01
Cleaned and joined data
Averaged and combined credit score, balance, and delinquency CSVs into county-level analysis tables.
Data analysis case study
County Level Access to Credit in the United States
An exploratory data science project using Opportunity Insights data to study how credit scores, credit card balances, and delinquency rates vary across county-race subgroups.
Scroll for project details
3
Opportunity Insights datasets joined for county-level analysis
3,225+
county identifiers represented in the source CSVs
6
race group labels used to compare subgroup-level patterns
92%
accuracy from the harder delinquency classification model
Overview
County-level credit outcomes, modeled across geography and race.
The project asks whether county-level credit outcomes reveal meaningful patterns across geography, income background, and race.
Methods
From raw Opportunity Insights files to interpretable models.
02
Summarized distributions
Compared means, medians, ranges, and IQRs to understand spread and skew across counties.
03
Mapped associations
Used scatter plots and correlations to evaluate relationships across score, balance, and delinquency.
04
Built classifiers
Created decision tree models to classify high versus low delinquency and tested a harder predictor set.
Evidence
Visual evidence linking credit scores, delinquency, balances, and subgroup patterns.
Credit score and delinquency had the strongest relationship.
The main visual shows a clear downward slope: higher average credit score is associated with lower average delinquency rate.
Balance and score moved together.
The association was positive, suggesting higher credit balances often appeared alongside higher average credit scores.
Balance and delinquency were weaker but still patterned.
The relationship was less linear, but the direction helped motivate the later classification model.
The first tree was intentionally simple.
This baseline model split directly on delinquency rate, making it useful for explaining the mechanics of a decision tree but less useful as a real predictor.
The second tree made the task more realistic.
After removing raw delinquency from the inputs, the model had to infer high or low delinquency from credit score, credit balance, and county features.
Delinquency was tightly linked to score.
Across county-race subgroups, higher delinquency was strongly associated with lower credit scores.
Race added context to the scatter plots.
Subgroup coloring revealed visible clustering patterns that could motivate deeper work on systemic credit access disparities.
The model still worked without the target leak.
After removing raw delinquency from the predictors, the classifier reached 92% accuracy using score, balance, and county features.
Artifacts
Source materials and reproducible analysis artifacts.
Reflection
What this project taught me about data, credit access, and model design.
The analysis showed how financial outcomes can look different once data is grouped by place and race. It also reinforced an important modeling lesson: a high accuracy score only matters when the features are chosen carefully and the model is not simply learning the answer directly.