| Mean | SD | Median | Min | Max | |
|---|---|---|---|---|---|
| Percent of Units With Eviction Filings | 6.40 | 5.02 | 4.86 | 0.01 | 20.39 |
| Percent of Black Residents | 19.49 | 22.05 | 8.83 | 0.00 | 79.75 |
| Median Household Income | 95373.16 | 46913.51 | 86744.00 | 18125.00 | 237188.00 |
Regression Analysis of Evictions in Boston
Background and Question
Evictions disproportionately impact poor neighborhoods and neighborhoods with higher concentrations of Black residents (Robinson & Steil, 2021). This study uses a generalized linear regression to model the log-transformed percentage of renter-occupied units with eviction filings between 2020 and 2023 by census tract, based on the share of Black residents and median income in Boston census tracts.
Specifically, this analysis seeks to answer three questions: (1) What is the relationship between the percentage of units with eviction filings and the percentage of Black residents? (2) What is the relationship between the percentage of units with eviction filings and median income? and (3) Holding constant the two independent variables, how does the relationship between eviction filings and each independent variable change?
Model
Three models are conducted 1) Bivariate regressions between: eviction and percent of Black residents 2) Bivariate regressions between: eviction and median income 3) Multivariate regression: eviction vs income and Black residents to capture the role of each variable while holding the other constant.
Data
This analysis uses data from Princeton’s Eviction Lab for eviction data and the American Community Survey (ACS) 5-Year Data (2018-2022) for demographic data. The total count of eviction filings between 2020 and 2023 are aggregated and merged with the ACS-survey dataset. The aggregated count of eviction filings is divided by the total number of renter-occupied units to obtain the outcome of interest: percent of renter-occupied units that have experienced eviction filings between 2020 and 2023. Of Boston’s 207 census tracts, those with no population or no renter-occupied units were removed, leaving 193 observations for the regression. Consistent with previous literature (Robinson & Stell, 2020; Desmond & Gershenson, 2017), the percentage of Black residents and median income are divided into tertiles to account for the possibility that the relationship between eviction and these two variables may be non-linear.
Key Takeaways
Census tracts with higher concentrations of Black residents and lower median incomes experience significantly higher eviction filing rates. The percentage of Black residents is the strongest and most consistent predictor of eviction filings in a census tract, pointing to systemic inequities and implicit biases within the housing system.
First bivariate model: Compared to tracts with the lowest percentage of Black residents (the first tertile), tracts in the highest tertile have an eviction filing rate 2.77 times (277%) higher, while those in the middle tertile have a 2.26 times (226%) higher rate.
Second bivariate model: Compared to the highest-income tracts, middle-income tracts have a 70% (0.7 times) higher eviction filing rate, while the lowest-income tracts show a 156% (1.56 times) higher rate,
Multivariate model: The coefficient for the highest percentage of Black residents remains highly statistically significant (p < 0.001) and retains the high magnitude. On the other hand, the predictive power of median income weakens, with both its coefficient magnitude and statistical significance falling.
Beyond investing in critical housing assistance programs such as emergency rental aid and housing vouchers, these findings underscore the need to address racial inequities in housing through initiatives like Right to Counsel and stronger anti-discrimination policies.
Data Distribution
All three variables exhibit a positive skew, particularly eviction filings and the percentage of Black residents. Between 2020 and 2023, in the 75th percentile (48 census tracts), at least 1 in 10 units experienced an eviction filing. The mean percentage of Black residents (19.49%) is more than twice its median (8.83%), with 5% of tracts (11 tracts) having 65% or more Black residents. Median income shows the least variation and skew, with a mean of $95,373 and a median of $86,744. Table 1 presents these variables’ summary statistics.
Correlation
Figure 4 & 5 shows two scatterplots depicting eviction filings against the two predictors, with the dashed lines representing the variable’s median value. The plot shows a stronger correlation between Black residents and evictions (Pearson Correlation Coefficient: 0.75) than median income and evictions (-0.46).
Spatial Distribution
The bivariate choropleth maps (Figure 6 and 7) display the percentage of renter-occupied units that experienced eviction filings between 2020 and 2023, and two predictor variables: percent of Black residents (Figure 4) and median household income (Figure 5). The data are divided into tertiles, representing the lowest (0% - 33%), middle (33% - 66%), and highest (66% - 100%) thirds. Census tracts are grouped according to their tertile combination for eviction filings and predictor variable.
