Canadian Social Environment Typology User Guide

1. Purpose

The purpose of this user guide is to define the concept of identical Dissemination Area (DA) clusters used in the Canadian Social Environment Typology (CanSET) and to give an overview of how the clusters can be used to explore DA level health and social inequalities. More detailed technical information on the formation of the social environment clusters can be found in the Canadian Social Environment Typology: Methodology Guide developed by the health inequality unit of the Centre for Population Health Data (CPHD).  

2. Background

In recent years there has been a growing demand for relevant health information at neighbourhood level. However, due to the lack of an available classification approach, it has been difficult to compare and contrast small areas within a city or among the cities in Canada. To overcome this shortcoming, the health inequality unit of the Centre for Population Health Data (CPHD) at Statistics Canada collaborated with the Urban Public Health Network (UPHN) to develop a Dissemination Area (DA) classification approach called the Canadian Social Environment Typology (CanSET). A methodology guide has been already in place, which describes the technical and methodological process involved in the development of the CanSET (available upon request). The main objectives of this document is to give basic information on how the CanSET was developed and also to provide detail information on how to use the CanSET classification to understand area level inequalities. 

The CanSET is a hierarchical clustering of DAs within Canadian Census Metropolitan Areas (CMA)^{Note 1} and Census Agglomerations (CA).^{Note 2} The typology generated through this cluster analysis can be taken as a social unit of analysis that indicates the geographic distribution of different kinds of combinations of population characteristics throughout CMAs and CAs. In other word, each social environment cluster includes similar DAs scattered across Canada. Therefore, the CanSET is not defining neighbourhoods but defining areas of common social composition which may be entirely different than the neighbourhoods defined by different local authorities in Canada. The new CanSET data can be used to better understand health inequalities in relation to the composition of sub-city geographies in Canada.

The main objective of this user guide is to facilitate the reader on how to use the new social environment clusters to understand social and health inequalities in more populous areas in Canada.  The social environment clusters are developed by using 30 different socioeconomic, demographic, and ethno-cultural variables from the 2016 Census aggregated at the DA level. All the variables for this product were taken directly from the 2016 Census of Population microdata except the population density variable, which was derived from Census of Population 2016 profile table.

3. Uses of the Canadian Social Environment Typology (CanSET)

The individual level socioeconomic, demographic and ethnocultural status are useful to understand health and social outcomes but are often not available in the most commonly used dataset like surveys or administrative data. Therefore, area based socioeconomic and demographic measures have been widely used to measure inequalities where the individual level information is unavailable or hard to reach. Area level analysis helps to understand whether living in a socioeconomically disadvantaged area gives additional health risk beyond the individual level socioeconomic status.

The new CanSET portrays the complex social composition of Canadian CMA and CA.  Unlike the previous studies that focused only on marginalization or health inequalities, the CanSET is more comprehensive as it includes a variety of dimensions of urban social environment and covers all populous areas of Canada. This is the first analysis to include all DAs from the CMAs and CAs simultaneously in a single typology rather than analyzing them separately for each CMA or CA. In addition, it includes the microdata from the Census of Population 2016 which covers all population of the study area. Therefore, the CanSET is very important for urban research as it allows cities to make comparison within themselves or with other cities and facilitates them to set benchmarks and track progress in combating health and social inequalities.

For instance, in light of the recent COVID-19 pandemic, the CanSET data can be used to understand which social environment cluster is performing better to minimize the rate of infection, hospitalization and mortality. It helps to understand which factors play important role in determining health behaviour and response strategies during a pandemic like the COVID-19.  This understanding could help in the planning and allocating of resources that are required for specific types of communities.

4. Data used to create the CanSET 

Data selected and analyzed for the CanSET was sourced from the Census of Population 2016 microdata for all the DAs within 35 CMAs and 117 CAs in Canada. The DA was used as a geographic unit of analysis. Statistics Canada defines a DA as a small, relatively stable geographic unit composed of one or more adjacent dissemination blocks with an average population of 400 to 700 persons (Statistics Canada, 2016). The DAs are the smallest geographic units for which Statistics Canada disseminates its Census data for public use. The DAs are easy to aggregate small geographic units as they can be combined to form other larger geographic units. Although there are 56,590 DAs in Canada (2016 Census Geography), only the 43,144 DAs that belong to the CMAs or CAs were included in the CanSET, mainly because of their high population densities, and also due to the availability of the variables of interest.

