As the Republican and Democratic national conventions draw near, Donald Trump and Hillary Clinton find themselves among the most disliked presidential candidates in U.S. history. Americans have registered their negative views for the candidates in poll after poll, and their dissatisfaction runs deep. Why is the dislike for the leading presidential candidates so widespread? And is it possible to change voters’ opinions?

While it may be tempting to point to the polarization of American politics as the answer, we think that more fundamental elements of human judgment are at work. Each candidate has a well-known past composed of both admirable actions and bad behavior. Our research suggests that people view these “mixed bags” in a systematic way: A bad reputation comes from doing bad deeds. A good reputation, in contrast, requires not only doing good deeds, but also not doing bad deeds. People tend to judge immoral behaviors harshly and judge moral behaviors with skepticism, according to our study recently published in the journal Social Cognition.

In a series of controlled experiments, we asked participants, mostly average Americans in their 30s and 40s, to read about and evaluate the actions of fictional people who behaved in either a moral or immoral way. We created the characters and the stories to reflect “small-dose” actions in everyday life.

As for Trump and Clinton, these findings help explain their stubbornly high negative ratings. Both candidates have impressive accomplishments: Clinton achieved top levels of public service as first lady, U.S. senator and secretary of state, and Trump has been touted as a modern business tycoon. But no matter how many personal selling points they may offer to their constituents, our findings (along with many others in psychological science) suggest that the human mind gives preferential weight to the bad things. After making comments insulting Muslims, Latinos and women, Trump has been unable to fend off charges of racisms and sexism. Clinton is dogged by voter mistrust stoked by her handling of classified State Department information on a private email server, the Benghazi hearings, and the long-ago Whitewater scandal.

Of course, people may think that some of these judgments are fairer than others. But regardless, our findings suggest that both presidential candidates will suffer from their past bad acts no matter how much good they advertise in the present, and therefore face an uphill battle in trying to change their reputations with voters before November.

It also explains why social stigmas are hard to shed. People who exhibit a few instances of socially unacceptable behavior are quickly labeled as deviant and have to commit disproportionately many more acceptable behaviors to restore their reputation.

In one experiment from our study, participants read about Barbara, who works in an office and behaves in a neutral manner—neither overly nice nor overly nasty. By the flip of a coin, some participants read that Barbara began engaging in small nasty behaviors, such as gossiping about others or cutting in line. Other participants read about a different Barbara, who began engaging in equivalently small, nice behaviors, such as holding the door for other people and giving compliments.

Participants could track Barbara for as long as they wanted, until they had “seen enough” to determine a meaningful change in her moral character. The results revealed an asymmetry in the point at which they tipped: people required about four unethical actions to decide that Barbara had appreciably changed for the worse, but about six equivalently ethical actions to decide that Barbara appreciably changed for the better.

This asymmetry proved consistent across a variety of situations, including generous and penny-pinching tipping in restaurants, and giving and taking money from strangers.

In another experiment, participants punished a hypothetical high school student after relatively few instances of bad behavior, but required markedly more instances of good behavior to grant the student an equivalent reward.

So understanding how people perceive moral tipping points is important because it illuminates how we as a society create different thresholds for rewarding versus punishing others. If it takes more evidence to accept a change for the better in someone’s character than it requires to believe someone has changed for the worse, then equivalent behaviors will warrant punishment while not qualifying for reward.

In everyday life, this stringent threshold can lead us to refuse to give other people a second chance after a negative first impression, and explains why a well-evidenced positive impression can so readily evaporate with a single stumble.