Boundaries of Free Speech: Profiling Support for Acceptance of Free Speech and Restrictions on Offensive Speech
Natasha Doré, Nicole Satherley, Kumar Yogeeswaran, Andrew James Vonasch, Maykel Verkuyten, Chris G. Sibley · International Journal of Public Opinion Research · 2022
In 2020, a talk by philosopher Peter Singer was canceled by a venue in Auckland, New Zealand, following public outcry over his position on euthanasia. Singer argues it should be morally acceptable for parents to euthanize infants suffering from severe disabilities that would significantly negatively impact the quality of life (Singer, 2011). While undoubtedly controversial, people were divided on whether it was the right decision to ban Singer from speaking. Some thought he should not be allowed to speak because his views were offensive to the disabled community; others thought that while such views may be distasteful, he should still be allowed to express them. The debate boiled down to the blurred boundaries between allowing free speech and restricting harmful or offensive speech. Specifically, should people be allowed to freely express opposing views? Or is censorship appropriate if someone’s views are considered offensive? The mixed response to the cancelation of Singer’s event exemplifies ongoing societal debates about tolerating speech one disapproves of versus banning speech deemed offensive to certain groups. While some people endorse free speech for a wide majority of cases and oppose restrictions on speech even when it is deemed offensive, others believe certain types of speech should be restricted or banned in public (Harell, 2010; Chong & Levy, 2018). New Zealand is ranked the fourth freest nation for freedom of expression (World Population Review, 2022). While it is unlawful to distribute threatening, abusive, or insulting content based on certain group memberships (New Zealand Human Rights Act, 1993, pp. 61–69), generally, everyone has the freedom to express varied opinions (New Zealand Bill of Rights Act, 1990, cl. 14). However, it is unclear whether the average New Zealander knows the legal regulations around free speech. An understanding of the nuanced ways in which the lay public distinguishes between accepting speech of different kinds and the size of these profiles within the wider population will be extremely informative for public policy-makers who are often expected to address public debate around these topics without sufficient data. The current research explores the extent to which New Zealanders support allowing speech they disapprove of and the censorship of offensive speech. As a secondary goal, we also examine how various personality traits predict speech endorsement. To investigate the subjective constellation of attitudes toward free speech and offensive speech, we examined responses to each type of speech taken from the nationally representative New Zealand Attitudes and Values Survey (NZAVS). We then use latent profile analysis (LPA) to explore different profiles within the population that may differ in their endorsement of free speech and their willingness to restrict speech deemed harmful or offensive to minority groups (which we call offensive speech). This analysis involves a person-centered approach by creating subgroups within the population based on people’s individual opinions on both issues simultaneously. In treating free speech endorsement and offensive speech suppression as separate factors, we can examine how endorsement of speech is informed both by whether speech contains content that (1) opposes one’s viewpoint, and (2) is perceived as harmful and offensive toward minority groups. This allows us to examine what percentage of the population tolerates disapproved speech, while opposing limits on offensive speech, compared to those intolerant of offensive speech, and any other subgroups that may exist within the population. Additionally, we examine the personality traits of people with differing speech profiles. While previous research has established a relationship between personality and tolerance (Sullivan & Transue, 1999), willingness to censor hate speech (Lambe, 2004), and political conflict (Grubbs, Warmke, Tosi, James, & Campbell, 2019), here we investigate whether certain personality traits are related to the likelihood of belonging to any potential profile that may emerge. Free speech is a cornerstone of most liberal democracies since the Enlightenment (McClosky & Brill, 1983), and widely endorsed—in the abstract. However, the extent to which people endorse free speech in specific depends on the content of the speech—while people support speech they like or consider morally good, they tend to be more intolerant of speech they dislike (Gibson, 2006), even if they try to maintain equal endorsement (Eftedal & Thomsen, 2021). As speech becomes more inflammatory, support for expression begins to taper off. For example, the global median support for allowing people to criticize government policies is 80% (Wike & Simmons, 2015). Yet offensive speech directed toward religious or minority groups is considered permissible by 35% of people. Public calls for violent protests garner even less support, with a global median of 25%. The more contentious the content of speech, the less people endorse free expression. People’s views on free speech may reflect a balance of two principles: free expression versus preventing harm. Whereas people typically wish to grant people the freedom to express views they