Abstract
Political parties have incentives to appeal to diverse voters to maximize support. This study examines the effect of politicians exploiting dogwhistles—expressions that convey hidden messages to specific audiences while escaping notice by the majority—for this purpose. Findings from a word replacement test and a survey experiment assessing the utility of racial dogwhistles in Sweden suggest that dogwhistling can aid politicians in gaining support from in-groups while retaining backing from out-groups; for example, “suburban gang” (dogwhistle) elicits more support than “immigrant gang.” Results suggest that dogwhistling is effective beyond the U.S. context and for different forms of political support.
The term dogwhistling denotes conveying, often controversial or norm-breaking, messages through words or expressions, understood by a select group while going unnoticed by others (e.g., Fear, 2007; Goodin & Saward, 2005). For example, “academic freedom” can be taken by the general public to be an anodyne statement of a scholarly ideal—that scholars should be independent of state interference in their intellectual work—but in politically radicalized discourse, it signals the existence of a conspiracy to mislead the public by restricting the range of scholarly outputs considered to have expert approval.
In theory, using these types of multivocal expressions can enable politicians and other actors interested in influencing public opinion to communicate contentious messages to a subgroup without alienating the broader public, potentially bolstering voter support (Fear, 2007). However, the impact of dogwhistling on political support is empirically under-researched. In this article, we explore whether political candidates can garner public support by employing racially coded dogwhistles and whether such effects are specific to certain groups.
The question of whether political actors can use dogwhistles to convey hidden messages to subsets of voters carries important democratic implications. Robert E.Goodin (2008), in his work Inventing Democracy, calls dogwhistling “perverse,” asserting that it undermines democratic deliberation: “Try deliberating with someone who says one thing to you and, at the same time, says something different to others in the discussion” (p. 224). Political dogwhistling has also been claimed to erode the legitimacy of a democratic mandate, as it obscures voters’ understanding of their voting options (Goodin & Saward, 2005).
The modern history of dogwhistling can be found in presidential campaigns largely run through the mass media. In the post-civil rights era, when explicitly racist rhetoric was less socially acceptable in the United States, for example, Republican politicians appealed to “law and order” to reach groups opposed to the civil rights movement, without alienating other voters (Drakulich et al., 2020; Haney-López, 2014). In the contemporary high-choice media environment, dogwhistling often occurs in various social media platforms and forums (Krämer, 2017) but can also travel to major mainstream news media (Åkerlund, 2022).
Previous research suggests that politicians can garner support on contentious subjects by dogwhistling covert messages, alluding to, for example, race (e.g., Schutten et al., 2022; Wetts & Willer, 2019; White, 2007). However, previous work seldom ascertains whether the investigated dogwhistles indeed possess different meanings for different people (but see Albertson, 2015). To identify dogwhistle phrases that convey different messages to different groups, we conducted a Lexical Replacement Test (e.g., Arefyev et al., 2020). Through this test, we identified four phrases where different groups provided a different meaning (one group provided a neutral/noncontroversial replacement, whereas another provided a replacement with a controversial, racially derogatory reference).
These phrases were subsequently included in a survey experiment, where Swedish citizens were asked to evaluate political statements that featured either a dogwhistle phrase (e.g., “suburban gang”) or its corresponding overt counterpart (“immigrant gang”). While we observed some variation in the effects, overall, the results indicate that politicians can gain higher support for political statements by employing racially covert language, by amplifying support from outgroup members while maintaining support from their intended ingroup.
While prior research has demonstrated that racially loaded dogwhistles can mobilize voters with racial biases by activating their political allegiances without overtly mentioning race (Haney-López, 2014), these studies primarily focused on a limited set of dogwhistle phrases and come almost exclusively from the United States. By studying covert anti-immigrant phrases in Sweden, we explore whether the findings transcend different political contexts and phrases.
Dogwhistling as a Form of Strategic Political Ambiguity
Research suggests that a perceptual disagreement about parties’ policy positions between voter groups can benefit the party in terms of support, as more voter groups may perceive the party as being close to their policy preferences (Somer-Topcu, 2015). This perceptual disagreement may be achieved by being strategically ambiguous in the communication of policies and policy platforms.
For example, parties may present political positions with different ideological slants, leading to a heterogeneous or ambiguous overall political platform (Bräuninger & Giger, 2018). They can also choose to be vague or avoid making statements on controversial issues (Rovny, 2012; Tomz & Van Houweling, 2009). These strategies, however, carry risks in that they can alienate voters (Somer-Topcu, 2015, p. 844) or cause uncertainty, discouraging risk-averse voters from supporting the party (Bartels, 1986; Downs, 1957; Shepsle, 1972).
Here, we investigate whether dogwhistling can generate political support by exploiting the inherent potential for ambiguity in human language. The term dogwhistle is used in everyday discourse referring to a wide range of phenomena. On one end, it can refer to code language that is only understandable to a very select group but meaningless to most people, for example, the use of “88” referring to “Heil Hitler” (as h is the eighth letter of the alphabet) (Michael, 2009). On the other end, some meanings are obvious to everyone but to some extent deniable. A contemporary example is the expression “family values,” which is strongly associated with hetero- and cis-normativity and a conservative view of society, but which is still not explicitly stating that the speaker subscribes to these values.
