M.S. Applied Data Science - Capstone Chronicles 2025

11

insurance through a previous workplace, having insurance through Medicaid, age, and educational attainment.

interaction between holding an associates or bachelors degree and being disabled brought another 18.8 point reduction. Overall, the benefit of holding this level of education is reduced from 139 points to 61.3 points. This means that being disabled reduces the benefit of an associates or bachelor’s degree by over 50%. A different model without any interactions studied educational attainment, disability, and hours worked per week showed that any benefit from educational attainment is reduced by 29.1 points when the person is disabled. Holding all other variables constant in this model, each one hour increase in work hours per week led to a predicted 2.8 point increase in percent of the poverty line. If someone is unable to work more hours in a week, they are limited in the amount that they can offset being disabled. A final model studying education, rurality, disability, and the interaction between rurality and disability showed the same conclusions as previous models, where disability reduces predicted poverty outcomes. This model also shows that disabled people in more rural areas are more negatively impacted. Each increase in the number of census tracts per PUMA reduced the predicted percent of the poverty line by 1.79 points. Compounding this, if someone is disabled, their predicted value reduces by 40.3 points for the disability, along with another 0.7 points for each increase in the number of census tracts per PUMA. While these models cannot account for much of the variability within the dataset, they clearly show that disability, along with rurality, reduces the benefit of higher education degrees. accuracy. However, classification models were unable to properly identify cases where individuals were truly in poverty. The highest sensitivity value came from the Linear Discriminant Analysis model, with 43.5%. Disability prevalence, low education levels, and weak labor-force participation consistently emerged as the most influential predictors. All modeling, validation, and hyperparameter tuning were executed in a fully reproducible RMarkdown Of the classification models, the Logistic Regression model had the highest accuracy score, with 91.3%

5 Results and Findings

The results from the modeling procedures closely aligned with the empirical patterns identified during exploratory analysis, confirming that disability prevalence and educational attainment jointly shape poverty outcomes across Illinois. The linear regression models indicated that higher education degrees, living in more populated areas, and increased hours of work per week all increased an individual’s percent of the poverty line. The highest performing regression model included all predictor variables and an interaction between educational attainment and disability status. This model received an adjusted R^2 value of 0.367, meaning that this model accounts for approximately 36.7% of the variability found in the data. Each increase in level of educational attainment raised the predicted standing of an individual by 23.42 points. However, being disabled brought down the predicted poverty outcome by 12.17 points. In this model, with an increase of 1 hour worked per week, an individual’s percent of the poverty line increases by 1.83%. Other predictors that reduced the predicted poverty outcome included having Medicaid, most non-white race indicators, and the number of tracts within a PUMA (which is a measure of rurality). This confirms the earlier hypothesis that higher education attainment helps raise individuals further out of poverty, but the reliance on hours worked per week also indicates that disability does impact poverty outcomes. In regression models with interactions containing disability status, a clear pattern of disability status reducing predicted poverty outcomes emerged. When looking at only educational attainment and disability status, increasing levels of education increased the predicted percent of the poverty line. Holding an associates or bachelors degree provided a 139 point increase when compared to a person with only a grade school education level. However, being disabled reduced this perceived increase by 58.9 points. The

186