Chapter 6 Results

6.3 Interacting Risk Cases

Above we have seen the differing experiences of changing toxicities by race. We find that toxicity is significantly improving for all, but relative improvement has begun to stagnate for minority groups at some levels. This begs the question of what the trends are for individuals with interacting risk cases. Given the multifaceted social identities individuals hold, any single metric will be missing part of the pattern. Ideally we would be able to represent the complex power structures and dynamic social categorizations in our society, but with limited data, and the inherent complexity of the problem, we can begin to broach this problem by looking at the intersection of two commonly discussed risk groups. In this case we examine how the trends of experienced toxicity vary for each race group when considering income levels.

To do so we split households in to low income (under 25,000), medium (between 25,000 and 75,000), and high income (over 75,000).

First we examine How the 5th, 50th and 95th percentiles differ for the black distribution. Dotted lines represent the low income group, solid lines represent the medium income group, and dashed lines represent the high income group. In this case the 50th and 95th percentile act as would be expected, with the high income group experiencing lower toxicity across the board. However, the 5th percentile shows the inverse pattern, with high income households experiencing higher toxicity. One reason this could be true is that high income individuals may be more likely to live in the more toxic suburban or urban areas, rather than rural areas.

In the white population we see a similar pattern. The 95th percentile continues to show the expected behavior at the 95th percentile, but in both the 50th and 5th percentiles the high income group has a significantly higher toxicity.

The Hispanic group shows exactly the opposite pattern of the white distribution, with the high income distribution experiencing higher toxicity in the 50th and 95th percentiles, and lower toxicity in the 5th percentile. In the 50th and 95th percentile, the differences between toxicities experienced by each income group are minimal. The 5th percentile, however, shows significantly lower toxicities experienced by the high income group.

Interestingly, the group with the largest differences between high and low income across all percentile groups appears to be the white distribution. This may indicate that housing discrimination and neighborhood sorting may be relevant areas to study. Given the rational sociopolitical reasons theorized for environmental justice origins, it may be that income is less predictive for minorities due to reduced choice in housing locations, meaning that even given the economic viability of escaping toxicity is less feasible.

Seeing these results together we find they aren’t internally consistent, and do not follow the expected pattern. As hypothesized in discussion of the black distributions, it may be that there are confounding factors that aren’t being addressed. Intersectionality requires a more complex view of identities as explanation of lived experiences, and these factors do not appear to be enough to understand the underlying patterns. Potential confounding areas to explore include: population density, geographic area, prevalence of local manufacturing jobs, level of political engagement, and education levels.

6.4 Exploration of Quantile Regression

Previous sections have focused on teasing apart patterns in the vast amounts of data we have. We find large differences in the toxicity experienced, but may be limited by the choices in division we make. Given we only investigate income as a relevant variable, there are likely patterns in the data that aren’t being examined.

In order to better tease out relationships in the data, and for easy interpretation of coefficients, a functionally similar analysis can be done with quantile regression, focusing on the computed parameters as indicators of the differences in experienced toxicity.

To illustrate the form this model could take, we build preliminary quantile regression models on the 5th, 50th, and 95th percentiles of the toxicity distribution. To predict log toxicity, we use the year, demographic information, population density and quadratic terms. This model operates on a tract level, using percentage of each population group, population density, and would ideally also include variables like unemployment rates, percentage renters, etc.

The predictions shown for each race or ethnicity group are fairly consistent with the results shown above. The distinct separation between the white and Hispanic distributions as compared to the black and ‘other’ distributions is mirrored in the model predictions. The model appears to follow the observed patterns fairly well, aside from a strange fit for the ‘other’ distribution in the 5th percentile.

By adding additional block level data to this regression, we may be able to discuss the magnitude of toxicity differences that are actually attributable to race, and will be informed on the other factors that hold importance in predicting toxicity. Above sections address the empirical reduction in toxicity and the reduction amount attributable to shifts of minority placement in the distribution. An approach able to take it a step further and find the differences attributable to race would add significant value.