On 24 September 2025, I fielded a survey using CloudResearch’s Engage platform for AI interviewing and Connect platform for participant recruitment. The sample was of 502 respondents in the U.S., census-matched to the U.S. population based on gender, age, race, and ethnicity. I also requested a balanced sample between Republicans, Democrats and Independents to allow for more detailed subgroup analysis
Each respondent was asked two closed-ended questions based on a report published by YouGov on 12 September 2025 (https://today.yougov.com/politics/articles/52960-charlie-kirk-americans-political-violence-poll). The responses to these two questions have been interpreted by politicians as indicating widespread support for political violence, especially among the political left (see e.g. https://www.politico.com/news/2025/09/15/vance-white-house-promise-to-crack-down-on-radical-left-lunatics-00564766).
I will note that, unlike some other YouGov reports, there was not a link to the questionairre that I could find. As such, these questions are constructed to exactly replicate the questions and response options from the report. This means there might be some slight differences in question wording or response sets if the YouGov questionairres differed from what was in the report, but these should not affect the conclusions.
The two questions were:
(Always or usually acceptable, Always or usually unacceptable, Not sure)
(Yes, violence can sometimes be justified; No, violence is never justified; Not sure; Prefer not to say)
After each of these two questions, respondents were asked a series of open-ended questions using Engage’s AI interviewing tool.
For the first question, the AI interviewer was instructed as follows:
“You are an expert on cognitive interviewing. As open-ended questions to understand how the respondent understood the question, why they answered the way they did, and what specific scenarios or stories they were thinking of when they answered the question.
As open-ended follow-up questions that encourage participants to share their emotions and detailed explanations. Probe for specific examples of scenarios or stories they thought of in their response. Please inquire more deeply about the following:
What do they think it means to be “happy” about the death of a public figure?
Why did they find being happy acceptable, unacceptable or they weren’t sure?
When answering this question, what specific scenarios were they thinking of where it would be acceptable or unacceptable?
Were they at all conflicted about their answer or can they think of a time they might have thought differently?
Listen for and probe on: Language and phrasing, Emotion, Tone.
Ask for clarification or more details when responses are vague or general.
As no more than five open-ended questions. Always ask one question at a time!”
For the second question, the AI interviewer was instructed as follows:
“You are an expert on cognitive interviewing. Ask open-ended questions to understand how the respondent understood the question, why they answered the way they did, and what specific scenarios or stories they were thinking of when they answered the question.
Ask open-ended follow-up questions that encourage participants to share their emotions and detailed explanations. Probe for specific examples of scenarios or stories they thought of in their response. Please inquire more deeply about the following:
What were they thinking “resort to violence” meant in the context of this question?
What scenario(s) or situation(s) were they thinking of when they said it could sometimes be justified, never be justified, said they weren’t sure, or preferred not to say?
Why did they choose the response they did?
Are there any conditions under which they could imagine changing their answer?
Listen and probe on: Language or phrasing, Emotion, Tone.
Ask for clarification or more details when responses are vague or general.
As no more than five open-ended questions. Always only ask one question at a time!”
This section reports the basic attributes of the sample. As noted above, the sample was matched to U.S. census numbers based on gender, age, race, and ethnicity. Because of this, there is some divergence in other areas. Nonetheless, there is still substantial diversity in other, non-matched areas like education, and income. To ensure more balanced partisan ID, we requested a balanced panel between Republicans, Democrats, and Independents. The tables below report these basic demographic characteristics of the sample.
