Firebreak, circuit break, or water break? The impact of metaphor on people’s perception and attitudes towards lockdown measures

Abstract

Metaphors can influence people’s reasoning because of their ability to highlight or hide features of the target domain. In this article, we investigate the extent to which different metaphorical frames lead to different policy recommendations that best fit with the structure of the frame, as well as the role of age and gender to account for variation in the responses. We rely on four naturalistic metaphorical frames used during the media coverage of the COVID-19 pandemic: fight, fire, machine, and water. A total of 203 Spanish participants were randomly shown one of the five experimental conditions and were asked to (1) rate their perception of control over the health emergency and (2) recommend policy measures to stop the spread of the pandemic. To assess the extent to which participants had noticed the metaphorical frames, a third question was added where they had to indicate the words that had been most decisive in their answers. Results indicate that the fight frame increases the perception of control over the situation, but mostly for men and older participants; they were also more likely to prefer restrictive measures, whereas women and younger participants favoured a balance between restrictive and preventive policies. Finally, fire keywords were the most likely to be remembered by everyone, unlike the keywords from other frames. These findings shed light on the role of age and gender in moderating the effect of metaphorical framing.

Introduction: The potential of metaphorical framings

One key assumption in metaphor theory is that metaphor can influence the way we reason about concepts and, therefore, has the potential to shape our actions accordingly. In a metaphor, partial material from a well-known concept (the source domain) is used to structure our reasoning of another, less graspable concept (the target domain). Since a metaphor is built upon a partial conceptual mapping between two ideas, it has the power to highlight and hide specific aspects of the target domain (Lakoff and Johnson 1980; see Valenzuela and Ibarretxe-Antuñano 2024, for an overview). The goal of this study is to gather more empirical evidence about the effect of metaphorical framing on cognition and behaviour, with a particular focus on the COVID-19 pandemic. It makes for an excellent case study as the pandemic led to an explosion of metaphors to cope with different aspects of the new ‘normal’: to understand the new threat, to process individual feelings, to coordinate collective actions, and ultimately, to legitimize policy (for examples, see Wicke and Bolognesi 2020; Olza et al., 2021).

One particularity of the way metaphors were used at the beginning of the pandemic was the ubiquity of the war frame to conceptualize almost every aspect of the pandemic, with the virus being associated with an enemy, health workers with soldiers, and vaccines with weapons to eliminate the enemy. Whereas there is no doubt about the usefulness of this frame at the initial stage of the pandemic because of its immediate power to evoke a sense of urgency and the need for drastic measures to survive in a life/death situation, there is also evidence of its detrimental effect on people if sustained over longer periods (Flusberg, Matlock and Thibodeau 2018). Alternative ways of speaking of the virus, for instance, in terms of ‘glitter’, may also be helpful to highlight the way it can easily spread and stick to surfaces. These alternatives can contribute to illustrating in a less stressful way the need to adopt careful hygienic habits, for example, in children (who are less likely to cope well with the pervasiveness of the war frame). Linguists from different countries gathered in the #ReframeCovid initiative (Olza et al., 2021) to call for the need to draw attention to other metaphorical frames to better explain different aspects of the pandemic as it evolved.

Interesting as this premise might be, it remains to be empirically confirmed. To what extent is the war frame limited to calling for action in a pandemic? Are there other frames that are more useful to shape people’s behaviour in a context of crisis, and if so, what are these? Is there a frame that connects with all the population in general, or can we identify specific frames that resonate best with different subgroups of the population? We set out a study to measure the impact of the following framings of the pandemic in the year 2021. The main predictor of this study was metaphorical framing, with five different levels: (1) war (pandemic as a struggle, counterattack measures to combat mass infection), (2) fire (pandemic as a fire, firebreak measures to quench mass infection), (3) machine (pandemic as a machine, circuit break measures to stop mass infection), and (4) water (pandemic as a tsunami, water break measures to contain mass infection). A total of 203 Spanish native speakers who were recruited online via Prolific participated in this study. Whereas war and fire were metaphorical frames that could be frequently found in the Spanish media, that was not the case of the water and machine frames, which were taken from the UK press. We did include these to factor out the role of familiarity with the metaphorical frame in testing people’s preferences. By including culturally ‘distant’ metaphorical frames and checking how they compare to commonly seen frames, we not only cover a wider array of possibilities for conceptualization but can see the real effect of metaphorical framing on people. All participants were shown the same text but with three different keywords belonging to each of the metaphorical frames under study.

In addition to the metaphorical frame, we also looked at variation across two additional predictors, gender and age, when analysing participants’ responses. These two variables are included for three different reasons. One is to avoid possible participant group bias; that is, depending exclusively on informants with similar sociolinguistic backgrounds to represent language processing can conceal the genuine diversity of language, culture, and their actual processing (Henrich, Heine and Norenzayan 2010; Ameka and Terkourafi 2019; Blasi et al., 2022). Another reason is to attend to possible differences in the understanding and managing of certain frames in a collective situation such as the COVID pandemic. Testing different populations is crucial in order to understand whether the frames proposed for the problematic situation (the pandemic) equally reached all populations or are simply the reflection of one part of society. During the pandemic, a noticeable divide emerged in the frames used by different genders: Male presidents such as Bolsonaro, Macron, or Trump often employed war-related frames, advocating for individualistic confrontation measures. In contrast, female presidents such as Ardern and Merkel adopted alternative frames, emphasizing collective actions as the primary approach (Olza et al., 2021). At the same time, the way the pandemic was experienced varied significantly across different age groups. Its impact on people’s lives differed depending on whether they were at the beginning or the later stages of their life (Eurofund 2022; Kleine et al. 2023). Finally, linguistic literature widely recognizes that men and women exhibit differences in both the innovation and maintenance of linguistic elements. Research on the relation between gender and language in the Labovian tradition (Labov 1972, 1990) unveiled a higher degree of awareness and proactive attitude for linguistic innovations in female speakers than in male speakers. In addition to this, variation by age has already been attested in the neological habits of Spanish speakers for COVID-related vocabulary (Zvereva 2022), with younger participants more likely to produce more neologisms than their more senior counterparts.

