A new working paper by the University of Chicago Booth School of Business found an increase of 2 to 3 percent in fatal accidents on the road since the introduction of ride-hailing services like Uber and Lyft.

The study, still in draft form, sampled cities nationwide with populations of at least 10,000 in 2010. It used data from Uber and Lyft, as well as the National Highway Traffic Safety Administration’s Fatal Accident Reporting System.

The paper’s authors found accidents had decreased in the United States from 1985 to 2010, when the first ride-hailing services launched. Since then, however, the trend has “reversed course and increased.”

The authors also say their evidence shows ride-hailing has caused increases in new car registrations, vehicle miles traveled on main roads, excess gas consumption and traffic congestion.

“Rideshare companies often subsidize drivers to stay on the road even when utilization is low, to ensure that supply is quickly available,” wrote the authors — John M. Barrios of the University of Chicago and Yael V. Hochberg and Livia Hanyi Yi of Rice University.

Both Lyft and Uber took issue with the study.

“Numerous studies have shown that rideshare has reduced DUIs, provided safe transportation in areas underserved by other options, and dramatically improved mobility in cities,” a Lyft spokesperson said in a statement. “The safety and protection of everyone on the road is our top priority.”

“Our data science team reviewed this report and found it to be flawed,” Uber said in a statement. “Uber has contributed to safety in many ways, and we take our responsibility to help keep people safe seriously.”

A study released earlier this year by the Chicago-based Moll Law Group examined the number of arrests for driving under the influence starting in 2009, before Uber’s launch, in 10 cities, including Chicago. That study found an 18 percent decrease in Chicago DUIs from 2009 to 2016, according to Chicago Police Department data.

The authors of the Booth School study could not be reached for comment.