Empathy—the capacity to share and understand others’ emotional states—is highly related to the subjective experience of emotions and mirroring others’ emotions (Decety and Jackson, 2006). Prior studies found that people with high relative to low empathy could capture negative emotions more quickly and experience them more deeply (Yan et al., 2016; Sun et al., 2018). Using facial electromyography, researchers further revealed that high (vs. low) empathy individuals induced increased corrugator supercilii activity in processing disgusting and fearful facial expressions, suggesting increased sensitivity of negative emotions in high empathy individuals (Rymarczyk et al., 2016). These findings suggest that high relative to low empathy people show more emotional reactions (e.g., more negative emotions and physiological responses) to process emotional stimuli conveying threatening/negative information (e.g., fear). However, little attention has been paid to the role of empathy in the modulation of emotions under a pandemic threat. Recently, a few studies uncovered that when compared to low empathy people, high empathy people experienced more emotional disorders during the COVID-19 pandemic [e.g., anxiety, depression, and poor sleep quality, Guadagni et al. (2020) and Petrocchi et al. (2021)]. Similarly, the association between empathy and negative emotions (e.g., anger) was replicated in an Eastern sample of 453 Chinese during COVID-19 (Ma and Wang, 2021). So far, most of these studies have focused on the impact of empathy on negative emotions, e.g., vigor, depression, distress (Guadagni et al., 2020; Van de Groep et al., 2020; Grignoli et al., 2021). Considering that positive emotions are of great significance in fostering subjective well-being (Livingstone and Srivastava, 2012) and help promote psychological resilience when facing an ecological threat [e.g., Yamaguchi et al. (2020) and Gurvich et al. (2021)], we are curious about whether empathy may help boost the public’s positive emotions to relieve stress. Furthermore, we are interested in whether empathy may modulate the positive and negative dimensions of personal emotions, the characteristics of the two dimensions (i.e., pattern similarity), and the relationship between the two dimensions.

The relationship (i.e., the ratio) of the two dimensions of personal emotions has been found to function as an indicator of psychological well-being (Larsen, 2009). Toward a better understanding of the relationship between positive and negative emotions, researchers have proposed different theoretical models and debated for a long time [e.g., the bipolar vs. bivariate models, Larsen et al. (2001); Schimmack (2001), and Ong et al. (2017)]. The bipolar model holds the view that positive and negative emotions are opposite to each other, that is, positive emotions and negative emotions are mutually exclusive and opposite to each other, and an individual can experience only one of them at the same time (Russell and Carroll, 1999; Liu et al., 2008). On the contrary, the bivariate model suggests that the two are not mutually exclusive (Larsen et al., 2001), and they could coexist and negatively correlate with each other (Liu et al., 2008). One possible account behind the debate could be that the sources (e.g., personal affairs and social contexts) that trigger individuals’ emotions are complex and diverse rather than singular. Considering that the complexity of determining the sources creates obstacles to understanding emotions and their patterns, it is worth mentioning that COVID-19, as a pandemic threat that has heightened the public’s collective emotions (Stanley et al., 2021), could provide a shared social context and contributes to a group perspective for understanding the relationship between positive and negative emotions and its underlying characteristics (i.e., pattern).

A previous study has shown that when exposed to a stressful event (e.g., a stressful speech), individuals’ positive and negative emotions exhibited more polarization—a higher level of negative correlation (Zautra et al., 2000). This finding suggests that a stressful event could lead to attentional narrowing, which may in turn influence personal emotional states to jointly help humans respond rapidly when facing threats (Hermans et al., 2014). Considering that COVID-19 is a severe stressor that has posed psychological burdens (Elbay et al., 2020; Gritsenko et al., 2020), we speculated COVID-19 might lead to a negative association between positive and negative emotions. Given that individuals with high (vs. low) empathy are more sensitive to negative social events (Yan et al., 2016; Sun et al., 2018), we predicted that people with high (vs. low) empathy might be more likely to show more polarization in the relationship between the two personal emotions.

Although previous studies have shown that people felt fearful about people and groups related to the COVID-19 disease, such as fear of interacting with people suffering or recovered from COVID-19, toward people related to COVID-19 (Bagcchi, 2020; Roberto et al., 2020; Abuhammad et al., 2021). Meanwhile, some studies have revealed that people would also have positive (or prosocial) feelings, i.e., sympathy, about people related to COVID-19 (Li et al., 2020). The emotional feelings toward others mentioned above are real (or expected) emotional experiences and reactions that people generate in real (or imagined) interactions with others, which are referred to as social emotions (Mercer, 2014). According to the previous research, we speculate that in the face of the pandemic, individuals’ social emotions toward disease-related groups are complex; that is, negative social emotions (fear) and positive social emotions (sympathy) coexist.