The study design features three response variables: perception of control, measured on a scale from 1 (no control) to 10 (total control); (2) measures to control the spread of the pandemic, which was formulated as an open-ended question which the authors later content-coded as ‘restrictive’ (e.g. police enforcement, lockdowns, and fines) or ‘preventive’ (educational programmes, public health messaging, and vaccination campaigns); and (3) recall of metaphorical keywords (we asked participants to motivate their recommendations, and we later annotated if any of the responses featured any of the keywords relating to the metaphorical frames, and the place in which these keywords appeared in the text: beginning, middle, or end position).

Three research questions were formulated to address this issue:

RQ 1: To what extent does metaphorical framing affect the perceived feeling of control over the pandemic? Is there any variation by age/gender?

This research question addresses the extent to which the different metaphorical frames help to highlight the idea of containing the pandemic. Given that at the time of collection of data (April–May 2021), the pandemic was still actively spreading in Spain, our working hypothesis is that the war domain might be more effective in evoking control than non-metaphorical statements and that male adults might be more prone to prefer such measures than female adults. However, we also expect the fire frame to be equally active to denote a situation of emergency because of the many fires in Spanish forests during that summer and because of the possible fatigue of the war domain after its intense use during the first year of the pandemic. No specific hypothesis is claimed with respect to age alone, but a tendency to prefer the war domain in the case of older speakers is expected.

RQ 2: To what extent does metaphorical framing affect the measures recommended to counter the situation? Is there any variation by age and/or gender?

In this research question, the measures recommended by participants to stop the spreading of the pandemic were examined. The possible interrelation between these measures and the participants’ responses in RQ1 was also studied. Our working hypothesis is that, on the one hand, those participants who were exposed to the war frames would be more likely to call for ‘restrictive’ measures; that is, more police vigilance in streets, higher economic penalties, and setting new lockdowns, among others. On the other hand, non-militaristic frames (fire, machine, and water) would call for more ‘preventive’ measures such as information campaigns, hiring more health workers, and mass vaccination, among others. Based on the differences reported in the literature about male preferences for individualist actions and female preferences for collectivistic actions, the working hypothesis is that male adults are likely to prefer ‘restrictive’ measures and female adults, ‘preventive’ measures. No specific hypothesis is claimed with respect to age alone, but a tendency to prefer the restrictive measures in the case of older speakers is expected.

RQ3: To what extent do people notice metaphors (and if so, does it matter where the metaphorical words appear in the text)? Is there any variation by age/gender?

This research question seeks to assess participants’ sensitivity and awareness of metaphor-related words when performing the tasks of this study. This question complements the previous ones in that it seeks to assess whether people process metaphors consciously or unconsciously. To measure this, three metaphorical keywords were placed at the beginning, middle, and end of the paragraph, respectively. At the end of the study, participants were asked to refer to the words that were most decisive in their responses to both experimental tasks. In line with previous research (Boers 2000), we predict that people will be likely to remember metaphorical words more easily than literal words; and that we will be able to replicate a serial-position effect by expecting participants to recall the first and third keywords more easily than the middle one when motivating their answers. Given the literature on metaphor processing related to our informants’ profile (Colston and Gibbs 2017), no specific hypotheses are claimed concerning either gender or age.

The outline of this article is as follows: Section 2 provides an overview of the theoretical underpinnings that support this study and offers a brief review of the literature on metaphorical framing, with a specific focus on war metaphors and metaphors for the pandemic. Section 3 discusses the methodology of this study and provides relevant information about the demographics and inclusion criteria of the participants in the study, details about the design of the experimental stimuli and procedure. Findings are discussed in Section 4. Conclusions, limitations of the present study, and suggestions for further research are presented in Section 5.

Theoretical framework

Metaphorical framing influences people’s reasoning (or not).

Frames are mental structures that shape the way we view the world. Frames have been the subject of study by many different disciplines, ranging from sociology (Goffman 1974) and artificial intelligence (Minsky 1974) to semantics (Fillmore 1975, 1985). The underlying substrate that unifies all the different accounts of framing is a general understanding of it as a portion of background knowledge that (a) highlights a specific view of the world, (b) is encoded with a specific choice of words, and (c) generates a specific set of expectations and attitudes. According to Entman (1993: 53), ‘to frame is to select some aspects of a perceived reality and make them more salient in a communicating text, in such a way as to promote a particular problem definition, causal interpretation, moral evaluation, and/or treatment recommendation’. In their Nobel-winning experiment, Tversky and Kahneman (1981) showed that there is a strong dependence on the emotions invoked by the framing in how a problem is formulated. They framed proposed solutions to the same problem (the outbreak of an Asian disease in the USA that was expected to kill 600 people) in four different ways: in terms of gains (people living) and losses (people dying), and in terms of sure options and probabilities (A: ‘200 people will leave’; B: ‘1/3 probability that 600 people will be saved’; C: ‘400 people will die’; and D: ‘1/3 probability that nobody will die’). Although all options are somewhat equivalent, the experiment showed that participants preferred the sure option (A) when talking about gains, and the risky option and (D) when confronted with losses. The groundbreaking finding of this experiment was that humans are not rational but emotional beings who are sensitive to how information is presented.

