Who has a free speech problem? Motivated censorship across the ideological divide
Ivar R. Hannikainen
Departamento de Filosofía I
Universidad de Granada
Recent years have seen recurring episodes of tension between proponents of freedom of speech and
advocates of the disenfranchised. Recent survey research attests to the ideological division in attitudes
toward free speech, whereby conservatives report greater support for free speech than progressives do.
Intrigued by the question of whether “canceling” is indeed a uniquely progressive tendency, we conducted
a vignette-based experiment examining judgments of offensiveness among progressives and
conservatives. Contrary to the dominant portrayal of progressives and conservatives, our study
documented ideological symmetry in their evaluations of offensive speech. When faced with utterances
whose content matters to them, both conservatives and progressives viewed outgroup speakers as more
offensive than ingroup speakers. A second contribution of this chapter is to provide a deeper understanding
of the cognitive mechanism implicated in evaluating outgroup speech as more offensive than ingroup
speech. Our results suggest that perception of offensiveness is mediated by ascriptions of intent: we tend
to attribute negative intent to the speaker whenever we deem their utterances to be offensive, even against
the explicitly stated speaker’s background attitudes.
Who has a free speech problem? Motivated censorship across the ideological divide
Recent years have seen recurring episodes of tension between proponents of freedom of speech and
advocates of the disenfranchised (see Meesala, 2020; Ramsay, 2021). This tension has also acquired a
political tint: Voices on the left assume a duty to protect underprivileged groups from insidious forms of
discrimination, that sometimes take the form of humor, science, or campaign messaging. Meanwhile, the
right reacts with resentment toward a growing climate of censorship, the so-called “cancel culture”,
which–from its perspective–has degraded intellectual life and stifled public discourse (Romano, 2021).
This phenomenon has also fractured academia: In 2020, a long list of writers signed the Harper’s
Letter, in which they denounced creeping restrictions on intellectual freedom throughout many
contemporary democracies (Harper, 2020). Shortly after, The Objective published a critical response in
which a second group of journalists and academics, emphasizing the signatories’ position of privilege,
questioned how a group of primarily cisgender white members of the cultural elite could expound on their
experience of intellectual repression while drawing attention away from the deeper and persistent
problems that divergent, minority voices face (The Objective, 2020).
Recent survey research attests to the ideological division in attitudes toward free speech, whereby
conservatives report greater support for free speech than progressives do (Pew Research Center, 2016,
2021). This alignment between ideology and endorsement of free speech has fostered distinct stereotypes
about liberals and conservatives: for instance, the idea that progressives are overly critical of
generalizations and differentiation involving sociodemographic groups, and unduly sensitive to
purportedly offensive language, and the idea that conservatives are callous in their treatment of
marginalized groups and even lack a moral compass.
Intrigued by the question of whether “canceling” is indeed a uniquely progressive tendency, we
conducted a vignette-based experiment examining judgments of offensiveness among progressives and
conservatives. Our study also builds on previous work in which we identified contextual factors that render
speech offensive, providing a clearer picture of the circumstances in which these effects arise and why.
Before presenting this research, we survey evidence from the published literature elucidating the factors
that shape our evaluations of offensive speech.
Whether a given statement is offensive or not depends evidently on what is being said; i.e., the
content of speech. Yet experimental research has established that a host of contextual elements also impact
our judgments of whether speech is offensive. For instance, their offensive meaning shifts between
professional and personal contexts (Fasoli, Carnaghi & Paladino, 2015; O’Dea et al. 2015; O’Dea &
Saucier 2016). Speakers with malicious intent are treated as more offensive than speakers with neutral
intent–even when making the same remark; and whether the audience reacts adversely or indifferently can
retrospectively determine whether a statement was offensive (Swim et al., 2003)
. Other studies have
uncovered an influence of the speaker’s identity: Listeners rapidly form an impression of the speaker from
their voice (Fourcart & Hartsuiker, 2021), accent (Cai et al., 2017), and gender (Fasoli, Carnaghi &
Paladino, 2015; O’Dea et al., 2015; O’Dea & Saucier, 2016)–which informs their interpretations of the
In our own research, we have focused primarily on whether the offensiveness of a statement about
some target group depends on the speaker’s relationship to that same group (Almagro, Hannikainen &
Villanueva, 2021). Our inspiration came from the observation that remarks that can be acceptable –merely
There is at least another possibility here: the offensive content of speech is determined by contextual factors. This
alternative formulation implies that the nature of offensiveness is semantic and not exclusively pragmatic. In other
words, it is what we say –rather than how we say what we say– that is apt to be offensive.
informative– when made by members of non-dominant groups are sometimes seen as offensive and
censurable when made by members of socially advantaged outgroups–a phenomenon to which our studies
attested. For example, in one of the scenarios we employed, a politician from the Spanish city of Ceuta,
an underprivileged region located in North Africa, asserts that “For Spain, losing Catalonia is not the same
as losing Ceuta”--a statement than can readily be understood as informative and/or offensive. We then
compared participants’ reactions in this control condition to two other experimental conditions in which
the same statement was made by an outgroup member. Our results showed that the statement was more
offensive and worthy of censorship when uttered by an outgroup member, e.g., a politician from a different
region in Spain, than by an ingroup member. Additionally, this effect held whether the speaker had a
higher social status or a comparable social status (e.g., belonging to an equally disadvantaged outgroup)–
which suggests that the offensiveness results from outgroup membership and not from elevated social
status. We refer to this finding as the speaker membership effect, and in the present study we seek to
understand whether it arises among conservatives’ and progressives’ assessments of offensive speech.