Previous evidence has shown that the psychological process of people’s social feelings toward others is closely related to their ability to empathize (Schipper and Petermann, 2013). Based on the definition of the two components of empathy, cognitive empathy reflects an individual’s ability to understand others’ emotions, and affective empathy refers to an individual’s capacity to share others’ emotions (Deutsch and Madle, 1975; Cox et al., 2012). High empathy people therefore may be better at sharing others’ emotions and understanding others’ situations. For example, high compared to low empathy individuals were more willing to offer money and time to assist people with difficulties (Lay et al., 2020; Xiao et al., 2021). Moreover, empathy-related neural activity (i.e., medial prefrontal cortex) contributed to the subsequent empathic concern toward the victim in need (Masten et al., 2011; Spencer-Rodgers et al., 2010; Perez-Bret et al., 2016). Consistently, high empathy people tend to predict high prosocial behavior under the COVID-19 threat (Taylor et al., 2020; Lv et al., 2021; Seong-Wook et al., 2021). Accordingly, we speculated that individuals with high vs. low empathy might generate more prosocial emotions (i.e., sympathetic) toward people in need (e.g., COVID-19 patients) during COVID-19.

On the other hand, recent evidence suggests that empathy may account for people’s fear and avoidance of disease-related groups who may carry the virus during the COVID-19 pandemic. For example, people tried to identify people who might pose a potential infection risk when facing a pandemic threat (Troisi, 2020). In line with this, high empathy people were observed to show more self-protection tendencies during COVID-19, such as engaging in physical distancing and wearing facial masks (Sassenrath et al., 2021). It is assumed that more fear of the disease-related groups and increased self-protection focus among high relative to low empathy people are served as a protective mechanism, which supports keeping away from danger when facing the threat of COVID-19 (Ramaci et al., 2020). Though the two seemingly contradictory aspects (i.e., fear and sympathy) of social emotions are related to empathy, so far, it remains unclear whether or to what extent empathy may simultaneously affect the two social emotions toward disease-related groups and their relationship under a pandemic threat.

The current study aims to examine the impact of empathy on personal and social emotions and their patterns under the threat of COVID-19. Previous research mostly focused on the mean or sum value of different items within the two-valence domains [i.e., calculate two values to represent positive and negative emotions, e.g., Gan and Fu (2022) and Matiz et al. (2022)]. Thus, it may hardly capture whether individuals’ emotions, as well as their feelings about others, may differ in their pattern characteristics in terms of similarity (Riberto et al., 2019). The above issues can be easily addressed by using representational similarity analysis (RSA)—a computational technique that utilizes pairwise comparisons of units to reveal the similarity pattern among a set of variables (Haxby et al., 2014). Unlike the traditional linear correlation testing of two-dimensional values, RSA has been gradually used in understanding behavioral as well as neural patterns in the field of social psychology (Brooks and Freeman, 2018), which helps uncover the pattern similarity and further provides direct comparisons across conditions (Luo et al., 2021).

In the current study, taking advantage of RSA, our first goal was to examine the pattern similarity of the public’s personal emotions (i.e., positive and negative emotions) and the pattern similarity between the two types of emotions during the COVID-19 pandemic. To address this question, we measured Chinese positive and negative emotions (e.g., inspired, active, nervous, and scared) when they were facing the COVID-19 pandemic. Our second goal was to reveal the pattern similarity of the public’s social feelings toward a variety of social groups (e.g., disease-related people, people violating moral norms, and healthy people) during the COVID-19 pandemic. As for social feelings, we first focused on the two sympathetic and fearful feelings toward the COVID-19-related people and groups (e.g., people infected by COVID-19, people who were suspected patients of COVID-19, and people who have been recovered from COVID-19), which have been demonstrated to be stigmatized during this pandemic [e.g., Baldassarre et al. (2020) and Ransing et al. (2020)]. In addition, we were curious about whether their feelings toward the other disease-related groups, such as people suffering from an infectious disease (e.g., AIDS) and people with mental illness (e.g., schizophrenic patients) may show more pattern similarity. Researchers believe that disease-related groups are usually viewed as social deviants who violate certain norms by the public (Phelan et al., 2008). We, therefore, included people who violate the moral norms (e.g., robbers). Finally, we added non-disease people (e.g., healthy people, natives) as control. Last but not the least, the third goal of the study was to test whether people who differ in the prosocial propensity (i.e., trait empathy) may elicit dissimilar patterns of personal and social emotions to help policymakers understand collective emotions in the context of such a crisis. To address this, we measured participants’ trait empathy using a Chinese version of the Interpersonal Reactivity Index (IRI) Scale (Zhang et al., 2010) and compared the pattern similarity of personal and social emotions between the high and low empathy people.