Most of the studies focusing on how lexical choices contribute to the activation of given frames have focused on metaphorical framing. What makes metaphorical framing unique is that, unlike other framing methods (e.g. gain-loss), the concepts being compared do not have a literal relationship to each other (Lakoff and Johnson 1980; Lakoff 1993). A canonical example is the study conducted by Thibodeau and Boroditsky (2011) who leveraged the power of metaphor to structure the knowledge about a complex issue such as crime and to develop several inferences and expectations about it. They asked people to imagine a ‘virus infecting a city’ or a ‘wild beast preying on a city’ to then explain the best way to solve the problem that they had imagined. Metaphorically framing crime as a virus creates a mapping between the two ideas by highlighting the similar structure; for example, in the same way, that doctors diagnose the cause of a viral infection, causes of crime can be identified by the proposed measures to treat them, such as prevention and education. At the same time, it hides other dissimilar structures that could point at alternative ways of addressing crime more likely to be activated when treating crime as a beast, such as police enforcement as the way to handle crime. Their findings show that participants favoured different measures depending on the metaphor they interacted with. Although participants were overall more likely to recommend enforcement measures (65%), their proposed solutions differed depending on the frame they encountered: 74% of the participants that were shown the crime-as-beast virus preferred enforcement measures, whereas this was the case only for 56% of the participants interacting with the crime-as-virus metaphor. It should be noted that, when factoring in the place where the metaphor was shown in the report (in line with our RQ3), participants exposed to the virus condition who saw the metaphor at the beginning of the report were more likely to recommend reform measures (such an effect was not observed when they read the metaphor at the end). This finding suggests that participants exposed to the beast condition were more influenced by the image of capturing the beast and then killing or caging it and preferred more restrictive measures to control the problem. In turn, participants interacting with the virus metaphor suggested investigating the source of the problem and implementing social reforms and prevention measures (such as educational campaigns to improve social hygiene).

One key criticism raised against most metaphorical framing studies is the lack of a control (non-metaphorical) experimental condition to compare the effect of metaphorical framing with (Hartman 2012; Steen, Reijnierse and Burgers 2014). This makes it difficult to discern whether the effect is due to the metaphor or to a general framing effect. Even though the findings of metaphorical framing experiments are somewhat inconclusive (for a detailed review, see Reijnierse et al., 2015), some indicators suggest that metaphors strengthen the impact of framing on people’s perceptions of a situation.

Besides psycholinguistic evidence attesting that metaphor (or figuration in general) (Citron and Goldberg 2014) is emotionally more engaging than their literal counterparts, little is known as to what makes a frame more engaging than others. A limitation identified by Boeynaems et al. (2017) in their review of studies dealing with the impact of metaphorical framing is the unclear idea of what makes an ‘apt’ frame. Research has shown that some variables may predict the success of a metaphorical frame:

• (1) Target domains that are inherently abstract or complex are more likely to be influenced by a metaphor. However, if the beliefs about the target concept are too strong, they may override the metaphorical mapping. For example, in Thibodeau and Boroditsky’s (2011) experiment, the virus/beast metaphor would be deemed ineffective if people had strong beliefs about how society should handle crime.

• (2) Prior knowledge of and attitudes on the source domain that refers to shared experiences (embodied or cultural) that people find easier to relate to. For example, in Ottati, Rhoads and Graesser (1999) study on the effects of using sports metaphors, only people who enjoyed sports were found to be more likely to be persuaded by the metaphor.

• (3) The valence of the metaphorical framing (Johnson and Taylor 1981), by which positive metaphors lead to more positive judgements, but only if people had prior knowledge of the subject framed.

• (4) The presence of the metaphorical framing first at the beginning of a text or speech (Thibodeau, Hendricks and Boroditsky 2017), as it activates mental images that help to interpret ambiguous information in the following text/speech that leaves room for interpretation that can be construed consistently with the metaphor.

Whereas there is an overall consensus regarding the fact that metaphorical frames are more engaging than literal ones, determining what makes a frame more adequate than others remains an open-ended issue, with several studies pointing at different features of metaphorical frames that make them more particularly suitable for different situations. Some studies have shown that there might perhaps be no one metaphorical frame that fits all situations and that people need a menu of metaphors to choose from (for reference, see ‘A “Metaphor Menu” for People Living with Cancer’, 2019). We elaborate on this idea in the following section.

