What drives job creation after SBA 7(a) loans?

QM 505 Data-Driven Decision Making · Walsh College · Spring 2025

What drives job creation after SBA 7(a) loans?

Does loan size, business type or borrower profile predict how many jobs a small business creates after an SBA 7(a) loan? She tested it on 5,000 FY2023 loan records, and the data says these attributes predict almost nothing.

Pythonpandasscikit-learnLinear regressionDecision treePivot analysis

Live demo

All 5,000 rows are embedded in this page. Pivots, regressions and holdout tests run in your browser.

1 · Pivot the loans

2 · Loan amount vs jobs

Points are nudged up or down by up to 0.3 jobs so overlapping whole numbers stay visible. The regression uses the exact values.

3 · Can a simple model predict jobs created?

Pick the inputs, fit a linear regression on a training split and score it on loans it has not seen. A holdout R² near zero (or below) means the model does no better than guessing the average.

Holdout R² across 20 random splits with the selected inputs. The highlighted bar is the split shown on the left.

Results

Computed live from all 5,000 loans.

Jobs created by loan size

How many jobs each loan created

How it works

  1. Loaded the 5,000 FY2023 7(a) loan records in pandas and checked them for missing values and duplicates (none of either).
  2. Built pivots of loan size, jobs created and jobs retained by business type, industry, state and borrower characteristics.
  3. Added an efficiency measure, jobs created per $100K lent, to compare groups on output per dollar.
  4. Trained a linear regression and a decision tree on a training split and scored both on held-out loans.
  5. Reported the result plainly: in this data, loan attributes explain almost none of the variation in jobs created (her report: linear R² ≈ −0.01, tree R² ≈ 0.01).

Data

  • SBA_7a_Loans_FY2023.csv:
  • Demographic fields (race, gender, veteran status) are shown for description only. Given the data-quality issues above, group differences here should not be read as evidence about lending fairness.

First 10 rows: