As many Uber users know, a driver with a poor rating will have difficulty being assigned rides by the ride-sharing platform, thus harming their ability to earn money, and in turn, maybe even make a living.

Uber has said the ratings system is designed to hold its drivers accountable and protect its passengers, which sounds effective in theory. But a new study from New York University found the value of ratings systems like Uber’s decreases over time because public pressure to give another person a high rating continually pushes the average up and up until it becomes irrelevant.

There’s a reason for this, and you wouldn’t be wrong if you guessed “guilt” for leaving a bad rating — but it’s more complicated than that. After all, people leave bad ratings for things all the time — look at Yelp, or Airbnb, or Amazon. Researchers found the difference in Uber’s case is that you’re not actually rating a product, you’re rating a person — and because of that, ratings on platforms like Uber are highly susceptible to inflation.

The paper, titled “Reputation Inflation,” cites an unnamed online marketplace that utilizes a gig economy, which relies on contract or freelance workers instead of full-time employees. The users of the service — the “employers” — can either leave “public” feedback visible to the “worker,” or “private” ratings and reviews that aren’t viewable by the worker or other users on the platform. The majority of workers, 82%, opted to leave a private score rating the service from one to five.

The study found the average worker was given a public rating of 3.74 when the company was started in 2007. By the end of that year, the score had increased 0.53 stars. Nearly a decade later — in May 2016 — the average shot up to 4.85.

In contrast, employers were more willing to give “unambiguously bad private feedback” once the service introduced the option in April 2013. Researchers found that between June 2014 to May 2016, roughly 15% of employers gave a negative review in private, but during that same time, only 4% of employers gave a public rating of three or less stars.

But then in March 2015, the company combined the private feedback ratings with the public stars in order to create an aggregate feedback rating for the workers. While the workers would not be able to identify exactly which employer gave them feedback, the new overall score would still be viewable on the worker’s profile, thus making the private feedback have a potentially consequential impact. The result? Negative feedback became rare and poor ratings were only given in truly exceptional instances.

That, in essence, is what happened to Uber. While riders may have taken the ratings seriously at first, it soon became common knowledge that poor ratings could put Uber drivers out of work, making the standard choice five stars unless the driver provides a legitimately awful or dangerous experience. People have an easier time rating products, or hotels, or restaurants, but when it comes to literally rating another person, users are not likely to be as genuine. Who can bare that guilt?