Frames for COVID-19: The pandemic beyond war

The effects of metaphorical framing are especially palpable when it is used to speak about hardship (Hendricks et al., 2018). This is so because of the need to provide novel but meaningful ways to reason and talk about complex problems. A case in point was the surge of metaphors during the pandemic of COVID-19. As reported by Pérez-Sobrino et al. (2022: 128), the virus was considered an invisible enemy that came in waves; health workers became heroes and soldiers at the frontline; herd immunity was the goal to be reached by means of social distancing and lockdowns … Rather than being a rhetorical adornment, the use of evocative images in metaphors helped governments to convey clear messages and to call for people’s action in an unprecedented scenario of global emergency. Hence, metaphors fulfil a very important pedagogical function (Semino 2008; Landau, Robinson and Meier 2014) as they help to bridge the gap between what we know and what we seek to know. They make the strange look familiar, or conversely, they provide a novel angle on what is familiar. Unsurprisingly, the invisibility of the virus to the naked eye and the lack of familiarity with such a situation posed a communication challenge for the population, who did not know how to react and behave. Evidence suggests that people experiencing different types of difficulties produce more metaphors when describing intense emotional experiences than when describing actions (Fainsilber and Ortony 1987). This leads to a higher density of metaphors in people’s accounts of hardship in the narratives produced by women suffering from cancer (see Gibbs and Franks 2002; and Semino et al., 2018, for a corpus study).

In the context of the COVID-19 pandemic, governments faced the need to communicate urgent and highly restrictive measures to the population. This resulted in the emergence of a wide range of diverse metaphors to reason about different aspects of the pandemic, as well as to heighten or mitigate its negative emotional impact. These metaphors were encoded in expressions such as fighting a war against an invisible enemy, doctors as soldiers in the frontline, hibernating economies that had to be restarted and countries activated after lockdowns. As conceptual devices, metaphors can also be manifested in images, such as COVID-19 curves that had to be flattened, traffic-colour-coded maps to indicate different degrees of restrictive measures, and depictions of the virus with varying human attributes to explain how it behaves and spreads.

Each metaphor highlights a different aspect of the pandemic and thereby neglects other aspects. The early stages of the pandemic were characterized by prominent use of metaphors coming from the war frame, which mainly portrayed the virus as an enemy to be beaten and society—and its different members—as the soldiers fighting against it (Gillis 2020; Wicke and Bolognesi 2020). The use of the war metaphor on social media reflected not only common uses when talking about health problems (Semino 2008: 175–89; Flusberg, Matlock and Thibodeau 2018) but also how political responses to the pandemic were being framed. War metaphors were of particular use at the beginning of the pandemic to convey the need for urgent and highly restrictive measures, and their choice reflects a commonly identified understanding of an illness as an enemy (Sontag 1978; Larson, Nerlich and Wallis 2005; Semino 2008). This correspondence between these two domains not only reflects the need of people to resist a virus while calling for a coordinated response among the members of society (Olza et al. 2021) but also allows for similarities to be established between the socio-economic consequences of the virus and those of war.

However, the sustained use of war metaphors over the course of the pandemic has triggered criticism both from the academic community and the media (Jenkins 2020; Musu 2020; Olza et al., 2021). The reasons for this criticism are clear: war metaphors can trigger possible negative effects. The current literature shows that metaphors can transfer perceptions and emotional associations (Landau, Arndt and Cameron 2018). Some of the perceptions associated with the war frame include the risk of ‘militarizing’ the public sphere (Flusberg, Matlock and Thibodeau 2018; Olza et al., 2021), the danger of identifying ill people or those who do not conform to the rules as culprits (Sontag 1978; Demjén and Semino 2016) which may eventually result in stigmatization of patients (Logie and Turan 2020) or an increase in fear and anxiety (Hulscher 2020; Páez and Pérez 2020) and, eventually, pandemic fatigue if used over a long period. The detrimental effects behind the abuse of militaristic metaphors have also been shown empirically; an experiment conducted in Germany and the USA (Schnepf and Christmann 2022) showed limited effectiveness of war metaphors, especially in the US context where non-military metaphors were more likely to lead to policy support than the military ones. They did not find evidence either that the battle frame increased participants’ fears or perceived spread of the virus. Other concepts with negative valence, such as ‘struggle’, lead to different connotations: in both countries, ‘struggle’ was perceived to trigger more of a sense of individual responsibility, compared to ‘war’, which was more likely to make people think that governments carried all the responsibility. Therefore, it is important to consider metaphors as a potential strategy to increase the quality of health communication (Peters, Ruiter and Kok 2013).

As the virus spread across the world, and countries and governments enforced different policies (with markedly different degrees of success), the use of ‘war’ metaphors lost significance in favour of alternative metaphors that equally conveyed the sense of shared responsibility and vigilance, but that carried less taxing effects on the population (Döring and Nerlich 2022, and references therein). To counter the (ab)use of war metaphors, there have been several proposals for alternative metaphors that convey a clear message of the need for a coordinated response between the different societal actors, such as fire-related metaphors put forward by Semino (2021) or the metaphors collected within the #ReframeCovid initiative (Olza et al., 2021), a crowdsourced collection of alternative non-military framings for the pandemic. The idea behind this collection (inspired by the ‘Metaphor Menu for People Living with Cancer’, 2019) was to offer a wide range of metaphors so that people for whom the war frame does not resonate well (children, the elderly, people with disabilities, etc.) can choose the one that best reflects their experiences, and therefore be able to better communicate their needs to doctors.

The present study seeks to find empirical confirmation of the (1) impact of metaphorical over non-metaphorical frames on people and (2) differences in the beliefs and attitudes triggered by the war frame vs. other metaphorical frames used during the pandemic, namely fire, water, and machine.

