As one of the most important issues for understanding election outcomes and political conflict in Western democracies in recent decades (Green-Pedersen and Otjes Reference Green-Pedersen and Otjes2019; Kriesi et al. Reference Kriesi2012), immigration is often considered a highly moralized topic. With statements about how ‘illegal immigration is destroying the fabric of our country’ (Trump Reference Trump2018) or, conversely, how ‘the people at the heart of this issue (…) deserve to be respected’ (Gambino Reference Gambino2015), politicians cast the issue as one that concerns fundamental beliefs about right and wrong, thus signalling its political gravity and their own righteousness. Yet, our understanding of the conditions that lead parties to claim the moral high ground on immigration is limited. In this article, we argue that moral language plays a central role in party competition over immigration and that parties’ incentives to (de)moralize the topic are driven by strategic concerns shaped by the political opinion climate.
Existing research on party competition over immigration is dominated by a focus on how the entry onto the political scene of extreme anti-immigrant challenger parties has resulted in mainstream party repositioning in an effort to win back votes (Abou-Chadi Reference Abou-Chadi2016; Abou-Chadi and Krause Reference Abou-Chadi and Krause2020; Akkerman Reference Akkerman2012; Bale et al. Reference Bale2010; Han Reference Han2015; Lutz Reference Lutz2019). This research has delivered important insights into the competitive pressures behind the right turn on immigration seen in several contemporary party systems. What has not been systematically considered so far is that strategic concerns are also likely to affect how parties communicate about the issue, since by playing up the moral significance of immigration and framing parties and their policies in a moral light, parties can attempt to assert themselves against competitors.
Understanding the political pushes and pulls of moralization is important, not least due to the divisive risks involved. While moral rhetoric has been shown to be an effective tool for bolstering policy support and mobilizing constituencies (Enke Reference Enke2020; Jung Reference Jung2020; Luttrell, Philipp-Muller, and Petty Reference Luttrell, Philipp-Muller and Petty2019; Simonsen Reference Simonsen2024), the activation of moral thinking in audiences is also associated with stronger antipathy and distance toward attitudinal out-groups (Clifford Reference Clifford2019; Garrett and Bankert Reference Garrett and Bankert2020; Ryan Reference Ryan2017; Simonsen and Bonikowski Reference Simonsen and Bonikowski2022), leading researchers to voice concern for ‘a poisonous cocktail of othering, aversion, and moralization [that] poses a threat to democracy’ (Finkel et al. Reference Finkel2020, 533). These concerns are particularly pertinent to ‘culture war issues’ such as immigration, which appear to generate ever deeper divides in many contemporary Western publics (Skitka et al. Reference Skitka2021). Since political actors play a central role in driving these trends, there is a need to understand the circumstances that make them more likely to moralize.
The study of politicians’ use of moral language has recently received significant interest and delivered important insights that we build on and test in the context of immigration discourse (Bos and Minihold Reference Bos and Minihold2022; Brady et al. Reference Brady2019; Kraft and Klemmensen Reference Kraft and Klemmensen2023; Lipsitz Reference Lipsitz2018; Neiman et al. Reference Neiman2016a, Reference Neiman2016b; Walter Reference Walter2020; Wang and Inbar Reference Wang and Inbar2021). However, with its theoretical focus on party-level factors associated with the use of moral language, this research potentially overlooks factors at the systemic level shaping parties’ rhetorical incentives. In addition, as empirical studies are typically single-case based and examine relatively short timespans, they offer limited variation on systemic factors, thus precluding assessments of how the political climate shapes parties’ moralization and restricting our ability to evaluate the cross-national and longitudinal generalizability of results. Finally, this work tends to analyze political discourse with a broad-brush perspective, meaning that neither the topics of debate nor the concrete employment of moral arguments are taken into account, raising questions about the transparency of results and introducing the risk that results are driven, for instance, more by a highly moralized topic taking over (or disappearing from) the agenda than by fundamental changes in the rhetorical terms of political debate.
Our study meets these challenges, first, by theorizing the role of elite polarization on immigration in driving parties to moralize, and, second, by developing and validating dictionaries to detect moral rhetoric in political discourse across five different languages and applying these dictionaries to a large corpus of immigration-related parliamentary speeches from eight Western democracies over up to six decades. Combining multilevel statistical regression based on computational text analysis with in-depth, inductive coding of highly moral immigration speeches, we offer transparent, cross-national, and rich insights into parties’ moralization of immigration, considering party and party-systemic factors.
Our findings show that as parties move further apart on immigration – and elite polarization on the issue thus increases – they all tend to make greater use of moral language. By contrast, we find no support for hypotheses derived from existing work on moral rhetoric concerned with party-level factors: extreme and centrist parties, anti- and pro-immigrant parties, and government and opposition parties are equally likely to use moral language on immigration. Our qualitative analysis reveals that moral rhetoric is used overwhelmingly by parties to attack one another, highlighting each other’s moral flaws. Together, these findings serve as a corrective to popular notions and academic theories associating particular parties with a stronger drive to moralize and instead underline the importance of the political climate in shaping rhetorical competition: a polarized political climate reduces parties’ concerns about the divisive risks involved in employing moral language and instead incentivizes them to draw sharp, uncompromising boundaries between their own goodness and the moral corruptness of their opponents. As a consequence, all parties contribute to making immigration a moral battlefield.
Existing Evidence on the Use of Moral Language in Political Communication
Recent research has spotlighted rhetorical style as a competitive element in party behaviour (Bischof and Senninger Reference Bischof and Senninger2018; Bonikowski and Gidron Reference Bonikowski and Gidron2016; Kosmidis et al. Reference Kosmidis2019; Osnabrügge, Hobolt, and Rodon Reference Osnabrügge, Hobolt and Rodon2021; Valentim and Widmann Reference Valentim and Widmann2021; Widmann Reference Widmann2021), with studies of moral rhetoric – language that appeals to fundamental beliefs about right and wrong – delivering important insights as part of this agenda. We start by synthesizing this work to formulate hypotheses about parties’ moralization of immigration. Common to earlier research – whether concerned with immigration specifically (Lamont, Park, and Ayala-Hurtado Reference Lamont, Park and Ayala-Hurtado2017; O’Brien Reference O’Brien2003; Esses, Medianu, and Lawson Reference Esses, Medianu and Lawson2013; Krzyżanowski Reference Krzyżanowski2020; Akkerman and Hagelund Reference Akkerman and Hagelund2007) or political communication more generally (Brady et al. Reference Brady2019; Lipsitz Reference Lipsitz2018; Wang and Inbar Reference Wang and Inbar2021) – is its emphasis on party-level factors, suggesting that parties with specific policy positions and political status are more likely to moralize: for parties that are anti-immigrant, extreme, and in an oppositional position, moral rhetoric should offer a strategic edge because it suits their messages particularly well, in turn prompting them to use it more frequently.
