Census Data Analysis

Income disparities are easy to describe and hard to interpret. The research question was which predictor, or combination of predictors, has the most significant impact on salary: race, sex, years of education, or age.

This was a five-person team paper. I co-authored it with Chris Loor-Luna, Cory Nguyen, Tam Khong, and Jeffrey Alvarez, contributing to the data preparation, analysis, visualizations, and written findings.

The team defined salary as a binary outcome (above or at most $50,000), one-hot encoded race categories, and treated years of education and age as continuous predictors. We used chi-square tests for categorical relationships, one-way ANOVA to compare means across groups, and Pearson correlation for continuous variables, supported by three visualizations.

• An age distribution histogram with a kernel density estimate.
• The share of people earning more than $50,000 at each education level.
• A correlation heatmap of education, age, sex, and race against income class.

Variables were defined first, then each relationship was tested with the method that fits its data type.

1. 11996 Census subset
2. 2Variable definitions + encoding
3. 3Chi-square · ANOVA · Pearson correlation
4. 4Visualizations
5. 5Written findings + limitations

Years of education showed the strongest positive relationship with earning more than $50,000, followed by age, consistent with the hypothesis that education is a key determinant of earning potential. Sex and race showed weaker relationships, with disparities between categories.

The data is a 1996 Census subset with a binary income threshold, so the results describe associations in that sample rather than current wages or causal effects. The paper notes non-response bias and the binary classification as limitations.

A clear statistical argument depends on explaining variables and uncertainty, not just ranking predictors.