Data Handling and Exploration

All the Data Used for Objective 1

The vulnerability index was developed using pre-determined indicators under each domain of the vulnerability index: exposure, sensitivity and adaptive capacity indicators (Table 1).

Table 1. Exposure, Sensitivity, and Adaptive Capacity Data Sources

Data Exploration - Exposure

Exposure (Table 2)
- Weather - annual mean temperature, annual cumulative precipitation
  - Both annual mean temperature and annual cumulative precipitation were transformed using log transformation for linearity in order to meet the assumption of using factor analysis at a later stage.
- Air pollution - nitrogen dioxide, ozone and delicate particulate matter
- Urban heat island index

Table 2. Raw data table (example) showing exposure indicators.

Figure 4. Scatterplots of correlated exposure indicators

Dimension Reduction - Factor Analysis on Exposure

The first two factors were extracted from the factor analysis on the exposure indicators, the selection of the 2 factors was based on the eigenvalue cutoff point of 1 (Figure 5).
- Factor 1 explained 23.7% of the total variance, including annual cumulative precipitation, annual mean temperature, nitrogen dioxide and PM2.5 (Table 3).
  - Higher annual mean temperature, nitrogen dioxide and PM2.5 were positively associated with factor 1, while higher annual cumulative precipitation was negatively associated with factor 1 (Figure 6).
- Factor 2 explained 20.7% of the total variance, including the urban heat island index (Table 3).
  - Urban health island index was positively associated with factor 2 (Figure 6).
- The two factors extracted in total explained 44.4% of the total variance.

Figure 5. Scree plot of exposure indicators to determine the number of factors. An Eigenvalue of 1 was used as the cut-off point in this analysis, and for exposure indicators, factors 1 and 2 were used to generate exposure dimension.

Figure 6. Factor analysis results on exposure variables. Factor 1 included annual cumulative precipitation, annual mean temperature, nitrogen dioxide and PM2.5. Higher annual mean temperature, nitrogen dioxide and PM2.5 were positively associated with factor 1 while higher annual cumulative precipitation was negatively associated with factor 2. Urban health island index was positively associated with factor 2.

Table 3. Loadings of each exposure variable, showing how each variable is related to the extracted factor of the exposure dimension. Factor 1 explained 23.7% of the total variance, and factor 2 explained 20.7% of the total variance. The two factors extracted in total explained 44.4% of the total variance.

Data Exploration - Sensitivity

Sensitivity (Table 4)
- Age - the proportion of people aged over 65 years in each DA. The higher the proportion of people aged over 65 years old, the more sensitive the area is when facing climate change and its related health threats.
- Income - the proportion of people with low-income status in each DA. The higher the proportion of people with low-income status, the more sensitive the area is when facing climate change and its related health threats.
- Immigration - the proportion of immigrants in each DA. The higher the proportion of immigrants, the more sensitive the area is when facing climate change and its related health threats.
- Education - the proportion of people who at least obtained a high school diploma in each DA. The higher the proportion of people who at least obtained a high school diploma, the less sensitive the area is when facing climate change and its related health threats. This indicator was further reverse-coded to match the same trend as other indicators.
- Visible minority - the proportion of people who are visible minorities in each DA. The higher the proportion of people who are visible minorities, the more sensitive the area is when facing climate change and its related health threats.
- Marital status - the proportion of people who are married or in common-law in each DA. The higher the proportion of people who are married or in common-law, the less sensitive the area is when facing climate change and its related health threats. This indicator was further reverse-coded to match the same trend as other indicators.

Table 4. Raw data table (example) showing sensitivity indicators.

Figure 7. Scatterplots of correlated sensitivity variables

Dimension Reduction - Factor Analysis on Sensitivity

The first two factors were extracted from the factor analysis on the sensitivity indicators. The selection of the two factors was based on the eigenvalue cutoff point of 1 (Figure 8).
- Factor 1 explained 31.4% of the total variance, including age, education, visible minority and immigration (Table 5).
  - A higher proportion of visible minorities and immigrants was positively associated with factor 1, while a higher proportion of people aged above 65 years and a lower proportion of obtaining a high school diploma were negatively associated with factor 1 (Figure 9).
- Factor 2 explained 28.0% of the total variance, including marital status and income (Table 5).
  - A higher proportion of low-income households and a lower proportion of people who are married or in common-law were positively associated with factor 2 (Figure 9).
- The two factors extracted in total explained 59.4% of the total variance.

