Counterfactual language, emotion, and perspective: a sentence completion study during the COVID-19 pandemic

1 Introduction

The state of global mental health dramatically deteriorated when the coronavirus disease 2019 (COVID-19) developed into a worldwide pandemic at the beginning of 2020 (Alzueta et al. 2020; Nikčević et al. 2021). In order to decrease the spread of the virus, governments imposed and reinforced regulations over long time periods. For instance, Germany and the United Kingdom implemented rules of social distancing, mask requirements, and lockdowns on public life from the beginning of 2020 (German Federal Government 2020a, b, 2021; UK Government 2020, 2021). Such drastic changes in social activity, work, and home life paired with the highly unpredictable nature of the virus increased negative emotional states, such as feelings of anxiety and symptoms of depression, in populations worldwide (Alzueta et al. 2020; Pandey et al. 2020; Racine et al. 2021; Schäfer et al. 2020; Wu et al. 2021).

Recent studies have found that changes of emotional well-being related to the pandemic are reflected in the language use of individuals. For instance, academic papers written during the pandemic contained more terms related to affect than those written before the outbreak, which suggests that writers were dealing with a crisis (Markowitz 2022). A longitudinal study described how narrative coherence was an enhancement factor to emotionally cope with stressful circumstances, such as the pandemic (Vanaken et al. 2022). Hence, language can be used to indirectly infer the emotional state and coping strategies of speakers or writers.

This raises the question of how other linguistic devices can be used to index emotional and cognitive processes. One example are the linguistic expressions of counterfactual thinking, which is a form of mental simulation. Counterfactual thoughts as reflected in language typically represent a juxtaposition to reality.

Among other forms, while talking about counterfactual thoughts, one often appeals to conditional sentences of the form If P, then Q, to express one’s belief in the consequent proposition Q supposing the truth of the antecedent proposition P (Liu 2019, 2022). Sentences such as (1) are referred to as indicative conditionals and those in (2) are subjunctive or counterfactual conditionals (von Fintel 2012). Unlike indicative conditionals, which typically consider possible situations, counterfactual conditionals consider alternative situations that are contrary to the facts, as illustrated in (2a) and (2b).

Counterfactual conditionals such as (2) are used to explore alternative outcomes (Sanna 2000; White and Lehman 2005), and have been an important area of interest for psychology of reasoning. If P, Q are statements which causally link two events or actions with each other (Roese and Epstude 2017). Through the process of mental simulation, an infinite set of such alternative events are possible. Since simulation is storage- and time-costly, certain aspects of reality, such as temporal relation, controllability, and expectability (or, following McCloy and Byrne (2000), exceptionality), restrict the set of events under consideration (Byrne 2015, 2002; McCloy and Byrne 2000; Van Hoeck et al. 2015). As a result, the chosen events or actions P for the form if P, Q reflect the alternative which matches best to these aspects, and thus has the strongest perceived significance leading to the outcome Q in the consequent (McCloy and Byrne 2000; Van Hoeck et al. 2015). In formal linguistics and analytical philosophy, different analyses have been under discussion for the semantics of counterfactuals; for example, on the similarity analysis, a counterfactual sentence if P, Q is true in the world of evaluation (w ₀) if all the P-worlds most similar to w ₀ are Q-worlds (Lewis 1973; Starr 2022). That is, (2) is true if all the worlds with the sky being overcast in the morning while being most similar to the actual world (where it was not) are worlds with rain in the afternoon.

As a central feature of human cognition, mental simulation, such as counterfactual thinking, associates with individuals’ emotional states and their coping strategies (Epstude and Roese 2008; Sanna 2000). Thus, counterfactuals are not only relevant for the psychology of reasoning, linguistics, and philosophy, but also for the psychology of emotion. The direct comparison between reality and the alternative event stresses the required conditions for improving or worsening the outcome (Broomhall et al. 2017; Epstude and Roese 2008), which is used to differentiate counterfactuals into two types (Markman et al. 1993). Upward counterfactuals involve mental simulation of better counterfactual worlds than reality, as seen in (3); downward counterfactuals involve mental simulation of counterfactual worlds worse than reality, as seen in (4) (Allen et al. 2014; Broomhall and Phillips 2018; Broomhall et al. 2017; Byrne 2002; Epstude and Roese 2008; Sanna 2000; White and Lehman 2005).