Focusing on policy-positional differences, case studies of immigration discourse have been preoccupied with anti- rather than pro-immigrant politicians’ communication.Footnote 1 Central to this work is its focus on the strategic employment of moralized metaphors to dehumanize immigrants, elevate the moral standing of their voter base, and present themselves as heroes capable of saving the country from moral decay (Akkerman and Hagelund Reference Akkerman and Hagelund2007; Esses, Medianu, and Lawson Reference Esses, Medianu and Lawson2013; Krzyżanowski Reference Krzyżanowski2020; Lamont, Park, and Ayala-Hurtado Reference Lamont, Park and Ayala-Hurtado2017; O’Brien Reference O’Brien2003). Although pro-immigrant parties can also use moral language – for instance, to emphasize moral obligations toward those in need and to recognize the moral worth of immigrants (Schall Reference Schall2022) – the overwhelming attention devoted to moral arguments in anti-immigrant communication suggests that this language is more characteristic of the anti- than the pro-camp, possibly because it contributes to sustaining the ‘moral panic’ narrative of anti-immigrant actors (Krzyżanowski Reference Krzyżanowski2020). That anti-immigrant parties should be more likely to contribute to (or even drive) the moralization of immigration resonates with media debates about the value-laden ‘us-them’ rhetoric of these politicians and how it serves to create moral boundaries between majority nationals on the one hand and threatening immigrant counter-citizens on the other (Gambino Reference Gambino2015; Rucker Reference Rucker2019). The first party-level hypothesis therefore states:
H1 Anti-immigrant parties are more likely to use moral language than pro-immigrant parties.
Whereas H1 is specifically derived from case studies of the immigration domain and does not assume a general tendency for conservative/right-wing parties to moralize more than liberal/left-wing ones, the following two hypotheses derive from general studies of moral rhetoric, which we apply to our topic: immigration.
The second hypothesis focuses on the association of moral claims with extreme standpoints. Examining campaign advertisements from the US in 2008, Lipsitz (Reference Lipsitz2018) demonstrates how candidates for the US presidency over the course of the election campaign tone down their use of language appealing to ideologically distinct moral values. She interprets this as evidence that distinct moral language is suited for communicating extreme policy standpoints to a party’s more extreme home base, but that politicians must moderate their rhetoric (that is, deemphasize distinct moral statements) as the audience broadens to the entire electorate and they become more concerned with appealing to the (less extreme) median voter. Given that Lipsitz finds no equivalent increase in politicians’ appeals to other (out-party-distinct) moral values, the overall effect of the rhetorical moderation pressure is a general drop in moral rhetoric.
The proposition that ideological extremity and moralization go hand-in-hand is corroborated by moral psychology research showing that individuals with high levels of moral conviction tend to hold more extreme attitudes about the issue in question (Garrett Reference Garrett2019; Skitka Reference Skitka2010), including survey experimental evidence demonstrating that exposure to moral rhetoric can drive audiences to take more extreme stances (Clifford Reference Clifford2019; however, see Simonsen and Bonikowski Reference Simonsen and Bonikowski2022). Parties with more extreme immigration positions might, therefore, use moral rhetoric to appeal to extreme voters or even seek to drive the opinions of potential supporters in a more extreme direction, in greater alignment with the party stance. This leads to the second party-level hypothesis:
H2 Parties with an extreme issue position are more likely to use moral language than parties with a centrist issue position.
While the first two hypotheses focus on party positions, the last perspective based on existing research concerns political power or status. As with H2, this hypothesis derives from general studies of moral rhetoric, which we supplement with recent literature on (anti-immigrant) challenger parties. Starting with the former, Wang and Inbar (Reference Wang and Inbar2021), also in the US, examine tweets over a two-year period and congressional speeches over thirty-six years to show that politicians are more likely to use moral language when they are in opposition. The authors argue that this finding is evidence that political actors ‘use moral language when they need to mobilize their supporters to regain a majority’ (p. 15), because moral language is effective in elevating group standing (Jung Reference Jung2020; Lamont, Park, and Ayala-Hurtado Reference Lamont, Park and Ayala-Hurtado2017) and increasing message diffusion (Brady et al. Reference Brady2017, Reference Brady2019).
Moreover, moral language holds the potential for shifting the terms of debate from policy accomplishments (toward which parties in government can more credibly claim to have contributed) to the more fundamental question of who is in the right or wrong; a potential that seems particularly fruitful for parties in an underdog or challenger role (Bonikowski and Gidron Reference Bonikowski and Gidron2016). This line of reasoning is corroborated by recent research examining (extreme anti-immigrant) challenger parties’ potential for disrupting ‘politics as usual’ by pushing governments on topics such as immigration through the use of anti-establishment rhetoric (De Vries and Hobolt Reference De Vries and Hobolt2020; Kollberg n.d.; Williams and Hunger Reference Williams and Hunger2022). Such rhetoric, employed to blame and shame ruling elites, is often considered to be highly moralized (Bonikowski and Gidron Reference Bonikowski and Gidron2016; Widmann and Simonsen Reference Widmann and Simonsen2024). In sum, the third hypothesis states:
H3 Opposition parties are more likely to use moral language than parties in government.
Together, the three hypotheses based on existing research on political uses of moral rhetoric contribute to the notion that extreme, anti-immigrant challenger parties are particularly likely to moralize immigration because they can use this rhetoric to draw moral boundaries against immigrants, mark out their extreme stances, and elevate their standing from an underdog position by challenging those in power.
The Polarization–Moralization Nexus: Considering the Political Climate
While existing research resonates with our proposition that parties’ use of moral rhetoric is driven by party-competitive concerns, we argue that by associating moral language with a particular party type, this work has tended to overlook the importance of the broader political context that surrounds parties and shapes political competition over a topic, with implications for their rhetorical incentives. In addition, this research typically emphasizes the political potentials rather than the pitfalls of moral rhetoric, meaning that the other side of the coin – party incentives to demoralize – has not been clearly theorized. Therefore, before proceeding to test H1–H3, we add a fourth hypothesis that focuses not on party-level factors (the focus of H1–H3) but on how the political climate around a political issue – specifically, the degree to which political elites are polarized or not – affects all parties. In developing this hypothesis, we take a step back to consider the semantic properties of moral language, which we combine with knowledge of the risk factors associated with activating moral thinking documented in moral psychology studies.