## # A tibble: 1 × 5
## mean_age stdev_age median_age max_age min_age
## <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 45.8 16.0 46 83 18## sex n percent
## Female 253 0.503984064
## Male 248 0.494023904
## Prefer not to say 1 0.001992032## race n percent
## American Indian or Alaska Native 5 0.009960159
## An ethnicity not listed here 10 0.019920319
## Asian Indian 2 0.003984064
## Black or African American 70 0.139442231
## Chinese 11 0.021912351
## Filipino 5 0.009960159
## Japanese 2 0.003984064
## Korean 2 0.003984064
## Samoan 1 0.001992032
## Vietnamese 2 0.003984064
## White 392 0.780876494## education
## Associate degree (for example: AA, AS)
## Bachelor's degree (for example: BA, AB, BS)
## Doctorate degree (for example: PhD, EdD)
## High school graduate - high school diploma or the equivalent (for example: GED)
## Less than a high school diploma
## Master's degree (for example: MA, MS, MEng, MEd, MSW, MBA)
## Prefer not to say
## Professional degree (for example: MD, DDS, DVM, LLB, JD)
## Some college, but no degree
## n percent
## 59 0.117529880
## 184 0.366533865
## 7 0.013944223
## 63 0.125498008
## 5 0.009960159
## 57 0.113545817
## 3 0.005976096
## 7 0.013944223
## 117 0.233067729## pid n percent
## Democrat 165 0.32868526
## Independent 129 0.25697211
## Prefer not to say 22 0.04382470
## Republican 166 0.33067729
## Something else 20 0.03984064## employment n percent valid_percent
## Business Owner 26 0.05179283 0.05349794
## Full-time 239 0.47609562 0.49176955
## Not in paid work (e.g., homemaker, disabled) 28 0.05577689 0.05761317
## Part-time 78 0.15537849 0.16049383
## Prefer not to say 8 0.01593625 0.01646091
## Retired 56 0.11155378 0.11522634
## Student 16 0.03187251 0.03292181
## Unemployed 35 0.06972112 0.07201646
## <NA> 16 0.03187251 NA## income n percent valid_percent
## $10,000-$19,999 26 0.051792829 0.053388090
## $100,000-$124,999 53 0.105577689 0.108829569
## $125,000-$149,999 30 0.059760956 0.061601643
## $150,000-$174,999 14 0.027888446 0.028747433
## $175,000-$199,999 13 0.025896414 0.026694045
## $20,000-$29,999 39 0.077689243 0.080082136
## $200,000-$224,999 13 0.025896414 0.026694045
## $225,000-$249,999 3 0.005976096 0.006160164
## $250,000 or more 13 0.025896414 0.026694045
## $30,000-$39,999 51 0.101593625 0.104722793
## $40,000-$49,999 40 0.079681275 0.082135524
## $50,000-$59,999 37 0.073705179 0.075975359
## $60,000-$69,999 37 0.073705179 0.075975359
## $70,000-$79,999 45 0.089641434 0.092402464
## $80,000-$89,999 21 0.041832669 0.043121150
## $90,000-$99,999 28 0.055776892 0.057494867
## Less than $10,000 9 0.017928287 0.018480493
## Prefer not to say 15 0.029880478 0.030800821
## <NA> 15 0.029880478 NAThis section reports the basline cross-tabulations between political ID and their responses to the closed-ended questions.
The first table below shows the overall percentages who find being “happy” acceptable or unacceptable. Overall, about 10.7% of our sample said they think it is always or usually acceptable, which compares with 8% in the YouGov sample. About 12% of respondents say they are not sure, compared with 14% in the YouGov sample. And about 77% say that being “happy” is always or usually unacceptable, which is the same as in the YouGov sample.
## happySupport n percent
## Always or usually acceptable 54 0.1075697
## Always or usually unacceptable 387 0.7709163
## Not sure 61 0.1215139The next table shows the relationship between partisanship and belief that it is/is not acceptable to be “happy” about the death of a public figure they oppose. Here there is less of a direct comparison with the YouGov results. Connect measures partisanship using traditional labels of “Democrat”, “Republican”, and “Independent.” YouGov reported their breakdowns using ideology, measured as “Very liberal,” “Liberal,” “Moderate,” “Conservative” and “Very conservative.” Nevertheless, we do see commonality in the breakdowns. About 20% of Democrats say it is acceptable in our sample, while 24% of those who identified as “Very liberal” and 10% of those who identified as “Liberal” did the same in the YouGov sample. About 8.5% of Independents in our sample said they found it acceptable in our sample, while about 7% of “Moderates” found it acceptable in the YouGov sample. In our sample, about 2.4% of Republicans said it would be acceptable, compared with 4% of those who identified as “Conservative” and about 2% of those who were labeled “Very conservative” in the YouGov sample. Despite the differing categories, the marginal breakdowns also generally seem quite similar between YouGov’s sample and our Connect sample.
