Airbnb guests staying in predominantly white neighborhoods are more likely to shop around at local businesses like restaurants than guests staying in neighborhoods that are mostly black or Hispanic, according to a new study from researchers at Purdue University.

People who use the home-renting platform to stay in a big city, like New York, for example, tend to eat at local neighborhood restaurants close to where they’re staying. But this effect doesn’t hold true when more than half of a neighborhood’s residents were black or Hispanic, according to the study’s authors.

“Airbnb has made repeated claims that it helps the local economy in black neighborhoods, especially in New York City,” said Mohammad Rahman, a Purdue University professor and co-author of the study, in a statement given to The Washington Post.

His team pulled data from the Bureau of Labor Statistics, the Census, and Airbnb for 42 neighborhoods across New York City — the most visited and active Airbnb city in the US — along with 10 years’ worth of Yelp reviews for 34,000 of the city’s restaurants. The goal was to see how trends in home sharing stacked against trends with local restaurant employment. They found that the promised economic spillover didn’t hold up across racial lines.

Neighborhoods with a booming population of Airbnb guests typically did see a growth in people working for local restaurants, along with a surge in the share of Yelp reviews, which was a measure the researchers used to reconfirm their employment findings. But these findings didn’t carry over into the neighborhoods that were predominantly populated by people of color, even if Airbnb rental numbers were just as high.

The reason, as Rahman told The Washington Post, might be because an “uncomfortable reality” is that some visitors might be less inclined to walk around and check out the local businesses in these neighborhoods, even though they might be drawn to how affordable the housing is. Instead, Rahman added, they might just use the Airbnb as a place to sleep at the end of the day, while ride-sharing services like Uber and Lyft allow them at travel to other neighborhoods and avoid exploring the one in which they’re staying. Though this first study was done only in New York, they subsequently expanded it and found similar patterns in Austin, Chicago, Portland, and San Francisco.

These findings undermine Airbnb’s long-standing reputation as a platform that gives an economic boost to local communities. According to the website, Airbnb guests “stay longer and spend more” in the “diverse” neighborhoods they visit, more than twice as much as a typical visitor. The company also claims that its NYC-based guests spend more than 30 percent of their money in the neighborhoods they stay in, and 95 percent of their hosts recommend local businesses.

This study (which is still a working paper) is far from definitive. There are other factors aside from proximity that influence what restaurant a person decides to visit, like the type of cuisine a person is in the mood for or what they might be accustomed to in their home city, state, or country. And even though Rahman used Yelp reviews to back up existing government data, the reviews are still limited. Not everyone who visits a restaurant leaves a Yelp review, and the study even pointed out that just eight neighborhoods of the 42 sampled counted for the overwhelming majority of the Yelp visitor reviews in 2015.

Cutting the study off at 2015 is another point of contention the company has with the findings. Airbnb’s seen swaths of new users across Queens and the Bronx over the past three years, it says. It has also taken steps to foster relationships with local businesses, including a recent partnership with the Queens Chamber of Commerce that brought out hosts from the Jamaica area to meet with owners of neighboring shops and restaurants.

“Airbnb undoubtedly boosts local businesses — Airbnb generated $2.8 billion in economic activity in New York City in 2016”, said company representative Nick Papas in a statement. “But using a subjective and voluntary input like Yelp reviews to draw conclusions in what purports to be a rigorous analysis is wrong.” As a result, he said, the study is “deeply flawed.”