M.S. Applied Data Science - Capstone Chronicles 2025

12

workflow, creating a pipeline capable of supporting future web-based deployment.

approximately 0.913, meaning the model correctly classifies poverty status the majority of the time. With reduced sensitivity values, though, these models do not properly account for all high-impact variables. Overall, these results confirm that disability undermines the economic returns to education and provide actionable insight for targeting ADA-aligned workforce investments, regional support programs, and geographically tailored antipoverty interventions. However, further research needs to be conducted on a precise geographical basis, to understand the full impact of disability and its interactions with other factors on poverty outcomes. 6 Discussion The results of this project provide strong evidence that disability fundamentally alters the economic returns to education, resulting in persistently higher poverty risk across all educational levels. The original research question asked whether educational gains translate into equivalent economic outcomes for disabled adults, and the modeling clearly showed they do not. Linear and logistic models demonstrated that disability prevalence remained a significant positive driver of tract-level poverty even after controlling for education, employment, and income normalization, while Geographically Weighted Regression identified where this relationship was most severe — particularly rural southern Illinois. These findings are consistent with national work by the Census Bureau and She & Carter (2023), which report that disabled adults experience lower earnings and higher unemployment even with comparable schooling. The spatial patterns extend prior research by pinpointing structural barriers — such as weaker labor markets, limited ADA-aligned employment, and fewer support services in rural regions — suggesting that education alone cannot close poverty gaps without parallel policy measures such as accessible workforce programs, targeted hiring incentives, and expanded disability services. Although ACS data include higher margins of error in small tracts, sensitivity checks showed the core patterns remain stable. Some interaction effects were weaker or spatially inconsistent, likely due to sampling variability.

5.1

Evaluation of Results

The evaluation of the modeling results supports the study’s original hypothesis that disability significantly elevates poverty risk even when educational attainment is held constant, and the findings remained consistent with the patterns revealed in the exploratory analysis. Regression models showed that disability directly impacted the predicted percentage of the poverty line a household would fall at, with coefficients ranging from -10 to -88 points. Disability also indirectly impacted poverty outcomes through hours worked per work. Disabled people that are able to work less hours than their able-bodied peers were restricted in their ability to earn higher incomes, even after adjusting for educational attainment, race, and source of insurance. One model focused specifically on educational attainment and its interactions with disability. This model showed the quantitative impacts of disability when holding educational attainment constant. Any educational attainment was strongly reduced by disability status and its interactions with educational attainment. In short, a higher education degree benefits able-bodied individuals much more than it does disabled people. Another regression model showed that disabled individuals in more rural areas were more negatively affected by disability than their non-rural counterparts. These regression models show that while educational attainment helps individuals move further above the poverty line, disability and rurality reduce these benefits by a large amount. These statistical outcomes aligned with the weighted poverty estimates: disabled adults with only middle school education faced a 39.2 percent poverty rate compared with 14.9 percent for non-disabled peers, and even among those with master’s degrees, pover ty remained 11.4 percent for disabled adults versus 3.3 percent for non-disabled adults. However, with a lowered R^2 values, the current regression models cannot account for the majority of variability in this data. Logistic regression achieved an accuracy of

187