As a type of language that frames political issues and policy positions in light of fundamental beliefs about right and wrong, moral rhetoric speaks in absolutes (Lakoff Reference Lakoff2002). This stands in contrast to non-moral language that engages in practical, pragmatic, and consequentialist reasoning (Jung Reference Jung2020). Once moral claims are invoked, there is no middle ground; only ‘antagonistic meanings’ (Lowe Reference Lowe, Hitlin and Vaisey2010). A somewhat overlooked and undertheorized implication is that moral language draws boundaries not only between what is right and wrong but also – implicitly or explicitly – between who is ‘good’ and who is ‘evil’ (Lamont Reference Lamont1992). We argue, therefore, that it is strategic for political parties to use (only) in situations where they have clearly identified opponents that they need to assert themselves against. In such cases, taking the moral high ground can be an effective means of mobilizing a sense of righteousness and moral superiority among the homebase (Jung Reference Jung2020). Importantly, this argument holds implications for situations where moralization is not strategic: under more consensual conditions, the drive to moralize should be dampened, since the boundary-drawing invoked in moral statements leaves little room for negotiation and accommodation and risks pushing away out-party supporters.
The argument that parties run a risk when using moral rhetoric is sustained by moral psychology research linking moral thinking to affective polarization: when holding attitudes with moral conviction, people are less willing to compromise on their opinions (Ryan Reference Ryan2017), they make strong, negative judgements of and show greater antipathy toward those who do not share their morally convicted views, while feeling even more strongly committed to their own group and its positions (Clifford Reference Clifford2019; Garrett and Bankert Reference Garrett and Bankert2020; Lamont Reference Lamont1992; Ryan Reference Ryan2014; Skitka Reference Skitka2010). Casting their disagreements with other political actors in a moral light, parties might seek to mobilize party supporters by emphasizing their own moral superiority (in contrast to their competitor’s moral corruptness) and the importance of standing behind the party’s position. By doing so, however, parties also risk pushing away other segments of the electorate (Simonsen Reference Simonsen2024; Simonsen and Bonikowski Reference Simonsen and Bonikowski2022), implying that they should only be willing to engage in moralization when there is minimal chance of winning over the supporters of other parties, which is the case under conditions of elite polarization where parties’ positions are already far apart.
In sum, when polarization on a topic is high, parties will know that there is little chance of persuading out-party supporters, and they will be more concerned with mobilizing their homebase: in this situation, moral rhetoric becomes strategically advantageous because parties can assert themselves against opponents by drawing boundaries between right and wrong, good and evil. While parties run the risk of spurring antipathy between partisan camps, they are likely to be willing to run this risk (only) when party positions are already far apart. Our fourth hypothesis about the political climate therefore states:
H4 When elite polarization on the issue increases, all parties are more likely to use moral language.
Methods
To evaluate the four hypotheses, we use computational text analysis based on a novel set of multilingual dictionaries developed for this study to establish rhetorical patterns in large amounts of political communication. Supplementing this analysis, we add transparent insights into the concrete applications of moral rhetoric based on an in-depth qualitative analysis. This combination of evidence aims to strike a balance between offering cross-nationally generalizable and longitudinal evidence to test the study’s hypotheses while simultaneously overcoming the pitfalls of broad-brush approaches to analyzing political text.
Data
We built a dataset of immigration-related parliamentary speeches in Austria, Canada, Denmark, Germany, the Netherlands, Sweden, the UK, and the US over up to six decades. By including eight Western democracies, we can address a central limitation of existing work since our study is better able to assess whether there is general support for the theoretical hypotheses across countries that vary in the make-up of their political and party systems. Beyond the variation in political systems, the eight countries differ in the size of the immigrant population and the types of immigrants they host, but the important point for our purposes is that immigration has been politicized to the extent that the topic has been central to party competition, resulting in large amounts of text from all countries devoted to debating the topic.
As mentioned, we analyze parliamentary speeches, an ideal text source for evaluating the use of moral language in party competition for several reasons. First, they represent how politicians communicate during political debate when directly faced with political opponents and their messages, meaning that strategic concerns in their communication are likely to be particularly central. The debate character of this text source is fitting for the study since our theorization emphasizes rhetorical pressures on parties to assert themselves against opponents. Such pressures are present in parliamentary speeches, making them similar to political debates on TV or social media communication, where parties can confront each other’s statements, whereas party communication that is more one-directional (such as it is on parties’ homepages or in their manifestos) may differ from the patterns documented here.
A second reason for analyzing parliamentary speeches is that they have been shown to be a crucial means by which parties are granted visibility in mass media (Proksch and Slapin Reference Proksch and Slapin2012; Tresch Reference Tresch2009). Thus, although parliamentary speeches are not always followed intensively by the public, the potential for quotes and clips from these debates of being covered in national newspapers and TV is likely to make parliamentarians conscious of an external audience.
Finally, parliamentary speeches allow us to study moral rhetoric over a long time span, which is central for ensuring within-country variation in polarization (to test H4), as well as variation in the party-level variables testing H1–H3 (both across and within parties). As shown in Table 1, the text data for some countries covers six decades, while two to three decades for others. Importantly, the data for all countries includes the period during which immigration has been most intensely politicized and debated (starting in most European countries and Canada in the 1990s and in the USA in the early 2000s) (Card et al. Reference Card2022; Gagnon and Larios Reference Gagnon and Larios2021; Green-Pedersen and Otjes Reference Green-Pedersen and Otjes2019).
Coverage and sources of text data by country
To identify speeches concerned with immigration, we followed prior research by using an immigration keyword string (Gessler and Hunger, Reference Gessler and Hunger2022; Ruedin and Morales Reference Ruedin and Morales2019); that is, a list of immigration words expected to be present in political speeches on immigration. Focusing on general immigration words that would reliably identify the topic across time, we included speeches in our corpus if containing three or more of the keywords (for example, three times ‘immigr*’ or three different words from the list). The English-language string is listed below (Online Appendix A contains strings in the other languages):
immigr* OR migr* OR asyl* OR refug* OR foreigner* OR ‘guest worker*’
To focus on party actors, we include only speeches given by regular Members of Parliament (MPs).