merely disagree with, they may oppose expressions they deem offensive to others when they think the harm of offense outweighs the benefits of free expression (Verkuyten, Adelman, & Yogeeswaran, 2020). There are also individual differences, for example, people with higher cognitive ability tend to support free speech even for those with opposing views (De Keersmaecker, Bostyn, Van Hiel, & Roets, 2020). Thus, it is an oversimplification to treat tolerance of free speech and intolerance of offensive speech as opposite ends of a bipolar construct. Strong support for freedom of speech is not the opposite of support for the suppression of offensive speech; a dichotomous view in which someone either endorses free speech or suppresses offensive speech may not capture nuances within the population. Therefore, the current research separately assesses people’s support for expressing opposing speech versus speech deemed offensive to minority groups for understanding the nuanced ways in which the public thinks about free expression. Research investigating the relationship between personality and tolerance provides insight into the importance of examining personality correlates of profiles involving the tolerance of varied speech. For example, Sullivan and Transue (1999) reveal that more neurotic and extraverted individuals are less politically tolerant, whereas those more open to experience are more tolerant of dissenting views. They also describe how those high in neuroticism are more responsive to threatening groups, as they are more sensitive to threat perceptions. Other research suggests that neuroticism is positively related, while openness is negatively related to a willingness to censor others (Lambe, 2004). It may be that those with a high support for the banning of offensive speech and the restriction of free speech may display a similarly high level of neuroticism and low level of openness. The personality traits mentioned above are part of the HEXACO Personality Inventory (Lee & Ashton, 2004), a measure of the six core dimensions of personality (humility, emotionality/neuroticism, extraversion, agreeableness, conscientiousness, and openness). Additional traits may also relate to these speech profiles. For example, highly modest people tend to underrepresent their accomplishments, while those low in modesty would not hesitate to display their strengths. It has been shown that status-seeking personality traits are associated with both moral and political conflict (Grubbs, Warmke, Tosi, James, & Campbell, 2019), suggesting that those low in modesty may be more likely to suppress speech they disagree with. For the sake of transparency, the current work examined how all personality traits available in the survey related to speech profiles. Beyond HEXACO, these include modesty (Ashton & Lee, 2005), vengeful rumination (i.e., the extent to which people think about past times they have been wronged; McCullough, Bellah, Kilpatrick, & Johnson, 2001), self-control (De Ridder, Lensvelt-Mulders, Finkenauer, Stok, & Baumeister, 2012), and perfectionism (Habke & Flynn, 2002). The present work uncovers profiles of speech endorsement by considering support for public expression of speech one opposes (free speech) and speech deemed offensive to certain groups (offensive speech). We then determine whether personality traits (the HEXACO Personality Index, vengeful rumination, self-control, modesty, and perfectionism) are associated with any of these profiles. We make no specific prediction as to how each of the personality factors may differ across any groups that may arise as the speech profiles that exist within the population are unknown. We use data from Time 11 of the NZAVS, an annual, longitudinal national probability sample of New Zealand adults using their residential location. All participants provided informed consent to take part in this hour-long survey. As a longitudinal survey, data are confidential, following ethical guidelines of the relevant institutional review board. This wave of the NZAVS contained responses from 42,684 participants drawn from the electoral roll of all New Zealand adults who are citizens or permanent residents, as registration is compulsory after the age of 18. Full details about the sampling procedure and sample recruitment can be found on the NZAVS website (NZAVS, n.d.). Of those who reported the relevant demographics, the average age was 48.60 years (SD = 13.86), 62.6% were women, and 9.8% were Māori (indigenous peoples), 2.2% Pacific Islanders, 5.3% Asian, and 3.8% reported other ethnicity, while 88.7% reported some European ancestry. Endorsement of free speech and suppression of offensive speech are measured using single items, using a scale from 1 (strongly disagree) to 7 (strongly agree). ‘Although I may disagree with the opinions that other people hold, they should be allowed to express those views publicly.’ ‘People who hold opinions that are harmful or offensive to minority groups should be banned from expressing those views publicly.’ HEXACO items were measured using the Mini-IPIP6 (Sibley et al., 2011), on a scale from 1 (very inaccurate) to 7 (very accurate). Participants were asked to rate how well each statement described them personally. The HEXACO model has been shown to have high reliability and validity (Lee & Ashton, 2004). Modesty items were adapted based on the work done by Ashton and Lee (2008), Campbell, Bonacci, Shelton, Exline, and Bushman (2004), and Sibley, Harding, Perry, Asbrock, and Duckitt (2010). Perfectionism was measured using items from the short form of the revised Almost Perfect Scale (Rice, Richardson, & Tueller, 2014). Items are measured from 1 (strongly disagree) to 7 (strongly agree). Self-control was measured using two items from the self-control scale (Tangney, Baumeister, & Boone, 2004). Participants are asked to “indicate how much each of the statements reflects how you typically are.” Vengeful rumination was measured using an item created specifically for the NZAVS, items adapted from work done by Berry, Worthington, O'Connor, Parrott, and Wade (2005), and the dissipation–rumination scale (Caprara, 1986). All items and the exact coding of variables can be found in the supplementary materials, alongside further demographic information (Table S2). This can also be found on the NZAVS website, alongside the syntax used (NZAVS, n.d.). LPA was conducted in Mplus 8.4. LPA requires multiple models with an increasing number of profiles to be specified. Models are then evaluated based on a combination of model fit, interpretability, and parsimony, to determine the final solution. Fit statistics for models with 2, 3, and 4 profile solutions for support of free speech and opposition to offensive speech items (r = −.228, p < .001) are displayed in Table 1. Decreases in akaike information criterion (AIC) and bayesian information criterion (BIC) values can be indicative of improvements in model fit, while Lo-Mendell-Rubin likelihood ratio tests formally compare a k profile solution with the k−1 profile solution (Lo, Mendell, & Rubin, 2001). Finally, entropy values (ranging from 0 to 1) indicate the clarity of separation between each estimated profile in the model, with higher values indicating better profile separation (Collins & Lanza, 2010). Fit Statistics and Comparisons of the Estimated LPA Models Note. LMR = Lo-Mendell-Rubin. *** p < .001 ** p < .01 * p < .05. Fit Statistics and Comparisons of the Estimated LPA Models Note. LMR = Lo-Mendell-Rubin. *** p < .001 ** p < .01 * p < .05. As shown in Table 1, the three-profile solution appeared to be a better fit than the two-profile solution across all indices. Furthermore, the four-profile solution also appeared to be a better fit than the three-profile solution, with an entropy value nearing 1. However, closer inspection of this model revealed that the fourth profile was achieved by splitting an existing profile in the three-profile solution based on free speech scores. This resulted in two profiles with very similar views on banning offensive speech, but one profile consisted entirely, or almost entirely of individuals who scored a 7 on support for free speech, and the other profile of individuals who scored 6. Because of the nature of the four-profile solution (very high entropy but inconsequential profile differences compared to the three-profile solution), we opted not to test models with even more profiles. Thus, the three-profile solution was deemed the most parsimonious and interpretable, while still achieving a very good fit to the data (indicated by high entropy). The good model fit is further indicated by the profile classification probabilities displayed in Table S1. For members of each latent profile, the classification probability of being assigned to that profile was very high (.936–.987), with classification probabilities for other profiles very low (.000–.060). The profiles identified in the three-profile solution are displayed in Figure 1. Profile 1, “censors,” was the smallest profile (2.6% of the sample). This profile was comprised of individuals with low average support for tolerating free speech (M = 2.520), and relatively high average support for banning offensive speech (M = 5.041). Profile 2, “moderates,” had moderate support for both free speech (M = 4.780), and banning offensive speech (M = 4.543), and represented 24.2% of the sample. Profile 3 represented “tolerators” and was the largest profile (73.2%). This profile expressed very strong support for tolerating free speech (M = 6.521), while expressing moderate to low support for banning offensive speech (M = 3.750). The extent to which members of each profile endorse both allowing free speech and banning offensive speech. Using a three-step logistic regression approach, we examined whether personality constructs are associated with latent profile membership while adjusting for demographic factors, with results displayed in Table 2. This approach allows the estimation of latent profiles first, with a variable formed of most likely profile membership for each observation, and the predictor variables are then used to predict profile membership while also accounting for classification error in profile membership (see Asparouhov & Muthén, 2014). For this analysis, predictor variables were normalized to a common scale of 0–1 for ease of interpretation of effects. Thus, the regression coefficients represent the effects of shifting from the minimum to the maximum value of a predictor. Regression Model of the Odds of Belonging to the Censor’s and Moderate’s Profiles, Relative to the Tolerator’s Profile Note. N = 38,725. based on age in years or over the reported are this is similar to other research in et al., As attitudes are likely by a and of factors, the of any one personality on profile membership will be sample size allows us to *** p < .001 ** p < .01 * p < .05. Regression Model of the Odds of Belonging to the Censor’s and Moderate’s Profiles, Relative