There is also no clear definition of dogwhistles in the academic literature. Most work on political dogwhistles typically refers to coded language or symbols tied to, for example, racial, cultural, religious, or ideological topics, to appeal to the core identities of certain constituencies (Fear, 2007; Henderson & McCready, 2019; Hindess, 2014). These phrases or symbols theoretically have the characteristic of being inaudible to potentially alienated groups (Fear, 2007; Goodin, 2008; Goodin & Saward, 2005). Dogwhistles, however, often develop over time (Sayeed et al., 2024) such that their meaning becomes more available to persons outside the intended in-group; in other words, specificity of the dogwhistle decreases. In that case, deniability becomes an important aspect of the dogwhistle, because it allows the speaker to breach social norms without incurring public opprobrium. Consider, for instance, the term “inner cities,” often cited as a dogwhistle in American politics, implicitly alluding to African American neighborhoods (e.g., Khoo, 2017; Wetts & Willer, 2019). When a politician wishes to convey a negative message about African Americans, using an overt statement leaves no room for anyone to agree with it without violating social norms about racial equality. Employing a coded racial term like inner cities, on the contrary, creates ambiguity, making it unclear whether the speaker intends the implicit racist meaning or the explicit definition of inner cities as densely populated urban areas (Khoo, 2017; Mendelberg, 2001; White, 2007). In essence, using the term inner cities allows the speaker to breach social norms while maintaining plausible deniability. If accused of being prejudicial or intolerant, they could claim: “I didn’t mean that, I meant this instead” (Fear, 2007, p. 5).
While dogwhistles, in theory, are expressions that hold different meanings in different audience groups, scholars studying political dogwhistles in the context of public opinion tend to focus on the effects of using certain expressions on, for example, political support, to empirically identify dogwhistles, where an expression is considered a dogwhistle when generating particular audience responses (this is often tested in an American context, with racial prejudice among Whites being exploited to gain support for conservative policies, e.g., Wetts & Willer, 2019). This definition allows the concept to encompass a wide range of linguistic phenomena, such as hinting and innuendo, as long as the messages are to some extent deniable and used in a political context. Here, however, we focus on the function of dogwhistles as expressions that trigger different inferences among different groups—the specificity of the expression—leaving the issue of deniability as part of a potential later phase of the dogwhistle “life cycle” (Sayeed et al., 2024). In other words, we refer to dogwhistles as expressions that have different alternative meanings among different groups, thereby enabling the speaker to convey a certain (controversial) message to some that will be less likely to be acknowledged by others. Given that dogwhistles function by triggering different inferences, the existence of different understandings of the meaning of the expression in the population or context of interest must be confirmed empirically before any broader effects on political support (or related political or social behavior) can be reliably inferred as being “dogwhistle effects.” This empirical confirmation can be obtained by performing, for example, a Lexical Replacement Test (e.g., Arefyev et al., 2020).
In short, we see dogwhistles as messages that, instead of appealing to different constituencies by presenting heterogeneous political platforms or by refraining from taking a stance on politically polarized issues, provides possibilities for political actors to reach different constituencies by communicating a message that itself has heterogeneous meaning. Via the covert payload meaning, political parties can at the same time present seemingly clear positions to an electorate, while sending different signals about their positions to different voter groups.
It should be considered here that political dogwhistles, in real-life politics, operate in a complex interplay of individual and issue-level intervening factors. At the issue level, the relevance of dogwhistling likely hinges on the salience of the topic within public discourse and the existence of established social norms with accompanying controversy, such as racism or homophobia. At the individual level, given that dogwhistles are partly used to evade outrage from groups opposing a controversial message, their effectiveness should be sensitive to the salience of social and political identities (e.g., Schroeder & Figueroa-Caballero, 2024). As political issues are increasingly sorted into partisan identities (Mason, 2015), people may, first, be more aware of political dogwhistles when they are strongly invested in their partisan identity, likely making the dogwhistles less effective on strong partisans. Similarly, it is likely that those who consider themselves the political opponents of groups associated with dogwhistles have high motivation to be informed of coded language and therefore are more likely to catch adversarial dogwhistles (Gvirsman, 2014). In addition, research on motivated political reasoning indicates that individuals with strong political positions are more resistant to any type of arguments or information (e.g., Kunda, 1990), and factors like political sophistication may contribute to a diminished impact of dogwhistles, as individuals with deeper political understanding may be better equipped to discern coded political language. In essence, there are reasons to expect that those with strong prior beliefs, political identities, and political knowledge should be less vulnerable to dogwhistle messages.
To summarize, in theory, politicians can use dogwhistles to make the same statement to all voters, while subtly signaling to some groups that they agree with a controversial position. The challenge lies in the potential for unintended recipients to decipher the controversial message.
Previous Research on Dogwhistles and Political Support
Research from the United States suggests that dogwhistling can help political candidates appeal to certain voter groups. A common phrase examined in this literature is the inner cities example. Hurwitz and Peffley (2005) examined support for prison spending and found that White participants with racial resentment were more supportive when referring to the phrase “violent inner-city criminals” compared to just “violent criminals,” while the effect was weaker among Whites with more positive racial attitudes. Relatedly, White (2007) investigated how racial cues affected racial thinking among White and Black Americans. By manipulating fictional news articles, he found that implicit racial references like “inner-city families” activated racial thinking among White participants, whereas explicit references like “black and African American families” evoked racial thinking among Blacks.
In another study, Schutten et al. (2022) asked American voters to choose between fictitious political candidates with randomly varied characteristics related to the phrase “gun control”—another potential dogwhistle alluding to race in the United States. They found that voters with racial resentment were more likely to support candidates ostensibly funded by the National Rifle Association (NRA) and who opposed gun control, while less racially resentful voters were averse to candidates with the same characteristics. As the authors note, these findings imply that the racial connotations of gun control are broadly understood and thus backfire among nonprejudiced voter groups—in essence, the “covert” meaning of the phrase is too widely understood for it to function as an effective dogwhistle in the United States.
Wetts and Willer (2019) examined both the inner city and the gun control phrase through two experiments with White Americans. In the inner-city experiment, they tested whether implicit references to racial stereotypes such as “inner-city unwed mothers” and families trapped in a “culture of poverty” yielded higher support for welfare programs than explicit references, such as describing blacks as “lazy, uneducated welfare recipients.” The second experiment largely resembled the first, but with “gun control” serving as the implicit racial cue. They found support for the hypothesis that implicit racist appeals could influence public support for policies, but, contrary to their expectations, the effects were notable among racially resentful White liberals, a group among whom they expected no or negative effects.