Development of Multilingual Moral Dictionaries
To measure the extent to which the speeches in our corpus contain moral rhetoric, we adopted a dictionary approach in line with previous research in the field (Bos and Minihold Reference Bos and Minihold2022; Brady et al. Reference Brady2017; Frimer Reference Frimer2020; Hackenburg et al. Reference Hackenburg, Brady and Tsakiris2023; Lipsitz Reference Lipsitz2018; Neiman et al. Reference Neiman2016b; Walter Reference Walter2020). The dictionary approach to computational text analysis is highly transparent because one can identify the words in a text that led the model to classify it as moral. In addition, dictionaries are computationally fast and cheap and easy to apply for other researchers, ensuring replicability and expanding opportunities for using our tools in other settings and studies.
Enabling the cross-national analysis of moral rhetoric, we developed multilingual moral dictionaries, taking point of departure in the English-language moral dictionary created by Jung (Reference Jung2020). We optimized Jung’s dictionary to ensure its precision, coverage, and nuance by removing stemming and adding additional words with clear moral connotations, among other things (for a detailed description of the steps taken, see Online Appendix B; all dictionaries are available in the replication materials, see Simonsen and Widmann Reference Simonsen and Widmann2025). Based on this optimized English dictionary, we then engaged with professional linguists who translated it into Danish, Dutch, German, and Swedish. The linguists, all native speakers of the relevant language, were instructed to offer the most direct, moral translation in their language, including all grammatical extensions. This task necessarily involves human judgement and culturally specific knowledge of the moral connotations of words, which we highlighted as important skills in the translation. In addition, paralleling our efforts at optimizing the English dictionary, we asked the linguists to add moral synonyms and word associations to the dictionary in their language. Together with our native- and multilingual-speaking team members, we quality- and cross-checked the dictionaries to ensure their completeness and precision as well as a high level of cross-linguistic comparability. This involved, for instance, multiple read-throughs of all dictionaries to check that the translated words were clearly morally connoted as well as comparisons across dictionaries to add synonyms identified in one language that were applicable to other languages.
To validate the performance of the multilingual dictionaries, we relied on crowd-coding (Benoit et al. Reference Benoit2016). Engaging with the crowd-coding platform Prolific and the survey company Epinion, we asked crowd-coders from the countries in focus of our study to evaluate the moral appeal of sentences from political communication in their language, summing up to almost 13,000 sentences in total. Online Appendix C describes the crowd-coding process and the different quality control measures we implemented. Previous studies using computational text analysis to measure moral rhetoric rarely present performance metrics, but, using the crowd-coded sentences, we can compare the classification resulting from the dictionaries to how most humans understand moral language. Table 2 presents common quality measures, including recall, precision, and F1 scores. Recall is the ratio of correctly predicted observations to the total amount of true observations, thereby indicating the number of false negatives. Precision is the ratio of correctly predicted observations to the total predicted observations, indicating the number of false positives. The F1 score is defined as the harmonic mean of recall and precision. To calculate performance metrics, we turned the crowd-coding and dictionary scores into binary variables. For the crowd-coding, a sentence is considered moral if a majority of crowd-coders judged it to be moral. For the dictionary, a sentence is considered moral if it includes at least one moral word from the dictionary.
Performance metrics for the dictionaries
As discussed in greater detail in Online Appendix C, our dictionaries achieve significantly higher performance scores than the original English-language moral foundations dictionary (Garten et al. Reference Garten2018; Graham, Haidt, and Nosek Reference Graham, Haidt and Nosek2009). This holds specifically for the precision scores, indicating that while our tools may be somewhat conservative in predicting moral rhetoric, they do so correctly to a large extent. Additionally, as we show in Online Appendix E (Model J; Figure E.7), the measure of moral rhetoric produced by our dictionaries is distinct from affective or emotional language. All in all, we believe that the multilingual moral dictionaries we have developed offer reliable tools to analyze moral rhetoric in a large, linguistically diverse cross-national set-up; something that prior research has been unable to do.
Applying the multilingual dictionaries provides a score representing the proportion of moral words in a text document. To give insight into the type of language identified as moral, consider the following examples drawn from our corpus of immigration speeches:
They [criminal foreigners] damage our security and cohesion and the belief in a peaceful and law-abiding life in Denmark’ (DK15, Conservative Party, 2017, moral score: 0.24).
[H]elping refugees is the right thing to do’ (UK20, LibDem, 2018, moral score: 0.13).
These sentences score high on the measure of moral rhetoric with the dictionaries identifying 24 per cent and 13 per cent, respectively, of the words in each sentence as moral compared to a mean of 2.5 per cent moral words across all immigration speeches in the corpus. Keeping in mind that moral words are relatively rare, the sentences illustrate how just a few moral words colour a sentence or speech and transform a statement from a pragmatic claim to a matter of normative judgement. Imagine, for instance, in the second sentence that the LibDems had exchanged the word ‘right’ for ‘reasonable’ or ‘efficient’. With the word ‘right’, the party engages in rhetorical boundary-drawing (implying that not helping refugees is ‘wrong’) and appeals to value-laden judgements of what one ought to do.
Statistical Analysis: Variables and Modelling Approach
We first present the statistical analysis used to test the study’s four hypotheses, after which we offer qualitative insights on the concrete application of moral language based on in-depth coding. In the statistical analysis, our main outcome variable, the moral language score, is the mean proportion of moral words in a party’s immigration-related parliamentary speeches for each year. We calculate this score by taking the proportion of moral words in each speech, then averaging across all speeches delivered by a party’s MPs in each year to produce the annual mean proportion for each party. Online Appendix D offers descriptives of the key variables included in the analysis.
To evaluate H1 and H2 concerning parties’ immigration position and positional extremity, we rely on data from the Comparative Manifesto Project (CMP) (Volkens et al. Reference Volkens2021). The CMP covers all the countries included in the study and the complete time period of the text data, which is in contrast to expert assessments such as the Chapel Hill Expert Survey. Based on hand-coded quasi-sentences from party manifestos, the CMP provides scores for party attention to various political issues, including whether their communication on the topic is negatively or positively valanced. To evaluate a party’s expressed position on a topic, the common approach is to subtract its positive attention from the negative attention it devotes to the topic.
