Perceived risk, emotions, and stress in response to COVID‐19: The interplay of media use and partisanship

Abstract

Relationships between risk perceptions, emotions, and stress are well‐documented, as are interconnections between stress, emotion, and media use. During the early COVID‐19 pandemic, the public responded psychologically to the threat posed by the pandemic, and frequently utilized media for information and entertainment. However, we lack a comprehensive picture of how perceived risk, emotion, stress, and media affected each other longitudinally during this time. Further, although response to the pandemic was highly politicized, research has yet to address how partisan affiliation moderated relationships between risk, emotion, stress, and media use over time. This three‐wave (N = 1021) panel study assessed the interplay of risk, emotion, stress, and media use for Americans with different political affiliations between March and May of 2020. Findings indicate that perceived risk, emotion, and stress at Time 1 predicted media use at Time 2, with predictors varying by type of media. Use of entertainment media and social/mobile media predicted later stress (Time 3), but news consumption did not. Later risk perceptions (Time 3) were not influenced by media use at Time 2. The predictors and consequences of different types of media use were notably different for Republicans and Democrats. In particular, risk perceptions predicted greater news use among Democrats but greater entertainment media use among Republicans. Moreover, social/mobile media use resulted in perceiving the risks of COVID‐19 as less serious for Republicans while increasing stress over time for Democrats.

1. INTRODUCTION

For many people, the COVID‐19 pandemic has been a significant life stressor. In the United States alone, COVID‐19 has caused over 91 million infections and more than 1 million deaths as of August 2022 (Centers for Disease Control & Prevention, 2022). On average, the public has perceived a high level of risk associated with COVID‐19 (Dryhurst et al., 2020), which has translated into a high level of stress (Yıldırım et al., 2022). Indeed, the prevalence of stress in the U.S. population became apparent early in the pandemic experience: in March 2020, 60% of Americans reported feeling stressed (McCarthy, 2020), while a third of U.S. adults experienced high levels of psychological distress in March and April of 2020 (Keeter, 2020). These reports are consistent with research showing that previous viral outbreaks increased stress in various populations (Buheji et al., 2020; E. Y. L. Cheung, 2015), even a year after an outbreak (A. M. Lee et al., 2007).

The early months of the COVID‐19 pandemic also witnessed a significant uptick in the use of various social and traditional media (Riehm et al., 2020). This was likely due, in part, to the public's escalating stress and corresponding effort to reduce negative feelings through media‐based coping (Eden et al., 2020; Pahayahay & Khalili‐Mahani, 2020), as well as the need to obtain information to address the risks associated with COVID‐19 (Goel & Gupta, 2020). Indeed, during the early pandemic many sources offered advice on how to use media as a stress‐buffering tool (Kopecki, 2020). However, recent studies also suggest that increased media use during the pandemic promoted higher risk perceptions, more negative emotions, and greater stress (Brailovskaia et al., 2021; Olagoke et al., 2020; Riehm et al., 2020). These effects also varied depending on the types of media and their use. For example, Pahayahay and Khalili‐Mahani (2020) found that self‐selected information or entertainment content helped users to cope with COVID‐19 stress, while consumption of negative news and social media use worsened it. Conflicting findings from these cross‐sectional studies suggest the importance of understanding the longitudinal interplay of different types of media use, risk perceptions, emotional responses, and stress.

Given the politicization of the pandemic, there is also reason to believe that the relationship between media use and psychological responses was moderated by political affiliation. In the United States, Democrats and Republicans differed in perceived risk associated with COVID‐19, emotional responses to COVID‐19, and stress (e.g., Bruine de Bruin et al., 2020). These differences appear associated with different patterns of media use. A large, nationally representative survey conducted in April and May 2020 showed that Democrats were more likely to obtain COVID‐19‐related news from MSNBC or CNN, whereas Republicans were more likely to obtain their news from the Fox network (Bruine de Bruin et al., 2020). The partisan groups also differed in their consumption of content from other media sources (Ali et al., 2020). To the extent that media use is moderated by political affiliation, there may be divergent relationships among media use, risk perceptions, emotional experiences, and stress responses.

To inform theorizing about the relationship between major societal stressors and long‐term health outcomes, and to guide responses to the ongoing pandemic and future public health crises, we designed the current three‐wave panel study to examine how risk perceptions, emotional responses, and stress both predicted and were influenced by media use during March and April, 2020, and how these associations varied as a function of political affiliation. Below, we detail the theoretical basis for our study design.

1.1. Risk perceptions, emotions, and stress

Stress arises when environmental demands are appraised as exceeding personal capacity to meet, mitigate, or alter those demands. As highlighted by the transactional model of stress and coping (Lazarus & Folkman, 1984), emotional responses are key for understanding stress. Positive emotions increase resilience and help individuals cope (Folkman & Moskowitz, 2000; Ong et al., 2006). In particular, research demonstrates that positive emotions not only help people recover from stressors (Tugade & Fredrickson, 2004) but they also predict less negative emotion and stress on subsequent days (Leger et al., 2020). In contrast, negative emotions are linked to more intense stress responses (Denson et al., 2009) and can diminish any positive emotions individuals experience when exposed to a stressor (Zautra et al., 2001). As described by appraisal theorists, emotions, in turn, are likely to stem from perceptions of risk (Frijda, 1986; Lazarus, 1991). Perceived risk from a threat is a function of susceptibility to that threat and the severity of consequences should the threat materialize (Griffin et al., 2008; Witte et al., 1996). Perceiving a risk as severe or likely typically fosters negative emotions like fear or sadness. However, anger can arise from a combination of lower risk and identification of parties to blame for the threat (Lazarus, 1991; Lerner & Keltner, 2001). Positive emotions like contentment or hope are likely to be experienced if risk perceptions are low and people believe the situation is likely to turn out well (Slovic et al., 2004).

Overall, the interconnections between risk perceptions, emotions, and stress are well‐documented, which provide a basis for understanding mental health during a crisis like the COVID‐19 pandemic. These psychological experiences, however, often interact in complex ways with social environmental factors, such as the media use that was particularly prominent during the COVID‐19 pandemic (Riehm et al., 2020).

1.2. The role of media use

1.2.1. Media use and risk perceptions

Risk perceptions promote media use by increasing information needs (Griffin et al., 1999). Media use, in turn, shapes risk perceptions in a wide variety of contexts (Koné & Mullet, 1994), depending on how the risk is framed (Priest et al., 2010) and the audience's dependence on media (Morton & Duck, 2001). Messages about risks can increase risk perceptions by increasing the mental availability of risk‐related information (Shrum, 2009), which comports with the recent observation that consuming more COVID‐19 news on mainstream media increased perceived COVID‐19 risk (Olagoke et al., 2020). However, viewing amusing media can decrease the perceived severity of issues (Zillmann et al., 1994) or lead to message discounting (Nabi et al., 2007). Indeed, research found that consuming humorous content related to COVID‐19 on social media led to reduced risk perceptions (Mohamad, 2020). Thus, the effects of media use on risk perceptions likely depend on the genre(s) of media consumed.

1.2.2. Media use and emotions

Media use is often driven by emotional states (Yang et al., 2018); media use, in turn, elicits a myriad of emotions in users (Bryant et al., 2003; Valkenburg & Peter, 2013). According to mood management theory, people are often subconsciously drawn to media that they have previously found to help them cope with negative emotional states (Knobloch, 2003; Zillmann, 1988). Relatedly, uses and gratifications approaches argue that people sometimes purposefully turn to certain types of media to help regulate their emotional states, to make them feel better, to empathize with others, or to distract themselves (Ruggiero, 2000). Conceptually, according to appraisal theory (Lazarus, 1991), any media message that shifts audience appraisals of a situation such that individuals perceive it as risky or threatening can elicit fear, while messages that foster appraisals of external blame for a threat can elicit anger and messages that encourage people to believe that positive outcomes are possible can foster feelings of hope or even contentment (Dillard & Nabi, 2006; Nabi, 2010).

1.2.3. Media use and stress

Media use is also related to stress: people use media when they feel stressed, and often find media to be an effective means of managing stress (Nabi et al., 2017). For instance, having virtual friends on social media increases perceptions of social support, which in turn decreases perceived stress (Nabi et al., 2013). However, there are also indications that stress responses are moderated by the type of media or platform used. One recent study found that consumption of self‐curated information or entertainment content helped to reduce COVID‐19‐related stress while general consumption of negative news and social media use increased it (Pahayahay & Khalili‐Mahani, 2020). Other research showed that stress was associated with more hedonic and less eudaimonic media use (Eden et al., 2020).

This body of work illustrates that media use is both shaped by and shapes risk perceptions, emotions, and stress. On this basis, we developed the conceptual model shown in Figure 1. In the model, perceived risk, emotions, and stress at Wave 1 were specified to predict media use in the 2‐week period between Waves 1 and 2, which in turn were specified to predict perceived risk, emotional responses, and stress at Waves 2 and 3. (As discussed further in Section 2, we were limited to a single assessment of media use, so the single media use assessment at Wave 2 functions as a proxy for media use between Waves 2 and 3.)

FIGURE 1.

The conceptual model

1.3. The moderating potential of partisanship

Partisan differences in perceptions of and responses to COVID‐19 emerged during the initial months of the pandemic (Collins et al., 2021). Not surprisingly, both psychological experiences and media use were contingent on political party identification. Democrats and Republicans differed substantially in how they consumed media in response to the pandemic (Bruine de Bruin et al., 2020; Freiling et al., 2021; Jurkowitz & Mitchell, 2020). For example, Democrats and Republicans differed in their use of media platforms for COVID‐related news: More Democrats obtained their news from news websites or apps, while more Republicans consumed television and radio news (Pew Research Center, 2020a). Across these platforms, Republicans were more likely than Democrats to consume right‐leaning channels and content (Bruine de Bruin et al., 2020). Such partisan differences were also observed with regard to the use of social media and entertainment media (Fioroni et al., 2022; Vogels et al., 2021).

These distinct patterns of media use led to different consequences for Democrats and Republicans. A March 2020 survey found that people who got most of their information from mainstream print and broadcast outlets had more accurate and higher assessments of the risks of COVID‐19 than did those who relied on conservative information sources (e.g., Fox News, Rush Limbaugh; Jamieson & Albarracin, 2020). Additional polling showed that nearly three in five Republicans reported feeling the news media were not working to benefit the public, while Democrats were twice as likely as Republicans to say news coverage of COVID‐19 was helping the country (Pew Research Center, 2020b). Correspondingly, party affiliation was associated with different views on virus‐related risk. In March 2020, Democrats and Democratic‐leaning independents were more likely than Republicans and Republican‐leaning independents to say that COVID‐19 was a major threat to their personal health and to public health (Pew Research Center, 2020c). Indeed, the elite discourse surrounding the pandemic has been sharply polarized since the beginning of the pandemic: Democrats highlighted its threats to public health and the workers, whereas Republicans, including then‐President Donald Trump, put more emphasis on national unity, China, and economic effects (Green et al., 2020).

The polarization in elite political discourse gave rise to polarized public opinions through partisan‐motivated reasoning and biased information exposure and processing (e.g., Bolsen et al., 2014; Leeper & Slothuus, 2014), leading Democrats to perceive greater risks associated with COVID‐19 than Republicans (Gratz et al., 2021). Higher risk perceptions in turn translated into more negative emotion and greater stress. Data collected between May and August of 2020 showed that Democrats consistently reported stronger negative emotions and higher levels of stress than Republicans (Bock & Schnabel, 2022). This is also consistent with evidence that, compared to Democrat‐trusted media, right‐wing media tended to promote misinformation that downplayed the severity of the pandemic and expressed more positive sentiment in their coverage of COVID‐19 (Guntuku et al., 2021; Lau et al., 2022; Motta et al., 2020).

Collectively, the evidence suggests that the relationships between media use and psychological responses to the pandemic differ for people with different political affiliations, but prior studies have not examined perceived risk, stress, emotion, media use, and political affiliation collectively in a longitudinal analysis. Accordingly, we proposed the following research question:

RQ: How does political party affiliation shape the interplay of risk perceptions, emotions, stress, and media use during the COVID‐19 pandemic?

2. METHOD

2.1. Participants and procedure

Qualtrics.com was employed to secure a national sample for a three‐wave survey. Qualtrics recruits participants for its panels from various online sources, including website intercept recruitment, member referrals, targeted emails, customer loyalty portals, and social media. Samples obtained through Qualtrics have demographic attributes that align with the 2010 Census data with approximately 10% variation (Heen et al., 2014). Potential respondents received an email invitation with a link to the questionnaire and information about participation incentives (typically worth $4–5 per survey). Wave 1 took place on April 20 (T1; n = 1021), Wave 2 between May 4 and May 8 (T2; n = 633), and Wave 3 between May 18 and May 22 (T3; n = 442), with a 2‐week lag between initiation of each wave. All procedures were approved by an institutional review board; participants provided informed consent prior to beginning the surveys. One prior analysis from this large data set has been published; it was focused on predictors of mental health and does not address media use (Zhou et al., 2020).

The average participant age was 45.30 years (SD = 16.46). Among participants, 52.30% were female (n = 534), 47.31% were male (n = 483), and 0.39% were nonbinary (n = 4). Most were White (n = 764, 75.05%) with 109 (10.71%) Black, 66 (6.48%) Asian, 24 (2.36%) more than one race, 17 (1.67%) American Indian or Alaska Native, 3 (0.29%) Native Hawaiian or other Pacific Islander, and 35 (3.44%) other or prefer not to answer. Additionally, 9.76% (n = 99) were Hispanic or Latino(a). Compared with participants who completed all three waves, those who did not do so were younger (M _diff = 7.11, p < 0.001) and more stressed at T1 ( M _diff = 1.35, p < 0.001). They also reported greater sadness (M _diff = 0.31, p < 0.01) compared to participants who completed all surveys. No other comparisons were significant.

2.2. Measures

Items used to measure each variable are described in detail in Table A1. Descriptive statistics for all variables are presented in Table 1.

TABLE 1.

Descriptive statistics

	T1			T2			T3
Variable	M	SD		M	SD		M	SD
Perceived risk	46.21	28.82		45.89	28.19		44.84	29.20
Negative emotions	4.65	1.50		4.43	1.56		4.23	1.62
Positive emotions	3.51	1.46		3.58	1.51		3.59	1.52
Anger	4.61	1.66		4.52	1.70		4.33	1.76
Stress	2.06	.086		1.97	0.83		1.88	0.80
News media	–	–		2.98	0.92		–	–
SNS/mobile media	–	–		3.05	0.91		–	–
Entertainment media	–	–		1.88	0.70		–	–

2.2.1. Risk perceptions

Perceived health‐related risk was measured with two 11‐point Likert‐scale items adapted from Griffin et al. (2008) asking about perceived susceptibility to and severity of COVID‐19. Risk perceptions were calculated as the product of the two items.

2.2.2. Emotions

Eighteen items on seven‐point Likert‐type scales were used to assess fear, anger, sadness, hope, contentment, and worry. Specifically, participants were asked to indicate to what extent they experienced these feelings about the COVID‐19 pandemic. Often, emotion items load onto either positive or negative dimensions when conducting a factor analysis. However, an initial confirmatory factor analysis (CFA) assuming two factors did not fit to the data: χ ² = 1973.96, df = 134, p < 0.001, CFI = 0.85, root mean squared error of approximation (RMSEA) = 0.116 (90% CI: [0.112, 0.121]), standardized root mean squared residual (SRMR) = 0.08. Factor loadings suggested that the items measuring anger did not load well on the negative emotion dimension, supporting previous research showing anger differs from other negative emotions and from positive emotions (Lerner & Tiedens, 2006). We thus specified them as a third dimension. Additionally, based on modification indices, we allowed the residual of the three items measuring fear to covary with each other. The residual of items measuring contentment were also allowed to covary with each other. This revised model showed acceptable fit: χ ² = 567.32, df = 123, p < 0.001, CFI = 0.96, RMSEA = 0.06 (90% CI: [0.055, 0.065]), SRMR = 0.06. The same analysis was performed on data from T2 and T3, and we observed the same improvement by specifying fear, sadness, and worry to load on a negative emotion factor, hope and contentment on a positive emotion factor, and anger on a third factor (T2: χ ² = 400.49, df = 123, p < 0.001, CFI = 0.97, RMSEA = 0.06 (90% CI: [0.053, 0.063]), SRMR = 0.05; T3: χ ² = 352.41, df = 123, p < 0.001, CFI = 0.96, RMSEA = 0.07 (90% CI: [0.057, 0.073]), SRMR = 0.05). A composite score was created for each dimension by averaging the corresponding items.

2.2.3. Stress responses

This variable was measured using the seven‐item stress subscale of the Depression Anxiety Stress Scale (Lovibond & Lovibond, 1995). CFA results showed a good fit with a unidimensional structure once the residual of two items (i.e., “I found it hard to wind down” and “I found it difficult to relax”) was allowed to covary (T1: χ ² = 23.56, df = 13, p < 0.01, CFI = 0.10, RMSEA = 0.04 (90% CI: [0.022, 0.055]), SRMR = 0.01; T2: χ ² = 38.81, df = 13, p < 0.001, CFI = 0.99, RMSEA = 0.06 (90% CI: [0.036, 0.077]), SRMR = 0.02; T3: χ ² = 14.28, df = 11, p = 0.22, CFI = 0.10, RMSEA = 0.03 (90% CI: [0.000, 0.060]), SRMR = 0.01). The residual of a third item, “I felt that I was using a lot of nervous energy,” was also allowed to covary with the two aforementioned items in the T3 model. A composite score was created by averaging the corresponding items.

2.2.4. Frequency of media use

Media use was measured once at T2 with 12 five‐point Likert‐scale items asking about the frequency of using the following types of media in the past 2 weeks: news media, social media, phone‐based texting, entertainment television, daytime talk shows, late night comedy, books, magazines, fictional movies, documentaries, podcasts, and talk radio shows. Results of an exploratory factor analysis showed that two items, entertainment television and books, cross‐loaded, and were therefore removed. The remaining 10 items loaded on three dimensions with social media and phone‐based texting on one dimension (labeled SNS/mobile media), news media on a second dimension (labeled news media), and daytime talk shows, late night comedy, magazines, fictional movies, documentaries, podcasts, and talk radio shows loaded on a third dimension (labeled entertainment media). A CFA conducted to assess this structure resulted in an acceptable fit: χ ² = 101.36, df = 33, p < 0.001, CFI = 0.96, RMSEA = 0.06 (90% CI: [0.045, 0.070]), SRMR = 0.03. A composite score was created for each dimension by averaging the corresponding items.

2.2.5. Party identification

This variable was measured at T1 by asking participants to indicate which of the following categories best described them: Democrat, Republican, Independent/no party, member of a third party, or decline to indicate. Among the participants, 424 self‐identified as Democrat, 307 as Republican, 261 as Independent/no party, 5 as member of a third party, and 19 declined to answer. Prior research indicates particularly stark differences between Democrats and Republicans in their opinions of and responses to the pandemic (Bruine de Bruin et al., 2020) and shows that compared to Independents and other groups, Democrats and Republicans are more vulnerable to motivated reasoning and biased processing (Lebo & Cassino, 2007). Therefore, the multigroup analysis was conducted with data from individuals who self‐identified as Republicans and Democrats.

2.3. Data analysis strategy

A cross‐lagged panel design with risk perceptions, emotional responses, and stress was used to conduct structural equation modeling (SEM). In addition, media use was specified to be predicted by the panel variables at T1 and to predict these variables at T2 and T3. We examined the moderating role of party identification using multigroup analysis. All analyses were conducted with the “lavaan” package (Rosseel, 2012) in R (R Core Team, 2020). Full information maximum likelihood estimation was used to deal with missing data across waves (Enders & Bandalos, 2001). Values of 0.90 or higher for CFI, 0.06 or lower for RMSEA, and 0.08 or lower for SRMR indicate acceptable model fit (Hu & Bentler, 1999; Kline, 2011).

3. RESULTS

3.1. Measurement invariance

We first estimated a series of models to establish measurement invariance over time. Estimating measurement invariance is essential for longitudinal SEM as it provides evidence that the meaning of the constructs remains unchanged across time. We started with an unconstrained configural model, followed by a weak invariance model where all factor loadings were constrained to be equal across time, and a strong invariance model where, in addition to factor loadings, intercepts were also constrained to be equal across time. The residuals of identical indicators were allowed to correlate across time in these models (Little et al., 2007). The chi‐square difference tests were performed to determine if the change in model fit is substantial. In addition, as the chi‐square test is sensitive to sample size, we also employed an alternative index recommended by G. W. Cheung and Rensvold (2002): The null hypothesis of invariance should not be rejected if the change in CFI (ΔCFI) is smaller than or equal to −0.01. The fit indices from the estimated models are presented in Table 2. Although results of the chi‐square test suggest that the strong invariance model for emotions was significantly worse compared to the weak invariance model, ΔCFI was smaller than −0.01, suggesting that a strong factorial invariance can be assumed (G. W. Cheung & Rensvold, 2002). Results of the chi‐square difference tests suggest that strong measurement invariance was also established for stress.

TABLE 2.

Summary of fit indices from measurement invariance model testing

Emotions	χ 2 (df)	RMSEA [90% CI]	CFI	SRMR	Δχ 2
Configural	3615.441 (1289)	0.042 [0.040, 0.044]	0.925	0.062	—
Weak	3640.033 (1319)	0.042 [0.040, 0.043]	0.925	0.062	p = 0.745
Strong	3717.657 (1353)	0.041 [0.040, 0.043]	0.923	0.063	p < 0.001
Stress
Configural	368.551 (165)	0.035 [0.030, 0.040]	0.981	0.027	—
Weak	382.324 (177)	0.034 [0.029, 0.038]	0.981	0.030	p = 0.316
Strong	396.964 (189)	0.033 [0.028, 0.037]	0.981	0.030	p = 0.262

Abbreviations: CI, confidence interval; CFI, comparative fit index; RMSEA, root mean squared error of approximation; SRMR, standardized root mean squared residual.

3.2. Model testing

Next, we estimated a measurement model that included all variables in the conceptual model. The model demonstrated acceptable fit: χ ² (3516) = 5832.93, p < 0.001, CFI = 0.92, RMSEA = 0.039, 90% CI = [0.038, 0.041], SRMR = 0.06. We then estimated a structural model to test the hypotheses and the conceptual model. The model also demonstrated a reasonable fit: χ ² (3544) = 7148.81, p < 0.001, CFI = 0.92, RMSEA = 0.032, 90% CI = [0.031, 0.033], SRMR = 0.06. Given that our primary interest lies in understanding the interplay of media use and individuals’ psychological responses, we examine the model with a focus on the predictors and consequences of different types of media use (see Table 3 for standardized path coefficients for all significant paths). Other significant paths in the model are presented in Table A2.

TABLE 3.

Predictors and consequences of media use

Predicting media use	β
RiskT1→Entertainment media	0.16**
PositiveT1→Entertainment media	0.29***
StressT1→Entertainment media	0.45***
RiskT1→News	0.12*
NegativeT1→News	0.14*
NegativeT1→SNS/mobile media	0.16*
AngerT1→SNS/mobile media	0.15*
StressT1→SNS/mobile media	0.32***
Predicting emotions at T2
Entertainment→PositiveT2	0.17***
Predicting stress at T2
Entertainment→StressT2	0.14**
Predicting emotions at T3
SSN/mobile media→NegativeT3	0.11*
News→PositiveT3	−0.09**
News→AngerT3	0.08*
SSN/mobile media→AngerT3	0.15*
Predicting stress at T3
Entertainment→StressT3	0.12*
SSN/mobile media→StressT3	0.16*

Note: Path coefficients were standardized. Only significant paths are presented.

^* p < 0.05;

^** p < 0.01;

^*** p < 0.001.

3.2.1. Perceived risk, emotions, and stress at T1 as predictors of media use

Results showed that risk perceptions, positive emotions, and stress at T1 predicted entertainment media use. Negative emotions at T1 predicted both news and SNS/mobile media use. Risk perceptions at T1 also predicted news use. In addition, both anger and stress at T1 predicted more SNS/mobile media use.

3.2.2. Perceived risk, emotions, and stress at T2 and T3 as consequences of media use

Results showed that entertainment media use predicted stronger positive emotions and greater stress at T2. Risk perceptions, negative emotions, and anger at T2 were not predicted by media use. At T3, SSN/mobile media use positively predicted negative emotions, anger, and stress. News use reduced positive emotions but increased anger at T3. In addition, stress at T3 was, again, positively predicted by entertainment media. Media use did not predict risk perceptions at T3.

3.3. Multigroup analysis

To investigate the moderating role of political party identification, particularly the identification with the Democratic and Republican parties, a multigroup analysis was conducted using party identification as a grouping variable. Participants who self‐identified as Democrat (n = 424) were compared with self‐identified Republicans (n = 307)1. We estimated (1) an unconstrained model where different structural paths were allowed between groups and (2) a constrained model where all structural paths were fixed to be equal across groups. The unconstrained model had a fit of χ ²(7158) = 12,136.25, p < 0.001, and the constrained model had a fit of χ ²(7253) = 12,314.14, p < 0.001. A significant χ ² difference test was observed: the unconstrained model fit the data significantly better (Δχ ²(95) = 177.89, p < 0.001), suggesting that the paths differed for Republicans versus Democrats. Given the number of paths involved in the model, rather than conducting a full‐fledged moderation analysis (i.e., conducting a χ ² difference test for each path), we consider the paths that were significant for one group but not the other, or the paths that had different signs across the two groups. Again, we focus primarily on the paths connecting media use and psychological responses (see Table 4). Differences in other paths are presented in Table A4. A series of post hoc t‐tests were run to see if Republicans reported using different types of media more or less often than Democrats, but all were nonsignificant, suggesting that partisan differences were not driven by the difference in frequency of media use.

TABLE 4.

Multigroup analysis on the predictors and consequences of media use

Psychological responses at T1 predicting media use	Model for Democrats (n = 424)	Model for Republicans (n = 307)
RiskT1→Entertainment media	0.12	0.25**
RiskT1→News	0.22**	0.08
NegativeT1→News	0.05	0.28*
AngerT1→SSN/mobile media	0.53**	−0.14
NegativeT1→SNS/mobile media	−0.01	0.51*
Media use predicting psychological responses at T2 and T3
News→RiskT2	0.02	0.12*
Entertainment media→PositiveT2	0.27***	0.05
Entertainment media→StressT2	0.19**	0.07
SSN/mobile media→StressT2	−0.38	0.48*
News→StressT2	−0.13**	0.05
SSN/mobile media→RiskT3	0.16	−0.77*
SSN/mobile media→NegativeT3	0.44**	0.01
News→AngerT3	0.20**	0.07
SNS/mobile media→AngerT3	0.51**	−0.10
Entertainment media→StressT3	−0.02	0.23*
SNS/mobile media→StressT3	0.66**	0.21

Note: Only paths that were significant for one group but not the other or had different signs across the two groups are presented. Path coefficients were standardized.

^* p < 0.05;

^** p < 0.01;

^*** p < 0.001.

3.3.1. T1 differences

At T1, perceived risk predicted news use among Democrats but entertainment media use among Republicans. Negative emotion predicted news use among Republicans but not Democrats. For SSN/mobile media use, anger had a positive effect among Democrats but not Republicans, while negative emotions had a positive effect among Republicans but not Democrats.

3.3.2. T2 differences

At T2, perceived risk was positively predicted by news use for only Republicans. For Democrats only, entertainment media use predicted higher levels of positive emotions. For Democrats, stress was positively predicted by entertainment media use and negatively predicted by news use. However, for Republicans, stress was positively predicted by SSN/mobile media use.

3.3.3. T3 differences

At T3, risk perception for Republicans was negatively predicted by SSN/mobile media use. For Democrats only, negative emotions at T3 were positively predicted SSN/mobile media use, and anger at T3 was positively predicted by news use and SSN/mobile media use. In addition, stress at T3 was positively associated with SSN/mobile media use among Democrats but not Republicans. However, for only Republicans, entertainment media use had a positive impact on stress at T3.

Overall, to answer the research question, political identification moderated the interplay of risk perceptions, emotions, stress, and media use such that the three psychological responses had differential impacts for Democrats and Republicans on their use of different types of media, which in turn led to varying psychological responses among the two groups at later times.

4. DISCUSSION

Our study employed a three‐wave panel design to assess the interplay of risk perceptions, emotions, stress, and media use in the United States during the early months of the COVID‐19 pandemic and to determine how political affiliation affected those associations. Our findings help to illuminate the impact of media use on stress during the early pandemic and provide insights on message design and targeting strategies for combatting pandemic‐related stress.

4.1. Interplay of risk perceptions, emotions, stress, and media use

We observed that different types of media use during the early pandemic were differentially influenced by risk perceptions, emotions, and stress levels. Negative emotions (e.g., fear, worry) and risk perceptions increased news media use. This is consistent with existing evidence that people who perceive higher risk and experience risk‐related negative emotions have greater information needs and are more motivated to actively seek information to cope with the risk (Griffin et al., 1999). Stress, negative emotions, and anger were predictors of greater SNS/mobile media use, consistent with the use of social media to vent negative feelings and escape from stressful experiences (Brailovskaia et al., 2020). At the same time, risk perceptions and stress drove audiences toward entertainment media, a finding consistent with increased viewing of at‐home streaming services like Netflix during the pandemic (Clarendon, 2020). It should be noted that our assessment of stress was general—not limited to stress from the health threats posed by COVID‐19. While the health risks posed by COVID‐19 were certainly a stressor for many people, other factors such as social and financial concerns were also likely to shape stress perceptions. For some individuals, high levels of stress might even be due to the “overly drastic” measures taken by the authorities in response to what they perceived as a hoax. The finding that stress had a similar effect as risk perceptions on entertainment media use thus suggests that entertainment media might be used to cope with risk‐ and non‐risk‐related stressors in similar ways. In line with research showing that positive emotions increased hedonic media use (Eden et al., 2020), our study reveals that positive emotions predicted greater entertainment media use, which probably reflects a preference for mood‐reflecting media to sustain positive states (Greenwood, 2010). These findings, of course, should be interpreted within the context of a pandemic where a stay‐at‐home order was enforced in most states during the time of data collection and the public had limited access to other entertainments. Records show that people have spent more time using media, particularly for entertainment purposes, since the pandemic's outbreak (Nielsen, 2020). Thus, the effect of psychological experiences on media use needs to be interpreted with the influence of lockdown measures in mind.

In addition, our study challenges the generalization that news and social media use during a pandemic exacerbate stress, while entertainment media alleviates it. Our findings suggest a more complex relationship between media use and stress. Entertainment media use predicted higher levels of stress at both T2 and T3, but also increased positive emotions at T2. This could be because people employed entertainment media to avoid reality, and such media use, while providing an escape and giving rise to positive emotion for a short time, does not prevent people from being confronted with their reality in the long run (Reinecke et al., 2014). Although news use was not associated with stress or fear responses at any time, it did predict more anger and less positive emotions at T3.

Our findings are clearer with regard to SNS/mobile media use, which predicted greater stress, negative emotion, and anger at T3. The prevalence of misinformation and partisan rancor on both news and social media at this point in the pandemic (Jamieson & Albarracin, 2020) may explain these findings. Indeed, much research has documented the intensity of public emotional response, particularly anger, fueled by constant dispute in the news and social media over virtually all aspects of the pandemic (Han et al., 2020; Smith et al., 2021). Individuals who viewed COVID‐19 as a real threat were angry about misinformation on these media that downplayed the severity of the pandemic. For these people, anger could also serve as a mechanism for coping with heightened distress stemming from risk information obtained from news and social media. At the same time, those who saw COVID‐19 as less risky experienced anger as a result of being exposed to media messages that “exaggerated” the threats of COVID‐19 and to narratives that placed blame for the pandemic on specific groups. The present study measured emotions as they related to the COVID‐19 pandemic overall, which constrains our ability to empirically examine the nuances involved in the relationships between media use and emotions. These speculations thus need to be tested by future research that assesses the specific sources and targets of individuals’ emotions. However, in line with prior research (Nabi et al., 2022), our findings indicate that media consumption played a prominent role in shaping the public's emotional experiences and stress during the early months of the COVID‐19 crisis.