Unlike the aforementioned studies that focused on racial appeals, Albertson (2015) studied the effects of a covert religious appeal: “I believe there is power—wonder-working power—in the goodness and idealism of the American people.” Through a survey experiment with two American voter groups: an “ingroup” (religious individuals who understood the religious appeal) and an “outgroup” (religiously diverse individuals who did not detect the religious connotations), she found that ingroup members were more positive to the political candidate when making overt and covert religious appeals compared to a nonreligious appeal. Members of the outgroup, instead, were negatively affected by the overt religious appeal but were not affected by the covert one.
In summary, prior research suggests that dogwhistling can enable politicians to appeal to selected subgroups of voters, whereas the impact on unintended voter groups (e.g., voters supporting racial equality) is more uncertain. This research provides important insights into the effects of dogwhistles, yet several questions remain unanswered. First, existing research has primarily been done in the United States, rendering it uncertain whether the effects apply in other political contexts. This study examines dogwhistle effects in Sweden, which diverges from the United States in several respects. Politically, the Swedish electoral system is characterized by proportional representation, involving competition between multiple political parties organized along a clear left-right dimension (Oscarsson & Holmberg, 2016).
Swedish political parties thus face a different competitive context than U.S. politicians, who operate within a two-party system. Swedish mainstream parties, like the Social Democratic party, must carefully navigate the middle ground to attract voters from both the left and right. For instance, if they focus on migration to appeal to right-leaning voters, they risk alienating their left-leaning base. In Sweden, dogwhistling thus theoretically enables “catch-all” parties to gain voters from one side without losing those from the other. In addition, Sweden enjoys consistently high voter turnout (between 80% to 86% since the 1990s; Statistics Sweden, 2023), while the United States struggles with lower turnout levels. Consequently, U.S. politicians may prioritize mobilizing their core supporters to vote over appealing to swing voters, which may encourage explicit and controversial messaging over implicit statements.
Another difference concerns the media system, where Sweden represents the democratic-corporatist model and the United States a prototypical liberal model (Hallin & Mancini, 2004). Sweden’s media landscape is also characterized by extensive and overlapping mainstream news use, strong public service broadcasting, and comparatively high media trust (Nord & Grusell, 2021; Syvertsen et al., 2014; Weibull & Jönsson, 2007). In contrast to the U.S. media, which have clear partisan divides in their readership (Jurkowitz et al., 2020), mainstream news outlets in Sweden are politically independent and not systematically ideologically biased in their news coverage (Johansson & Strömbäck, 2019), and they enjoy a broad public readership across ideological divides (Oscarsson et al., 2021, p. 13). These factors enable Swedish politicians to communicate broadly across different voter groups through the same mainstream news channels, potentially incentivizing the use of dogwhistling. In sum, the party system, voter turnout, and media landscape all speak in favor of dogwhistles playing a more prominent role in Swedish politics compared to the United States; however, other factors may also matter, such as political culture and rhetorical traditions.
Second, previous studies assume that certain expressions, like “inner cities” or “gun control,” have the linguistic features of a dogwhistle (i.e., a covert meaning understood only by some); yet these assumptions are rarely tested. An exception is Albertson (2015), who distinguished an ingroup (recruited based on religious identification indicated in a previous survey) from an outgroup (general U.S. citizens) based on their understanding of a religious dogwhistle phrase. Only individuals who expressed familiarity with the phrase “wonder-working power” and correctly mentioned its religious origin were included in the ingroup, while participants in the outgroup were excluded on the same basis. This approach confirms the multivocal nature of the dogwhistle phrase under examination but comes with an unfortunate caveat: it may have influenced the participants’ responses by drawing attention to the phrase before the experiment was conducted.
Third, prior studies have focused on a limited set of phrases, like the aforementioned “inner cities” and “gun control.” Our contribution is thus threefold: (a) we examine dogwhistle effects in a non-U.S. political and media context, (b) we investigate multiple dogwhistle phrases not previously investigated, which we (c) pretest for their multivocal meaning using a lexical replacement test with a different sample to avoid priming effects.
Hypotheses
Based on theory and previous research, we anticipate that political candidates using dogwhistles can garner support from diverse voters by conveying a controversial message that appeals to a specific subgroup while going unnoticed by the majority.
Theoretically, the effectiveness of dogwhistle communication lies in its precision—the ability to convey an implicit message to ingroups while avoiding detection among outgroups. Ideally, the covert meaning of a dogwhistle is understood by 100% of ingroup members but none of the outgroup members. In reality, however, such perfection is unlikely. Nevertheless, even if some members of the outgroup are aware of the alternative meaning, as long as these members are relatively few, we anticipate that the outgroup as a whole will perceive the dogwhistle as less objectionable than an explicit message. On the contrary, the impact on the intended ingroup is expected to be marginal compared to explicitly expressing the controversial message, as this group should be attracted also by such a statement.
We formulate three hypotheses about the effects of using dogwhistles on political support, differentiating between effects on the intended ingroup and outgroup. In defining ingroups and outgroups, we consider whether the groups are likely to be drawn to or repelled by and discern or fail to discern the covert meaning of the dogwhistle: intended ingroup = “likely to endorse and recognize the covert meaning,” and intended outgroup = “likely to find the covert meaning repulsive but unlikely to recognize it.”
