A series of recent events has raised alarm about the rising levels of political violence in the US. These episodes include the assassination of political activist Charlie Kirk on September 10, 2025; the murder of a Democratic Minnesota state legislator and her husband in June 2025; and two attempts to kill Donald Trump during the 2024 presidential campaign.
Some polls have reported that large numbers of Americans are willing to support the use of force for political purposes, or believe that political violence can sometimes be justified.
My research is in political science and data analysis. I conducted surveys for almost 25 years. Over the past three years, I have studied new techniques that leverage artificial intelligence to conduct and analyze interviews.
My own recent polls, which use AI to ask people why they give their answers, show that the surprisingly high level of support in response to these questions is likely the result of confusion about what these questions are asking, not actual support for political violence.
Lack of communication
Why would several surveys get incorrect answers to this important question? I think the cause is a problem called response error. It means that respondents do not interpret a question the way the researcher thinks they will.
As a result, the answers people provide do not truly reflect what the researcher believes the answers show.
For example, asking whether someone would support the use of force to achieve a political objective raises the question of what the respondent believes “use of force” means in this context. It could be interpreted as violence, but it could also be interpreted as using legal means to “force” someone to do something.
Such response errors were a concern for surveyors since survey research began. They can affect even seemingly simple questions.
What did you mean by that?
To avoid this problem, I used an AI interview system developed by CloudResearch, a well-known survey research company, to ask respondents some of the same questions about political violence from previous surveys. I then used it to ask them what they were thinking when they answered those questions. This process is called a cognitive interview.
I then used AI to review these interviews and categorize them. Two brief reports summarizing this process applied to both surveys are available online. These analyzes were not peer-reviewed and the results should be considered very preliminary.
However, the results clearly demonstrate that respondents interpret these questions in very different ways.
You may be interested in: An accidental post by Trump was a message to Pam Bondi to go after political enemies
Nuances matter in political violence
For example, in my poll, about 33% of Democrats agreed with the statement that “the use of force is justified to remove President Trump from office.” However, when asked why they agreed, more than 57% gave answers like this: “I wasn’t thinking physically, but rather in the sense that he, the president, might need to be ‘fired’ or forced out of office due to rules or laws.” Others imagined future scenarios in which a president illegally seized power in a coup d’état.
Once these different interpretations of the question are taken into account, the AI only coded about 8% of Democrats as supporting the use of force in violent terms under current conditions.
Even here, there was substantial ambiguity, for example this type of response was not unusual: “The ‘use of force’ language was too broad for me. I couldn’t justify killing Trump, for example, but less extreme uses of force were valid in my eyes.”
Similarly, 29% of Republicans agreed that “the use of the military is justified to stop protests against President Trump’s agenda.” However, almost all respondents who agreed with this statement anticipated the National Guard interceding nonviolently to stop violent protests and riots. Only about 2.6% of Republicans gave comments supporting the use of the military against nonviolent protests.
Almost everyone who agreed that the use of the military was justified expressed thoughts like this: “I see the military coming and acting as a police force to stop or prevent demonstrations that turn violent. Peaceful protesters should be allowed to exercise their right to free speech.”
When is political violence justified?
Even questions that explicitly ask about political violence are open to wide interpretation. Take, for example, this question: “Do you think it is ever justified for citizens to resort to violence to achieve political goals?”
The lack of a specific setting or location in this question invites respondents to engage in all kinds of philosophical and historical speculation.
In my survey, almost 15% of respondents said that violence could sometimes be justified. When asked about examples they were thinking of, respondents cited the American Revolution, the anti-Nazi French Resistance, and many other incidents as reasons for their answers. Only about 3% of respondents said they were thinking about US stocks right now.
Furthermore, almost all respondents stated that violence should be the last resort when all other peaceful and legal methods fail.
One respondent illustrated both problems with one sentence: “The (American) settlers first tried petitions and negotiations, but when those efforts failed, they resorted to armed conflict to gain independence.”
A call to understanding
Even these numbers probably overestimate Americans’ support for political violence. I read the interviews, reviewed the AI system’s labeling, and concluded that, if anything, it was overestimating support for violence.
Other factors may also distort reports of public support for political violence. Many surveys are conducted primarily online. One study estimated that between 4 and 7% of respondents in online surveys are “fake respondents” who select arbitrary answers. Another study reported that these respondents dramatically increase positive responses to questions about political violence.
Respondents may also be willing to adopt anonymous attitudes online that they would never say or do in real life. Studies have suggested that “online disinhibition effects” or “survey trolling” can affect survey results.
In summary, my preliminary research suggests that response error is a substantial problem in surveys of political violence.
Americans almost universally condemn the recent political violence they have witnessed. Recent poll results showing the opposite are more likely due to confusion about what the questions are asking than actual support for political violence.
*Ryan Kennedy is Timashev Professor of Data Analytics and Professor of Political Science at The Ohio State University.
This article was originally published on The Conversation/Reuters
Do you use Facebook more? Follow us to always be informed