Moral conviction and the contours of speaker identity norms
Throughout this broad base of experimental research, differences between progressives’ and
conservatives’ response patterns have either been left unexamined, limited in scope, or secondary to the
researchers’ objectives. In the present study, we direct our emphasis onto this exact question: Do
conservatives and progressives differ in their tendency to oppose discordant messages and deem them
offensive? In recent years, research in political psychology has systematically examined whether various
psychological dispositions are symmetrical (manifesting equally among progressives and conservatives)
or asymmetrical (i.e., emerge selectively or more strongly in one ideological group; see e.g., Crawford,
A historical perspective calls into question the view that freedom of speech is a fundamentally
conservative principle. Early appeals to freedom of expression were made from the left, in an effort to
protect labor unions’, abolitionists’ and suffragists’ right to protest from the suppression of government.
These considerations provide some intuitive support for the symmetry view, and point toward a particular
explanation: Perhaps, listeners oppose free speech that threatens their moral convictions (Epstein et al.,
2008; Skitka et al., 2021). When the exercise of free speech violated tradition and undermined authority–
characteristically conservative values (Graham et al., 2009)–then it was progressives who defended free
speech and conservatives who called for its containment. Nowadays, as free speech seemingly challenges
the progressive values of justice and equality, conservatives have come to adopt a more favorable attitude,
and progressives a more restrictive attitude, toward free speech (see Epstein et al., 2008). From this
perspective, the contemporary politicization of free speech is primarily a product of opportunism and not
principle (see identity-protective cognition, Kahan 2017; Kahan, Braman & Jenkins-Smith, 2021)–i.e., a
reflection of the shifting issues at hand and not of stable political values of the left versus right.
In the present chapter, we hone in on this question: namely, whether progressives and
conservatives selectively “cancel” speakers who threaten their worldview while upholding the freedom of
speech of those who promulgate it. This makes a further prediction about the evolution of historical
debates as ideological division wanes (Hernández, Anduiza & Rico, 2021). To assess the latter hypothesis,
we also contrast attitudes toward contemporary or active political disputes (regarding Covid-19,
immigration and gender inequality) with attitudes toward historical or inactive debates (regarding divorce,
same sex marriage, and the Spanish national anthem). Conservative and progressive views about the
permissibility of divorce or same sex marriage, though once in stark conflict, have largely aligned over
time in the direction of the historically progressive stance. So we reasoned that a closer look at judgments
of offensiveness in the context of historical issues for which ideological division has waned would help
dissociate the contributions of ideological conviction and political identity to evaluations of offensive
Speaker identity norms as effects of inverse planning
A second contribution of this chapter is to provide a deeper understanding of the cognitive
mechanism implicated in evaluating outgroup speech as more offensive than ingroup speech. In our
previous studies, participants often denied the relevance of speaker identity when asked in abstract terms,
despite the fact that it exerted a robust influence on their case-based evaluations (Almagro, et al., 2021).
This helps rule out a particularly straightforward explanation: namely, the reason why people selectively
censor outgroup speakers is not due to the endorsement of a corresponding explicit norm.
Partly inspired by previous research on the reappropriation of certain slurs (Gibson, Epstein &
Magarian., 2019), we considered a rather different explanation: i.e., that contextual cues regarding the
speaker’s identity and prior attitudes could help establish whether what they said was offensive by
informing listeners’ representations of the speaker’s intent.
For instance, outgroup speakers may be
ascribed more harmful intent (i.e., the desire to offend the target group) than ingroup speakers, who might
be ascribed a neutral intent (such as the goal of describing the target group)–when making the exact same
statement. This intention ascription could then help explain the speaker membership effect on
offensiveness, since mental state inferences are known to exert downstream effects on prescriptive
evaluations, such as blame and wrongness (Cushman, 2008; Monroe & Malle, 2017). On this view, a
speaker’s identity has an immediate effect on representations of their intention, and indirectly on the
degree to which their utterances are deemed offensive.
This hypothesis could also link people’s judgment patterns to their abstract principles. If indeed speaker identity
influences ascriptions of intent, then the endorsement of speaker intent as the most relevant criterion could help
account for the large impact of speaker identity on perceptions of offensiveness.
Materials, data and analysis scripts are publicly available on the Open Science Framework at:
To estimate our required sample size, we conducted a power analysis (f2 = .10, alpha = .05, power
= .90) for a linear regression model with 7 numerator degrees of freedom, corresponding to the three main
effects (statement orientation, speaker membership, and speaker intention) and all their possible
interactions. A target N of 182 provided 90% power to observe a significant effect of magnitude f2 = .10.
Since we were interested in conducting two separate studies (one among progressives and one among
conservatives), our target sample size was 364 participants.
We recruited 400 Spanish natives in partnership with a survey research firm. The sample was
divided into four strata according to political orientation and sex, with 100 participants in each stratum:
Men-Left, Women-Left, Men-Right, and Women-Right. Mean sample age was 49 years old (M = 50 left-
leaning participants, M = 48 right-leaning participants). In addition to sex, the sample was also nationally
representative for age group (six strata: 18-24, 25-34, 35-44, 45-54, 55-64, 65-99) and geographical region
(nine regions: northeast (Catalonia and Balears), east, south (Andalusia), center, northwest, north center,
Canary Island, Barcelona metropolitan area, Madrid metropolitan area). Furthermore, the sample included
self-declared supporters of each of the five primary political parties in Spain (Unidas Podemos: n = 36,
PSOE: n = 73, Ciudadanos: n = 37, PP: n = 68, Vox: n = 42, Other: n = 66, No answer: n = 78). By
comparison to recent nationally representative polls of voting intention (Centro de Investigaciones
Sociológicas, 2022), our study appeared to overrepresent conservative voters (i.e., Ciudadanos, PP and
Vox), which could be explained by our 1:1 left/right stratification.