As (3) and (4) show, the same antecedent (P) can result in different directions of the counterfactual sentence. The speaker of (3) indicates that in a world where they had worked harder, they also would have reached their most favorable goal of getting the job (Q). In contrast, the same antecedent would result in a less favorable outcome for the speaker of (4); thus, this alternative world would be worse for the speaker. The directionality of the counterfactual thought depends on the causal relationship between the antecedent and the consequent, reflected in the mental simulation of an agent. This causal relationship can also have an impact on the resulting emotions. Accordingly, comprehenders can rely on the desirability of the consequent (Q) to infer about that of the antecedent (P) and of the simulated world of P and Q; subsequently, they can understand the agent’s emotion. Upward counterfactuals, due to the contrast between the actual world versus a better counterfactual world, evoke negative emotions, such as regret and shame; downward counterfactuals, due to the contrast between the actual world versus a worse counterfactual world, evoke positive emotions, such as relief. That is, the labels of upward and downward counterfactuals are based on the direction of counterfactual reasoning rather than the distinction between positive or negative emotions, which admittedly can be counterintuitive for the reader who encounters this terminology for the first time.

During the COVID-19 pandemic and ever since, newspaper articles frequently have employed counterfactual sentences to explore different outcomes of the pandemic. The quoatation in (5) is an example of an upward counterfactual, relating to a negative affect, while (6) an example of a downward counterfactual, relating to a positive affect.

Both types of counterfactual thoughts have specific functions related to behavior and affect regulation. Upward counterfactuals focus on the problem at hand that influenced the conceivably unwanted outcome; they can thus have an impact on future behavior in terms of performance improvement (Roese and Epstude 2017; White and Lehman 2005). Moreover, in the case of upward counterfactuals, the perspective can reveal the direction of responsibility (Broomhall et al. 2017); when taking an egocentric (self-focused, according to Epstude and Roese 2008) perspective, the learning effect is additionally enhanced by reflecting on self-improvement in contrast to non-egocentric (other-focused) perspectives (Epstude and Roese 2008; Roese and Epstude 2017).

Experimental evidence has shown that generally more upward counterfactuals are produced than downward counterfactuals, suggesting that the performance-improvement function is more often employed (Summerville and Roese 2008). It has also been shown that the self-improvement function (namely upward counterfactuals with self-referent) predominates over downward counterfactuals (Allen et al. 2014; Epstude and Roese 2008; Summerville and Roese 2008). As for downward counterfactuals, no difference in the use of self- or other-focus has been found, unlike with upward counterfactuals (Epstude and Roese 2008).

Besides behavior improvement, upward counterfactuals also serve a preparatory function (Eisma et al. 2020; Roese and Epstude 2017; White and Lehman 2005). The contrast between a more favorable alternative outcome and reality causes negative affect, which serves as potential motivation for future behavior (Medvec et al. 1995). Thus, upward counterfactuals serve as a problem-focused coping mechanism and enable a learning process from past events to regulate future behavior under the cost of amplifying the current negative affect (Allen et al. 2014; White and Lehman 2005). While upward counterfactuals occur more frequently after situations in which the outcome was changeable, downward counterfactuals appear more frequently when the outcome was inevitable (Epstude and Roese 2008) or caused a high level of stress (White and Lehman 2005). Here, the focus lies on emotion-focused coping: from the experience of an unfavorable outcome, positive emotions, such as relief, arise by imagining a worse alternative (Allen et al. 2014; Broomhall and Phillips 2018; Byrne 2002; Epstude and Roese 2008; Sanna 2000; Van Hoeck et al. 2015; White and Lehman 2005). They thus reduce negative emotions and serve an affective function (Byrne 2002; Eisma et al. 2020; Epstude and Roese 2008; Roese and Epstude 2017; White and Lehman 2005). In terms of behavior regulation, downward counterfactuals imply the preservation of the status quo since the event or one’s own actions were good enough (Epstude and Roese 2008).