To measure party immigration position, then, we follow prior studies to construct a pro-anti-immigration scale (Abou-Chadi Reference Abou-Chadi2016; Careja Reference Careja2016; Helbling, Reeskens, and Wright Reference Helbling, Reeskens and Wright2016; Schmidt and Spies Reference Schmidt and Spies2014). Pro-immigration positioning includes an anti-national way of life and pro-multiculturalism (CMP items 602 and 607). Anti-immigration positioning includes a pro-national way of life and anti-multiculturalism (CMP items 601 and 608). To arrive at a party score, we subtract a party’s pro-immigration positioning from its anti-immigration positioning, meaning that negative values indicate more pro-immigration stances, whereas positive values indicate more anti-immigration stances. The mid-point (taking the value of ‘0’) indicates a neutral stance.
Party positional extremity is measured using the immigration position scale but folding it at zero so all values are positive (or zero in the case of a party expressing pro- and anti-immigration stances in equal measure) and higher scores indicate a more extreme position in either direction (mathematically, this amounts to first squaring the immigration position score and then taking the square root).
The government-opposition variable is a binary indicator of whether a party is in government in a given year. In years of government change, the variable is 1 if a party was a member of the government for more than half of the year, 0 otherwise.
To measure elite polarization on immigration policy standpoints we use the immigration position variable described above. We follow the common approach in the literature to, for each year, calculate the standard deviation of immigration positions across parties, resulting in a measure of the spread of policy standpoints in a country in a given year (Ezrow Reference Ezrow2007). We use an unweighted polarization measure (that is, not weighted by the vote/seat share of parties), because even small parties (with few seats in parliament) may contribute to expanding ‘the menu of policy choices’ (Ezrow Reference Ezrow2007, 186) and influence political discourse, in turn affecting the rhetorical incentives of other parties. In Online Appendix E, we use a seat-weighted polarization measure to test robustness as well a more expansive measure of elite polarization that includes other cultural issues such as gender, LGBTQ+ issues, and nationalism in addition to immigration.
Finally, we add two control variables to ensure that any potential effects of the hypothesized explanatory variables on moral rhetoric are not driven by some unobserved third factor. We therefore control for the salience of immigration in a party system, as the attention given to the topic might affect parties’ immigration positions as well as parties’ likelihood to use particular rhetorical strategies such as moralization. Salience is calculated by taking the proportion of immigration-related speeches to the total number of parliamentary speeches in each country in each year. We also include a dummy variable indicating whether a given year is an election year since proximity to an election can potentially influence politicians’ positioning and communication (Lipsitz Reference Lipsitz2018).
Because most explanatory variables are only available on a yearly (or less frequent) basis, all variables are aggregated to the year level. In the case of less frequent data (variables based on the CMP data), we use linear interpolation to fill the gaps between data points (Zeileis and Grothendieck, Reference Zeileis and Grothendieck2005). Following the logic of filling in the gaps between data points, we similarly extend the first and last data points of each data series with two years using linear interpolation to predict each variable’s values in the preceding and following two years. To enable the easy comparison of effect sizes, all variables are standardized (mean 0, standard deviation 1).
Since the hypotheses concern within-country developments, we use a country-fixed effects OLS regression. We also test the robustness of our findings in various ways. First, we use a categorical variable (instead of a scale) to test the influence of parties’ immigration positions (distinguishing between pro, anti, and moderate parties). Second, we use the two alternative polarization measures (using weighting and including other cultural issues) mentioned above. Third, we test two alternative model specifications: one applying party-fixed effects to assess H1–H3 based only on within-party developments (disregarding variation across parties) and one applying a one-year lag to the dependent variable to gauge the time sequence of potential effects. Fourth, we examine whether the patterns change with the introduction of television broadcasting or social media, since – as we have emphasized in our discussion of the usefulness of parliamentary speeches – the existence of an (implicit or explicit) audience is likely to reinforce the competitive logic behind the hypotheses. Finally, we add a control for sentiment to ensure that the findings hold when considering the potential overlap between highly emotive and moral language. As we will describe where relevant below, none of these alternative specifications alter the conclusions we draw based on the main model (full models in Online Appendix E).
Findings
Figure 1 presents the results of a country-fixed effects regression of party moral rhetoric in relation to immigration on the four variables testing H1–H4: immigration position (H1), positional extremity (H2), government-opposition (H3), and elite polarization on immigration (H4), along with controls for salience and election year.
When do parties use moral rhetoric?
Note: Based on a country-fixed effects regression on moral rhetoric in parliamentary immigration speeches. Lines around point estimates represent 95 per cent confidence intervals. All variables are standardized.
While we do not find support for the party-level hypotheses (H1–H3), we find evidence to support the notion that the political climate shapes parties’ incentives to moralize (H4). As Figure 1 illustrates, when parties move further apart on immigration, they engage more in moralizing the topic (support for H4). The effect size for polarization corresponds to a 0.43 per cent average increase in moral language across parties when polarization increases from its minimum to its maximum levels, amounting to one additional moral word in a short 230-word paragraph (approximately the length of this paragraph from beginning to end), or slightly more than three additional moral words in a typical immigration speech (with an average length of 735 words). While one additional moral word per paragraph may not appear to be much, this result should be considered in light of the significant change of tone that just one moral word can make to an argument, as illustrated in the example sentences from the methods section. Prior research has found political advertisements containing a few moral words to elicit stronger emotional reactions than advertisements without moral language, in turn making the political message more memorable to audiences (Lipsitz Reference Lipsitz2018). The rhetorical impact of adding three moral words to a speech, then, is likely to be noticeable, not least because moral words are otherwise a relatively rare encounter, making up only 2.5 per cent of the words in our corpus of immigration speeches.
The finding for polarization indicates how clearly distinct issue positions do not dissolve the need for parties to employ moral rhetoric; quite to the contrary, it seems that interparty disagreements make room for moralization since parties can contrast their own moral superiority against the moral failings of their opponents. Conversely, when parties begin to move closer to each other (that is, when polarization wanes), they become less likely to moralize, possibly because moralization – due to the antagonisms it invokes – strains political collaborations and risks igniting antipathy between political camps, as well as backlash from out-party supporters (Clifford Reference Clifford2019; Garrett and Bankert Reference Garrett and Bankert2020; Lamont Reference Lamont1992; Ryan Reference Ryan2014, Reference Ryan2017; Simonsen and Bonikowski Reference Simonsen and Bonikowski2022; Skitka Reference Skitka2010). Whether using the weighted polarization measure or the more expansive cultural polarization measure, this finding replicates as seen in Online Appendix E (Models B and C; Figures E.1 and E.2).
Turning to the three unsupported hypotheses, Figure 1 shows that anti- and pro-immigrant parties are equally likely to use moral rhetoric (lack of support of H1). It appears, then, that anti-immigrant parties are not in fact more likely to moralize than pro-immigrant parties, despite the intensity with which research has examined the strategic use of moral arguments in the discourse of the anti-camp. This finding should motivate research to devote more attention to understanding how moral rhetoric is used by pro-immigrant parties.
Adding to the picture, parties with more extreme immigration positions are no more (nor less) likely to use moral language when engaging in debate about the topic than parties taking a more centrist approach to immigration (lack of support for H2). This finding goes against the idea that extreme parties use moral rhetoric as a signalling tool to express and potentially push voters in the direction of their more extreme stances, whereas more moderate parties must tone down appeals to notions of right and wrong. Instead, the null finding falls in line with recent research demonstrating the inability of moral rhetoric to make audiences to take more extreme positions (Simonsen and Bonikowski Reference Simonsen and Bonikowski2022), suggesting that extreme parties cannot rely on moral boundary drawing as a tool for shaping public opinion in their favour. In effect, this removes the strategic edge they could have gained from using moral rhetoric in comparison with more moderate parties. The null findings for H1 and H2 replicate when using a categorical variable to measure party positions as well as under the party-fixed effects specification (see Online Appendix E, Models D and K; Figures E.3 and E.8), showing that there is no change either in the moral language used by parties that move their immigration position in a more (or less) anti-immigrant or extreme direction.
Beyond the lack of differences in the level of moral rhetoric used by parties across the immigration position spectrum, we find government and opposition parties to be equally likely to moralize immigration (lack of support for H3). This means that moral language is not only, or primarily, used by parties outside of the corridors of power; government parties are equally driven to appeal to notions of right and wrong in their communication. We replicate this finding in the party-fixed effects specification, indicating that parties losing (or winning) government power do not change the degree to which they moralize immigration (see Online Appendix E, Model D).
In sum, the lack of support for H1–H3 serves as a corrective to the notion that extreme, anti-immigrant opposition parties drive the moralization of immigration. This conclusion holds even when excluding system-level variables from the model (see Online Appendix F, Table F.1) as well as across all levels of elite polarization: exploring the interaction effects of each of the party-level factors and the system-level factor of elite polarization on immigration, we find that most of the party-level factors’ association with moralization remain statistically insignificant, with the exception of a small, negative interaction effect for the immigration position variable in the opposite direction of H1. In other words, there is no scenario in which anti-immigration parties use more moral rhetoric on immigration than their pro-immigration counterparts (see Online Appendix F).
In addition to the test for H1–H4, we find that one of the control variables, increased topic salience, contributes to parties’ use of moral language to the same degree as polarization. This result might suggest that the ‘poisonous cocktail of othering, aversion, and moralization’ that researchers have expressed concern about (Finkel et al. Reference Finkel2020) is particularly likely to be set in motion when immigration is both high on the agenda and parties begin to move apart on the issue (we note that salience and polarization are distinct factors, only weakly correlated with a mean correlation across countries of 0.17). Recall, in this connection, that salience too is a party-system variable, further underlining the need for considering factors in the political climate when theorizing parties’ rhetorical incentives.
Considering the robustness check for television broadcasting, it does indeed appear that the competitive dynamics are more pronounced when parliamentary speeches become occasions for speaking directly to the public (Gennaro and Ash Reference Gennaro and Ash2023). As can be seen in Online Appendix E (Models F and G; Figure E.5), while none of the hypotheses find support in the period before television broadcasting, polarization (but none of the other variables) is a significant predictor of moralization in the period after. We cannot rule out other time-variant factors that co-occur with the introduction of television broadcasting that might also explain this difference, but the fact that polarization is more strongly associated with moralization in the later rather than the earlier period suggests that parties’ strategic behaviour in terms of the rhetoric they use is muted as long as they are ‘screaming into the void’. However, when they have the opportunity to echo across the crowd, they are likely to become more attentive to the rhetoric they use. We see no similar moderation effect for social media, indicating that the advent of social media did not contribute further to the publicity effect beyond television broadcasting.
Finally, while the nature of our design prevents a strict test of the causal direction of effects, and it is theoretically possible that polarization and moralization co-occur or that moralization leads to further polarization, the lagged dependent variable specification reported in Online Appendix E (Model E; Figure E.4) examines whether the results from our main model replicate when the hypothesized explanatory variables are to predict moral language use a year later, rather than instantaneously. This model produces similar results to those reported in Figure 1, offering at least initial support for a time-sequence interpretation of the relationship between polarization and moralization.
Qualitative Insights on the Uses of Moral Language
Having evaluated the study’s hypotheses, we now turn to inductively generated qualitative findings offering insight into the concrete uses of moral language. In this part of our study, we zoom in on speeches high in moral language to understand how this rhetorical style is employed and shapes the political arguments being made. To do so, we took a random sample of twenty highly moral immigration speeches from each country (160 in total) and, through an inductive analytical approach, developed a coding list to identify the uses of moral rhetoric (details on our procedure can be found in Online Appendix G). The uses we identified include evaluative statements about parties and their policies in the form of a) moral attacks on political opponents, condemning these parties and/or their policies or questioning their moral capabilities, and b) moral self-praise by which parties present themselves or their policies as morally good. Among the uses not concerned with party-focused statements, parties make c) claims about what is morally right or wrong in relation to immigration, such as in statements about immigrants’ (un)deservingness, and d) moral calls to take political action on immigration, often presented as a way to redress a morally untenable situation. Finally, parties use moral rhetoric to e) justify political decisions on immigration, f) condemn, or g) praise non-party political actors, such as state leaders from other countries or civil society organizations.
Table 3 gives an overview of the percentage of speeches coded per category (for coding examples and sub-codes, see Online Appendix G). Within one speech, moral language can be used for multiple purposes (why the total of the right-hand column does not sum to 100). For instance, a politician may open their speech by attacking an opponent party for being morally wrong, then make a contrasting statement praising the good moral qualities of their own party.
Uses of Moral Rhetoric in Immigration Speeches
Note: Based on qualitative coding of a sample of 160 highly moral immigration speeches (20 per country). Each speech may contain multiple codes (uses of moral rhetoric).