The materials for this study were based on three historic or ‘closed’ issues (divorce, same sex
marriage, and the Spanish national anthem), and three contemporary or ‘open’ issues (Covid-19,
immigration and gender inequality).
For each issue, we wrote a total of eight variants, in order to complete the 2 (orientation: whether
the remark affronts progressive or conservative values) x 2 (identity: whether the speaker was a
member/non-member of the target group) x 2 (attitude: whether the speaker has a negative/positive
attitude toward the target group) matrix. These manipulations are illustrated in Table 1.
Table 1. Example Stimuli.
During a TV show, several guests discuss the consequences of the
law that regularized divorce in Spain, approved in 1981.
Ana, a woman who is the spokesperson for a feminist group (/Ángel,
and who in the past has always been uncomfortable (/comfortable)
in the presence of married people, says the following:
“Most divorced people feel freer, safer, happier and more dignified
than during marriage”
(/”Children of divorced families are, statistically, more prone to
delinquency and antisocial conduct”).
In each scenario, a verbal statement was made that was defined by its target group (i.e., the political
group that could potentially take offense). For the sake of our present example, let us suppose that the
speaker states that “Children of divorced families are, statistically, more prone to delinquency and
antisocial conduct” and, therefore, would have progressives as their target group.
After the opening sentence, we provided information about the speaker’s identity, by randomly
presenting either a characteristically conservative (e.g., a bishop
) or progressive (e.g., a spokesperson for
a feminist group) speaker. This manipulation allowed us to code for two further properties of the scenario:
whether the speaker belongs to the target group or not (which we refer to as membership), and whether
they belong to the participant’s ideological ingroup or outgroup (which we refer to as ingroup status).
Next, we convey the speaker’s background attitude toward the target group, which could be
positive or negative (depending on condition assignment). And lastly, we displayed the statement (which,
as previously stated, could be geared toward offending progressives or conservatives).
In a balanced incomplete block design, participants were randomly assigned to one of eight groups
and viewed a battery of six consecutive scenarios in a random order. In each group, participants viewed
six of the eight factorial combinations in the 2 (speaker membership: ingroup, outgroup) × 2 (speaker’s
background attitude: offensive, neutral) × 2 (statement orientation: left, right) matrix paired with a
different scenario on each trial. Thus, no participant viewed the same scenario or factorial combination
twice. Collapsing across groups, we achieved balance in the scenario-by-condition matrix (n per cell: Mdn
= 52, Min = 41, Max = 64).
After the six scenarios, participants self-reported the relevance of the same four considerations in
determining whether speech is offensive: the speaker’s identity and intent, as well as the statement’s
orientation, and the salience of the issue in question.
It is important to note that, in Spain, the catholic religion is associated with conservatives.
Offensiveness and intent ratings
On each trial, participants evaluated the statement’s offensiveness and the speaker’s intent.
Offensiveness was assessed through four items:
[O1] the speaker “is offensive,”
[O2, reverse-scored] the speaker “simply provides information about” the situation,
[O3] the speaker “should not say that kind of thing”, and
[O4, reverse-scored] “I see no problem with the speaker saying that kind of thing”.
Attributions of intent were made in a pair of items:
[I1, reverse-scored] the speaker “merely wanted to provide information about” the situation, and
[I2] the speaker “believed that her statement would be offensive”.
Every item was assessed on a continuous scale from 1: “Strongly disagree” to 7: “Strongly agree”
in decimal increments. The four offensiveness items formed a composite score with good reliability
(Cronbach’s α = .77). The composite intent score, however, exhibited questionable reliability (Cronbach’s
α = .60), perhaps due to the low number of items.
Participants reported their emphasis on (1) the speaker’s identity (“who the speaker is”), (2) the
speaker’s background attitude (“whether or not the speaker holds a pejorative attitude toward the group
he/she talks about”), (3) the statement’s orientation (“whether the topic under discussion is relevant to the
left or to the right”), and (4) the salience of the issue in question (“whether or not the topic is politically
salient”). These ratings were provided using single items and recorded on a continuous scale from 1: “No
importance/relevance” to 7: “Absolute importance/relevance”.
We also recorded participants’ age, sex, political orientation, and voting preference.
The results section is divided into four subsections: In Parts 1 (on offensiveness) and 2 (on intent), we
analyze the responses of left-leaning and right-leaning participants separately. These subgroup analyses
are followed by moderation analyses on the full dataset, to assess whether judgments of offensiveness and
intent depend on participants’ ideology. In Part 3, we introduce a plausible causal model of intent and
offensiveness and conduct mediation analyses to evaluate its fit to the data. In Part 4, we report
supplementary analyses of differences in reactions toward active debates (such as Covid-19, immigartion,
or gender inequality) versus inactive or historical issues (such as divorce, same sex marriage, or the
Spanish national anthem).
Part 1: Offensiveness
As shown in Table 2, in the left-leaning sample, we observed main effects of statement orientation,
speaker membership, and speaker attitude. Statement orientation interacted with speaker membership,
whereas there were no interactions involving the background attitude factor. Replicating previous
research, the main effect of attitude revealed that negative background attitudes rendered speech more
offensive (M = 4.07, 95% CI = [3.35, 4.80]) than positive background attitudes (M = 3.81, 95% CI = [3.08,
4.53]), B = 0.26, t = 3.02, p = .003.
We followed up on the membership×orientation interaction by examining the pattern of marginal
effects: Affronts to a progressive worldview were seen as more offensive when coming from non-members
(M = 4.61, 95% CI = [3.89, 5.32]) than members (M = 3.95, 95% CI = [3.23, 4.66]), B = 0.66, t = 5.35, p
< .001. The pattern reversed for affronts to a conservative ideology: They were seen as more innocuous
when coming from progressive non-members (M = 3.44, 95% CI = [2.73, 4.16]) than conservative
members (M = 3.75, 95% CI = [3.04, 4.47]), B = -0.31, t = -2.55, p = .011.