Methodology

To test our hypotheses, we set out an experiment to measure the impact of metaphorical frames vs. literal texts, and also to assess the likelihood of each metaphorical frame to recommend restrictive or preventive measures to control the spread of the pandemic. Metaphorical words belonging to four different frames (war, fire, water, and machinery) to assess the extent to which they shape people’s perceptions of the pandemic were included in three different places within the paragraph: beginning, middle, and end. We focus on gender and age as potential sources for variation in the responses.

Participants

An a priori power analysis conducted with G*Power (Faul et al., 2007) yielded that we needed 196 complete surveys to detect a medium effect (f = 0.25; Cohen 1992) with a power of 0.80, and alpha set at 0.05. We set our inclusion criteria before collecting data: participants should be Spanish citizens and native speakers of Spanish between 18 and 65 years old. We relied on the screening filters of Prolific (www.prolific.co) to recruit online a representative sample of male and female participants (93 men and 110 women) from different ages (98 participants between 18 and 39 years old, M = 23.9, SD = 5.28; and 105 participants between 40 and 69 years old, M = 46.3, SD = 5.39). In total, 203 surveys were collected.

Participants provided informed consent before accessing the survey and were informed that their data would be treated anonymously and that they could withdraw at any time during the study. They were paid according to the duration of the survey, consistent with Prolific’s policy of rewarding above the minimum wage.

Design and materials

Our main predictor (independent variable) was the metaphorical framing used in the text to refer to the pandemic, with four levels: fight, fire, machinery, and water. Four versions of the same text in Spanish (see Table 1 for texts and English translations) were generated with relevant metaphorical keywords to refer to different aspects of the frame placed in three positions within the paragraph: (1) initial sentence, to frame the pandemic as a problem; (2) halfway through the paragraph, to frame the measures to control the pandemic; and (3) final sentence, to frame the purpose of such measures. In addition to this, a fifth version without metaphorical keywords was added as a control condition to serve as a baseline (which, as pointed out by Reijnierse et al. (2015), is not usually present in many framing experiments, thus limiting the significance of the results). Although texts were fictional and generated for the purposes of this study, all the metaphorical keywords—numbered and in bold in Table 1—inserted in each text were naturalistic examples found in the media coverage of the pandemic.

Experimental procedure

Data were collected online through a questionnaire designed in LimeSurvey (https://www.limesurvey.org). After an introductory page with information relevant to the consent and storage of data, participants moved on to a new screen where one of the five texts was randomly shown. Once they had read it, all participants were taken to a new screen where they had to answer the same three questions:

Question 1. ‘In your opinion, to what extent can mass infection in Vetusta be controlled?’ This question was designed to measure participants’ perception of control on a 10-point Likert scale ranging from 0 = no control to 10 = absolute control.

Question 2. ‘What measures should the Vetusta government take to improve the situation in the city? Why?’

In this question, participants were asked to freely recommend measures to control the spread of the pandemic in their own words. In line with the guidelines put forward by Thibodeau and Boroditsky (2011) and Reijnierse et al. (2015), participants’ qualitative responses were annotated as ‘restrictive’ if they focused on punitive actions to control the pandemic once it is on the rise, such as setting higher economic penalties, setting lockdowns, and increasing police street vigilance; and ‘preventive’, if they focused on actions that were to be taken before the spread of the pandemic, such as recommending mass vaccination, designing better educational programmes and information campaigns, etc. Table 2 offers some illustrative examples from participants’ responses.

When both restrictive and preventive measures were recommended, we coded the one that was more strongly recommended. This decision was taken because more individual measures were going in that direction or because of the manner in which the recommendations were verbalized, e.g. with intensifier adverbs and/or more words.

Finally, participants were asked to motivate their answers by indicating the words that they remember to be most influential in their responses (Question 3). We later coded their answers by annotating whether they had referred to any of the metaphorical keywords present in the text, and whether they appeared at the beginning, middle or end. They also indicate their age, gender, nationality, and native language. They were thanked for participating and were redirected to Prolific to collect their remuneration.

Statistical tests

The questionnaires, data, and R script are available in a publicly accessible repository: https://osf.io/wkv62/

All the statistical analyses were conducted in the R statistical environment (R version 4.0.2) within the linear model paradigm. For the first research question, we fitted a linear regression model with one main categorical predictor (metaphor condition, five levels: war, fire, water, machinery, and control) and a numeric outcome (perception of control over the health emergency, from 0 to 10). The age and gender of the participants were fitted as interaction variables. It was not necessary to add any random items because this study did not involve any repeated measures.

For the second research question, we looked at the same categorical predictor (metaphor condition, five levels), but this time we entered a categorical binary outcome based on our coding of participants’ qualitative responses (restrictive or preventive measures), which is why we opted for a binomial generalized linear model. Age and gender were also fitted as interaction variables, and no random effects were entered into the model.

For both the first and second research questions, the significance of the overall model was computed using likelihood ratio tests (i.e. performing an ANOVA of the full model against the null model to see if the inclusion of the predictor triggered any significant effects).

For the third research question, we shifted from the linear modelling paradigm to the chi-square test of independence, followed by an inspection of the standardized residuals. This decision was taken due to the lack of any a priori assumptions as to what was the cause and what the effect was. Unlike the two previous questions, this third question simply sought to explore if people reported having noticed the different metaphorically related words present in the stimuli (five levels) and, if so, where these words were more likely to be located (three levels—beginning, middle, and end of the paragraph).

Results

Research question 1. Perception of control

Figure 1 shows the ratings for perceived control (from 0 = no control to 10 = absolute control) triggered by the different frames in each of the four experimental conditions, plus a control condition (‘neutral’) without any metaphorical framing that serves as a reference level (vertical line indicates the mean).

A likelihood ratio test of the model with the experimental condition effect against the model without the experimental condition effect revealed a significant difference between models (χ²(1) = 40.865, P < .05). In other words, this finding indicates that the ratings for perceived control for each experimental condition were different from each other in a statistically significant way. When looking at specific effects, we observed that it was mainly because of the stark contrast between the neutral condition, namely, the literal text, and the fight condition. Participants who read the text including keywords related to fighting were more likely to rate the perception of control over the situation higher (6.72/10) with respect to the control condition (5.62/10) in a statistically significant way (P < .05). The model did not return any additional statistically significant differences between the other three metaphorical framings and the literal text.

The next step is to study whether these results change if the two further sources of variation, that is, age and gender, are considered. Figure 2 shows a breakdown of the ratings displayed in Figure 1 by age group.

The model returned significant statistical differences when fitting age as an interaction with the experimental condition as opposed to the null model (χ²(1) = 76.77, P < .05). However, in a closer look at the specific effects, we observed that such statistical significance was due to age as main effect (P < .05) and also for the fight metaphorical frame (fight: young—6.05, old—7.35, P < .05) for which we observed a greater gap between the ratings provided by the two age groups. In other words, this means that by and large, the explanation for the variation in the data was that younger participants were overall less likely to perceive control over the situation than old participants across all experimental conditions in a statistically significant way (mean values for the different conditions: water: young—5.55, old—6.13; fire: young—5.37, old—5.43; and neutral: young—5.00, old—6.25). The only exception was machine, for which younger informants reported slightly higher ratings of perceived control (young—6.20, old—6.05).

Interestingly, a statistically significant difference was observed for the specific case of the fight framing (P < .05). This difference could be interpreted as the most polarizing framing of all four under scrutiny. In that sense, both the 18–39 and the 40–65 age groups reported similar perceptions of control when exposed to fire, machine, and water-related metaphors.

Finally, it should be noted that when contrasting the type of framing used (literal vs. metaphorical, collapsing all four metaphorical framings), there were indicators that young participants are comparatively more sensitive to metaphorical framings. The younger group reported higher ratings for perceived control in response to all four metaphors with respect to the literal condition, whereas the responses of the elder group were not as consistent. Although this finding only borders significance (P = .07), further research should look at this pattern in a more detailed manner to see if it is a relevant trend.

As far as gender is concerned, Figure 3 shows the role of this source of variation in the participants’ ratings for perceived control over the situation. Much like in the case of age, there are statistically significant differences that were attributable to the gender of the participants as main effect (χ²(1) = 85.57, P < .01). As can be seen in Figure 3, women were overall less likely to perceive the situation to be in control (water: female—5.50, male—6.12; machine: female—5.81, male—6.71; and fight: female—6.58, male—6.93) with the only exception of the fire framing where both females and males reported equivalent ratings (female—5.4, male—5.4).

The specific differences between female and male ratings reported for each experimental condition show that the main difference was due to the broad gap in the literal condition (female—4.68, male—6.48), which was statistically significant (P < .05). This is because metaphorical framings increased the perception of control over the situation for women in a statistically significant way (P < .05), but not for men, whose ratings were similar regardless of the type of framing used. In sum, metaphorical framings unified the perception of control over the situation by both men and women (and in some cases, it even increased such perception with respect to the literal condition).

Overall, results show that, although there seems to be a positive effect of the fight metaphor to increase the perception of control over the pandemic, the picture changes when we break down the ratings by age and gender. Women and younger informants perceive to be less in control over the situation, for both literal and metaphorical framings. However, metaphorical framings help to increase the perception of control of all participants, perhaps because metaphors offer a sense of structured response to the spread of the pandemic. Interestingly enough, such an increase was steeper in the case of women, thus helping to narrow the gap with men in the way they think the spread of the pandemic is controlled. Younger speakers could also be more sensitive to metaphorical framings than their older counterparts, but our findings are inconclusive, and therefore, caution should be exercised.

Research question 2. Restrictive or preventive measures

The second research question explored the measures recommended by people to control the health emergency described in the text and the extent to which they were more likely to be of a restrictive or preventive nature depending on the metaphorical framing used. Figure 4 shows the distribution of preventive and restrictive measures by experimental condition. In all five cases, participants preferred restrictive measures (N = 139) over preventive measures (N = 46). A chi-square goodness-of-fit test showed that the distribution of restrictive vs. preventive measures across conditions was statistically different from the one expected by chance (χ²(1) = 46.751; P < .001). However, none of the four metaphorical framings revealed any distinct effect from the control condition (χ²(1) = 0.86, P = .93), as the distribution of restrictive and preventive measures was kept more or less stable across the five experimental conditions.

However, even though both age groups were overall more likely to recommend the adoption of restrictive measures to control the spread of the pandemic (and therefore, no interaction effect was observed), as Figure 5 shows, the gap between the number of restrictive and preventive measures provided by participants was narrower in the case of the 18- to 30-year-olds compared to the group of 40- to 65-year-olds (number of measures provided: preventive: young—30, old—16; restrictive: young—58, old—81), the latter being, in general, more likely to favour restrictive measures (χ²(1) = 7.71, P < .01).

