Negative advertising has long been a cornerstone of contemporary American politics and has a storied history at the forefront of election campaigns.

The cornerstone of Lyndon Johnson’s 1964 reelection campaign was a tv spot entitled “Daisy,” which implicitly suggested that the election of his opponent Barry Goldwater could result in nuclear war. More recently in 2004, an anti-John Kerry group called the Swift Boat Veterans for Truth put out an ad smearing Kerry for allegedly exaggerating war crimes he witnessed in Vietnam, as well as his own service record. It was so iconic that the term “swiftboating” has since been used as a moniker for sweeping negative ad campaigns.

The 2010s have brought more of the same.

In 2012, both the Obama and Romney camps spent far more on negative messaging than they did on positive ads, according to the Atlantic. Meanwhile, CNN found 92 percent of ads in the final days of the 2016 presidential election had some level of negative messaging. That comes out to over $100 million spent on negative ads in just a week.

Now, a group of researchers have published a study shedding light on why negative advertising is so prevalent in our elections.

Michael Lewis of Emory University, Yanwen Wang of the University of British Columbia, and David A. Schweidel of Georgetown University initially planned to look at using social media as a tool for predicting election results. But as social media rapidly became commonplace in elections, they shifted their focus towards the impact and efficacy of negative advertising, a staple of elections.

“For forever, voters have expressed disgust with the level of negative advertising,” Lewis explained, “but we see a lot of it. So, [the question was] does it actually work?”

According to the data their study produced, yes, it does. But under certain conditions. Looking at correlations between the volume of negative ads and the vote shares achieved by U.S. Senate candidates in 2010 and 2012, the researchers found that “while positive political advertising does not affect two-party vote share, negative political advertising has a significant positive effect on two-party vote shares.” However, they also found that the source of the ads makes a difference in the ads’ efficacy, noting “negative advertising sponsored by PACs is significantly less effective than that sponsored by the candidate or party in affecting two-party vote shares.”

One potential reason for this is that PACs tend to have less trust among the general public than candidates. “While PACs may go to market with unbiased-sounding names,” the study asserts, “these names are likely new to the vast majority of voters,” making viewers less likely to trust that source. This phenomenon is referred to as the “source credibility theory.”

“There’s a certain logic to it that makes sense,” Lewis says of that finding. “When you’ve got these unknown entities with names that are little more than platitudes and feel-good collections of patriotic phrasing, then it makes sense that those messages don’t have a lot of impact versus something that someone puts their name behind.” Examples of these PAC names include “Committee for a New Start in the Right Direction,” “Faith Family Freedom Fund,” and “Americans For a Better Tomorrow, Today”

The source credibility theory is further strengthened by the finding that “positive advertising by candidate or party has a negative impact on two-party vote shares, while positive advertising by PACs has no effects on two-party vote shares.” Both the study and Lewis credited this to voter-perceived “obvious self-interest” by campaigns, while PACs have less skin in the game and are therefore seen as being driven more by ideals and self-gain.

But PACs, like campaigns, are always more likely to go negative. This begs the question of what steps, if any, PACs might take to adapt to these findings and make their negative ads more impactful.

“You may end up seeing more shadow political parties instead of these one-off PACs, whose business is to take in donations and sponsor candidates, to build up that credibility over time,” Lewis posited. “Maybe [they’ll] become these hyper partisan organizations whose mission is to gin up the base.”

Finally, the study looked at the effects of ads on more competitive races. “negative advertising from PACs and super PACs plays a limited role in competitive senatorial races,” the study found, because “heightened media attention results in a more informed and engaged electorate and therefore a greater likelihood that voters utilize more detailed, central route processing.” In other words, the more voters already know about the candidate, the less they need to learn from ads.

Take the most extreme example. In the 2016 Presidential election, Hillary Clinton and Donald Trump were universally known to voters and were the subjects of wall-to-wall coverage on all levels of news. For that reason, according to Lewis, “You could’ve given Hillary Clinton another billion dollars and it wouldn’t have made a difference, because at that point there’s so much media coverage and the candidates are so well known that it’s complete overkill.” This is not necessarily the case in a race like a non-competitive Senate election or a House race. “Advertising is more impactful as you go down the pecking order,” he said.

The takeaway from this study? “The data tells me you go negative,” Lewis concluded, “People say they hate it, but you go negative. That’s the best way to win these elections.”