Among right-leaning participants, we found a main effect of statement orientation, which was
qualified by a two-way interaction with speaker membership–as in the progressive sample (see Table 2).
No main effects of speaker membership or attitude were observed.
The two-way interaction mirrored the result obtained among progressives (see Figure 1): Affronts
to a conservative ideology were seen as more offensive when stemming from non-members (M = 4.14,
95% CI = [3.36, 4.92]) than members (M = 3.79, 95% CI = [3.01, 4.56]), B = 0.35, t = 2.97, p = .003. In
contrast, affronts to a progressive ideology were evaluated comparably whether coming from a member
(M = 3.79, 95% CI = [3.01, 4.57]) or non-member (M = 3.71, 95% CI = [2.93, 4.49]) of the target group,
B = -0.08, t = -0.69, p = .49.
Table 2. Model Comparisons: Offensiveness.
Left (n = 200)
Right (n = 200)
F(1, 983) = 4.08, p = .044
(AIC = 4509)
F(1, 996) = 2.31, p = .13
(AIC = 4470)
F(1, 1060) = 8.68, p = .003
F(1, 1068) = 2.06, p = .15
F(1, 1087) = 58.17, p < .001
F(1, 1100) = 7.51, p = .006
F(1, 1050) = 0.00, p = .93
(AIC = 4480)
F(1, 1076) = 0.66, p = .42
(AIC = 4469)
F(1, 1023) = 31.90, p < .001
F(1, 1042) = 6.73, p = .010
F(1, 1122) = 0.25, p = .62
F(1, 1133) = 0.01, p = .91
F(1, 1043) = 4.15, p = .042
(AIC = 4479)
F(1, 1038) = 0.29, p = .59
(AIC = 4470)
Summary & Comparative Analyses
Statement orientation interacted with speaker membership in both progressives’ and
conservatives’ ratings of offensiveness. In both groups, outgroup speakers were seen as more offensive
than ingroup speakers when making a statement hostile to participants’ own ideology.
To jointly analyze progressive and conservative responses, we recoded the independent variables:
(1) ideological fit, which takes the participants’ political orientation and the statement’s content and codes
whether these are concordant or discordant, and (2) ingroup/outgroup, which takes the participant’s
political orientation and the speaker’s identity and codes whether the speaker belongs to the participant’s
ingroup or outgroup.
We conducted separate analyses for concordant and discordant statements. In comparative analysis
of concordant messages, this membership effect was significant, F(1, 885) = 4.85, p = .028, and was not
moderated by participants’ political ideology, F(1, 876) = 1.72, p = .19. For discordant messages, the effect
of speaker membership was highly significant, F(1, 885) = 35.29, p < .001, and was moderated by
participants’ political ideology, F(1, 830) = 4.29, p = .039. Specifically, outgroup speakers were perceived as
more offensive by liberals than by conservatives, B = 0.47, t = 3.61, p < .001, while ingroup speakers were
evaluated similarly in both groups, B = 0.12, t = 0.91, p = .36 (see Figure 1). Thus, while we observed the
membership effect among both progressives and conservatives (suggesting ideological symmetry), the
magnitude of the effects was significantly different (suggesting asymmetry).
When looking separately at progressives and conservatives, the effect of speaker attitude attained
statistical significance only in the left-leaning subsample. Yet a comparative analysis revealed a main
effect of attitude, F(1, 2143) = 8.60, p = .003, that was not moderated by participants’ political orientation,
F(1, 2143) = 0.61, p = .44–suggesting that the discrepancy in the subgroup analyses was not meaningful and,
in turn, that the effect of background attitudes on offensiveness is symmetrical across the political divide.
Figure 1. Marginal effect of background attitudes (top) and speaker membership (bottom) on
offensiveness and intent ascriptions. In the top panel, we display the main effect of attitudes, collapsed
across statement orientation. In the bottom panel, we display the membership effect separately for
progressive (red) and conservative (blue) statements.
Part 2: Attributions of Intent
Among left-leaning participants, we observed main effects of statement orientation, speaker
membership, and speaker intent. The effects of statement orientation and speaker membership were
qualified by a two-way interaction (see Table 2). Once again, speaker attitude exerted a main effect–such
that speakers with a negative attitude (M = 4.07, 95% CI = [3.54, 4.61]) were ascribed more harmful intent
than speakers with a positive attitude (M = 3.86, 95% CI = [3.33, 4.40]), B = 0.21, t = 2.44, p = .014.
Regarding the two-way interaction between statement orientation and speaker membership,
marginal effects revealed a strong influence of speaker membership for conservative remarks. Outgroup
speakers (M = 4.61, 95% CI [4.08, 5.14]) were ascribed more harmful intent than were ingroup speakers
(M =3.78, 95% CI [3.24, 4.31]) when making the same statements hostile to a progressive worldview, B
= 0.83, t = 6.89, p < .001. Interestingly, the effect of speaker membership appeared to reverse for
progressive remarks: In other words, outgroup/conservative speakers (M =3.86, 95% CI [3.33, 4.39]) were
ascribed marginally more harmful intent than ingroup/progressive speakers (M =3.63, 95% CI [3.10,
4.16]), t = 1.95, p = .051–despite being members of the target group which the remark could ostensibly
Right-leaning participants exhibited a similar pattern of effects (see Table 2): We found that the
speaker’s identity, and attitude (but not statement orientation) impacted attributions of harmful intent. The
two-way interaction between statement orientation and speaker membership (observed among progressive
participants) also emerged in the conservative sample.