We then looked at the specific differences by experimental condition to investigate the specific responses given by the two age groups. As shown inFigure 6, the pattern was stable across all metaphorical framings, as there was no variation attributable to the experimental condition (at least, in a statistically reliable way in our data). The only exception was the machine framing for which the difference in the recommendation of preventive measures by younger and older participants (one instance) returned a statistically significant difference (χ²(1) = 7.27, P < .01).

To close this section, we turn one more time to the role played by gender in accounting for the variation in the recommendation of control measures. Much in the same line as the findings reported above for age, as shown in Figure 7, the model showed that gender was a statistically significant predictor for differences in participants’ responses (χ²(1) = 6.93, P < .01). Although both women and men preferred restrictive measures to control the spread of the pandemic (preventive: women—31, men—16; restrictive: women—59, men—81), women were overall less likely to recommend restrictive measures and were more in favour of preventive measures.

Figure 8 shows the breakdown of preferred recommendations by gender and experimental measure, which supports the preference for restrictive over preventive measures across experimental conditions and in a fairly similar proportion in both women and men. The exception is, once again, the machinery framing: whereas women were more likely to recommend a balance between preventive and restrictive tools, men chose exclusively restrictive options (χ²(1) = 8.42, P < .01).

The predictions of our models provide evidence that the differences in the recommendations provided by participants are better motivated by personal variables, such as age and gender, rather than by the metaphorical framing they were exposed to. People preferred restrictive over preventive measures in general, although we saw a clearer gap in older speakers and men who displayed a greater preference for more restrictive measures to control the spread of the pandemic. Young participants and women were more likely to recommend a combination of preventive and restrictive measures. Starker differences were observed for the specific case of the machine framing. This framing seemed to increase the gap between both age and gender groups, making the elder and male groups select almost exclusively restrictive measures. This behaviour may be due to the inferences triggered by the machine framing, such as the possibility of activating commands and buttons that automatically shut down a machine. In application to the pandemic, these participants may have associated stopping the circuitry of a machine with the most restrictive measures and, therefore, the most successful in containing the spread of the virus. This is nevertheless an open question for future study.

Research question 3. The effect of metaphorical keywords

In terms of keywords, only 60 keywords coincided with any of the three metaphorical keywords featured in the paragraph they read. A total of 66 keywords referred to any of the other literal words in the paragraph (such as ‘pandemic’ or ‘virus’), and 89 words were not keywords present in the text participants read (see the full list of keywords in the repository).

A chi-square test for independence revealed a significant relationship between the type of keyword (literal, metaphorical, and not a keyword) and the framing condition (χ²(1) = 44.765, P < .001), meaning that some specific relationships differed from what could be expected by chance. Indeed, as shown in, Figure 9, the analysis of the Pearson residuals shows that metaphoric expressions related to fire were particularly influential in people’s decisions (N = 26, χ²(2) = 44.77, P < .001; highlighted in blue). Such findings could not be replicated for the rest of (in grey).

Given these overall results, it is necessary to zoom in on the keywords that instantiate the respective metaphorical frame in the stimuli materials. Recall that metaphorical keywords were embedded in three parts of the text (first position, to frame the pandemic; second position, to frame the measures to control the pandemic; and third position, to frame the result of such measures). A chi-square goodness-of-fit test could only partially confirm the serial-position effect (χ²(2) = 19.966, P < .001), as participants were more likely to refer to the metaphor-related keyword that appeared in the first position (N = 32) of their paragraphs to motivate their answers, followed by the keyword in the middle position (N = 24); but only very few referred to the keyword that appeared in the last place (N = 4).

As Figure 10 shows, in the breakdown of metaphorical keywords by experimental condition and by positioning in the text, the vast majority of metaphorically related keywords belonged to the fire (N = 26) and water conditions (N = 20), and over half of them appeared in the first place (N = 14 and N = 12, respectively), followed by the metaphors embedded in the middle of the text (N = 8 in both cases). In turn, participants did not seem to have noticed the metaphorical keywords in the machine condition or (interestingly enough) the war condition in any of the three positions, in spite of its documented pervasiveness across the first months of the pandemic in 2020 (or, perhaps, because of that ubiquity, that made them go unnoticed for participants while completing the experimental task).¹ This finding explains that the priming effect was highly indebted to the type of metaphorical frame used, with some of them being more impactful than others.

Conclusions

This article has explored the extent to which the variation of people’s perceptions of the pandemic and their recommendations to control the spread can be attributed to different metaphorical framings. In this study, participants were randomly exposed to one out of five texts (four containing metaphorical keywords and a fifth literal version to use as control) and were asked three questions designed to assess the impact of the metaphorical frames on their responses.