Experimental evidence with Olympic athletes confirmed this tendency (Medvec et al. 1995). Participants were shown videotapes of interviews with bronze and silver medalists from the Olympic Games in 1992 and rated how much they are perceived to focus on what could have gone worse (downward) and how much on how they could have done better (upward). Silver medalists were found to focus more on upward thoughts than winners of the bronze medal. In addition, participants judged whether the athletes’ thoughts focused on (i) downward counterfactual scenarios compared to competitors who finished behind them (downward comparison), (ii) upward counterfactuals in comparison to competitors who finished ahead (upward comparison), or (iii) their own accomplishment without engaging in comparison with others (neutral no comparison). Downward counterfactual thoughts were rated as less frequent in both medalists. Here, silver medalists were found to focus more on upward comparisons than bronze medalists. The latter were rated to engage in more self-focused thoughts (neutral no comparison) than silver medalists.

Similar results were found in a study from the Olympic Games in 2016 (Allen et al. 2019). Participants were shown static photos of the athletes, and similarly, they rated silver medalists to express more counterfactuals in general, particularly in the upward direction. The authors concluded that the extent of counterfactual thought correlates with relative height on the podium; silver medalists focus on the failure of not winning a gold medal, which could potentially serve as a motivation to improve their performance.

The coping effect of counterfactual thinking has been experimentally studied in the context of depression and anxiety. A longitudinal study with participants who had experienced the loss of a family member in the past three years showed a correlation between the type of counterfactual used as well as the perspective taken (the egocentric perspective “I” or non-egocentric perspective involving other people) and depression. While downward counterfactual use was not found to associate with depression symptoms, both other and self-referent upward counterfactuals related to depression and longitudinal depression symptoms. Thus, focusing on one’s own actions in the context of loss increases self-blame and guilt (Eisma et al. 2020). Predominant upward counterfactual thinking was also found in individuals with anxiety symptoms, such as women who experienced a miscarriage (Callander et al. 2007). Moreover, meta-analyses suggest there is a reliable relationship between negative affect, such as depression, and the extensive use of upward counterfactuals in general (Allen et al. 2014; Broomhall et al. 2017; Epstude and Roese 2008).

While coping mechanisms, such as counterfactual thinking, appear in everyday life, a global pandemic that has caused the most deaths since the Spanish flu represents an extraordinary situation. Due to the pandemic, negative emotions were reported as being more central to individuals’ emotive states in general (Whiston et al. 2022). Studies from the very start of the pandemic suggested a deterioration in mental health, such as severe distress causing depression and anxiety symptoms as a result of isolation measures imposed by governments (Alzueta et al. 2020). However, some demographic characteristics (e.g., young age, female gender, higher country income classification), and certain personality traits (for example, neuroticism, a disposition to experience negative affects such as anger, anxiety, emotional instability, and depression; Widiger and Oltmanns 2017) proved to be risk factors for a stronger decrease in mental health (Alzueta et al. 2020; Nikčević et al. 2021). In a global survey study with participants from 59 countries worldwide (Alzueta et al. 2020), the authors used, among others, the Epidemic-Pandemic Impact Inventory (EPII; Grasso et al. 2020) and the 21-item subset of the Depression, Anxiety, and Stress Scale (DASS21; Lovibond and Lovibond 1995b). Most participants reported normal anxiety symptoms (80.5 %) and mild depression symptoms (74.6 %). Both symptoms correlated with being female, young, and from a country classified with higher income, among other factors; thus, subjects with these characteristics were more likely to report high depression and anxiety levels. In terms of the effect of age, younger adults could potentially be more exposed to COVID-19-related media, be more affected by financial instability, and might have less efficient coping mechanisms than older adults. Moreover, the authors found evidence that the level of negative affect was robustly predicted by COVID-19-related effects on private life.

