Moderate voices are the quietest on Duke University’s campus, according to a study that maps self-censorship among undergraduates. The research, conducted by Cory Dowd of Duke University (now with the New York City Mayor’s Office of Operations), alongside Peter Arcidiacono of Johns Hopkins University and Matthew Lilley of the Australian National University, was circulated as a working paper by the National Bureau of Economic Research.
Self-censorship describes when individuals with private opinions withhold or soften what they think in social settings, even without formal rules or penalties. The study notes that campus debates are often shaped by visible expressions rather than private beliefs, and that questions about silence have intensified amid political polarisation and shifting trust in higher education.
“As a graduate student at Duke, I noticed how conversations in the classroom were often dominated by certain types of views to the detriment of our shared learning and understanding,” Dowd told PsyPost. “When I talked to students individually, they would frequently share opinions not brought up in class and complain about not feeling comfortable having difficult conversations with their peers. While this phenomenon was widely discussed, it was poorly understood and arguments too often relied on anecdotes and conjecture.”
“The limitations in the existing literature are what motivated me to research this topic for my Master’s in Public Policy capstone Master’s Project, and to partner with Peter and Matthew to analyze and share the results,” he added. “The goal was to contribute to the conversation with specific quantitative data on political beliefs and self-censorship behavior that could be used to ground conversations. The survey used an innovative methodology to more accurately identify which views were being censored, map detailed Overton Windows and identify specific consequences, all along ideological and demographic populations. The result enables us to make more concrete assessments of behavior and have more practical interpretations of the experience for students.”
Between February 7 and March 3, 2023, the researchers surveyed undergraduates at Duke University in North Carolina, emailing invitations to a random sample of 1,398 students (37 per cent of whom completed the questionnaire) and posting a link through a campus behavioural research centre. Out of 656 total submissions, 498 met the eligibility criteria of being enrolled undergraduates aged 18 and older who answered a minimum number of questions. Every student who completed the survey received a £20 gift card, whether or not their response was included in the final analysis. Using university records on race, sex, and class year, the authors weighted the eligible sample to match Duke’s undergraduate demographics.
Participants evaluated eight contentious topics: 2020 presidential election fraud claims, abortion access, Black Lives Matter, gun control, the role of the U.S. government, puberty blockers and hormones for transgender-identifying youth, whether controversial invited speakers belong on campus, and equality of opportunity versus equity of outcome.
Each topic paired opposing statements on a 0-to-100 scale, standardised so 0 represented the most conservative stance and 100 the most progressive. On each issue, students rated their own belief, estimated the average Duke student’s belief, and marked the lowest and highest scores they would tolerate on an ideal campus, defining their Overton window.
Respondents also rated each topic’s personal importance, their confidence in their stance, and how often they shared their opinion (never, sometimes, usually, or always). Separate items asked how often they shared political and social views in the classroom, elsewhere on campus, off campus with Duke peers, and off campus with non-Duke people. Finally, students selected reasons for holding back from a checklist and could add written comments.
Like peers at other selective universities, Duke undergraduates leaned strongly progressive: among respondents who reported an ideology, about 70 per cent identified left of centre, 20 per cent right of centre, and roughly a tenth in the middle. Average private opinions leaned distinctly progressive on five topics, led by rejecting 2020 election fraud claims, supporting abortion access, and backing Black Lives Matter. Views sat near neutral on puberty blockers for youth (averaging 56) and tilted slightly conservative on allowing any invited campus speaker and favouring equality of opportunity over equity of outcome.
“Support for Black Lives Matter was not synonymous with support for Equity of Outcome over Equality of Opportunity,” Dowd noted. “While the average student expressed strong support for Black Lives Matter, the question about Equity or Equality had much more diverse responses, with a ‘conservative’ average opinion (closer to Equality than Equity on the spectrum).”
When the authors examined who regularly spoke up (usually or always sharing), self-censorship was not simply about being conservative. Across most topics, sharing followed a J-shaped curve: progressive students spoke up the most, strongly conservative students spoke up somewhat less, and moderate or lean-right students spoke up the least. By overall ideology, only 19 per cent of lean-right students and 33 per cent of centrists regularly shared opinions across issues. By comparison, 45 per cent of left-wing students and 48 per cent of right-wing students regularly voiced their views.