In response to the first research question, which sought to know the relationship between the different metaphorical frames and people’s perception of control over the pandemic, our findings show that metaphorical framings have a direct effect on the way we think and behave when faced with hardship. Overall, the fight frame seemed particularly fit to increase people’s perception of control. However, when looking closer at the data, this only applied to men and older participants. War metaphors did not seem useful for women and young informants, and thus, our data support the assumption that war framings are the most polarizing in that they make people react in more divergent ways than what was observed for other metaphors. Interestingly enough, the other metaphorical frames (fire, water, and machine) were more likely to increase the perception of control reported by women and young speakers and thus helped to narrow the gap between all participants’ ratings. These findings support, in an empirical way, the driving assumption behind many theoretical and qualitative works (e.g. Koller 2004; Flusberg, Matlock and Thibodeau 2017, 2018; Hendricks et al. 2018) supporting the detrimental effect of sustained use of militaristic terminology across a wide range of domains and for some specific individuals, and highlighting the need for alternative framings such as the ones collected in the ‘Metaphor Menu for People Living with Cancer’ (Semino et al. 2017), and the #ReframeCovid collection (Olza et al. 2021) to ensure meaningful and ethical communications between institutions, mass media, and citizens.

With respect to the second research question, i.e. the extent to which each metaphorical frame led to recommend restrictive or preventive measures, all participants were more likely to favour restrictive measures, such as imposing higher economic penalties, stricter lockdowns, or having more police patrolling the streets. However, when accounting for age and gender, this trend was influenced by the responses of men and older participants. Although women and young participants also showed a preference for restrictive measures, their responses also reflect a balance with other preventive measures, such as setting up educational campaigns or mass vaccination programmes. A further question for future study will be to explore whether the date of data collection may have some influence on these results. Data were collected right after the surge of a new wave of COVID-19 in the late spring of 2021 and may have motivated the preference for restrictive measures. It is virtually impossible to disentangle people’s actual perceptions of the pandemic from the ones triggered by the experimental stimuli if the study is conducted during the phenomenon being scrutinized. Further studies should look into this issue by attempting to replicate findings now that the pandemic has been declared controlled.

The last research question addressed the extent to which the metaphorical keywords had been relevant in determining participants’ answers. Results show that this was the case exclusively for the fire frame, at least in a consistent way. In addition, results did not confirm a serial-position effect, with no observable difference between the first or last keywords with respect to the middle ones. This may be due to the design of the experiment (participants were not able to go back and re-read the text) or because people were already heavily influenced by the actual situation of the pandemic external to the experiment and, therefore, did not need to rely on the text to answer the questions. However, despite the limitations of conducting a study on people’s perception of the pandemic during an ongoing health emergency, we still find that relying on naturalistic authentic examples of metaphorical frames has more research benefits than disadvantages in terms of generalization of the results and potential applications for policymakers.

Apart from these lines of improvement, there are two further paths of future research. The present study did not cover the political affiliation of the participants, a factor that has been shown to be decisive in similar studies. For example, prior research found a connection between right-wing voters and a preference for restrictive measures in response when exposed to the war frame (Reijnierse et al. 2015; Panzeri, Paola and Domaneschi 2021). It remains to be seen whether these results can be replicated, and whether when taken into account together with political affiliation, such effects are moderated by gender (specifically in female participants) and age (younger participants).

Finally, another aspect that should be taken into consideration is the cross-linguistic scope of these results. In other words, whether these materials can be replicated and be apt in other languages and cultures. The notion of aptness has been empirically shown to vary from country to country. In a study to underscore audience perceptions of nine metaphorical frames for COVID-19 (including war, ship, dance, and fire among others) (Brugman et al. 2022) , showed that the choice of source domain can influence whether metaphors are considered ‘apt’ and credible by people, and therefore, to potentially impact people’s attitudes and behaviours in times of crisis. Otherwise, the choice of the least apt source domain could hijack the communicative objectives. Interestingly, this study shows that even though metaphors were considered overall less apt than the literal condition, the literal condition was not appreciated or understood better or found more credible than the metaphorical frames. Among all the metaphors involved in the study, the dance and ship metaphors ranked as the most apt, possibly because they resonated better with the need of people to constantly adjust their behaviours in an ever-changing COVID-19 context. For a similar empirical study of people’s attitudes and perceptions of metaphors for COVID-19 in the US context, see (Burnette et al. 2021). Future studies should compare these results from Spanish speakers with speakers from other places.

All in all, this research confirms the affordances of metaphorical framing in influencing people’s perceptions and decisions, but also its limitations when accounting for age and gender. Overall, this study concludes that there is not a single frame that works in the same way for everyone. Therefore, institutions and mass media reporters need to bear in mind that relying on different frames may be a good strategy if they want to develop more meaningful and ethical communications with the different members of society.

Notes on Contributors

Paula Pérez-Sobrino is an Associate Professor in the Department of Modern Languages at the University of La Rioja. Her work deals with how metaphors and other types of figurative language help or hinder cross-cultural communication, both from a theoretical and an experimental perspective.

Iraide Ibarretxe-Antuñano is a Professor in the Department of Linguistics and Hispanic Literatures at the University of Zaragoza and an elected member of the Academia Europaea. Her research focuses on the relationship between language, cognition and communication from a cognitive, typological and psycholinguistic perspective.

They are both founding members of the research group Iberus Cognition in Action (ICON).

Acknowledgements

We are grateful to two external reviewers for their extensive feedback on an earlier version of this manuscript, which have been key to improve the final report of our study. Any remaining mistakes and inaccuracies are solely the authors’ responsibility. We would also like to thank Andrea Ariño Bizarro for her help in setting up the questionnaire in Limesurvey.

Funding

Funding for this study has been granted by the Spanish Ministry of Science, Innovation and Universities/FEDER (PID2020-118349GB-I00, PID2021-123302NB-I00), and the Government of Aragon (Psylex H11-17R; ICON group).

Footnotes

References