## happySupport Democrat Republican Independent
## Always or usually acceptable 20.00% (33) 2.41% (4) 8.53% (11)
## Always or usually unacceptable 64.24% (106) 92.17% (153) 78.29% (101)
## Not sure 15.76% (26) 5.42% (9) 13.18% (17)
## Something else Prefer not to say
## 20.00% (4) 9.09% (2)
## 60.00% (12) 68.18% (15)
## 20.00% (4) 22.73% (5)Turning now to the second question, which asks whether political violence is ever justified. The table below shows the marginals for this question in our sample. In our sample, about 14.5% of all respondents saying it could sometimes be justified, 72.5% saying it is never justified, 10.5% saying they were not sure, and 2.4% saying they’d prefer not to say. This compares with 11% saying it could sometimes be justified, 72% saying it could never be justified, 12% not sure, and 6% preferring not to say in the YouGov sample. Here again, we see a pretty close correlation in the topline responses.
dta %>%
mutate(vSupport = factor(vSupport, levels = c("Yes, violence can sometimes be justified",
"No, violence is never justified",
"Not sure",
"Prefer not to say"))) %>%
tabyl(vSupport)## vSupport n percent
## Yes, violence can sometimes be justified 73 0.14541833
## No, violence is never justified 364 0.72509960
## Not sure 53 0.10557769
## Prefer not to say 12 0.02390438Finally, the table below looks at these responses, broken down by political ID. In our sample, about 20.6% of Democrats, 13.8% of Independents, and 8.4% of Republicans agree that violence can sometimes be justified. This compares with 25% of the “Very liberal”, 17% of the “Liberal,” 9% of the “Moderates,” 6% of the “Conservatives,” and about 2% of the “Very conservative” respondents in the YouGov sample. Again, these numbers are not too far off from each other. At the very least, there is nothing to indicate reason for concern that our sample participants are overly different in terms of their baseline responses from those in the YouGov sample.
## vSupport Democrat Republican Independent
## Yes, violence can sometimes be justified 20.61% (34) 8.43% (14) 13.18% (17)
## No, violence is never justified 62.42% (103) 84.34% (140) 74.42% (96)
## Not sure 14.55% (24) 6.02% (10) 10.08% (13)
## Prefer not to say 2.42% (4) 1.20% (2) 2.33% (3)
## Something else Prefer not to say
## 20.00% (4) 18.18% (4)
## 55.00% (11) 63.64% (14)
## 20.00% (4) 9.09% (2)
## 5.00% (1) 9.09% (2)The problem with previous reporting of these results is that there are several aspects of the questions that are ill-defined.
For example, the question asking whether it is acceptable to be “happy about the death of a public figure they oppose” has several vague terms that invite a range of interpretations. To begin with, we should note that saying such feelings are acceptable does not mean they support political violence. It is unclear the degree to which the two might or might not actually correlate. The question also does not specify any context. While many people were quick to suggest that this was directed at Charlie Kirk or President Trump, the question does not make the person who dies explicit. As we will see below, respondents cited a broad range of historical or non-U.S. figures as examples. Third, the question does not define what is meant by “happy.” Specifically, does this refer to a sense of private relief that is felt, perhaps guiltily, or does it refer to celebrating that the person is dead? Finally, does the person seem to place conditions on their answer? Political scientists have long recognized that people are ambivalent on most issues, meaning that they have feelings both ways that might result in changing answers, depending on context.
To look at what people meant by their answers, I asked follow-up questions using an AI interviewing system with the prompt described above.