“Self-censorship among students is a very common experience across the board in every context and with every audience, but the students who report self-censorship the most don’t necessarily align with common narratives,” Dowd explained. “Typically, it’s the moderate and lean-right voices that choose not to share their opinions in the classroom more often than very liberal or very conservative ones.”
“I was initially surprised that conservatives were nearly as vocal as liberals, although I shouldn’t have been,” he continued. “As a student, I often described a phenomenon that I referred to as the ‘moderate dog whistle.’ Students interested in having more nuanced conversations and sharing moderate opinions would use phrases like ‘I mostly agree with ____’ or ‘I’d like to think about it ____ a little more.’ There’s nothing risky or controversial about these statements, but they draw the interest of curious moderates who also want to explore potential disagreements on topics.”
That quiet middle tracked issue importance and personal certainty. Students at both ideological poles tended to see contentious topics as more important and felt more confident in their beliefs than moderates, who saw less reason to enter debates. Campus free speech was the sole topic where conservative students shared views more often than progressives, matching the greater importance and confidence right-leaning students reported on that issue.
Groups traditionally viewed as marginalised on campus did not self-censor more than majority peers; instead, female, Black, and non-straight students tended to voice opinions more often, especially on topics closely tied to their groups. On Black Lives Matter, 68 per cent of Black students regularly shared their views versus 37 per cent of white students. Women were similarly more likely than men to speak up on abortion (61 per cent versus 37 per cent), and non-straight students were over twice as likely as straight classmates to share views on puberty blockers. Overall, Hispanic and non-Christian students also shared their views more frequently than Asian and Christian peers.
“The levels of self-censorship are very high across the board, but there are also significant differences between groups in terms of causes, circumstances and consequences,” Dowd emphasised.
Comparing settings indicated that right-of-centre silence at Duke was specific to campus rather than a general reluctance to discuss politics. Overall, classrooms were the most inhibited venue, with 34 per cent of students regularly sharing opinions in class versus 54 per cent elsewhere on campus. Only 12 per cent of lean-right students regularly spoke up in Duke classrooms, yet 49 per cent did so off campus with non-Duke people, whereas left-leaning students spoke up slightly less off campus with non-Duke acquaintances than in class.
Left- and right-leaning students also cited different motives for staying quiet. Left-identifying students mainly pointed to internal doubts, including worrying about misstating their position or lacking confidence in their opinion, selected by 48 per cent of left-wing students versus 15 per cent of right-wing students. Right-identifying students emphasised external social costs: 66 per cent cited fear of isolation or exclusion, over triple the 20 per cent rate on the left, with written comments mentioning “cancel culture” and feared grade penalties.
“It’s worth highlighting that survey respondents were able to select ‘Far Left,’ ‘Left,’ ‘Lean Left,’ ‘Center,’ ‘Lean Right,’ ‘Right’ and ‘Far Right,’ but we often combined ‘Far Left’ with ‘Left,’ and ‘Far Right’ with ‘Right,’ partly because there were so few students who reported being on the ‘Far Right,’” Dowd pointed out.
Because progressive students spoke up more often while moderates and conservatives held back, publicly voiced opinions at Duke sat further left than private beliefs on seven of the eight topics. On puberty blockers for youth, the average view among vocal speakers sat 15 points further toward the progressive end of the scale than the overall campus average, and across abortion, Black Lives Matter, gun control, and puberty blockers, vocal students scored 14 to 20 points more progressive than classmates who kept quiet.
Students also misjudged peer opinion on several issues. On campus free speech, where higher scores meant keeping controversial guests from speaking and lower scores meant letting any invited speaker appear, students guessed a peer average of 55 when the actual campus average was 45, overestimating classmate support for speaker restrictions.
To test whether conservative concerns about peer intolerance matched campus attitudes, the authors analysed the width of students’ Overton windows, adjusting for demographics, confidence, and opinion extremity. Holding more progressive views on an issue was associated with a narrower window of acceptable debate, yet attitudes toward campus free speech carried much more weight. Favouring restrictions on controversial speakers was linked to narrower Overton windows on other topics with an effect more than twice the size of a student’s own view on the topic itself, and adjusting for free-speech attitudes cut the direct link between progressive views and narrower windows by roughly a third.
