If you live in any city long enough, chances are you’ll come across a rat. Few people enjoy the experience, yet many DC residents don’t report every rat they see. So, DC’s 311 service request system only helps identify some of the places where rats live. To help the DC Health Rodent Control team find rats that go unreported, we built a statistical model to predict the locations of rat burrows. While our model wasn’t good at predicting where the Rodent Control team would find unreported rat burrows, it was very good at prioritizing which 311 complaints were most likely to lead to the discovery of a rat burrow. However, the Rodent Control team found that this information would not substantially improve their operations.
Lab staff on a ride along with the Rodent Control Team. (Credit: The Lab @ DC)
Why is this issue important in DC?
Rats are a persistent health risk,1 and rat colonies grow and spread if they are not treated. Identifying unreported rat burrows could help the Rodent Control team limit the growth and spread of rats across the city.
What did we do?
We developed a statistical model to predict the likelihood of finding rat burrows in a given census block in the District. The model relied on data from the District’s 311 service request system and open data related to known rodent living patterns. For example, we included data on building age, size, and condition, population density, and the mix of residential and commercial property.
In October and November 2017, we conducted a field assessment to test our statistical model. The Rodent Control team completed proactive inspections on 100 randomly-selected locations around the city where our model predicted they were likely to find rat burrows. The team collected data on whether or not they found burrows via a digital form. We also compared our predictions to the outcomes of 311 requests for rodent abatement over the same time.
What have we learned?
Our model was good at predicting where rats would be found when the Rodent Control team responded to new service requests. However, inspectors only found rat burrows at about half of the locations where we sent them for proactive inspections, regardless of the model’s predictions.
“One of the big things was making sure that I’m not just creating a predictive model in a vacuum. But we wanted to make sure that this is something that the team could use in the field.”
What comes next?
Because the model was not able to predict where unreported rat burrows were very likely, the Rodent Control team decided not to use it to send their busy inspectors on proactive inspections. Our findings illustrate the importance of testing predictive models in the field.
What happened behind the scenes?
The best part of this project was working closely with the DC Health Rodent Control team. The Lab joined the Rodent Control team on a “rat ride-along” around the city to learn more about how the team finds and treats rodent infestations.