Examining Table 3, we first note that both of the semantic properties discussed in our theorization play out in the immigration speeches: parties use moral language to talk about what is right and what is wrong as well as to draw boundaries between who is good and who is evil (as seen in their attacks on other parties and other political actors, and in their praise of themselves or other actors). In addition, parties draw implications of their moral statements, either by calling for action to remedy a moral wrong or by using moral language to justify actions taken. Existing research has tended to be vague with regards to how moral language is used or theorized it mainly as a device for parties to highlight their values on a topic. However, our qualitative analysis indicates that moral language is often more confrontational and that moral judgements of other party actors may be equally if not more central to understanding the role of moral rhetoric in political discourse. We start by considering uses of moral rhetoric for making claims about right and wrong before we take a closer look at the party-focused and more confrontational uses.Footnote 2
About half of the highly moral speeches contain statements proclaiming, for instance, that ‘[w]e have a responsibility towards refugees. We do not have the right to treat them this way’ (CA18, New Democratic Party, 2011) or that an ‘unbelievable abuse of asylum is being carried out’ (AT16, FPÖ, 2007). Such statements about what is right and what is wrong in relation to immigration send signals about what, in the message sender’s opinion, being in the right on immigration entails, and, as such, they are a logical precondition for engaging in more direct moral-rhetorical struggles with opponents. In addition, the rhetorical clarity and punch of such statements – speaking as they do in absolutes (Lakoff Reference Lakoff2002) and underlining the importance of the topic – are likely to be part of a more general attempt by a party to raise its profile on immigration. As demonstrated by Brady et al. (Reference Brady2017, Reference Brady2019) in the US context, moralized discourse is picked up by audiences and shared with others to a higher extent than non-moral statements, suggesting that parties that frame their messages in a moral light have a strategic advantage in subsequent moral battles because their messages have already had greater reach.
Moving on to the confrontational uses of moral rhetoric, which we define as those in which parties explicitly position themselves against each other, either through highlighting their own policies or deeds as morally good or through moral attacks on opponents and their policies, the most striking pattern to note is the dominance of uses related to moral attacks on political opponents. A substantive majority of the speeches in our highly moral sample (79 per cent, to be exact) fall into this category, making speeches containing attacks around twice as frequent as speeches with moral self-praise. The high prevalence of moral attacks is significant because these speeches put the emphasis on the ‘evil’ of other parties and thus use moral rhetoric to negatively draw a boundary to the other side.
That parties use moral rhetoric more often to condemn other parties than highlight their own righteousness may be interpreted as a reflection of the fact that moral rhetoric is more likely to be used when parties move further apart on immigration (H4). Under the condition of increasing polarization, voters are likely to be aware of parties’ distinct positions, and the strategic concern is, therefore, less about emphasizing where they stand and more about why other parties are wrong. In this case, attacking opponents serves to play up the moral significance of picking the ‘right’ side. Moral language, then, transforms disagreement into a conflictual situation.
To illustrate this interpretation, we zoom in on a 2018 speech from the Alternative für Deutschland in which it attacked the governing SPD–CSU/CDU coalition (and, more generally, all established political parties: ‘all of you, from the Left Party to the FDP’) by questioning their moral capabilities to protect those in true need:
You have ensured that the truly persecuted in this country are once again exposed to persecutors from their home countries. You have perverted the right of asylum, turned it into a system of mass immigration and thus turned asylum seekers’ homes into crime hotspots. Are you actually proud of having perverted a policy that was once created to protect genuinely persecuted people? (DE13, AfD, 2018)
Claiming that the government coalition, with the policies it had implemented, had in fact ‘perverted’ the values supposed to guide asylum policy, the speech switches the terms of debate from the pragmatic business of discussing policy accomplishments (which were seemingly successful in that the government coalition had ‘ensured’ something) to the more fundamental question of whether such policies were based on the ‘right’ values and thus whether the government was righteous. The AfD’s rhetoric suggests that parties should be measured not on whether they are able to implement policies (that is, ‘handle’ the issue) but on whether they are concerned with doing good. As such, the example illustrates the usefulness of moral rhetoric in a divided political situation, because it has the rhetorical potential to circumvent other central logics of party competition (for example, issue competence) (Bonikowski and Gidron Reference Bonikowski and Gidron2016).
A counterattack from 2019 indicates that moral language exerts an escalatory bind on all parties: Once immigration discourse has been moralized and other parties begin to engage in moral attacks, it seems that the only rhetorically viable path forward is to respond in the same register. Here, the CDU/CSU framed the conflict between the coalition and the AfD as a moral one by expressing explicit distrust in the sincerity of the AfD:
You are not interested in talking about refugees and their fate. Nor are you concerned about the host society. You are concerned with injecting poison into society through a false narrative (…) the people who are here with us deserve dignity and decency and respect, and that is what you lack towards the people. (DE17, CDU/CSU, 2019)
In an effort to (re)establish moral superiority over the opponent, the CDU/CSU transform their disagreement with AfD into a conflict between good and evil, leaving voters with a binary choice and a clear message about which side to pick.
Occurring less frequently than moral attacks but still appearing in many of the highly moral immigration speeches, moral self-praise (occurring in 40 per cent of the highly moral speeches) and moral calls for action (occurring in 39 per cent of the highly moral speeches) are employed to highlight a party’s ambition and moral superiority. For instance, in a speech from 2000, the UK Labour Party took credit for driving norms around racism in a more progressive direction; a change that the party, and its voters with it, ‘ought to be proud of’. Appeals to a sense of pride and, more generally, rhetorical efforts to place a party on a moral pedestal serve strategic purposes because party identifiers are likely to be mobilized by the positive feelings they generate (Jung Reference Jung2020; Skitka, Hanson, and Wisneski Reference Skitka, Hanson and Wisneski2017). Combined with the frequent use of moral attacks, however, there is a risk of such moral elevation often being established on the basis of strong negative moral judgements of others (Lamont Reference Lamont1992).