When the researchers modelled sharing frequency against classmates’ Overton windows without including a student’s own stance, peer disapproval worked asymmetrically: falling outside the acceptable range of classmates to one’s right had no statistically significant link to speaking up, whereas falling outside the windows of classmates to one’s left was strongly linked to staying quiet. Including both a student’s own view and left-side peer disapproval in the same pooled model left neither statistically significant on its own, as the two measures closely track each other.
Comparing how often the same student shared opinions across different topics, an approach that holds individual talkativeness constant, made those patterns sharper. Within the same person, a student was less likely to speak up on issues where they perceived the campus average to sit further left, and less likely where their opinion fell outside left-leaning classmates’ Overton windows. Once that left-side measure entered the model, a student’s own ideological position on the topic no longer carried a detectable independent link to speaking up. In the authors’ reading, progressive majorities and narrower left-leaning speech boundaries channel conformity pressure mainly onto moderate and conservative views.
Comparing responses across class years showed little variation between older and younger cohorts on nearly every issue except campus speech, where seniors leaned roughly five points more progressive than non-seniors. “There is a common narrative that institutions in higher education are indoctrinating students with liberal ideas, but most class years shared similar views across the topics,” Dowd said. “There is also an observation that juniors would have been incoming freshmen in the year 2020; they did have slightly more conservative beliefs than the other years, but they also had a smaller class size.”
These findings among undergraduates align with an earlier survey of U.S. psychology professors, which found that faculty frequently held back controversial views out of fear of social ostracism and reputational harm, while younger, female, and more left-leaning professors expressed greater support for suppressing contentious scholarship. That earlier research, however, assessed professors withholding empirical research conclusions rather than undergraduate students self-censoring personal and political opinions.
“The perception is that self-censorship remains a pressing issue on many college campuses, particularly elite campuses such as Duke,” said Christopher J. Ferguson, a psychology professor at Stetson University who was not involved in the study. “This new study fits well with previous observations and suggests that this issue isn’t going away, even as other aspects of American life have adjusted back from the 2020-2023 ‘cancel culture’ period.”
“I think the good side is that some of the evidence here suggests that many students are self-censoring due to internal reasons … lack of confidence or fear of being misunderstood,” Ferguson added. “Though fear of social exclusion remains a concern particularly for conservative students. But these are things that could be worked on within the campus environment. This seems to be consistent with students primarily worrying about the reaction of other students, but this is still something professors could model in terms of a culture of open inquiry.”
There are a few things to keep in mind when interpreting these results. The study was circulated as a working paper and has not yet undergone peer review, and although responses were weighted to match Duke’s demographics, voluntary participation means respondents could still differ from non-participants. The survey also took place at one selective private university in early 2023, where seven in ten of those reporting an ideology identified left of centre, a political balance very different from predominantly conservative or religiously affiliated campuses.
“This data is specific to Duke University,” Ferguson cautioned. “It may be tempting to generalise it to all universities, but the situation may be different at small liberal arts, regional comprehensive state schools or community colleges where the campus culture is far different. I wouldn’t broadly say, from this, that all campuses are hotbeds of self-censorship.”
Because the survey captured beliefs and self-reported behaviour at one point in time, the design cannot establish cause and effect, such as demonstrating that narrower tolerance windows among progressive peers directly cause conservative classmates to stay silent. As the authors note, college self-selection may also play a role: conservatives who choose a predominantly progressive school like Duke may arrive already unusually comfortable with opposing viewpoints, helping explain their wider Overton windows.
“Our study was conducted only among undergraduates at one school,” Dowd remarked. “We hope future research will use this survey design on other student body populations and be measured over time to understand how civic discourse differs on various campuses and is evolving over time. There’s also an opportunity for future research to examine the consequences of self-censorship in the classroom deeper to understand how this phenomenon impacts students and our society beyond the classroom.”
Ferguson suggested that future work could also compare perspectives on both sides of the classroom. “I still think it would be interesting to look at student versus professor perceptions of bias in education,” he said.
The study, What Drives Censorship on College Campuses?, was authored by Peter Arcidiacono, Cory Dowd, and Matthew Lilley.
