Calorie counting is useful to lose weight, but a fitness tracker could sabotage your efforts.

The devices are very popular. The leading brand, Fitbit, has sold at least 30 million. The company promises that Fitbits “track steps, distance, calories burned, floors climbed, active minutes & hourly activity.” Others – such as PulseOn, Apple Watch, Basis Peak, Samsung Gear S2 and Microsoft Band – promise the same.

A team of Stanford University researchers, however, recently called foul after testing these trackers. Although the devices purport to help users track calories – daily energy expenditure – the number is often markedly incorrect, the scientists said in a paper published in the Journal of Personalized Medicine.

The least accurate, PulseOn, was off by an average of 93 percent. The most accurate, Fitbit Surge, was off by an average of 27 percent, the Guardian reported.

In a statement to NPR, PulseOn said the extremely high level of inaccuracy may “suggest that the authors may not have properly set all the user parameters on the device.”

The consequences of such large margins of error could, of course, be significant.

“People are basing life decisions on the data provided by these devices,” said Euan Ashley, a professor of cardiovascular medicine at Stanford and co-author of the study.

Let’s say some users check their device at the end of a long day and discover that they burned 1,000 calories when they actually only burned 730. They might have an extra dessert or glass of wine since they think they’ve met their goal.

Over time, that adds up. In this scenario, that’s 1,890 extra calories each week the users don’t know about. Each pound of fat consists of 3,500 calories.

“It’s just human nature,” said Tim Church, professor of preventative medicine at Pennington Biomedical Research Center at Louisiana State University.

He wasn’t involved in the study but told NPR: “People are checking these inaccurate counts, and they think they’ve earned a muffin or earned some ice cream and they’re sabotaging their weight-loss program.”

Some margin of error is inevitable, but the scientists said it should be far lower.

“For a lay user, in a nonmedical setting, we want to keep that error under 10 percent,” said Anna Shcherbina, a Stanford graduate student and study co-author, in a news release.

“It’s very hard to train an algorithm that would be accurate across a wide variety of people because energy expenditure is variable based on someone’s fitness level, height and weight, etc.,” Shcherbina said.

The study participants included a “diversity of ages, male and female, and then also we looked at diversity of skin tone, and then size and weight to try and represent the population generally,” Ashley told the Guardian.

The devices proved most accurate for white women who were already fit, meaning “for those for whom it might matter the most, who are trying to lose weight, the error was actually greater,” Ashley told NPR, speculating that perhaps the companies only test the devices on a narrow group of people.

While the energy expenditure numbers were woefully off, Shcherbina said it’s much easier to assess heart rate, which can be measured directly and not through proxy calculations.

Indeed, Ashley said, “The heart rate measurements performed far better than expected.” Most were off by only about 5 percent.

A multiyear study published last September in JAMA split into two groups almost 500 people hoping to lose weight. One used fitness trackers; the other did not.

Those with the trackers lost about 50 percent less weight than those without.

Lead study author John Jakicic, a researcher of health and physical activity at the University of Pittsburgh, at the time said it likely had to do with people incorrectly interpreting the trackers.

“These technologies are focused on physical activity, like taking steps and getting your heart rate up,” Jakicic told NPR.

“People would say, ‘Oh, I exercised a lot today, now I can eat more.’ And they might eat more than they otherwise would have.”

The Stanford study, though, suggests that perhaps the participants were merely working with faulty data.