President-elect Donald Trump built his campaign on promises to put a wall on the Mexican border, “utterly destroy” the Islamic State, and accelerate economic growth to heights never before seen outside of wartime. Days before he won the presidency, Trump told his supporters he would give them “every dream you ever dreamed for your country.”

Trump’s penchant for sweeping promises — and the likelihood that he may have trouble keeping them — has Republicans concerned about what would happen if he doesn’t or can’t follow through. “If we’re given the White House and both houses of Congress and we don’t deliver,” Texas Sen. Ted Cruz said recently, “I think there will be pitchforks and torches in the streets.”

Although Cruz’s vision of a violent uprising may be an exaggeration, my research suggests that Trump would indeed face a backlash if he fails to deliver on key promises.

Losses outweigh gains in the human mind. What does that mean for politics?

That conclusion rests on one of the most robust theories of modern psychology, prospect theory. Prospect theory argues that in our minds, perceived losses outweigh perceived gains in ways that profoundly affect our decision-making.

In a political context, this means that when the president surprises you by doing something you like, you’re happy about it. But that happiness is not nearly as powerful as the disappointment — or even sadness or anger — that you experience when the president does something you hate. One implication is that the backlash a president faces for breaking a promise to his supporters may be much stronger than whatever positive reactions come from voters who are pleasantly surprised by his decision not to pursue that campaign pledge.

For Trump, a shift away from some of the radical positions he has staked out may in fact please even a majority of Americans. But any positive reaction will likely be muted, while the disappointment of his original supporters will be amplified. He could find himself losing some of his supporters without picking up the same number from the other side — which could leave him even more unpopular than he already is.

Let’s take Medicare, for example. Trump highlighted his deviation from Republican orthodoxy by campaigning against cuts and privatization. But his appointment of Rep. Tom Price (R-Ga.), a privatization champion, as Health and Human Services secretary suggests he may be open to an overhaul.

If Trump embraced privatization or another significant reform, he could risk angering people who voted for him and who like Medicare as it is. Assuming that most Americans expect Trump to keep his promise not to fundamentally alter the program, those who would approve (certainly nothing close to a supermajority) would likely discount it as a gain, while opponents would view it as a loss. The backlash could be stronger than the enthusiasm among backers of privatization.

Here’s how I did my research

Survey data from the American National Election Studies (ANES) help highlight this dynamic. Over various periods in the ’70s, ’80s, and ’90s, the ANES repeatedly asked the same people where they thought the president and presidential candidates stood on a variety of issues. These people were also repeatedly asked where they themselves stood on these issues, and how they rated the president or candidate on a feeling thermometer.

I examined cases in which a president is perceived as shifting from a slightly liberal position to a moderately conservative one on the issue of government services, analogous to a potential Trump shift on Medicare. In those cases, liberals’ negative reactions are 77 percent stronger on average than conservatives’ positive reactions; if liberals and conservatives were equal in number and they all perceived such a shift, Trump’s approval rating would drop, by between 3.6 and 9.9 points as I have modeled it. The only way a shift on Medicare would not harm Trump’s approval rating is if the public favored his policy by at least a 60-40 split.

A similar story is told by the data on most other issues, regardless of the direction of the shift and from where the president is moving. A shift of moderate magnitude on defense spending costs the president 7.4 points on average when the public is split on the issue. On welfare policy, the average hit to a president’s approval is 9.2 points.

These are big league effects for just one policy change. The consequences of disappointing a president’s supporters can be massive, while the positive reactions of voters he might be hope to win over sometimes fail to register at all. These “expectation costs” are a result of having led the electorate to expect certain policies that then shape how they evaluate future outcomes.

It is important to note that these effects cannot be explained by simultaneous changes in the voter’s opinion of the president due to other factors, nor are they altered by media exposure or political knowledge. So long as the individual has a consistent opinion on the issue and does perceive the president as being inconsistent on it, this costly dynamic appears inescapable.

Trump’s supporters take him seriously but not literally. But on which promises?

With Trump, one question is what issues are most likely to generate disappointment among his supporters. This is especially relevant since many of the people who voted for him seem to take him seriously, but not literally. Do they really expect him to keep any of his promises? And if so, which ones?

The key factor seems to be whether the promise relates to means or to ends. If the president merely changes how he will go about securing some outcome, then people will not care much one way or the other (unless, of course, they really care about pursuing that outcome in a certain way).

So far, Trump’s reversals have largely taken this change-of-methods form — just recently, after all, he claimed that, whatever reforms he eventually decides on, the outcome will be “insurance for everybody” (and, indeed, reversals on the issue of government insurance do not seem to generate measurable expectation costs, given the available data).

The reaction to Trump’s reversals thus far suggests that, much to the chagrin of orthodox partisans on both sides of the aisle, most voters do not care much about specific policy tools like deregulation or building physical walls against immigration. Expectation cost theory strongly suggests, though, that Trump’s promises do matter, because they generate strong expectations regarding certain outcomes.

If Trump fails to deliver on them one way or another, then he will pay an additional and substantial cost for having made them.

David Hunter Walsh is a PhD candidate in political science at Rutgers University, and a veteran of the U.S. Marine Corps.