Once these interviews were completed, I analyzed the responses of those who said they thought being “happy” was “Always or usually acceptable.” I utilized GPT-4.1 on Microsoft Azure to code the interviews in terms of what the respondent understood “happy” to mean, why they thought it might be acceptable to feel this way, what specific scenarios they were thinking of when they answered, and whether they felt conflicted about their answer. I also used Engage’s insight platform to verify the consistency of coding between AI systems and I read each interview to ensure the coding was credible.
To begin this analysis, we note that a significant number of respondents expressed unprompted, explicit condemnation of violence, drawing a clear distinction between how one might feel after someone dies versus approving the conditions that brought about their death. Almost 28% of those who thought being happy was always or usually acceptable explicitly said they opposed political violence. None of the respondents said the acceptability of feeling happiness was related to support of violence. If we only include those who did not explicitly mention opposition or acceptance of political violence, this drops to about 7.6% of the sample. As one respondent stated, “Do I think it’s okay for a person to get murdered? No. But that wasn’t the question.”
## anti_violence n percent
## Expressed opposition to violence 15 0.2777778
## No explicit mention of either 39 0.722222238/502## [1] 0.07569721Moreover, a large number of these respondents express ambivalence about the question, suggesting that happiness in this instance is conditioned on other contextual factors. More than half of respondents expressed suggested that there are serious ethical considerations about being happy, but suggested, on balance, that people should not be punished for their feelings. When those who demonstrate ambivalence on the question are excluded, only about 4.8% of total respondents are unequivocal in their statements that being happy is acceptable.
## conditional n percent
## No 24 0.4444444
## Yes 30 0.555555624/502## [1] 0.04780876Next is the question of what scenarios respondents were considering. Looking through these responses, there is a wide variety of answers. In addition to those who mentioned Charlie Kirk or President Trump, there were a substantial number of respondents who pointed out that people were very happy by the passing of Saddam Hussein, Osama bin Laden, Adolf Hitler, or Joseph Stalin. Henry Kissinger, who passed away from natural causes at 100 years old, was even mentioned by two respondents. Looking at our sample, a little over 56% of respondents said they were thinking of either Charlie Kirk or President Trump. This is only about 5.3% of the total sample.
## include_kirk n percent
## Mentioned Kirk or Trump 27 0.5869565
## Only mentioned others 19 0.4130435## [1] 0.05378486There was also a lack of clarity about what being “happy” entailed. Almost 41% of these respondents were explicit that they were thinking of “happy” more in terms of relief than celebration. As one respondent put it, “Maybe it’s more of a feeling of relief. If an evil person dies, they can do no more harm.” This is an important distinction. Relief suggests a release from anxiety or distress, not necessarily suggesting expression of enjoyment in an event. Less than 5% of total respondents seem to suggest that celebrating the passing of a public figure is acceptable.
## what_happy n percent
## both 16 0.29629630
## celebrating 13 0.24074074
## relieved 20 0.37037037
## unclear 5 0.09259259## [1] 0.04780876Combining the last two categories to find out how many respondents seem to be suggesting it is acceptable to celebrate the death of Charlie Kirk or President Trump, only about 16 total respondents fall into this category. This is only about 1/3 of those who said being happy was acceptable.
## what_happy Mentioned Kirk or Trump Only mentioned others
## both 33.33% (9) 31.58% (6)
## celebrating 22.22% (6) 21.05% (4)
## relieved 37.04% (10) 42.11% (8)
## unclear 7.41% (2) 5.26% (1)## [1] 0.2962963Bringing these numbers back to the full sample, only about 3.2% of the sample seems to be suggesting it is acceptable to celebrate the death of a political figure like Charlie Kirk or President Trump. This is far lower than what has been implied from the YouGov survey initially.
## [1] 0.03187251There is a similar trend when looking at those who say that political violence can sometimes be justified. Reporting on this question seems to suggest that respondents are saying that some of the violence taking place in the U.S. is justified. But the question does not actually state that it is talking about the U.S., nor does it specify a time frame or scenario for consideration. In other words, asking someone whether they think violence to achieve political purposes is ever justified can trigger a wide range of thought processes that are not close to suggesting it is justified in current U.S. politics. Indeed, by asking whether political violence is “ever justified,” the question seems to encourage respondents to brainstorm extreme cases.