The maps illustrate the data’s skewed distribution geographical clustering. Black residents are highly concentrated in the southwest region. Median income also shows geographical clustering, though to a lesser extent than the share of Black residents.Areas with higher percentage of Black residents are concentrated in regions with the highest eviction filing rates. Lower income also displays some concentration in the highest eviction filing areas but less so than the percentage of Black residents.
Results
Since the outcome is on the log scale, the results are exponentiated to interpret them in the original scale of eviction filings. The intercept represents the expected value of eviction filings on the logarithmic scale when all predictor variables are at their reference (baseline) levels. The reference levels represent the lowest eviction rates for ease of interpretation (e.g., the highest income tertile and the lowest tertile for Black residents). The other coefficients represent additive effects on the log scale relative to the reference group.
First bivariate model (Table 2, first column): Holding all else equal, the model predicts an eviction filing rate of 3.08% for tracts with the lowest percentage of Black residents. This rate increases to 4.55% and 11.6% for tracts in the second and third tertiles, respectively. These predicted values represent a 47% (0.47 times) and 270% (2.7 times) higher eviction filing rate compared to tracts with the lowest concentration of Black residents. Both coefficients are statistically significant at the highest level (p < 0.001). Figure 8 presents these predicted values with 95% confidence intervals at each tertile.
Second bivariate model (Table 2, second column): Tracts in the highest-income group are expected to have an eviction filing rate of 3.64%, which increases to 6.17% in middle-income tracts and 9.37% in the lowest-income tracts. Compared to the highest-income tracts (the reference group), middle-income tracts have a 70% (0.7 times) higher eviction filing rate, while the lowest-income tracts show a 156% (1.56 times) higher rate. These coefficients are also statistically significant at the most stringent level (p-value < 0.001). See Figure 9 for the predicted values with 95% confidence intervals for each income tertile.
The multivariate model (Table 2, third column): In the highest income tracts, eviction filing rates are predicted to be 3.82% for tracts with a moderate percentage of Black residents and 9.02% for tracts with the highest percentage. These correspond to a 38% (0.38 times) and 226% (2.26 times) higher eviction filing rate, respectively, compared to tracts with the lowest percentage of Black residents. These findings remain statistically significant, though the coefficient for the second tertile becomes less significant, with the p-value increasing from less than 0.001 to 0.005.
For tracts with the lowest percentage of Black residents, eviction filing rates are predicted to be 3.71% in the lowest-income tracts and 3.44% in middle-income tracts. These expected values represent increases of 34% (0.34 times) and 25% (0.25 times), respectively, compared to the highest-income tracts. However, these coefficients’ magnitude and statistical significance decrease substantially in this model compared to the bivariate model.
| Bivariate: Eviction vs Race | Bivariate: Eviction vs Income | Multivariate: Eviction vs Income + Race | |
|---|---|---|---|
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | |||
| Tertile 2 - Black Residents (4.24% - 18.63%) | 0.390*** | 0.325** | |
| (0.109) | (0.113) | ||
| Tertile 3 - Black Residents (19.68% - 79.75%) | 1.326*** | 1.183*** | |
| (0.109) | (0.127) | ||
| Tertile 2 - Middle Income ($70,655 - $111,477) | 0.528*** | 0.221+ | |
| (0.130) | (0.113) | ||
| Tertile 1 - Lowest Income ($18,125 - $70,385) | 0.941*** | 0.295* | |
| (0.129) | (0.127) | ||
| Intercept: Baseline for Reference (Lowest Percent of Black Residents and Highest Income) | 1.125*** | 1.292*** | 1.016*** |
| (0.077) | (0.092) | (0.091) | |
| Num.Obs. | 193 | 193 | 193 |
References
Desmond, M., & Gershenson, C. (2017). Who gets evicted? Assessing individual, neighborhood, and network factors. Social Science Research, 62, 362–377. doi:10.1016/j.ssresearch.2016.08.017
Robinson, D., & Steil, J. (2021). Eviction dynamics in market-rate multifamily rental housing. Housing Policy Debate, 31(3-5), 647-669. https://doi.org/10.1080/10511482.2020.1839936