We operationalize support for a political candidate using three indicators: (a) agreement with the statement, (b) intention to vote for the candidate, and (c) intention to endorse the statement online. 1 Agreement with the statement, or, more precisely, agreement with the position expressed in the statement, is a straightforward way of measuring the dogwhistle effect, and similar measures have been used in previous studies (e.g., Hurwitz & Peffley, 2005; Wetts & Willer, 2019). However, it is one thing that voters agree with a dogwhistle statement and another thing that politicians benefit politically from communicating it. We therefore also include vote intention (cf. Albertson, 2015; Schutten et al., 2022), which is a more concrete way of measuring political gains. While we acknowledge that voting solely based on a single statement, without further information about the candidate, differs from real election scenarios, vote intention serves as the primary indicator reflecting the anticipated rationalization for using political dogwhistles (that politicians strategically use dogwhistles to enhance their chances of being [re-]elected). In addition, vote intention represents a higher level of political commitment compared to lower-cost rewards such as endorsing statements on social media. To encompass a broad spectrum of potential support, we thus also include intentions to endorse the statement on social media (likes/shares). This, in turn, captures online engagement, which is gaining increasing interest in research on political participatory behavior (e.g., Gil de Zúñiga et al., 2012). Social media interaction can benefit political candidates in terms of for example attention, at the same time as political actors who want to convey controversial messages without being penalized for it (e.g., via content moderation) have been shown to engage in dogwhistle communication online (Bhat & Klein, 2020; Jhaver et al., 2018). A limitation of this indicator is that not everyone uses social media, another that people may like or share posts (or withhold doing so) for reasons other than (dis-)agreement with the content, such as what reactions they anticipate from other users.
Together, our three indicators cover various aspects of political support. Our theoretical expectations when it comes to dogwhistle effects are the same, however. The first hypothesis addresses overall support obtained by using a dogwhistle. The subsequent hypotheses address support obtained from tentative outgroups and ingroups, respectively:
In the case of H3, we expect that the use of a dogwhistle (as opposed to the explicit message) will not impact the degree of support from the ingroup members bcause these members will recognize the covert meaning and, hence, the meaning of the implicit and explicit message will be the same.
Method and Design
Hypotheses are tested through a survey experiment with Swedish citizens, comparing the effects of phrases with a covert racist appeal on support by an ingroup (identified as supporters of the anti-immigration party the Sweden Democrats, SD) and outgroup (not supporters of SD), to those of an explicit anti-immigration message.
Immigration has received a great deal of attention in Swedish politics and the news media lately (Johansson & Strömbäck, 2023). While Sweden was once seen as a liberal outlier when it comes to immigration politics (Demker & Odmalm, 2022; Green-Pedersen & Odmalm, 2008), along with the 2010 entry into the parliament and continued electoral successes of the SD, the Swedish political discourse has progressively become similar to that of other European countries in terms of a more negative stance (Dahlstedt & Neergaard, 2019; Elgenius & Rydgren, 2019). Yet, immigration remains a contentious issue in Swedish politics, as well as in the public (Demker, 2022). Politicians from mainstream parties and the SD alike thus have incentives to appeal to individuals with anti-immigrant positions when communicating their policies, while at the same time avoiding alienating other voter groups.
Before designing the survey experiment, we performed a lexical word replacement test to identify phrases that have the characteristics of racially loaded dogwhistles in Sweden.
Lexical Replacement Test
To identify phrases with an alternative racial derogatory meaning among some, and a neutral/nonracial meaning among the majority, we performed a Lexical Replacement Test, a method used in computational linguistics (McCarthy & Navigli, 2009) and Natural Language Processing applications (Arefyev et al., 2020) to empirically explore how people understand words and expressions. In 2021, we invited a subsample of 1,780 members from a web-based Swedish citizen panel, administered by the SOM Institute at the University of Gothenburg, Sweden. The invitees were prestratified on sex, age, and education to resemble the Swedish population. Of the invited, 900 participated in at least 80% of the survey, resulting in a net participation rate of 55%. Of those who participated, 17% were supporters of the SD, and the remaining 83% were supporters of some of the other seven parliamentary parties. The study was approved by the Swedish Ethics Review Authority on January 19, 2021.
Participants were presented with 12 sentences related to salient politicized topics in Sweden, including immigration. For each sentence, they were tasked with substituting a potentially multivocal word or phrase with one or a few other words while retaining the overall message. This method allowed participants to freely choose suitable replacement words while avoiding the disclosure of the implicit meaning conveyed by the dogwhistle (McCarthy & Navigli, 2009). The responses were manually coded using an iterative process. Three coders coded a smaller subsample of each word until acceptable intercoder reliability was achieved, after which one of the coders completed the rest of the coding. For details on the coding process, including the annotation scheme, see Online Appendix A.
Based on the results from the replacement test, four phrases were selected for the survey experiment, considering: (a) whether the words had different distinctive meanings across members of the sample, of which one had explicit racial derogatory connotations, and (b) whether supporters of the SD were more likely to replace the words/phrases with the explicit racist meaning when compared to non-SD supporters.
The first phrase from the replacement test chosen for the experiment is “suburban gang” [förortsgäng]—a term that featured in the SD’ 2018 election manifesto (Sverigedemokraterna, 2018). In recent years, Sweden has witnessed an increase in gang violence and shootings, primarily occurring in socio-economically disadvantaged suburbs of larger cities. Given that individuals with immigrant backgrounds tend to be overrepresented in these areas, the term “suburban gang” may trigger inferences about the ethnic composition of these gangs, akin to the American term “inner cities” (e.g., Hurwitz & Peffley, 2005). Our replacement test supported this suspicion. While the most common replacements did not explicitly refer to immigration (e.g., teenage gangs or gangs in socio-economically disadvantaged areas), “immigrant gangs” appeared frequently (21%). Dividing the sample into SD supporters (17% of the sample) and non-SD supporters revealed a noteworthy higher proportion of SD supporters using the replacement “immigrant gangs”; 49% compared to 16% among non-SD supporters.