The DAs for which Census data (i.e., either short form or long form) were not released due to confidentiality or data quality issues were not included in the creation of the CanSET. The DAs associated with Indian reserves were excluded from the analysis because some Census questions were either not asked or the concepts were not applicable or they were incompletely enumerated. 



5. Methodology

The CanSET was created using hierarchal cluster analysis of DAs into three levels of nested social environment clusters.  Cluster analysis attempts to assign observations to groups (clusters) based on their similarity or differences using a measure of statistical (i.e., Euclidian) distance from each other. Observations within each group are found to be similar to one another with respect to variables or attributes of interest. In other words, the goal is to group the observations into homogeneous and distinct clusters. A hierarchical cluster analysis method was applied for cluster analysis with a ‘Fast Ward’ option in JMP® 13 (SAS Institute Inc., 2016) analytical software. The Fast Ward method applies an algorithm that computes Ward’s method more quickly for large size data, therefore this method was used considering the data size for this analysis. Details of the methodology can be found in the “Canadian Social Environment Typology: Methodology Guide”, available upon request.

5.1 Variable Selection

The selection of variables was the result of a consultation with UPHN members. An extensive review of the past literature associated with neighbourhood typology was also performed before selecting the variables. Variables describing the demographic, socioeconomic, and ethnocultural determinants of health within the DAs across Canada were used in the clustering algorithm to group similar DAs. Numeric variables were selected that were reliable and readily available from the Census of population 2016. The variables chosen for creating the CanSET cover a wide range of subjects including: demographic structure (age structure, family size, etc.), socioeconomic status (income, education, labour market status, housing condition, etc.) and ethno-cultural background status (Aboriginal status, immigration status, visible minority status, etc.). Health-related variables were deliberately not used in the creation of the typology to make the typology equally applicable in other fields. General socioeconomic and demographic variables were derived directly from the short form Census, whereas specific variable that were not available in short form Census (e.g., level of education, immigration status, occupation etc.) were derived from the long form Census and weights were applied. Some variables were available at person level whereas others were available at census family, economic family or household level. A preliminary list of 93 diverse variables were selected from the microdata of the Census of Population 2016. However, after an extensive review of the variables and after a round of preliminary analysis, some variables that were strongly correlated with each other were excluded. In addition, some similar variables were grouped together that resulted into 30 final variables as shown in Appendix A. All the variables were aggregated at DA level geography and median values were used.

The variables selected for cluster analysis were measured on different scales, or on a common scale with differing variances. Therefore, they were standardized in order to mitigate the effect of these differences among the variables. All the 30 variables were standardized with mean 0 and variance 1 prior to performing the cluster analysis. Some variables with highly skewed distributions were still much skewed even after normalization. So, all variables were capped at their 99^th percentile. 

5.2 Number of Clusters

The optimal number of clusters were six, ten and twenty for the first, second and third level of hierarch respectively. The clusters are defined as the set of DAs that are similar in terms of the selected characteristics (variables).  Davies-Bouldin Index (DBI) for hierarchical clustering was used to determine the optimal number of clusters nested in three hierarchical levels. The smaller value of the DBI means a better clustering solution. Using the DBI, three different optimal ‘k’ values (number of clusters) were determined.

The optimal ‘k’ values were 6 in the range of 2 to 9 clusters, 10 in the range of 10 to 19 clusters and 20 in the range of 20 to 30 clusters as the Davies-Bouldin index was the smallest for these solutions. Therefore, a nested hierarchy of six, ten and twenty clusters was created as outlined in Table 2.



5.3 Cluster Names, Characteristics and Uses

Three levels of nested clusters were created for the CanSET using hierarchical clustering method (Table 2). The optimal number of clusters were determined at six, ten and twenty. The median values were calculated for each variables for each cluster solution and specific characteristics were developed based on the position of the median value of each cluster in a quintile distribution of all the DAs in the analysis.

The CanSET data table comes with Dissemination Area Unique ID (DAUID) and associated cluster membership for each of the three cluster solutions. For example, one DA may fall into cluster A when we use six cluster classification but the same DA may fall into cluster A3 and A32 respectively for ten and twenty cluster classification. Users can select the cluster solution that distinguishes their area of interest and can use the respective value to classify the neighbourhoods. Users can select the cluster solution to use depending on the socioeconomic, demographic and ethno-cultural composition of a CMA or CA of interest. For instance, if users wants to use CanSET data to understand health inequalities in large CMAs like Toronto, Montreal or Vancouver, they may use all six, ten or twenty cluster solutions as all type of DAs are present in large CMAs having large population. However, only a few clusters types may be present in small CMAs or CAs. In that case, they may end up using only six or ten clusters to compare health outcomes. 

