It was May 27, 2017—another late night for Brian Sweis. The 26-year-old MD/PhD neuroscience student had been running lines of code and analyses on tens of thousands of rows of data, dating back to experiments conducted in 2012. The goal: to better understand “sunk cost,” the idea that the more you invest in something, the harder it becomes to abandon it, even in cases when it is in your best interest to do so. A long-term relationship gone awry, for example, is harder to leave than a short-term one. Even when inadvertently choosing a longer line at the grocery store, people struggle to back out of the decision and move into a shorter one.

Sweis and his colleagues at the University of Minnesota were specifically interested in how sunk cost impacts decision-making—not just in humans, but in rats and mice, too. Alex Kacelnik, a behavioral ecologist at the University of Oxford who was not involved in the study, puts it this way: “If you find that one species makes a systematically bad choice, you can argue it’s an accident of history. But if you find a very different species does the same thing—then, come on, there must be some reason why evolution doesn’t eliminate this form of behavior.”

Sweis and his team decided to investigate the phenomenon in different species. Some of their most recent findings, published Thursday in Science, show rats, mice and humans all exhibit sunk cost behavior. Their analyses also suggest both rodents and humans have separate decision-making processes that are not all susceptible to sunk cost. The time one spends making a decision, whether or not to leave that relationship or grocery line, is not considered part of the sunk cost; only time invested after the decision is made is taken into account. When Sweis made this realization, he e-mailed his advisor late that night in May, calling the findings “flipping huge.”

But it all began several years ago, when neuroscience graduate student Adam Steiner at Minnesota devised an experiment to see how rats make decisions. Dubbed “Restaurant Row” by the researchers, the experiment presented rats with four meal options with different flavors. The rats scurried around the food-court-style maze and chose whether to wait a certain amount of time to get a specific food or continue onward to another “restaurant” with a shorter wait time. Once a rat had made a decision, it was given the option to change its mind and leave the area if it no longer felt the wait was worthwhile.

As the team considered how best to analyze their results and test similar decisions in humans, studies on sunk cost landed on the front pages of some of the field’s most prestigious journals. Some found rodents are susceptible to this phenomenon; others found they are not. Research in other species, including in birds, also yielded mixed results. For the Minnesota team , the key was not only to test sunk cost in rodents but to gauge how these decisions played out in humans as well. “The goal of the field is ultimately to improve human behavior, so being able to translate rodent findings across to humans is sort of a gold standard,” says Shelly Flagel, a behavioral neuroscientist at the University of Michigan who was not involved in the study.

Three years, at least four failed study designs and countless hours of insignificant analyses later the team found what it had been waiting for: “Kitten videos,” says Samantha Abram, a graduate student in psychology at Minnesota and a co-author of the study. Just as rats and mice forage for food, Abram says, humans forage for entertainment and information online. The researchers presented undergraduates with four video options, and watched as each student navigated the selection and decided whether to wait longer for a video they preferred. Some videos took longer to load than others, and the students had the chance to leave the video- loading page if after their initial choice they decided the wait was not worth the reward. On average, people hesitated to leave a video even if it did not match their interest because of the time they had already invested in waiting for the reward.

Taken together, the findings of the experiments suggest rats, mice and humans are all vulnerable to sunk costs. If they made a decision that did not match their preferences, it was a struggle for each individual to reverse course quickly. This raised the question of whether the process is irrational or is, in some way, a purposeful evolutionary advantage. The findings also indicated that sunk cost is not applied to all aspects of a decision. “A lot of explanation for the sunk cost effect has been that people don’t like wasting precious resources,” Sweis says. “But what we found is that no matter how much time was wasted during the initial decision, once the offer was accepted, that initial amount invested had no effect on your likelihood to stay committed once waiting.”

Flagel and others find the results exciting. But they warn against drawing sweeping conclusions from the study. “There are always going to be caveats, especially when drawing parallels between rodent and human behavior,” Flagel says. One of the central limitations of this research is that it doesn’t look at consistent behaviors across humans, mice and rats. Although foraging for online information and food might be related activities, they are distinct—making the translation from rodent to human less certain. There is also the gender component: the mice studied were all male whereas females made up a majority of participants in the human study.

The Minnesota researchers remain wary as well. “A study is a data point. This is one data point,” says David Redish, a neuroscience professor at the school who co-authored the paper. Redish and his colleagues have, however, published five studies based on the restaurant row and video gallery experiments, some that further demonstrated commonalities between rodents and humans in decision-making. A Nature Neuroscience study in 2014 found evidence rats experience regret; another in PLOS Biology in February showed , as the title says, “Mice learn to avoid regret.”

Other work has corroborated the idea of distinct decision-making processes that are not all vulnerable to sunk cost. Sweis, Redish and another colleague published a study in June in Nature Communications that examined how withdrawal from cocaine and morphine impacts decision-making in mice. The results showed cocaine alters deliberative behavior whereas morphine disrupts the re evaluation of the initial choice. “This actually has implications for other things, like addiction,” Sweis says. If there are two distinct processes, further research could illuminate which one is active in relapse, and how to then modify treatment to address that nuance.

The team is also looking to do more work on the neural activity underlying this behavior. “Can we actually go into the brain and chase these processes down?” As brain-related technology advances, Redish sees a chance to intentionally target—and potentially alter—these decision-making processes. But he says it is up to society to decide whether it is responsible to do so.