The Lab @ DC
Stronger Evidence for a Stronger DC

Can we predict housing code violations?

Can we predict housing code violations?

Project Summary
Everyone wants a safe home. Housing codes are rules that are meant to ensure all homes in DC are safe to live in, but there are thousands of rental properties in the District and limited inspectors in the Department of Consumer and Regulatory Affairs (DCRA). Many properties are safe, so if we knew which rental properties were most likely to have a life-threatening violation, then inspectors could focus their efforts on those buildings. We developed a statistical model to predict how likely it was that a rental property would violate a housing code and compared our predictions to the results of actual inspections of building exteriors. Our model was not good at predicting where the DCRA inspectors would find housing code violations on inspections, but in creating the model, we also helped the agency streamline their process for selecting random properties to inspect. Based on the results of these inspections, DCRA is not currently using the model, but they are using the streamlined process we developed.

Why is this issue important in DC?
Violating housing codes can hurt people’s health or even risk their lives. Along with responding to tips and complaints from tenants, DCRA chooses random rental properties to inspect for housing code violations that could affect people’s safety. Identifying buildings where housing code violations are most likely would help DCRA find and address code violations before they hurt residents.

What did we do?
We developed a statistical model to predict the likelihood of finding a housing code violation. We based the predictions on a combination of census information, tax records, and records of permits and previous violations.

In 2018, DCRA inspected the exteriors of 260 rental properties where the model predicted they were likely to find a housing code violation. DCRA could only inspect the exteriors because interior inspections need to be scheduled with the landlord. We compared the results of these inspections to the model’s predictions.

What have we learned?
DCRA only found housing code violations on 24% of external inspections. During random inspections, DCRA finds housing code violations 50% of the time. It is strange for a model to predict so poorly. There are several possible reasons for the poor performance, including a change in data entry procedures or that most violations cannot be detected from the exterior, but we cannot know for sure with the data we have.

What comes next?
We recommended that DCRA stop using the model for inspections of building exteriors for the time being, but we hope to revisit this issue and our perplexing results in the future.

What happened behind the scenes?
Before this project, a DCRA employee chose the buildings to inspect randomly, by manually selecting them from a spreadsheet of thousands of properties. As part of this project, The Lab developed a new process that automated the selection. DCRA still uses this automated process to choose properties for its inspections!