Here again, I utilized AI interviews to ask respondents what scenario they were envisioning when they responded to this question, and I utilized AI labeling to classify the responses. And, again, the respondents who expressed conditional support for political violence were, on the whole, not saying it was justified in the current U.S. situation. Almost all of the respondents who gave positive answers to this question were suggesting extreme situations, most of which occurred outside the U.S.. Moreover, the overwhelming majority said it was only to be used as a last resort and many expressed ambivalence about it.
Let’s start by looking at the temporal context of the responses. Here it becomes immediately obvious that most respondents were not thinking about the current U.S. situation in their responses. Only 21 respondents, about 4% of the total sample, seem to be suggesting that political violence could be justified in the current setting. Respondents came up with numerous scenarios, including the American Revolution, the French resistance to the Nazi invasion during World War II, the Pullman Railway strikes, and a number of others. For future scenarios, people talked about military takeovers or a potential authoritarian government. One respondent gave the following example to show when political violence may be justified, “the American Revolution, whereby colonists tried for years to resolve grievances through petitions, assemblies and appeals, but Britain repeatedly dismissed these efforts by enforcing taxes and military control.”
## when n percent
## current 13 0.17808219
## past 21 0.28767123
## future 10 0.13698630
## current and future 6 0.08219178
## past and current 2 0.02739726
## past and future 17 0.23287671
## unclear 4 0.0547945221/502## [1] 0.04183267Moreover, a substantial proportion of respondents gave examples outside of the U.S. context. More than 38% gave examples that were outside of the U.S. context to illustrate when political violence might be justified.
## where n percent
## inside the U.S. 36 0.4931507
## outside the U.S. 19 0.2602740
## both 9 0.1232877
## unclear 9 0.1232877When we combine these two together, only about 15 total respondents were thinking of scenarios in the U.S. during the current time. This comprises less than 3% of the total sample. Put simply, the question asked by YouGov was viewed by a substantial proportion of respondents as more of a philosophical hypothetical rather than a question specifically related to current events in the U.S.
## when inside the U.S. outside the U.S. both unclear
## current 25.00% (9) 15.79% (3) 0.00% (0) 11.11% (1)
## past 33.33% (12) 31.58% (6) 22.22% (2) 11.11% (1)
## future 13.89% (5) 15.79% (3) 0.00% (0) 22.22% (2)
## current and future 13.89% (5) 5.26% (1) 0.00% (0) 0.00% (0)
## past and current 0.00% (0) 5.26% (1) 11.11% (1) 0.00% (0)
## past and future 13.89% (5) 26.32% (5) 66.67% (6) 11.11% (1)
## unclear 0.00% (0) 0.00% (0) 0.00% (0) 44.44% (4)## [1] 0.02988048Beyond this problem, one might reasonably ask the conditions under which someone might say political violence could be justified. The overwhelming response from those who said it could sometimes be justified was that it could only be so as a last resort when facing severe oppression and a lack of other legal avenues. More than 86% made clear in their interviews that violence was to be a last resort, and only 3 respondents (less than 1% of the full sample) had interview responses suggesting it was a tool that could be used before other options were exhausted.
## resort n percent
## last resort 63 0.86301370
## tool to be used 3 0.04109589
## unclear 7 0.09589041Finally, more than 45% of the respondents who thought violence could sometimes be justified demonstrated ambivalence towards this response in their interviews.
## ambivalent n percent
## Yes 33 0.4520548
## No 30 0.4109589
## Unclear 10 0.1369863As with my last report, I would also note that some additional caution is warranted when we are talking about numbers in the single digits with online surveys. Pew Research has suggested that most online survey panels, even those considered “high quality” in terms of respondents passing attention checks, have anywhere from 4% to 7% of respondents they characterize as “bogus respondents” (https://www.pewresearch.org/methods/2020/02/18/assessing-the-risks-to-online-polls-from-bogus-respondents/). These bogus respondents are likely to simply agree with any statement, even statements that contradict each other. For both of these questions, “agree” responses were indicative of greater use of force. In other words, we cannot discount that a significant portion respondents who did respond in a way consistent with the reporting around these numbers are roleplaying what they think they should have said, given that they had already agreed with the prior question. At a minimum, the valence of such important questions should be reversed to see how much of these percentages might stem from such “bogus” responses.