The main effect of intent revealed that speakers with offensive intent (M = 3.93, 95% CI = [3.43,
4.43]) were ascribed more offensive intent than speakers with a positive attitude (M = 3.74, 95% CI =
[3.25, 4.24]), B = 0.19, t = -2.27, p = .023. To probe the two-way interaction, we examined the marginal
effects of speaker membership separately for each statement orientation: When uttering statements hostile
to a conservative worldview, outgroup speakers (M = 4.14, 95% CI [3.65, 4.64]) were ascribed more
harmful intent than were ingroup speakers (M = 3.63, 95% CI = [3.14, 4.13]), B = 0.51, t = 4.44, p < .001.
No difference between ingroup and outgroup speakers arose for statements concordant with a conservative
worldview, B = 0.06, t = 0.53, p = .60.
Table 3. Model Comparisons: Intent.
Left (n = 200)
Right (n = 200)
F(1, 983) = 12.34, p < .001
(AIC = 4472)
F(1, 995) = 7.10, p = .008
(AIC = 4428)
F(1, 1053) = 5.76, p = .017
F(1, 1047) = 4.98, p = .026
F(1, 1078) = 25.80, p < .001
F(1, 1072) = 2.25, p = .13
F(1, 1044) = 0.37, p = .55
(AIC = 4436)
F(1, 1055) = 0.46, p = .50
(AIC = 4421)
F(1, 1021) = 39.89, p < .001
F(1, 1035) = 12.48, p <
F(1, 1119) = 0.51, p = .48
F(1, 1124) = 0.11, p = .74
F(1, 1041) = 2.41, p = .12
(AIC = 4436)
F(1, 1031) = 0.49, p = .49
(AIC = 4422)
Summary & Comparative Analyses
Both conservatives and progressives tended to ascribe more harmful intent to outgroup speakers
than ingroup speakers–when interpreting statements that defy their own ideology. In a comparative
analysis, this membership effect on discordant statements was highly significant, F(1, 824) = 65.70, p < .001,
and appeared to differ when comparing progressives to conservatives, F(1, 824) = 4.32, p = .038. Specifically,
when making ideologically discordant statements, outgroup speakers were ascribed more harmful intent
by liberals than by conservatives, B = 0.44, t = 3.44, p < .001, while ingroup speakers were ascribed
comparable intent, B = 0.10, t = 0.81, p = .42 (see Figure 1). The corresponding effect on concordant
statements was non-significant, F(1, 881) = 2.62, p = .11. Meanwhile, the effect of background attitudes, F(1,
2111) = 10.91, p < .001, did not vary by participants’ political orientation, F(1, 2112) = 0.01, p = .94.
Part 3. Mediation Analyses
The previous sections documented two consistent effects across intent and offensiveness models:
what we refer to as the effect of background attitudes, and of speaker-orientation congruence.
Figure 2. Scatter Plot of Offensiveness by Intent. The panels display assessments of progressive (left) and
conservative (right) speakers issuing statements that cohere with progressive (top) or conservative
(bottom) values. Sample means and 95% confidence ellipses, revealing discrepancies between progressive
and conservative participants’ assessments, are overlaid.
These effects on offensiveness can be parsimoniously explained by positing a common
mechanism: When observing an ambiguous intentional action (in our present case, a remark that could be
construed as either merely informative or offensive), individuals may seek contextual or environmental
cues to help interpret the behavior. In particular, individuals may be asking themselves why the behavior
was performed: To this end, information about the speaker’s relation to the social group–in terms of their
prior attitudes toward the group, as well as their membership or non-membership–might help to infer the
agent’s goal or intent (Baker et al., 2008).
Then, perceptions of whether the speaker had malicious or neutral intent can be expected to
derivatively impact evaluations of their offensiveness–as documented in abundant previous research on
the downstream effects of culpable mental states (Cushman, 2008; Malle & Knobe, 1997). This
mechanism can be depicted as a causal diagram in which ascribed intent mediates the effects of attitudes
and speaker-orientation congruence (see Figure 3).
Figure 3. Causal Model of Verbal Offense: The experimental effects of speaker-orientation congruence
and background attitudes on offensiveness are mediated by ascriptions of intent.
To evaluate this model, we aggregated the data from progressives and conservatives and employed
the recoded factors as independent variables: ideological fit (whether the statement coheres with the
participant’s political orientation or not), ingroup/outgroup (whether the speaker belongs to the
participant’s ingroup or outgroup), and attitude. Thus, the fit×outgroup interaction captures the selective
effect of outgroup derision, i.e., when outgroup members make ideologically discordant remarks.
As a preliminary step in our mediation analysis, we ask whether entering intentionality ascriptions
into a model of offensiveness ‘blocks’ the experimental effects of attitudes and (fit×outgroup) congruence.
In a model without intentionality ratings, the effects of attitudes, F(1, 2149) = 9.33, p = .002, and fit×outgroup
congruence, F(1, 2005) = 6.73, p = .010, on offensiveness were both significant. Then, entering intentionality
ratings into the model rendered both effects non-significant (i.e., attitudes: F(1, 2141) = 0.93, p = .33,
congruence: F(1, 2010) = 0.28, p = .59). In this same model, the effect of intentionality attributions was
highly significant, F(1, 2391) = 2051, p < .001.
We then conducted two separate mediation analyses with 5000 quasi-Bayesian simulations to
evaluate the direct and indirect effects of each experimental treatment. (1) The effect of background
attitudes on offensiveness was mediated by perceived intent (mediated/total = .76), ACME = 0.14, 95%
CI [0.06, 0.22], p < .001, leaving no direct effect from attitudes to offensiveness, p = .32. Similarly, (2)
the effect of congruence on offensiveness was mediated by perceived intent (mediated/total = 1.15),
ACME = 0.36, 95% CI [0.20, 0.52], p < .001, with no remaining direct effect, p = .57.