The second word is “enrich” [berikar]. This term was initially used to express support for multiculturalism in Sweden and beyond; however, far-right political groups adopted it sarcastically, turning it into a derogatory phrase “cultural enricher” to implicitly convey hostility toward immigrants and ridicule people positive to multiculturalism (Åkerlund, 2022). While this derogatory usage emerged in far-right fringe forums online, it has since made its way into more mainstream discourse (Åkerlund, 2022). In Sweden, a Sweden Democrat politician and former MP, for example, defended using cultural enricher in an opinion piece published by the Swedish Public Service Television (Ekeroth, 2010). To avoid direct association with far-right jargon, we thus used a modified version, omitting “culture.” The sentence read: “It is enough to open a newspaper to get examples of how immigrants enrich Sweden.” Despite the unequivocally positive lexical meaning of the word enrich, 12.5% of all respondents replaced it with words or phrases implying that immigrants harm society, such as “destroy” or “ruin,” indicating an understanding of the word’s derogatory racial connotation. Among SD supporters, the proportion replacing enrich with a negative term was notably higher compared to non-SD supporters: 30% versus 10%.
The last two phrases selected for the experiment are “re-migration” [återvandring] and “good order” [ordning och reda]. In the replacement test, also these phrases were substituted with distinctly different alternatives, one of which had explicit racial connotations. In these cases, the tendency to replace the phrases with racial alternatives was however not as clearly linked to support for SD. The inclusion of two types of racial dogwhistles—some associated with support for an anti-immigration party and others less so—enables us to investigate whether effects on support are conditioned by the ingroup (outgroup) (not) understanding the racial connotations, or whether the mere fact that the word is multivocal alleviates negative effects on outgroups compared to the explicit message.
Starting with re-migration, its lexical meaning refers to immigrants’ voluntary return to their country of origin. However, in Sweden, it also carries an implied alternative connotation of enforced deportation. In a tweet posted before the 2022 national election by the SD politician Tobias Andersson, a picture of a tram adorned with SD advertisements accompanied the text: “Welcome to the re-migration train. You have a single ticket. Next stop, Kabul!.” In our replacement test, 34% of respondents substituted “re-migration” with “deportation” or equivalent, while the remaining respondents used phrases indicating a voluntary return to one’s home country. The inclination to use “deportation” as the replacement term differed only slightly between SD supporters and non-SD supporters: 40% versus 33.6%.
The fourth and final phrase is “good order” [ordning och reda]. Calls for “good order in migration policy” appeared during the 2015 refugee crisis when Europe, including Sweden, faced a substantial influx of refugees (Petersson & Kainz, 2017), and have persisted in Swedish political debate since then. What made us suspect this phrase to have multiple meanings was its usage among politicians with opposing positions on the immigration issue. For instance, during the 2022 National Swedish election campaign, “good order in migration policy” was used in official communication by both the pro-immigration Center Party (Centerpartiet, n.d.) and the SD (Sverigedemokraterna, 2022). In the replacement test, approximately 27% of all respondents replaced “good order” with terms implying a stricter immigration policy whereas most others replaced it with terms implying things like better structure and clarity. The share replacing it with restrictive policies was only marginally higher among SD supporters: 31% versus 26.6% among non-SD supporters.
Survey Experiment
A total of 3,200 members of the citizen panel at the SOM Institute, who had not participated in the replacement test, were invited to participate in the survey experiment. Of these, 1,948 completed at least 80% of the survey, resulting in a net participation rate of 61%. The study was fielded in November 2022, and the sample was prestratified on age and education to resemble the Swedish population. The study was approved by the Swedish Ethics Review Authority on January 7, 2021. Hypotheses and a preanalysis plan were registered at the Center for Open Science online platform on June 6, 2023, before the data were released to the authors (Lindgren et al., 2023).
Compared to population parameters, our sample closely aligns with the Swedish population in terms of sex and education but entails some underrepresentation of individuals under 30. In addition, supporters of smaller left-leaning parties are overrepresented whereas supporters of the three major parties are somewhat underrepresented compared to a high-quality probability-based survey conducted by Swedish Statistics. These imbalances limit the generalizability of our results to the overall Swedish population. However, since our focus is on identifying dogwhistle effects and mechanisms rather than the exact population dispersion of the effects, we do not apply any adjustments to address the imbalances. Detailed information on sample distribution and population benchmarks is available in Online Appendix B.
Each participant was asked to read four fictitious social media posts from anonymous political candidates and indicate their willingness to support the candidates. Three of the posts explicitly pertained to immigration, and one pertained to crime, with each one testing the impact of a different dogwhistle phrase. We employed a between-subject design with one experimental factor, “wording of post,” with two levels: (a) implicit racially charged dogwhistle and (b) explicit derogatory racial counterpart. For each post, participants were randomized to receive either the dogwhistle version or the explicit message. The posts were counterbalanced by randomizing the order in which they appeared to the subjects. Results from randomization checks are presented in Online Appendix C.
The four sentences are presented in Table 1. The dogwhistle phrases are in bold; their explicit counterparts are in italics.
Treatments.
Dependent Variables
We include three indicators of the dependent variable, political support. The first indicator, agreement with the statement, is measured by the question: “To what extent do you agree with the post?” Respondents answered on a 7-point scale ranging between 1 (Do not agree at all) to 7 (Agree completely). The second indicator, intentions to vote for the candidate, was measured by the question: “Based on the social media post alone, how likely is it that you would vote for the politician?” Respondents answered on a 7-point scale ranging between 1 (Not likely at all) to 7 (Very likely). The third indicator, intentions to endorse the postonline, is measured by the questions: “How likely is it that you would . . . (1) like . . . (2) share . . . the politician’s post on social media?” Respondents answered on 7-point scales ranging between 1 (Not likely at all) to 7 (Very likely), with an opt-out option of “I don’t use social media.” For the analysis, the responses were added to an index, normalized to range from 0 (Very unlikely) to 1 (Very likely). 2 People who chose the opt-out option were treated as missing.