In a related way, certain personality traits were shown to affect an individual’s experience of negative affect during the COVID-19 pandemic (Nikčević et al. 2021). High levels of neuroticism were assumed to be a vulnerability factor, increasing the level of experienced negative affect. Extraversion and conscientiousness were assumed as possible protective factors, thus reducing the risk of experiencing negative affect. In a survey study (Nikčević et al. 2021) using among other instruments, the 10-item subset of the Big Five Inventory survey (BFI-10; Rammstedt et al. 2012), the results showed that high generalized anxiety and depression levels are predicted by certain personality traits and age. Personality traits, such as extraversion, agreeableness, and conscientiousness, related negatively with generalized anxiety; thus, the higher their scores, the lower the anxiety level. Neuroticism was found to positively relate to general anxiety and is directly connected to COVID-19-related anxiety and high depression levels. Moreover, younger age was found to associate with higher levels of anxiety. Nikčević et al. concluded that neuroticism acts as a vulnerability factor for generalized and COVID-19-related anxiety, while conscientiousness was confirmed as a protective factor. Additionally, agreeableness and extraversion highly predicted low COVID-19-related anxiety symptoms. Nikčević et al. suggested that these traits might favor certain coping mechanisms which alleviate emotional distress.

Outside the context of the pandemic, certain personality traits have also been reported to be associated with a predominant use of a certain counterfactual direction. For instance, neuroticism has been found to positively relate to upward counterfactuals (Allen et al. 2014; Ardakani et al. 2015), while high levels of extraversion and openness lead to less frequent use of upward counterfactuals (Allen et al. 2014).

In the current study, we were originally interested in (i) the relationship between counterfactual thinking and perspective-taking in general, and (ii) how counterfactual thought processes relate to personality traits, emotional states, and the perceived situation during the COVID-19 pandemic. However, a post hoc power analysis we ran shows that the study was underpowered for (ii). Due to the nature of the study, further data collection is not possible; thus, we only provide the descriptive statistics of the data for (ii) in the appendix. In the following, we will report the details of (i).

Based on the literature, then, upward counterfactuals are used more frequently than downward counterfactuals due to their improvement function and use as a problem-focused coping mechanism, while upward counterfactuals appear more frequently with self-reference. Thus, we hypothesized that the direction of counterfactuals relates to the perspective-taking in such sentences. In order to test this, we assessed counterfactual thinking with a COVID-19-themed sentence completion task in Germany and the UK. The pandemic served as a common topic which led to similar influences on people’s lives in both countries due to similar regulations (i.e., lockdown, mask requirements) during the time of data collection. More specifically, we had the following hypotheses:

Furthermore, we investigated for exploratory purposes whether the number of upward and downward counterfactuals differ in the non-egocentric condition.

2 Methods

3 Results

3.1 Experiment 1

In the dataset of the sentence completion task, 10 sentence completions were ungrammatical, and 160 counterfactuals were classified as neutral: 108 were egocentric sentence completions, 52 non-egocentric ones. One can raise the question why there were more egocentric ones who were neutral, which we are unable to answer in the current study. In total, 16 % of the 1212 sentences were removed. Thus a sample of 1018 counterfactual sentences entered further analysis (see Table 1 and Figure 1 for an overview).

Table 1:

Number of counterfactual sentences analyzed in Experiment 1, categorized by type of conditional and perspective. The values in parentheses represent the percentage in comparison to the total dataset.

Perspective	Type of conditional		Total
	Upward	Downward
Egocentric	317 (31.1 %)	162 (15.9 %)	479
Non-egocentric	262 (25.7 %)	277 (27.2 %)	539
Total	579	439	1,018

Figure 1:

Results of the sentence completion task of Experiment 1 (Germany) and Experiment 2 (UK) in comparison. The x-axis shows the counterfactual types. The y-axis depicts the number of upward and downward counterfactuals produced, and their percentages with respect to the specific dataset provided in the bars.