These results add
plausibility to the model depicted in Figure 3, and derive support for the broader thesis that contextual
elements drive perceived offensiveness as listeners spontaneously mentalize about the speaker’s intent.
Moderation by Explicit Norms
In our previous research, multiple approaches revealed that participants endorsed the speaker's
background attitude and denied the relevance of the speaker’s identity (Almagro et al., 2021), when asked
Both the outcome and mediator models also included the main effects of fit and membership, enabling us to test
whether there were indirect and direct effects of these two variables. The effect of ideological fit was direct (p <
.001) not indirect (p = .87), and the main effects of membership were weak and inconclusive (direct p = .12,
indirect p = .085).
to report on their reasoning processes. Additionally, our analyses repeatedly found no relationship between
participants’ overt principles regarding offensive speech and the influence they had on their particular
judgments of offensiveness.
In the present study, participants viewed the speaker’s background attitude as the most relevant
principle (replicating Almagro et al., 2021); which significantly exceeded the relevance of every other
principle in pairwise comparisons: identity, salience, and orientation.
Furthermore, the rank order of these
principles was the same among conservatives and progressives.
Motivated by the finding that intent ascriptions fully mediated the effects of both attitudes and
congruence, we reasoned that participants’ beliefs about the importance of speaker intent, and not identity,
might moderate these effects. In other words, the effects of attitudes and congruence on offensiveness
might be stronger among participants who consider a speaker’s intent (and not identity) relevant in
deciding whether a statement is offensive.
To evaluate this hypothesis, we conducted the mediation model at two different levels of the
moderator: low relevance (the 1st quartile), and high relevance (the 3rd quartile)–and qualitatively
examined the overlap in the confidence intervals. This exercise revealed that participants’ beliefs about
the relevance of speaker intent moderated the magnitude of both indirect effects, i.e., of background
attitudes and congruence (as shown in Table 4).
Speaker intent was more important for progressives (M = 5.51, SD = 1.34) than for conservatives (M = 5.18, SD
= 1.51). Speaker identity (M = 4.75, SD = 1.78; M = 4.53, SD = 1.79), topic salience (M = 4.44, SD = 1.69; M =
4.44, SD = 1.78), and statement orientation (M = 4.38, SD = 1.85; M = 4.40, SD = 1.76) mattered equally to
progressives and conservatives.
Table 4. Moderated Mediation Results: Indirect and Direct Effects at Low (First Quartile) and High
(Third Quartile) Values of the Moderator.
0.03 [-0.07, 0.12]
0.31 [0.18, 0.44] **
0.10 [-0.01, 0.21]
-0.04 [-0.17, 0.09]
0.19 [-0.00, 0.37]
0.57 [0.34, 0.80] **
-0.06 [-0.29, 0.17]
-0.07 [-0.35, 0.20]
0.40 [0.23, 0.58] **
0.26 [0.06, 0.45] **
-0.04 [-0.24, 0.16]
-0.10 [-0.34, 0.15]
Part 4. Active versus Inactive Debates
Finally, we analyze the differences in reactions toward the three active debates vs. the three
inactive issues. Statement orientation significantly interacted with topic salience in both progressive, F(1,
897) = 14.38, and conservative groups, F(1, 810) = 17.02, both ps < .001. When looking at contemporary
disputes, the effect of ideological fit was present among both progressives and conservatives: Progressives
viewed conservative remarks (M = 3.77, 95% CI = [2.74, 4.79]) as more offensive than progressive
remarks (M = 3.46, 95% CI = [2.44, 4.49], B = 0.30, t = 2.34, p = .020. Conservatives deemed progressive
statements more offensive (M = 3.78, 95% CI = [2.67, 4.89]) than conservative statements (M = 3.20, 95%
CI = [2.09, 4.31]), B = 0.59, t = 4.70, p < .001. Thus, we observed symmetry in relation to active debates–
whereby both ideological groups viewed discordant remarks as significantly more offensive than
When considering historical or inactive debates, the pattern was distinctly asymmetrical:
Progressives viewed regressive statements (M = 4.79, 95% CI = [3.76, 5.81]) as much more offensive than
progressive statements (M = 3.73, 95% CI = [2.71, 4.76]), B = 1.05, t = 8.18, p < .001. The effect of
ideological fit was absent among conservatives. If anything, the difference trended toward greater
opposition to regressive (M = 4.30, 95% CI = [3.19, 5.41]) than progressive (M = 4.14, 95% CI = [3.03,
5.25]) statements, B = 0.16, t = 1.30, p = .19. In other words, progressives found reactionary statements
more offensive than progressive statements, while conservatives judged them equally.
One explanation for this is that contemporary conservatives may not view regressive views on
historical debates as any more concordant with their ideology than progressive views on that set of issues:
It is in the nature of the historical debates included here that they have yielded a general consensus in
favor of the progressive view: For example, what was once the progressive stance toward divorce (i.e.,
pro-divorce) may now be the norm–embraced by progressives and conservatives alike. This line of
reasoning could explain why conservatives are not selectively offended by the defense of progressive
values on these historical disputes –what was once a paradigmatically conservative position is no longer
identified as a stance that is associated with their political ideology.
Recent calls to limit freedom of speech have reified certain stereotypes about conservative and progressive
worldviews: Progressives express greater reservations about the exercise of free speech, and conservatives
report greater support (Pew Research Center, 2016). This represents a striking reversal of the attitudes
conservatives and progressives held historically–when free speech rights emerged as a tool for government
criticism and dissidence.
In this chapter, we sought to contribute to our understanding of verbal offense, by examining who
sees speech as offensive (i.e., whether progressives and conservatives exhibit similar reactions), when or
under what circumstances, and why, through which specific mechanism.