to the Tolerator’s Profile Note. N = 38,725. based on age in years or over the reported are this is similar to other research in et al., As attitudes are likely by a and of factors, the of any one personality on profile membership will be sample size allows us to *** p < .001 ** p < .01 * p < .05. Relative to the “tolerators” profile, neuroticism was associated with higher of belonging to the profile = = p = as was vengeful rumination = = p < the level of neuroticism was associated with a in the of belonging to the profile to the while the level of vengeful rumination was associated with a in compared with the level of vengeful openness to experience = = p = and modesty = = p < .001) were associated with of belonging to the profile (i.e., and the to higher of = = p < = = p < = = p = openness to experience = = p < and modesty = = p < .001) were associated with of belonging to the to and in the compared to the level of each predictor. the of neuroticism = = p < .001) and vengeful rumination = = p < .001) were associated with and higher of belonging to the to demographic variables were also associated with profile The of belonging to the and profiles were and than women, to “tolerators” profile This not that were to be “tolerators” as this profile of the and a majority of participants were people were less likely to to both the and profiles to the “tolerators” profile, with of belonging to these profiles being and the age in the compared to the people and were less likely to to both these profiles to those over suggesting that the relationship between the profiles and age is For example, people in their or may be more likely to to the and while those or are more likely to to the “tolerators” on the were associated with higher higher the compared to the of belonging to the Furthermore, more individuals were more likely to to the “tolerators” profile, as to the and profiles. who reported a New Zealand European were less likely to to the profile to people who reported any other Finally, more politically people were less likely to be in the and profiles compared to “tolerators” profiles. Using a national the current research examined profiles within the population on their willingness to free speech (i.e., support for speech they disagree and restrict offensive speech (i.e., support for banning speech deemed offensive to certain The between these two items (r = the importance of treating these as constructs and not that high support for free speech one disapproves of is related to suppression of offensive speech. Using a person-centered approach, revealed profiles in New and closer of differences on both constructs that the between the profiles in their support for tolerating free speech. While was between the of support for banning offensive speech across profiles (M = the differences in for free speech across the groups were much (i.e., = This suggests that while people low to moderate of support for restricting offensive speech, differences arise when it to tolerating opinions one with. is no group with high support for allowing free speech and high support for banning offensive speech. This may be because all profiles for the restriction of offensive speech when considering their endorsement of free speech. We also found that some personality traits were more associated with certain profiles over for demographic For example, were less more and more in vengeful rumination than were more more in vengeful rumination, less less less and in self-control than support previous work suggesting that neuroticism and openness a willingness to censor hate speech (Lambe, 2004), but also insight into the correlates of personality with the tolerance of free speech and intolerance of offensive speech. Relative to were also more likely to over previous times they have been this the likelihood of them to on those who they believe have it to that the effects identified here are relatively and they indicate how personality to membership in the may be more such as of or the of to for in The that was no between and “tolerators” in is similar to work suggesting that low of may be associated with the of as to the in censorship 2020). we measured people’s toward opposing speech, than their a different emerge. research investigate support for of such as people for offensive As the items used reflect attitudes on speech, individuals may have differing attitudes about specific or that should be For example, tolerance research that intolerance is in different for different for it from the and for from the & 2002). research be to determine any between people’s about the expression of opposing views and restrictions on offensive speech, and of the work should also investigate whether a similar of results in differing as the current work is specific to one national While use of a national probability sample is an by results with high national also from us to on speech research this work using more of speech. the current research provides insight into the nuanced ways in which people think about free expression. people between people’s right to express opposing and to restrict speech deemed This is with divided public attitudes toward like Peter to free speech, but it to offensive views. The NZAVS is by a grant from the They had no in or analysis and interpretation of data. We have no conflict of to As the NZAVS is a longitudinal data be However, the Mplus syntax used to test all models reported in this is available on the NZAVS to the being it was not