to understand our urban areas. With this
approach we see the city as a series of demographic, economic
and social patterns; patterns that reflect the impact of urban policy
and planning decisions over the past 60 years.
One of the most important notes is that the variables selected for this study,
and the subsequent patterns they create, are under the subjective influence of the researcher.
For this project, six domains representing key factors in the residence patterns of the
city were chosen, each domain containing its own six related indicators drawn from the
2015 Canadian Census. While these domains are all sourced from academic research and
the indicators are all questions from the census, the ultimate choice of these domains and their
indicators reflect the researcher's decisions and internal biases. It's important to note this subjectivity;
the neighbourhood cluster results would certainly differ if another individual were to choose the indicators to drive this study.
This research approach is a factorial analysis followed by a cluster analysis. Data was organized into census tracts, which contain on average between 2,500-8,000 people. A factor analysis is conducted for each of the six domains (each domain containing six indicators). This factor analysis produces three correlations, or patterns per domain. For example, one correlation pattern for the House Tenure might be a presence of high homeowners, low mobility, few major repairs needed to housing, and few owners paying 30%+ of their income towards housing. This pattern would indicate general housing security. Through this factor analysis we now have a total of 18 correlation patterns (3 per domain). Each census tract now has a value for each of the 18 correlation patterns. If a census tract has a positive value for that housing security pattern, it indicates lots of homeowners, low mobility, good housing condition, etc. If the census tract has a negative value it indicates many renters, high mobility, poor housing condition, and housing financial stress.
Next, a cluster analysis is conducted. Each census tract has all 18 of its factor analysis values tossed into a cluster analysis formula. This formula clusters census tracts together based on their similarity. Census tracts with generally similar values are placed into the same category. Within each grouping, general similarities for housing, income, labour, citizenship and household pop up and provide the definition that you see for each of these clusters, or cities. This cluster analysis is also dependent on the researcher, who can choose how many clusters are formed. For this analysis, cities in Ottawa and Toronto formed three broad cluster groups, with Toronto forming ten total "cities" and Ottawa nine. This forms the clustering of cities that represent a multi-variate analysis of these large urban areas, and hopefully starts a conversation about how our cities are organized, and how we can use this perspective to work towards more equitable solutions for all residents.
To read more about this methodology, please read the methodology chapter of my graduate thesis, The Ten Cities of Toronto.
Study Indicators
A visualization of the methodology described above can be found in the image below. This model also includes the 36 variables used for this study. All variables are sourced from the 2016 Canadian Census. Of these variables, 32 indicators are taken directly from the Canadian Census. These full definitions can be found in the Census Dictionary link below. The remaining four custom indicators are calculated using census variables, and are defined below:
Gini Coefficient of Income Inequality: an indicator measuring the level of distribution between all income classes. The higher the value, the larger the divide between different income levels in a neighbourhood. This can either manifest as a large income gap between a lower and upper income class, or at one end of the income spectrum, such as income disparity between fairly high income households and very high income households. A low Gini Coefficient would indicate a high homogeneity of income for all households within that neighbourhood.
Walks Index of Income Polarization: this indicator is similar to the Gini Coefficient, but instead of measuring the distribution of income between various groups, this index measures the level of polarization between low and high income classes. The benefit of having this second indicator is that it can distinguish when there is a large wealth gap within the neighbourhood itself. A neighbourhood with a high polarization index would indicate the presence of both many low and high income households.
Standardized Average Income: this indicator is a standardized value for after-tax household income. The value has been standardized, as the large value of an annual income would distort the factorial analysis. A 0 value indicates the exact average. Positive values indicate an income above average, while negative values indicate an income below average.
Racial Diversity Index: this indicator uses racial indicators provided by the Canadian Census, and uses a formula to indicate how diverse a neighbourhood is in terms of individuals identifying with a given ethnic place of origin, as defined by Statistics Canada. A higher value indicates a greater level of racial diversity within a neighbourhood. It should be noted that this is an indicator especially affected by a researcher's subjectivity, and by the variables contained within the 2016 Census.
Methodology Structure