The results of the models 1, 2a, and 2b are depicted in Table 2. First, the Poisson regression revealed a significant difference between the number of upward and downward counterfactuals, in that there were more upward counterfactuals (model 1: β ^  = 0.28, χ ²(1) = 19.32, p < 0.0001).

Table 2:

Output of the analysis for Experiment 1 using generalized linear models for a Poisson regression (model 1), and generalized linear mixed models for a binomial regression with logit-link function (model 2a/b).

	Fixed effects				Model comparison
	Estimate	β ^	Std. error	z Value	χ 2(1)	p Value
Model 1	Intercept	1.50	0.05
	Type	0.28	0.06	4.37	19.32	< 0.0001
Model 2a	Intercept	−0.09	0.44	−0.21
	Egocentric	1.48	0.40	3.66	9.52	0.002
Model 2b	Intercept	1.03	0.55	1.87
	Non-egocentric	−1.19	0.18	−6.69	46.71	< 0.0001

Second, for model 2a of the binomial regression, the model with random subject as well as item intercepts with random slopes for the egocentric perspective fit the data best. The result showed that the odds ratio of producing upward counterfactuals was significantly different in the egocentric versus non-egocentric perspective (model 2a: β ^  = 1.48, χ ²(1) = 9.52, p = 0.002). The transformation from log-odds resulted in an 80.0 % probability to obtain upward counterfactuals in the egocentric condition, that is, more upward counterfactuals than downward counterfactuals.

For model 2b, the model with random subject as well as item intercepts fit the data best. The result showed that the odds ratio of producing upward counterfactuals was significantly different in the non-egocentric versus egocentric perspective (model 2b: β ^  = −1.18, χ ²(1) = 46.71, p < 0.0001). The transformation from log-odds resulted in a 47.7 % probability to obtain upward counterfactuals in the non-egocentric condition, that is, fewer upward counterfactuals than downward counterfactuals.

4 Discussion

The current study examined the relationship between counterfactual thoughts, emotions, and perspective-taking during a real-life crisis (the COVID-19 pandemic). We ran two experiments, in Germany and in the UK, which had similar restrictions on people’s day-to-day life during the time of data collection. More specifically, experimental participants read COVID-19-themed scenarios containing an incomplete conditional sentence, with the conditional antecedent establishing a counterfactual situation in which the pandemic itself did not happen or certain, pandemic-related regulations were not in place. Using the sentence completions that were produced, we tested whether and how counterfactual direction (i.e., upward or downward) relates to the perspective-taking (i.e., egocentric or non-egocentric) of the statement.

The results showed that a significantly higher number of upward than downward counterfactuals were produced in the German dataset, while there was no such difference in the English dataset. Thus, hypothesis H1 was only partially confirmed. With regard to perspective manipulation, we found converging evidence. In both datasets, we found an effect in that more upward than downward counterfactuals were produced in the egocentric condition. This confirms our hypothesis H2, providing experimental evidence for the related claim in the literature. Furthermore, we found in both datasets that fewer upward counterfactuals were produced than downward counterfactuals in the non-egocentric condition.

Overall, our findings provide an interesting insight into the influence of perspective on counterfactual thoughts and emotions. Even though there was no significant difference between the number of upward and downward counterfactuals in the UK dataset in contrast to the German data, the direction of counterfactual thoughts is similarly perspective-dependent across both datasets, showing a significant upward versus downward difference of opposite distribution in the egocentric versus non-egocentric condition. People produced more upward than downward counterfactuals when they talked about themselves and fewer upward counterfactuals when they talked about others.