Symmetry and asymmetry in offensive speech norms
Participants in our study were confronted with relatively subtle acts of verbal discrimination–of
the kind that blur the boundary between innocuous and offensive speech. In one case, a hypothetical
speaker stated that “many women claim that they do not suffer any harassment or discrimination”. This
statement, uncongenial to a progressive worldview, was–according to conservative participants–protected
by freedom of speech. In contrast, progressives considered statements of this kind offensive and
censurable, particularly when made by an outgroup member. In this way, our study was able to recreate
the attitudes that prevail in today’s cancel culture debate.
We experimentally reversed the orientation of several such offenses so they would constitute
affronts to a conservative worldview instead. e.g., “the results of implicit bias tests show that men have
more gender biases than they are aware of”. In a demonstration of ideological symmetry, progressives saw
this type of statement as an innocuous exercise of free speech, while conservatives perceived it as
offensive and deserving of censorship (once again, especially when coming from an outgroup member).
Recent surveys have found that the right purports to uphold freedom of expression, while the left
expresses greater reservations (Pew Research Center, 2021). Contrary to the idea that this constitutes a
principled difference between conservatives’ and liberals’ beliefs, our findings documented motivated
appeal to free speech rights among both conservatives and liberals: Participants upheld the freedom to
express messages concordant with their own worldview, but took offense at outgroup speakers whose
utterances could be construed as targeting their ideological convictions.
These results can be fruitfully explained by theories of identity-protective or cultural cognition
(Kahan, 2017; Kahan et al., 2021), and suggest that the current politicization of free speech reflects a
contingent fact: i.e., that most recent episodes of intergroup disagreement have involved conservative
affronts to progressive values (and not vice versa).
Another possibility, compatible with the previous one, is that most judgments within the culture
war are a way of expressing adhesion to a political ideology (Funkhouser, 2020; Ganapini, 2021; Williams,
2021). Maybe those who complain about “cancel culture” are mainly signaling their political identity.
This would mean that the central issues in the culture war are issues for which the parties are highly
affectively, but not ideologically, polarized (Bordonaba & Villanueva, 2018; Mason, 2018; Iyengar et al.,
Lastly, although the presence of offensive speech norms was symmetrical, their magnitude was
asymmetrical. For instance, the tendency to view outgroup speakers as more offensive when making
ideologically discordant remarks was significantly stronger among progressives than among
conservatives. Our final analyses indicated that this discrepancy was driven by the inclusion of historical
issues. These historical issues have, over time, been settled in favor of a liberal stance. From this
observation, two assumptions may follow: (1) that modern-day conservatives may not align ideologically
with the historically conservative stance, and (2) that modern-day progressives may imbue these historical
victories with much stronger moral conviction than their attitudes regarding ongoing culture wars. In this
regard, a model according to which a difference in ideological conviction is not accompanied by an actual
difference on how freedom of expression mediates our particular judgments of offensiveness can
accommodate evidence of asymmetry regarding historical social issues.
Cognitive mechanism: The role of intent
In our previous research, we observed that a speaker’s identity impacted participants’ perceptions
of offensiveness–though they then espoused the speaker’s intent as the primary guiding principle behind
their judgments (Almagro et al., 2021). The present work replicated both these findings, and offered a
tentative explanation of the seeming discrepancy.
When judging third-party behavior, we spontaneously form representations of the agent’s intent
(Knobe, 2003) and these representations have downstream effects on attitudes of blame (Cushman, 2008;
Kirfel & Hannikainen, 2022; Malle & Knobe,1997). Inspired by this literature, we recorded participants’
ascriptions of intent and learned that speaker intent played an important mediating role.
When considering an ideologically discordant statement, participants appeared to avail themselves
of contextual cues with which to inform their representations of the speaker’s intent (see Baker et al.,
2008; Gibson et al. 2019). For instance, the statement that children of divorced families are “statistically
more prone to delinquency and antisocial behavior” could be made either with neutral intent in order to
inform and describe, or with a harmful intent. In this predicament, information about who the speaker is
played a decisive role in disambiguating perceived intent among the target group (in this case,
progressives): The statement was seen as intended to harm when uttered by a conservative individual, but
intended to convey information when made by a fellow liberal; and this phenomenon too was symmetrical
across the political divide. In turn, perceived intent appeared to fully account for the effect of identity (and
background attitudes) on the degree to which such remarks were considered offensive.
These results could dissolve the apparent incongruity between participants’ behavioral focus on
identity, and their self-reported emphasis on intent (and disregard of speaker identity; Almagro et al.,
2021). Moreover, the process by which variation in speaker identity is encoded as differences in intent
(and derivatively in offensiveness) appeared to be within executive control, since participants’ self-
reported emphasis on intent did in fact determine the extent to which outgroup derogation was seen as
offensive (via ascriptions of intent).
Contrary to the dominant portrayal of progressives and conservatives, our study documented
ideological symmetry in their evaluations of offensive speech. When faced with utterances whose content
matters to us, and somehow threaten our ideological stance, both conservatives and progressives react as
if the utterances were in fact offensive. Support for free speech does not seem to have an impact on our
actual judgments of offensiveness. These processes can be interpreted as evidence of identity-protective
or cultural cognition in response to outgroup affronts to one’s values. To explain these results, we
advanced an exploratory model of offensive speech judgments in which representations of the speaker’s
guiding intent occupy a central role. Even against explicitly stated background attitudes, we tend to
attribute negative intent to the speaker whenever we deem their utterances to be offensive. In future work,
this model should be confirmed and extended to a broader range of linguistic phenomena.
Almagro, M., Hannikainen, I. R. & Villanueva, N. (2021). Whose words hurt? Contextual determinants
of offensive speech. Personality and Social Psychology Bulletin.