Moderator variable: Ingroup/Outgroup
When defining the ingroup and outgroup, we considered the likelihood of members in the groups (a) endorsing the covert racial derogatory message of the dogwhistle, and (b) recognizing the covert message. In our study, we use support for the SD as a proxy for being an ingroup member and nonsupport for SD as being part of the outgroup. We do this for the following reasons: (a) SD is the only Swedish parliamentary party that campaigns primarily on the immigration issue and takes an unequivocal critical stance on the same, (b) have made use of all the four dogwhistles in their communication with voters, and (c) supporters of SD were the group most likely to recognize the racial connotations of the dogwhistle phrases in the replacement test.
We classify the respondents into SD- and non-SD supporters based on their responses to the question: “Which political party do you like the most today?” Response options included all eight Swedish parliamentary parties (The Left Party, The Green Party, The Social Democrats, The Center Party, The Liberals, The Christian Democrats, The Moderate Party, The SD), “Other Party,” and an opt-out option “Prefer not to answer.” The variable was re-coded into a dichotomous variable where 1 = SD-supporter (14.30% of the sample), and 0 = non-SD supporter (chose any of the other parliamentary parties or “other party”) (82.04% of the sample). Individuals who chose the opt-out option (3.92%) were coded as missing.
Analytical Strategies
Our first hypothesis (H1), which targets the main effect of dogwhistles on support, is assessed using two-sample t-tests of the difference in means between the treatment group (dogwhistle) and the group exposed to the explicit, racial message. One test is performed for each measure of support, resulting in 12 tests in total.
The second and third hypotheses (H2 and H3) inquire about conditional treatment effects depending on in-versus outgroup membership (operationalized as support/not support for the SD). These hypotheses are tested by ordinary least squares (OLS)-regression models, including support for the SD as an interaction variable. Specifically, for each measure of support, and each experimental trial (version of dogwhistle), we estimated a baseline model with only the experimental variable as the independent variable (Model 1), one model including SD-support and an interaction term (Model 2), and one model adding a series of selected pretreatment covariates to model 2 (Model 3). We include models with covariates in the tests of moderation, given that our moderator variable (support for SD) is a nonrandomized pretreatment variable, and we can therefore not rule out that it is some third variable, related to our moderator variable (e.g., attitudes toward immigration, sex) that conditions the treatment effects. For brevity, we present only the results from the second models in the manuscript (excluding covariates), but tables with results from all models (specified in the pre-analysis plan) are presented in the Online Appendix D (Table D1–D12)), including the formal models for the regression analyses, and a description of the covariates. All in all, the results are robust when controlling for covariates, but it should be noted that the level of statistical confidence differs slightly in a few cases. These are highlighted in footnotes in the results section along with the non-adjusted results. None of these discrepancies make any substantive differences in terms of interpretation of the outcomes and conclusions.
For further probing of H2 and H3, the marginal effects of the treatments for respectively SD-supporters and non-SD-supporters were assessed after each model 2.
Results
To test whether politicians can obtain more support from voters by using racially loaded dogwhistles (H1), we compare the means of the support measures between the groups exposed to overt racial messages and the groups exposed to the dogwhistle versions. t-test statistics (see Table 2) reveal that in most cases, the use of the dogwhistle message attracted higher total support from our participants than did the overt racial message. The finding was consistent across all measures of support in two of the four experiments (“suburban gang” vs. “immigrant gang” and “enrich” vs. “destroy”). Referring to “good order” instead of “restricting” when speaking about immigration politics generated a higher agreement with the post on average but did not differ in terms of vote intention or online endorsement. Referring to “re-migration” rather than “deportation,” finally, resulted in an average higher agreement with the post and vote intentions but did not impact the intention to like/share the post online. Hence, H1 is partially supported: it received consistent support concerning agreement with the post, in three of four cases concerning vote intention, but only in half of the trials for online endorsement.
Effects of Dogwhistles on Agreement With the Post, Vote Intention, and Online Endorsement.
Note. Entries are mean values with standard deviations in parentheses. Statistical significance of differences in means between treatment (dogwhistle) and control group (racial message) are obtained through two-sample t-tests.
p < .05. ***p < .001.
Different Support for Ingroups and Outgroups
To examine whether the higher support from the dogwhistle messages comes mainly from the outgroup (H2) while the ingroup remains unaffected (H3), we regressed each support measure on the treatment groups, adding an interaction term for treatment by support for the SD. The results are summarized in Table 3.
Effects of Dogwhistles by Ingroups and Outgroups on Agreement With the Post, Vote Intention, and Online Endorsement.
Note. Entries are unstandardized beta coefficients (B) with standard errors (SE). The estimates presented in this table come from nonadjusted models (covariates excluded). Results are all in all robust when controlling for covariates (see details in Online Appendix D, Tables D1–D12). The level of confidence deviates slightly in two cases; the interaction term for good order by SD-support goes from significant at a 90% confidence level to significant at a 95% level, whereas the opposite is the case for the interaction term for re-migration by SD-support regressed on online endorsement.
p < .1. *p < .05. **p < .01. ***p < .001.
The interaction terms for agreement with the post indicate that being exposed to the dogwhistle message has a different effect on ingroups and outgroups in the cases of “suburban gang,” “enrich,” and “good order.” When it comes to voting intention, the analyses suggest a difference in the dogwhistle effect among ingroups and outgroups in the case of “enrich,” whereas differences in treatment effects on online endorsement are indicated in the cases of “enrich” and “re-migration.”
To obtain a closer understanding of when and where ingroup and outgroup membership conditions the effects of the dogwhistles, we continue by probing the marginal effects of the treatment for SD and non-SD supporters. The predicted marginal effects are illustrated in Figures 1 to 3 (significance tests of all pairwise comparisons are presented in Online Appendix E).
Agreement With the Post: Marginal Effects of Dogwhistle for Ingroup and Outgroup
Vote Intention: Marginal Effects of Dogwhistles for Ingroup and Outgroup.
Online Endorsement: Marginal Effects of Dogwhistle for Ingroup and Outgroup.