Figure 8. Scree plot of sensitivity indicators to determine the number of factors. An eigenvalue of 1 was used as the cut-off point in this analysis, and for exposure indicators, factors 1 and 2 were used to generate sensitivity dimension.

Figure 9. Factor analysis results on exposure variables. A higher proportion of visible minorities and immigrants was positively associated with factor 1, while a higher proportion of people aged above 65 years and a lower proportion of obtaining a high school diploma were negatively associated with factor 1. A higher proportion of low-income households and a lower proportion of people who are married or in common-law were positively associated with factor 2.

Table 5. Loadings of each sensitivity variable, showing how each variable is related to the extracted factor of the sensitivity dimension. Factor 1 explained 31.4% of the total variance, and factor 2 explained 28.0% of the total variance. The two factors extracted in total explained 59.4% of the total variance.

Data Exploration - Adaptive Capacity

Adaptive Capacity (Table 6)
- Active living environment - active living environment index. It was transformed using log transformation for linearity in order to meet the assumption of using factor analysis at a later stage. The higher the index of the DA, the more adaptive capacity that area had.
- Greeness - Normalized Difference Vegetation Index (NDVI), mean within 1 km of DA centroid. It was transformed using log transformation for linearity in order to meet the assumption of using factor analysis at a later stage. The higher the index of the DA, the more adaptive capacity that area had.
- Facilities - number of non-profit organizations, healthcare facilities, clinics, EMS and ambulance services within 1 km of DA centroid. All these variables were divided by the population in that DA and then log-transformed for linearity in order to meet the assumption of using factor analysis at a later stage. The more facilities that are accessible in a DA, the more adaptive capacity that area has.
- Dwellings in need of major repairs - Major repairs needed ÷ Total - Occupied private dwellings [25% sample in the DA]. The higher the proportion of houses that needed major repairs, the less adaptive capacity that area has. This indicator was further reverse-coded to match the same trend as other indicators.

Table 6. Raw data table (example) showing adaptive capacity indicators.

Figure 10. Scatterplots of correlated adaptive capacity variables

Dimension Reduction - Factor Analysis on Adaptive Capacity

The first two factors were extracted from the factor analysis on the adaptive capacity indicators. The selection of the two factors was based on the eigenvalue cutoff point of 1 (Figure 11).
- Factor 1 explained 62.0% of the total variance, including dwellings that needed major repairs and the number of non-profit organizations, healthcare facilities, clinics and EMS (Table 7).
  - Dwellings that needed major repairs were negatively associated with factor 1. In contrast, DAs that have higher access to or more significant numbers of non-profit organizations, healthcare facilities, clinics and EMS are positively associated with factor 1 (Figure 12).
- Factor 2 explained 13.9% of the total variance, including active living environment and greenness (Table 7).
  - The active living environment was associated positively with factor 2, and greenness was negatively associated with factor 2.
- The two factors extracted in total explained 59.4% of the total variance (Figure 12).

Figure 11. Scree plot of adaptive capacity indicators to determine the number of factors. An Eigenvalue of 1 was used as the cut-off point in this analysis, and for adaptive capacity indicators, factors 1 and 2 were used to generate the adaptive capacity dimension.

Figure 12. Factor analysis results on exposure variables. Dwellings that needed major repairs was negatively associated with factor 1, while DAs that have higher access to or larger numbers of non-profit organizations, healthcare facilities, clinics and EMS are positively associated with factor 1. The active living environment was associated positively with factor 2 and greenness was negatively associated with factor 2.

Table 7. Loadings of each adaptive capacity indicator, showing how each variable is related to the extracted factor of the adaptive capacity dimension. Factor 1 explained 62.0% of the total variance, and factor 2 explained 13.9% of the total variance. The two factors extracted in total explained 76.0% of the total variance.

Health Data for Objective 2

The vulnerability index was further validated using health events between 2015-2018 from administrative databases (Table 8 and Table 9).

Respiratory diseases (total cases of 34,351 among all DAs)
Cardiovascular diseases (total cases of 90,817 among all DAs)
Mental health, all kinds (total cases of 88,037 among all DAs)
Injury events (total cases of 30,529 among all DAs)

Table 8. Sources of health outcome data for validation

Table 9. Raw data table (example) showing the count of each health outcomes in DA.

Page updated

Report abuse