Our final observation concerns the relatively infrequent use of moral language to justify policies and political actions against criticism (occurring in 16 per cent of the highly moral speeches); a finding that may seem surprising since moral rhetoric has been proposed as ideally suited for legitimizing purposes (Appell Reference Appell1980; Haidt Reference Haidt2012). In a party-competitive light, however, the finding is less surprising. Where attacks, self-praise, and calls for action signify to audiences that the speaker has political agency and moral authority, parties engaged in rhetorical justification implicitly admit a need for defence; a strategically difficult position to take because it acknowledges the criticism and thus the moral flaws of the party. By contrast, by engaging in a moral attack, and thus rhetorically turning the direction of criticism 180 degrees, or disregarding the criticism entirely by engaging in moral grandstanding, a party can seek to appear beyond reproach.
Discussion
Why does immigration discourse sometimes come to take shape as a moral battlefield? Relying on a combination of quantitative and qualitative pieces of evidence, our findings show that elite polarization creates the conditions that allow and incentivize parties to claim the moral high ground, often through attacks on their opponents for being morally wrong. Importantly, moral rhetoric is not reserved for a specific party type, and no party is immune from the competitive pressures that polarization places on them to moralize: the language of right and wrong, good and evil is used by parties positioned at extreme ends as well as those in the middle of the ideological spectrum on immigration, by those promoting anti- as well as pro-immigrant viewpoints, and by incumbent and opposition parties alike. These findings help us correct the common notion that extreme, anti-immigrant opposition parties are the main moralizers; instead, we show that the political context creates competitive pressures that affect all parties.
That all parties engage in moralization does not mean that they do it all the time, however. Our theorization and findings lay bare that existing research has tended to underestimate or insufficiently consider the risks involved in employing moral rhetoric, since the focus has been on what parties might gain from using this language. Our study emphasizes that parties employ moral rhetoric when they have moved sufficiently far apart to discourage rhetorical efforts to de-escalate conflict. Under such conditions, the risk of backlash and the conflict escalation associated with moral language no longer serve as the dampening factors they otherwise appear to be, a proposition that is underlined by the findings from our qualitative analysis on the dominance of moral attacks.
The importance of the political opinion climate over party-level factors that our study highlights stands in contrast to existing research on parties’ use of moral rhetoric. Here, we consider potential explanations and implications of this divergence. First, the single-case and relatively shorter-term perspective of most prior work limits its ability to assess the influence of systemic factors compared to our cross-national and long-term study. While it may be the case that specific parties stand out as more moralizing than others in some instances, for example, because external shocks allow them to act as (rhetorical) challengers (Widmann and Simonsen Reference Widmann and Simonsen2024), our findings suggest that such interparty differences are unlikely to be stable or cross-nationally generalizable. In the longue durée, the political opinion climate seems to be the driving force of moralization according to the results of this article.
Second, it is relevant to consider the medium of communication that this study relies on. We have argued that the use of parliamentary speeches is ideal because the debate character of this text source underlines the competitive dimensions of political rhetoric that we have theorized are essential for driving political parties’ moralization of immigration. While we cannot rule out that the party-level factors in the focus of H1–H3 might shape more one-directional forms of political communication, such as party manifestos or party homepages, parties are always (if at times more implicitly) engaged in competition with each other, which, under conditions of elite polarization, should make them turn up their moral rhetoric, even on their own platforms. This argument is sustained by research demonstrating high campaign consistency across communication channels (Kang et al. Reference Kang2018), suggesting that political actors align their messaging – and potentially their rhetoric – across platforms.
Finally, our study differs from most others on politicians’ use of moral rhetoric by zooming in on one topic – immigration – rather than applying a broad-brush approach. This was motivated by a wish to understand the rhetorical dynamics of the immigration debate specifically, but our focus on one topic also has methodological implications: As we have argued, the risk involved with broad-brush analyses is that they might misinterpret patterns driven by a particular topic taking over the agenda as evidence of a fundamental change in the rhetorical terms of debate across topics. Here, we take this opportunity to reflect on the potential for transferring the insights generated from our study of immigration to other political topics.
Due to the contemporary association of the issue with debates about cultural values, immigration might tend to be seen as lending itself more easily than others to moralization. Indeed, the mean moral score in immigration-related speeches is higher than in the remainder of (non-immigration-related) parliamentary speeches in our corpus (see Online Appendix D), underlining the suitability of immigration as a case to study moralization dynamics. At the same time, we believe this study’s findings are likely to be transferable to other topics. For instance, research shows that survey participants consider topics such as unemployment, education, and same-sex marriage as equally central to their moral convictions as immigration (Ryan Reference Ryan2014), suggesting that politicians might as easily moralize these issues. This is supported by recent evidence showing that even topics traditionally considered to be non-moral, such as taxation and public administration, are at times as highly moralized in political rhetoric as immigration (Simonsen and Widmann Reference Simonsen and Widmann2023). What is more, a central finding from the qualitative analysis is that much of the moral rhetoric applied in immigration discourse concerns parties and their policies, indicating that our findings are not only or even primarily driven by communication related to the issue substance (what is right and wrong with respect to immigration) but, equally importantly, by discourse focused on parties. We therefore believe that the competitive dynamics documented in our study are likely to extend beyond immigration, and we welcome future research that tests our claim about the centrality of elite polarization in the context of other political topics.
To conclude, our study highlights the affordances of a cross-national, multilingual, and topic-specific approach to the study of (moral) rhetorical competition. Although developing and validating multilingual text tools is resource- and time-consuming, efforts to do so are crucial for assessing the generalizability of existing research as well as for expanding opportunities for understanding the centrality of factors located at the party-systemic level. With the publication of this article, our dictionaries are made available for other researchers to use, and we hope that our study will inspire further research on how politicians use moral language in multilingual contexts, integrating quantitative and qualitative modes of analysis to ensure the transparency and validity of findings.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007123425000122.
Data availability statement
Replication data for this paper can be found in the Harvard Dataverse at https://doi.org/10.7910/DVN/HAJHSG (Simonsen and Widmann Reference Simonsen and Widmann2025).
Acknowledgements
We would like to thank the following for helpful feedback on earlier versions of the manuscript: Roman Senninger; Hauke Licht; participants in the CIS seminar at ETH Zurich; participants in the CSDC Speaker Series Montreal; members of the Migration, Integration, Transnationalization research unit at the WZB Berlin; and participants in the Migration & Ethnicity Workshop at Aarhus University. We also thank the three anonymous reviewers and editor Ruth Dassonneville for constructive comments during the review process.
Financial support
Kristina Bakkær Simonsen acknowledges funding for this research through a Sapere Aude Research Leader grant from the Independent Research Fund Denmark (grant 0167-00002B).
Competing interests
None
Open access
