Results
The findings we obtained from our analysis showed that domestic violence appears to decrease in the year following lockdown (Map 1). However, it is important to problematize a simplistic reading of this. While there might have been a decrease in domestic abuse rates during the pandemic, we know that domestic abuse reporting rates have decreased due to stay-at-home orders, increased isolation of victims, increased exposure to exploitable relationships, and reduced options for support (Usher, Bhullar, Durkin, Gyamfi, & Jackson, 2020). Consequently, decreased reporting may be one of the reasons we see this pattern. In addition, blue markers were added to represent locations of domestic violence shelters and resources (shelter, legal, counselling), indicating where domestic violence victims are able to seek support. It is evident that these resource centres are located in areas where instances of domestic violence have appeared to decrease between March 21, 2019, and March 22, 2021, and absent from the few areas we observe an increase in domestic violence. The resulting fact is that these resources are not allocated adequately across the city to help the most vulnerable populations.
Map 1: Percent Change of Domestic Abuse Rates Pre- and During-COVID-19 Pandemic
Exploratory Regression:
​
The candidate demographic variables used in our regression analyses were: percentage of housing crowded, percentage of households below the poverty line, percentage of age 16+ unemployed, percentage of age 25+ without high school diploma, and per capita income. We used the exploratory regression tool to evaluate all possible combinations of candidate variables, to find the best model to fit our topic of interest, and to determine predictors of importance.
The exploratory regression results for March 21st, 2019 to March 21st, 2020 confirmed percent housing crowded, percent aged 16+ unemployed, and percent aged 25+ without a high school diploma to be the best explanatory variables for counts of domestic battery. These variables had the highest adjusted R-squared (0.25) and the lowest AICc (1113.59), two statistics that represent model performance and goodness of fit. Re-running the exploratory regression with these variables yielded the exact same results. The exploratory regression for March 22nd, 2020 to March 22nd, 2021 resulted in the same three variables being significant. However, the adjusted R-squared and AICc values were slightly different: 0.24 and 1089.86 respectively.
Using the same five base demographic variables, we conducted an exploratory regression in the same manner as before, investigating their effect on the percent change in cases between both years. Both exploratory regressions found percent aged 16+ unemployed and per capita income as the best explanatory variables. These variables had an adjusted R-squared of 0.08 and an AICc of 618.15.
Generalized Linear Regression:
​
Using the three assigned demographic variables for each study year and the two for the percent change of all years, we conducted a GLR on all three data sets. Across most of the city in both years there is a minimal relationship between variables (Map 2). The teal and dark green regions of the map show that the demographic variables have an effect on the counts of domestic violence. For both study years, the explanatory variable with the most significance was the percent aged 16+ unemployed, with an R-squared of 0.23. The areas where this has the strongest effect are notoriously poor areas. There was a change in only 4 community areas between pre- and during-COVID-19 years. In the northern region, one area became slightly more correlated but increased by only one standard deviation. In the southern region, three areas became darker purple for the during-COVID-19. Meaning, during COVID-19 the variables were even less correlated in those areas.
Map 2: GLR Results for Domestic Violence and Select Demographic Variables Pre- and During-COVID-19 Pandemic
On the left side of the layout (Map 3), we notice that the GLR output for percent change in domestic violence cases across both years of study in relation to demographic variables. We see a similar pattern to the count GLR outputs but with slightly less relation between variables in the northern region. Interpreting the results of this map shows no variables are statistically significant in predicting the percent change in domestic battery instances. The highest R-squared value was 0.02 and between percent change and per capita income. This map also includes the GWR output to investigate if space was a factor. There is no visual or numerical change present, confirming that space is also not a factor and the demographic variables we used were not able to explain percent change.
Map 3: Comparing GLR and GWR Results for All Years Between Percent Change in Domestic Violence and Select Demographic Variables
Geographically Weighted Regression:
The GWR results (Map 4) between both pre- and during-COVID-19 years used the same variables defined in the exploratory regressions for each year. During-COVID-19 years, we note that many areas on the map changed from showing very little correlation (light purple) to a slight correlation (white). The areas of significance in dark green remain consistent with the GLR results. Interpreting some of the output tables shows one significant variable relationship in both years: count of domestic battery vs. percent 16+ unemployed (Same as GLR).
Map 4: GWR Results for Domestic Violence and Select Demographic Variables Pre- and During-COVID-19 Pandemic
The charts below show the scattering of residuals for this relationship (Graph 1; Graph 2). However, these result tables when compared to the GLR results are exactly the same. There was a slight decrease in the relationship between the study years, but geography is not a factor in this relationship. The demographic data is a much better predictor for count instances of domestic battery than it is for percent change across all years. However, we were unable to find a spatial relationship within our data. Therefore, we are clearly missing key explanatory variables of incidents of domestic violence.
Graph 1: Residuals Relationship Pre-COVID-19
Graph 2: Residuals Relationship During-COVID-19