We should also be careful about interpreting statements online completely literally, since there are other reasons someone might respond with support for violence in an anonymous online setting. We have long known that people are willing to say things in anonymous online settings and suggest they would support actions that they would never advocate, support or do in real life - the “online disinhibition effect” (https://www.liebertpub.com/doi/abs/10.1089/1094931041291295?casa_token=4FVXj89kV8MAAAAA%3ANCAUlrxCcQ1FI-XEdGEljLZlUZHmK6H7rSZa2TxJxMWAf2OCt6kxl63GULi8zxEHh4f8AE5Vm2I). So we must be careful not to assume that these are statements of support for violence, rather than emotional expressions of general frustration. Put another way, these could be “speech acts,” where the respondent is using this online format to express frustration rather than to state their support for a real world action if they were given the choice to actually do it.
Finally, we have little to no idea how firm these attitudes are. As John Zaller, among many others, have noted, survey respondents can respond very differently over time, depending on the salience of particular examples. Even responses that seem firmly held during one interview can reverse quickly, based on what examples, scenarios or stories are salient to the respondent when they are taking the survey. Among those who responded positively to use of the military to stop protests against President Trump’s agenda, there was clear indication among almost all respondents that the way they interpreted recent protests, as portrayed in the media sources to which they were exposed, as quite violent affairs. Their responses seem likely to be highly contingent on this media exposure, rather than general statements about support for such actions.
These, however, enter into the realm of speculation and should be read with caution by the reader. For an issue like support for political violence, where inaccurate poll results hold the potential to throw gasoline into an already polarized political fire, such caution seems warranted.
The bottom line is that the YouGov numbers, at least as it has been reported in the media, wildly exaggerate support for violence. While there is plenty of legitimate evidence that political violence is an increasing problem in the United States, support for such actions is not widespread.
I will end by noting a few caveats to these findings. First, while I have reviewed the interviews and generally found the AI-generated labels to be credible (indeed, my evaluation made me think that the AI was still over-estimating support for violence), these results should still be considered highly preliminary. Language is ambiguous and I will be taking additional steps to verify the accuracy of the AI-generated (or human-generated for that matter) labels, including using multiple human coders, multiple runs with different LLMs (though I should note that I did run the labeling procedure five times and the overall percentages remained consistent), and inviting other scholars to review the results. Normally, I would take all of these steps prior to release of the data, but, in this case, the urgency of pointing out that these widely quoted statistics were not nearly as solid as they are being portrayed seemed to take precedence.
Second, I am sure that a critical reader will argue that people are less likely to express their true feelings directly rather than indirectly. This is likely part of the reason why YouGov likely did not get very specific about the scarios they were asking about. This is certainly possible - social desirability bias is a well-noted phenomenon. Nonetheless, the inconsistency I note here between how YouGov’s results have been interpreted in public and what respondents said they meant when responding still stands as prima facie evidence that the public interpretation has been incorrect. Absent any evidence to the contrary, beyond pure conjecture by those doing the interpretation of vague multiple-choice answers, this is the best evidence we have for how people are interpreting these questions.
Finally, I should discuss the expected error of the percentages displayed above. While I could use statistical procedures to place estimates of sampling error on these estimates, there are multiple reasons to take them as guidelines rather than formal estimates of population parameters. The sample used here may not be as carefully drawn as that YouGov selected, though it is difficult to determine this. Also, as with any qualitative research technique, generalizing from semi-structured interviews poses issues. Nonetheless, the percentages are highly informative and, at a minimum, suggest large portions of respondents are not interpreting the questions in the way that they are publicly being presented. There is no reason to assume that these differences in interpretation are any less prominent in YouGov’s panel than in our CloudResearch sample.