Agreement With the Post
The results reveal support for H2 in all trials. There was a positive effect of the treatment among members of the outgroup in the case of “suburban gang” (t = 8.29, p < .001), “enrich” (t = 21.13, p < .001), “good order” (t = 6.58, p < .001), and “re-migration” (t = 4.99, p < .001) when compared to being exposed to the overt racial counterparts.
When it comes to effects among the ingroup, the hypothesis of no differences between implicit and explicit messages (H3) is supported in two of the four trials. There are no effects of being exposed to the term “suburban gang” versus the overt racial message “immigrant gang” (t = .82, p > .1), of being exposed to “re-migration” compared to “deportation’ (t = 1.41, p > .1), 3 or of being exposed to the phrase ‘good order’ compared to ‘restrict’ (t = .05, p > .1) on agreement with the post within this ingroup. In the case of ‘enrich,’ however, there is a negative effect on agreement when being exposed to the dogwhistle compared to the overt message (t = –5.69, p < .001).
Vote Intentions
The results reveal support for H2 in three of the four experimental trials. In line with the hypothesis, there was a positive effect of the treatment among members of the outgroup in the case of “suburban gang” (t = 3.75, p < .001), “enrich” (t = 14.67, p < .001), and “good order” (t = 2.23, p < .05). In the case of “re-migration,” however, there was no effect of being exposed to the dogwhistle compared to the overt message “deportation” among members of the outgroup (t = 1.01, p > .1).
When it comes to the ingroup, the hypothesis of no dogwhistle effect (H3) is supported in three of the four trials; “suburban gang” (t = .50, p > .1), “good order” (t = 1.59, p > .1) and “re-migration” (t = –.08, p > .1). In the case of “enrich,” there was an unanticipated negative effect on vote intention when being exposed to the dogwhistle compared to the overt message stipulating that immigration destroys Sweden (t = –4.88, p < .001). The implications of the latter finding are discussed in the concluding section.
Online Endorsement
The results reveal support for H2 in one case. There was a positive effect of the treatment on intentions to like/share the posts on social media among members of the outgroup in the case of “enrich” (t = 11.42, p < .001). However, there was neither an effect of “suburban gang” (t = 1.85, p < .1), 4 “good order” (t = 1.10, p > .1), nor of “re-migration” (t = 1.65, p < .1) in this group.
For the ingroup, the hypothesis of no dogwhistle effect (H3) is supported only in the case of “suburban gang” (t = 1.01, p > .1) when it comes to online endorsement. In this case, there was a positive effect among ingroup members of being exposed to the phrase “good order” compared to “restrict” (t = 2.79, p < .01) and of “re-migration” compared to “deportation” (t = 2.49, p < .05), whereas there was a negative effect of being exposed to the term “enrich” compared to “destroy” (t = –3.97, p < .001).
Concluding Discussion
Scholars have raised concerns about how political dogwhistling can undermine representative democracy by obscuring politicians’ real policy intentions (Goodin, 2008; Goodin & Saward, 2005). Our study in Sweden on anti-immigration dogwhistles generally aligns with previous findings from the United States, suggesting that dogwhistling can indeed be a powerful tool to gain politically by communicating different messages to different voter groups. We found that using implicit racial messages, as opposed to explicit ones, increased overall voter willingness to agree with a political statement, vote for a candidate, and endorse the statement on social media.
Our results thus confirm that dogwhistle effects extend beyond the U.S. context, specifically, it seems to apply also in a European multiparty, and public service-oriented media system context. The study furthermore utilizes a novel method for identifying the multivocal nature of dogwhistles. Unlike most previous studies on dogwhistle effects, which did not empirically test the multivocal nature of the phrases they investigated, our research empirically demonstrates it beforehand and with a different sample of individuals to avoid priming effects. This strengthens the conclusions we can draw on the specific role that the multivocal nature plays when dogwhistles impact people’s reactions to political appeals; we encourage future studies to consider using this, or similar methods to verify their experimental treatments.
Looking at the results from the study, there was a general pattern, suggesting that implicit anti-immigration dogwhistles gained more outgroup support compared to explicit anti-immigration statements, while ingroup support in most cases remained roughly equal. However, some specific results are worth discussing. A dogwhistle that stood out in the analyses is “enrich,” where the implicit message received significantly less support than the explicit one among ingroup members, who otherwise did not distinguish between the two. One possibility is that the tentative ingroup—SD-supporters—took the term enrich literally, not necessarily because they were unaware of its alternative derogatory application in relation to immigration but due to a lack of further contextual information. While all social media posts used in the experiment lacked information about the candidate’s political affiliation, the sentence containing “enrich” also lacked a general sentiment, unlike the other treatments, which referred to negatively tainted topics (e.g., public violence [suburban gang] and integration problems (good order)).
Another explanation could be that too few of the ingroup respondents picked up on the implicit derogatory message. The replacement test showed that few outgroup members substituted “enrich” with terms like “destroy,” whereas this was three times more common among the ingroup (10% versus 30%). Nevertheless, this implies that 70% of ingroup members may not be aware of the implicit derogatory meaning behind the statement “Immigrants enrich Sweden.” This highlights the importance of maintaining a balance between specificity and sensitivity for effective dogwhistling. As sensitivity increases (the ingroup’s ability to decode the implicit meaning increases), specificity tends to decrease (more outgroup members detect the implicit meaning)—an inevitable trade-off for those who intend to use dogwhistles as a means to maximize political support.
