Since Shelby County is an extreme outlier in its bankruptcy rate, MLK50: Justice Through Journalism analyzed which demographic characteristics separate high-bankruptcy U.S. ZIP codes from low-bankruptcy ones. Along with analyzing total personal bankruptcies, we explored how demographics impact the type of bankruptcy filed — Chapter 7 or Chapter 13.

To do this, we used 2022 bankruptcy data from the Federal Judicial Center website and 2021 five-year American Community Survey data from the Census Bureau. The bankruptcy data includes one row for every bankruptcy filed during 2022. Columns included the type of bankruptcy and the ZIP code connected with each debtor. The Census Data included the following data for each ZIP code:

  • The Labor Force Participation Rate (the percentage of adults who are either working or looking for work)
  • The Unemployment Rate (the percentage of the labor force that isn’t working)
  • The Homeownership Rate (the percentage of households that own their residence)
  • The percentage of residents with bachelor’s degrees
  • The percentage of residents who are Black
  • The population in each ZIP
  • The median income in each ZIP

We chose most of these factors because they’re important economic characteristics. We chose the percentage of Black residents because prior research has shown that African Americans are more likely to file Chapter 13 than Chapter 7.

Using machine learning techniques, we explored the importance of different demographic variables. We primarily fit linear models to the data, but we also used the Random Forest algorithm. 

In our linear models for the total personal bankruptcy rate and the Chapter 13 bankruptcy rate, race, homeownership and bachelor’s degrees proved to be the strongest predictors. For the percentage of bankruptcies that are Chapter 13, college education lost most of its predictive power, and both the unemployment rate and the labor force participation rate became more important.

Here are the full results from our model for the total personal bankruptcy rate: 

Termestimatestd.errorstatisticp.value
<chr><dbl><dbl><dbl><dbl>
(Intercept)1.570.12312.81.573- 37
BlackPercent2.380.076431.13.443-205
LabForceParticipation-o.615o.153-4.035.70e- 5
HomeOwnRate1.010.093610.85.86e- 27
UnRate0.01900.3520.05419.57e- 1
MedIncome-0.03680.00781-4.712.54e- 6
BachPercent-1.440.123-11.71.26e- 31

The Random Forest algorithm determines the predictive power of each variable. While the linear models consistently saw large effects caused by similar variables, the Random Forest importance rankings were much different from each other, depending on whether total personal bankruptcies, Chapter 13 bankruptcies or percentage of Chapter 13 bankruptcies were being predicted. For total bankruptcies, it listed the bachelor’s degree and race variables as most important and homeownership rate as least important. For Chapter 13 bankruptcies, it listed the labor force participation rate as most important and race as least important. For the percentage of bankruptcies filed as Chapter 13, homeownership and race proved most important. 

Here are its importance rankings for the percentage of bankruptcies filed as Chapter 13:

rf variable importance Overall

HomeOwnRate            100.00

BlackPercent            95.78

UnRate                  29.06

BachPercent             28.02

LabForceParticipation    3.44

MedIncome                0.00

Both the linear and Random Forest models were able to predict bankruptcy rates in our testing data set with low root mean square errors.