We collected two waves of data during the recovery period of COVID-19 (one is from May 2 to July 12, 2020; and the other is in December 2021, the severity of the outbreak is similar to the former in terms of the number of locally confirmed cases in China), using convenience sampling method (via WeChat spreading) through the online platform like Qualtrics.¹ According to Meade and Craig (2012)’s recommendation for detecting careless data, we ruled out data with wrong answers in the probe questions (e.g., if you notice this question, please select the option “6”; if you choose any other options, your data will be invalid) (N = 32) and invalid data (only completed the informed consent part of the questionnaire) (N = 21). We excluded 21 participants from Hubei Province, considering that their perceptions of COVID-19 related groups might be different from those who were from other Chinese provinces because the Hubei province had the most serious outbreak of COVID-19 in China during the period that we conducted this study, and people in Hubei had mostly experienced lockdown. Additionally, previous research has shown that people in the areas with severe epidemics have different risk perceptions of the COVID-19 pandemic than those from other areas (Wen et al., 2020). The final sample was 345 (120 males and 224 females, mean age 25.13 ± 7.85 years). Before the survey, all participants were given informed written consent. The present research was approved by the ethics committees of the Institute of Psychology, Chinese Academy of Sciences the institute. The current research was from a big project on COVID-19 and mental health. The dataset of social emotion was from a previously published study (Zhu et al., 2022). In the current study, we reanalyzed the data using representation similarity analysis to address the empathy modulation on emotions, which was not investigated by the previous study.

We first analyzed the distribution of all emotions (Supplementary Results and Supplementary Figure 2). Generally, participants experienced a low frequency of negative emotions but a medium frequency of positive emotions during the COVID-19 pandemic (Figure 1A). The RSM of the two personal emotions revealed that the positive and negative emotions were inversely correlated with each other in general (Figure 1B).

We calculated the distribution of the social emotions of the disease, control, and deviant clusters (Supplementary Results and Supplementary Figure 3). A medium-to-high rate of sympathetic feelings and a similar rate of fearful feelings were observed for the disease cluster (Figure 1C) and lower rates of the two social feelings were shown for the control clusters (Figure 1D). To further examine the pattern similarity of the social feelings, we calculated the sympathy RSM and the fear RSM (Figure 1E) for the two clusters separately. Results showed that the sympathy RSM was correlated with the fear RSM in the disease (r = 0.39, p < 0.05) and control clusters (r = 0.87, p < 0.05).

First, the distribution results of personal emotions showed that the high (vs. low) empathy group reported a higher frequency of positive and negative emotions (Supplementary Results and Supplementary Figure 4). Consistently, the independent t-tests on each personal emotion item showed that the high (vs. low) empathy group reported a significantly higher frequency of the following emotions: active, upset, nervous, jittery, irritable, scared (2.00 < ts < 3.35, ps < 0.05, Supplementary Table 1). We then calculated the RSMs of the personal emotions and found more pattern similarity of the negative RSM in the high (vs. low) empathy group [t₍₂₀₎ = 3.19, p < 0.01, Figure 2A]. No difference in the pattern similarity of the positive RSM between the two groups [t₍₅₎ = 1.15, p = 0.30]. Moreover, the positive-negative RSM of the high empathy group was more dissimilar than that of the low empathy group [t₍₂₇₎ = −12.90, p < 0.001, Figure 2A], indicating that less pattern similarity between the two personal emotions was induced in the high vs. low empathy group during the pandemic.

The distribution results showed that the high (vs. low) empathy group showed a higher rate of fear and sympathy for the disease cluster (Supplementary Results and Supplementary Figure 5). Specifically, the independent t-tests showed that the high (vs. low) empathy group reported a higher rate of fear toward COVID-19 patients, AIDS patients, and Flu patients (1.98 < t < 2.53, ps < 0.05) and a higher rate of sympathy toward all disease-related people (2.05 < t < 3.93, ps < 0.05, Supplementary Table 2). Paired sample t-test showed that the high (vs. low) empathy group showed less pattern similarity in the fear RSM for the disease [t₍₂₀₎ = −2.96, p < 0.01] and control clusters [t₍₅₎ = −12.14, p < 0.001, Figure 2B]. In addition, the pattern similarity of the sympathy RSM of the high (vs. low) empathy group was lower for the disease cluster [t₍₂₀₎ = −3.14, p < 0.01, Figure 2C] but not for the control cluster [t₍₅₎ = −1.80, p = 0.13]. These results suggest that high empathy people may induce more dissimilar and mixed social feelings about the disease-related groups to help them better adjust when facing a pandemic threat.