Another finding worth mentioning is that the effects of the dogwhistles differed between indicators of support. While we did not formulate separate hypotheses for agreement with the statement, vote intention, and online endorsement, comparing the outcomes on these indicators offers interesting insights. First, they yield slightly different results on H1, that dogwhistles generate higher overall support. Agreement with the statement is the only indicator consistently supporting H1. If politicians aim toward maximizing agreement with a claim across a diverse constituency, then dogwhistles thus appear beneficial. When it comes to voting intention, the prediction held in three out of the four cases (enrich, suburban gang, and good order), whereas for online endorsement, it only held in two cases (enrich and suburban gang). When it comes to online endorsement, the failure to support H1 in two cases could potentially be the result of a so-called floor effect; overall, only a few respondents expressed a willingness to share or like any of the posts, potentially suppressing the difference between implicit and explicit messages. Interaction with social media, however, also constitutes a different type of political support compared to voting intention and statement agreement, in the sense that it is public. Hence, the extent to which (a) the derogatory term is perceived as controversial and (b) the extent to which the derogatory racial message is perceived to be plausibly deniable in the implicit version, may both matter more for people’s responses on this indicator than on the other two.
Another point we want to raise concerns the fact that we included two different types of dogwhistles in the experimental study: discriminatory dogwhistles, where the covert meaning was understood mainly by the ingroup (“suburban gang” and “enrich”), and nondiscriminatory dogwhistles, where the understanding of the implicit message was less distinctly distributed between ingroup and outgroup members (“good order” and “re-migration”). Here, the results suggest a dogwhistle effect in several instances also for the two nondiscriminatory dogwhistles. This implies that the multivocal nature of the words, which gives plausible deniability, itself can leverage support. The role of plausible deniability when it comes to dogwhistle effects on political support is worth investigating more closely in future studies.
This study holds implications for political journalism and journalists. While politicians may use dogwhistle expressions strategically, we posit that the dogwhistle effect is independent of the speaker’s intentions; its impact hinges on how the audience interprets the expression. Consequently, dogwhistles may have repercussions when disseminated by actors like journalists reporting on interviews or covering statements, speeches, or debates from prominent politicians—with or without their knowledge of the double meaning. Furthermore, as the “fourth estate,” mass media and journalists bear the responsibility of scrutinizing political rhetoric and alerting voters to instances where a political statement may carry double messages. Given the multivocal and ambiguous nature of dogwhistles, whether communicated by a political actor or indirectly disseminated through secondary channels (Haney-López, 2014), it is imperative for journalists and other actors involved in communicating political news and information to the public, to be aware of the dogwhistle phenomenon, stay updated on current dogwhistle expressions, and understand their potential impact on the public.
Before concluding, some limitations of this study should be noted. First, while the lexical replacement test is valuable for identifying potential dogwhistle candidates, it has limitations. It may not fully capture unconscious word associations that may arise in real-life situations. Moreover, some respondents might consciously avoid expressing implicit meanings; some may be hesitant to exhibit fluency in anti-immigrant or right-wing populist vocabulary, others might avoid it to express refusal of the alternative meaning. In addition, the method may overlook multiple interpretations of words; as respondents are restricted to providing only one interpretation, alternative associations, less salient to the respondent at the moment of the test, might not be captured.
Second, there was a considerable time gap between the replacement test (spring 2021) and the survey experiment (fall 2022), potentially allowing shifts in public understanding of the dogwhistle phrases. Third, the word replacement test shares common biases with most surveys on controversial topics, such as social desirability bias, that may hinder respondents from replacing the words with phrases with explicit racial derogatory meanings. Fourth, as a limitation on external validity, party affiliation is a well-known factor conditioning peoples’ reactions to political appeals (e.g., Armstrong & Wronski, 2019). Participants in our experiment viewed brief social media posts from fictional political candidates without any information on the candidate’s background or affiliation; it thus remains unclear how foreknowledge of the candidate’s profile would alter the effects.
Fifth, while there are likely individual-level factors that condition the effectiveness of dogwhistles such as political interest and sophistication, as well as issue-level factors such as salience and social norms, our study design focused on assessing effects on a general level. Future research should investigate how such factors affect the political utility of dogwhistles.
Finally, we recognize a distinction between internally endorsing provocative statements and expressing them publicly, which may be central to which effects we can see of dogwhistles. In the current experiment, the smaller effect found for the “social media endorsement” indicators of support, compared to the “vote intention” and “agreement with the statement” indicators, could potentially be attributed to the public visibility of the social media actions. However, our current experimental design does not allow us to conclusively determine whether this is the case. The conditional role of publicity could thus be explored in future research, for example, by measuring endorsement of a statement, or voting intentions, while experimentally varying whether the action is made publicly or anonymously.
In conclusion, this study contributes to the dogwhistle literature by demonstrating that (a) racially loaded dogwhistles are effective beyond the U.S. context, (b) these effects apply to different forms of political support (particularly agreement and voting intent), and (c) they occur in cases of both exclusive ingroup understanding (discriminatory dogwhistles) and wider audience comprehension (nondiscriminatory dogwhistles). Future research should explore these effects in different political settings and when more information about the dogwhistler, like party affiliation, is known.
Supplemental Material
sj-docx-1-jmq-10.1177_10776990241280373 – Supplemental material for Coded Appeals and Political Gains: Exploring the Impact of Racial Dogwhistles on Political Support
Supplemental material, sj-docx-1-jmq-10.1177_10776990241280373 for Coded Appeals and Political Gains: Exploring the Impact of Racial Dogwhistles on Political Support by Elina Lindgren, Björn Rönnerstrand, Gregor Rettenegger, Ellen Breitholtz, Robin Cooper and Asad Sayeed in Journalism & Mass Communication Quarterly
Footnotes
Acknowledgements
The authors are grateful to the SOM Institute at the University of Gothenburg, for their assistance with data collection, to the participants of the 2023 APSA panel Racial Cues and Dog Whistles, and to the editor and anonymous reviewers of Journalism and Mass Communication Quarterly for valuable feedback on earlier versions of this manuscript.
Author Note
A pre-analysis plan is available at the project’s Open Science Framework page (Lindgren et al., 2023).
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors gratefully acknowledge funding from the Marianne and Marcus Wallenberg Stiftelse (grant no: MMW 2019.0214).
Supplemental Material
Supplemental material for this article is available online.
Notes
Author Biographies
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
