Minutes into our interview with one of the most experienced managers at the New York City Department of Buildings, he confessed: “To put it bluntly, it’s just too complicated.” The manager was referring to the DoB’s process to collect data about accidents on construction sites.
We, a team of five graduate students at Harvard Kennedy School, are working with the DoB on a research project to determine how the department can reduce construction accidents.
Throughout the previous week, our team had researched how accident and safety data moves through the DoB. We discovered that they have an extensive accident database, but the information in that database is neither complete nor useful.
We called twelve DoB managers at the department to find out what changes they would like to see in the database to make it useful for them.
Reflecting on decades of experience, there is consensus among most senior DoB managers: they would love to see leading indicators of high-risk types of construction. For example, is a construction site more dangerous in Queens than in Manhattan? Are construction sites above a height of ten stories more hazardous than sites below ten stories? If they had those trends, most managers said, they could purposely send their inspectors to those higher-risk types of constructions. But right now, DoB managers cannot see any trends in the data, because the database is useless for that purpose.
Three key problems come up in our conversations again and again:
- The accident database is often described as “confusing”, “cryptic”, “messy”, and “unclear”.
- Most people in the DoB are focused on enforcing the building code and issuing permits, not on preventing accidents or increasing worker safety.
- Over the years, the majority of accident information has been captured and shared internally at DoB via email, despite the existence of a accident database. Lacking a useful database, DoB managers have no basis to inform them where best to send their limited number of inspectors.
Next Steps: Brainstorming and Prototyping
One option our team is now considering is to build a predictive tool on incomplete and messy data. A more productive route, however, might be to make the accident database more usable for DoB. Our next step is to explore a technical solution that will enable DoB to bring order to the chaos of its accident database.
Anthony Arendt, Dan Bacon, Howaida Kamel, Kirsten Rulf and Daniel Wagner