Baker, C. L., Saxe, R., & Tenenbaum, J. B. (2009). Action understanding as inverse planning. Cognition,
Bordonaba, D. & Villanueva, N. (2018). Affective polarization as impervious reasoning. In Philosophical
Perspectives. The 13th conference of the Italian Society for Analytic Philosophy: Italian Society
for Analytic Philosophy.
Cai, Zhenguang G., Rebecca A. Gilbert, Matthew H. Davis, M. Gareth Gaskell, Lauren Farrar, Sarah
Adler, Jennifer M. Rodd (2017). Accent modulates access to word meaning: Evidence for a
speaker-model account of spoken word recognition. Cognitive Psychology.
Crawford, J. T. (2014). Ideological symmetries and asymmetries in political intolerance and prejudice
toward political activist groups. Journal of Experimental Social Psychology, 55, 284-298. DOI:
Cushman, F. (2008). Crime and punishment: Distinguishing the roles of causal and intentional analyses
in moral judgment. Cognition, 108(2), 353-380.
Epstein, L., Parker, C. M., & Segal, J. A. (2018). Do justices defend the speech they hate? Journal of Law
and Courts, 6(2), 237–262.
Fasoli, F., Carnaghi, A., & Paladino, M. P. (2015). Social acceptability of sexist derogatory and sexist
objectifying slurs across contexts. Language Science, 52, 98–107.
Fourcart, Alice & Hartsuiker, Robert J. (2021). Are foreign-accented speakers that ‘incredible’? The
impact of the speaker's indexical properties on sentence processing. Neuropsychologia, 158 (30).
Funkhouser, E. (2020). A tribal mind: Beliefs that signal group identity or commitment. Mind &
Ganapini, M. B. (2021). The signaling function of sharing fake stories. Mind & Language.
Gibson, J. L., Epstein, L., & Magarian, G. P. (2019). Taming uncivil discourse. Political Psychology, 41,
Harper (2020). “A Letter on Justice and Open Debate” (Published 2020). Harper’s Magazine.
https://harpers.org/a-letter-on-justice-and-open-debate/ (Last access: 02/03/2022).
Hernández, E., Anduiza, E. & Rico, G. (2021). Affective polarization and the salience of elections.
Electoral Studies, 69. dio: https://doi.org/10.1016/j.electstud.2020.102203
Iyengar, S., Lelkes, Y., Levendusky, M., Malhotra, N., & Westwood, S. J. (2019). The origins and
consequences of affective polarization in the United States. Annual Review of Political Science,
22, 129–146. https://doi.org/10.1146/annurev-polisci-051117-073034.
Kahan, D. M. (2017). The expressive rationality of inaccurate perceptions. Behavioral and Brain Sciences.
Kahan, D. M., Jenkins-Smith, H., & Braman, D. (2011). Cultural cognition of scientific consensus.
Journal of Risk Research, 14(2), 147–174.
Kirfel, L., & Hannikainen, I. R. (2022). Why Blame the Ostrich? Understanding Culpability for Willful
Kneer, M., & Bourgeois-Gironde, S. (2017). Mens rea ascription, expertise and outcome effects:
Professional judges surveyed. Cognition, 169, 139-146.
Malle, B. F., & Knobe, J. (1997). The folk concept of intentionality. Journal of experimental social
psychology, 33(2), 101-121.
Mason, L. (2018). Uncivil agreement: How politics became our identity. University of Chicago Press.
Meesala, S. (2020). “Cancel Culture: A Societal Obligation or Infringement on Free Speech?” (published
2020). UAB Institute for Human Rights Blog.
infringement-on-free-speech/ (Last access: 02/03/2022).
Monroe, A. E., & Malle, B. F. (2017). Two paths to blame: Intentionality directs moral information
processing along two distinct tracks. Journal of Experimental Psychology: General, 146(1), 123–
O’Dea, C. J., Miller, S. S., Andres, E. B., Ray, M. H., Till, D. F., & Saucier, D. A. (2015). Out of bounds:
Factors affecting the perceived offensiveness of racial slurs. Language Science, 52, 155–164.
Pew Research Center (2016). In ‘political correctness’ debate, most Americans think too many people are
easily offended. https://www.pewresearch.org/fact-tank/2016/07/20/in-political-correctness-
debate-most-americans-think-too-many-people-are-easily-offended/ (Last access: 15/03/2022).
Pew Research Center (2021). How Americans feel about ‘cancel culture’ and offensive speech in 6 charts.
and-offensive-speech-in-6-charts/ (Last access: 15/03/2022).
Ramsay, A. (2021). “Culture wars: It’s the Right that is trying to cancel free speech” (published 2021).
Open democracy. https://www.opendemocracy.net/en/opendemocracyuk/culture-wars-its-the-
right-that-is-trying-to-cancel-free-speech/ (Last access: 02/03/2022).
Romano, A. (2021). “The second wave of “cancel culture”” (published 2020).
Swim, J. K., Scott, E. D., Sechrist, G. B., Campbell, B., & Stangor, C. (2003). The role of intent and harm
in judgments of prejudice and discrimination. Journal of Personality and Social Psychology, 84(5),
The Objective (2020). “A More Specific Letter on Justice and Open Debate” (Published 2020). The
debate/ (Last access: 02/03/2022).
Viciana, H., Hannikainen, I. R., & Gaitán, A. (2019). The dual nature of partisan prejudice: Morality and
identity in a multiparty system. PloS one, 14(7), e0219509.
Williams, D. (2021). Signalling, commitment, and strategic absurdities. Mind & Language.
Young L., Scholz J., Saxe R. (2011). Neural evidence for “intuitive prosecution”: the use of mental state
information for negative moral verdicts. Social Neuroscience , 6(3), 302–15.