The finding that people engaged in more egocentric upward counterfactual thinking reveals that the pandemic caused severe changes in people’s habits and plans. In the datasets, people more often reported on what they missed out on because of regulations than on positive aspects. Example (9) from Experiment 2, the British dataset, shows a contrast between upward and downward counterfactuals. While (9a) focuses on the impression that there was nothing to appreciate, (9b) shows that even in such a crisis, the person found a positive aspect which supposedly would not have happened without a pandemic. Moreover, example (10) shows how emotive words are used in the context of the pandemic. While the upward counterfactual in (10a) shows how the lockdown in Germany had an emotional impact on the participant in a negative way, the downward counterfactual in (10b) shows how the participant would have been negatively effected if measures had not been imposed.

Before we finish the paper, we would like to briefly address the scope and limitations of the current study. Firstly, the classification of the counterfactuals into upward and downward was done manually and is thus at least partially subjective. In future work, asking participants after they complete a sentence whether it represents an improvement of the situation, or, as was done in the study by White and Lehman (2005), for the reasons behind those thoughts, could further improve the results and avoid the removal of numerous neutral sentences. Secondly, our items related to situations which were not in the control of the participants; for example, whether or not the pandemic happened was out of the control of the individual. Given that the comparison between controllable and uncontrollable events did not form part of our experimental design, this requires further research. Thirdly, there might be more constraints on the causal relation between the antecedent and the consequent proposition in the mental simulation. The emotional valence of an antecedent proposition might have an effect on the sentence continuation: for example, Wenn die Pandemie nicht gewesen wäre ‘If the pandemic had not happened’ (item 1) might more strongly prime upward completions (due to the negative nature of the pandemic), relating to negative emotions. In contrast, Wenn Homeoffice nicht möglich gewesen wäre ‘If working from home had not been possible’ (item 2) might more strongly prime downward completions (possibly, due to the positive connotation of the modal verb möglich in a “permission” reading), relating to positive emotions. For exploratory purposes, we looked at the descriptive statistics of the items for both the German and the UK study. They show that for item 1, there were 80 upward and 8 downward counterfactuals in the German study and 74 upward and 12 downward counterfactuals in the UK study. For item 2, there were 15 upward and 70 downward counterfactuals in the German study and 37 upward and 48 downward counterfactuals in the UK study. As the counts are our measure, we are not able to draw conclusions about the nature of the antecedent propositions used in the items. Due to the time-sensitivity of the present study, we will leave this aspect to be considered in future research of similar kinds, which can include, for example, a norming pretest about the emotional valence of the antecedent propositions (positive vs. negative).

With these limitations taken into consideration, our study used a factorial design with strictly controlled contexts, in contrast with previous studies using open production measures such as sentence production in interviews (see Medvec et al. 1995; White and Lehman 2005). The main finding of the study is that more upward counterfactuals indicating negative emotions were produced in egocentric perspective-taking; this provides convergent evidence for previous research in this regard. The opposite pattern in the non-egocentric perspective that the study found has not been reported previously to our knowledge. Last but not least, in the realm of conditionals, researchers have related optatives (e.g., If only I had caught the train!) to expressing speaker emotions such as regret about having not caught the train in this case (Grosz 2012). Our study provides empirical evidence that not only optatives but also regular counterfactual conditionals can signal speaker emotions.

5 Conclusions

To sum up, the COVID-19 pandemic has affected our life drastically since it started in 2020; approaches to deal with it differed, however, from country to country, and from person to person. Focusing on the deteriorating mental health in the population, we conducted a study in the UK and Germany at a critical point in time when both countries were in lockdown. We used a sentence completion task to measure the direction of counterfactual thoughts in egocentric and non-egocentric sentence contexts. Results show that in both samples, the direction of counterfactuals, as the index of direction of emotions, is perspective-dependent, in that more upward counterfactuals (relating to negative emotions) were found in the egocentric condition, with fewer upward than downward counterfactuals in the non-egocentric condition. We conclude that the direction of counterfactual thinking is perspective-dependent. We hope that the method we developed is applicable and extendable to access counterfactual language, thoughts, and emotions in general.

Supplementary Materials (data availability statement)

The datasets and code files generated for this study are available on the OSF repository: https://osf.io/tyh8p/.