5.3.1 Six clusters solution

Level one of the hierarchy creates six clusters as the optimum cluster solution. The number of DAs in each cluster ranges from 1,328 to 19,127 whereas the median population density ranges from 2,176 to 8,160 people per square kilometre. In this solution, each DA is given a value of 1 to 6 based on their cluster membership. Following are some of the major characteristics of each cluster for this solution.



5.3.2 Ten clusters solution

Level two of the hierarchy creates ten clusters as an optimum cluster solution (Table 2). The number of DAs varies from a minimum of 1,260 to the maximum of 7,787. The median population density ranges from 1,834 to the maximum of 8,160 per square kilometre. The ten clusters developed are nested in the six clusters described above (see Table 2). Therefore, the clusters in each level share some characteristics with the higher level of cluster to which they belong. The ten clusters have the characteristics as outlined below.



5.3.3 Twenty clusters solution

Level three of the hierarchy creates twenty clusters as an optimum cluster solution. The twenty clusters developed are nested within the ten clusters which in turn are nested within six clusters described above (see Table 2). Therefore, the clusters in each level share some characteristics with the previous level of cluster to which they belong. In the twenty cluster solution, the number of DAs in each cluster ranges from 847 to 6,650 whereas the median population density ranges from 32 to 19,750 per square kilometre. The twenty clusters have the following characteristics.



6. How to link the CanSET with your data

The CanSET is designed to use with any dataset for which the DA or postal code information is available.  The CanSET data can be easily linked with other data sources using the dissemination area unique identification (DAUID) number from the Census of population 2016. DAUID from any other Census of population may need a correspondence file to match them with the 2016 Census DAUID. The data set that do not have DA information but have the postal code information can also be linked with CanSET data using the Postal CodeOM Conversion File plus (PCCF+) produced by Statistics Canada.

The CanSET data comes with a table that includes DAUID and hierarchal cluster membership for each of the three cluster solutions. The table contains four columns of DAUID, Cluster 6, Cluster 10 and Cluster 20 with numeric value of 1 to 6, 1 to 10 and 1 to 20 respectively as shown in a sample table (Table 6) below.



7. Summary and Conclusion

This user guide provides an overview of the methodology used to develop CanSET and provides information on how to use these social environment clusters with the users own health or social data. To create the CanSET Canadian DAs were grouped together based on their demographic structure, socioeconomic status, and ethno-cultural background. The CanSET used 30 different variables from the Census of Population 2016 to classify Canadian DAs into three nested levels of clusters using the Fast Ward hierarchical clustering method. The optimal number of clusters were determined at six, ten and twenty using the Davies-Bouldin index. The median values were calculated for each variables for each cluster solution and specific characteristics and names were given to each cluster based on the value of the variables in each clusters.

The new CanSET portrays the complex social composition of Canadian urban geography.  The CanSET is more comprehensive as it includes a variety of dimensions of urban social environment and covers all populous areas of Canada. This study is unique as it include all DAs from the CMAs and CAs simultaneously in a single typology rather than analyzing them separately for each CMA or CA. The analysis is based on the microdata from the Census of Population 2016 which covers all population of the study area. Therefore, the CanSET data allows cities to make comparison within themselves or with other cities and facilitates them to set benchmarks and track progress in combating health and social inequalities.

8. References

Davies, D. L., & Bouldin, D. W. (1987). A Cluster Separation Measure. IEEE Transactions on Patern Analysis and Machine Intelligence, 1(2), 224-227.

Johnson, R., & Wicheren, D. (2007). Applied Multivariate Statistical Analysis (6th ed.). Prentice Hall.

SAS Institute Inc. (2016). JMP 13 Multivariate Methods: Hierarchical Cluster. Cary, North Carolina, USA: SAS Institute Inc.

Statistics Canada. (2016). Dictionary, Census of Population. Ottawa: Statistics Canada. Retrieved from http://www12.statcan.gc.ca/census-recensement/2016/ref/dict/geo021-eng.cfm

Statistics Canada. (2018). Postal Code Conversion File Plus (PCCF+) Version 7B, Reference Guide. Ottawa: Statistics Canada.

Appendix A