Interestingly, media use had no significant relationship with risk perceptions at T2 or T3. Rather, across the three waves, risk perceptions at each subsequent wave were positively predicted by negative emotions but negatively predicted by anger at each previous wave. Prior research indicates that whereas fear is associated with a sense of uncertainty and lack of control, anger is associated with a sense of certainty and control; this difference affects risk perceptions (Lerner & Keltner, 2001). Our findings echo this observation. In addition, risk perceptions at T3 were also positively predicted by stress at T2, which aligns with the evidence that feelings of stress may increase perceived risk by serving as heuristic information (Slovic et al., 2004; Sobkow et al., 2016). Importantly, however, we found no effect of media use on risk perceptions, contradicting findings from prior research (J. Lee et al., 2021; Olagoke et al., 2020). This is likely because we examined general—rather than COVID‐specific—media use in the current study. Moreover, in light of the highly politicized nature of the pandemic, individuals likely engaged in selective exposure to attitude‐confirming information and avoided information that was incongruent with their pre‐existing beliefs (Knobloch‐Westerwick & Meng, 2009). As such, instead of a consistent main effect, media use may exert varying impacts on individuals’ risk perceptions depending on their prior attitudes toward the pandemic, the content they consume, and the way they process the information. The observed moderating effect of political party affiliation (which we discuss in detail in the next section) supports this possibility. Nonetheless, as demonstrated by the findings, media use can still indirectly influence risk perceptions by shaping individuals’ emotions and stress and therefore it remains consequential for the adoption of health‐protective behaviors.

4.2. Partisan differences

Examining the moderating role of political identification on the interplay of psychological responses and media use, the current study highlights several key differences between Republicans and Democrats. Notably, negative emotions predicted Republicans’ use of both news and SNS/mobile media, whereas perceived risk and anger seemed to drive Democrats to these media. It is well documented that negative emotions such as fear and worry tend to increase information insufficiency and thereby promote information seeking (Griffin et al., 1999; Kahlor, 2010). The failure of negative emotions to predict media use among Democrats might be due to a ceiling effect, as Democrats tended to follow information about COVID‐19 very closely—much more than Republicans (Jurkowitz, 2021). It is also possible that Republicans used news and SNS/mobile media for content that minimized the severity of the pandemic and thereby assuaged negative emotion; this was unlikely to be helpful for Democrats. In fact, the prevalence of severity‐minimizing misinformation was likely a source of anger for many Democrats (Borah et al., 2021). Hence, as suggested by the findings, they were motivated by anger to use SNS/mobile media, potentially to disseminate information consistent with their risk perceptions and to engage in debates with their right‐wing counterparts. Moreover, our observation that risk perceptions only promoted more entertainment media use among Republicans echoes the finding that Republican‐identified individuals are more likely to take an avoidant coping approach to deal with psychological discomfort (Haltinner & Sarathchandra, 2018; Nam et al., 2013).

With regard to the effects of media use on psychological responses, a striking finding was that SNS use was unrelated to perceived risk for Democrats, but for Republicans, the more social or mobile media they used, the less of a health risk they later perceived COVID‐19 to be. As noted previously, this is likely due to the greater prevalence on right‐wing social media of misinformation downplaying the severity of the pandemic (Havey, 2020; Peters, 2020). It also illustrates how efforts to educate people about COVID‐19 risk can easily be undermined by social media messages connected with political and ideological affiliations. Although news use did increase risk perceptions for Republicans, their inattentiveness to COVID‐19 news (Jurkowitz, 2021) likely prevented news from raising their risk assessment to a level that stimulates the adoption of protective behaviors.

We also observed some interesting patterns regarding the influence of media use on stress: SNS/mobile media use only predicted Republicans’ stress levels at T2, and Democrats at T3. Entertainment media use, in contrast, increased Democrats’ stress at T2, and Republicans at T3. News consumption decreased Democrats’ stress at T2. The differential impact of the use of different types of media on stress over time might be related to the different stressors experienced by Democrats and Republicans and the corresponding media environment at the time. For example, research shows that communication on social media was dominated by objective information and facts at the beginning of the pandemic. Later, this was replaced by messages that expressed personal beliefs and opinions, many based on misinformation (Wicke & Bolognesi, 2021). This might help to explain why only Democrats, who were more concerned about the threat of COVID‐19, reported greater stress at later times as a result of using SSN/mobile media. Democrats’ stress, on the other hand, was relieved by watching news, at least during the early weeks of the pandemic, which again suggests that being (accurately) informed about the crisis helped this group deal with the type of stressors they experienced. This may also explain why entertainment media use, a potential escape from reality, increased Democrats’ stress level at that time. For Republicans, although entertainment media use seemed to have provided a temporary buffer, the effect did not last. Instead, as suggested by the positive effect of entertainment media use on stress at T3 for this group, it backfired later when they encountered the reality that strict measures were being enforced and society was significantly disrupted by a virus that they believed was not much of a threat—a likely source of stress for this group. Overall, these patterns suggest that interventions to reduce stress during a pandemic may need to target different groups via different channels based on how, exactly, the issue is being framed by these groups at a given point in time.

The influence of media use on emotions also differed by party affiliation. Entertainment media use resulted in greater positive emotions among Democrats at T2, but not Republicans. The use of news media and SNS/mobile media also positively predicted anger at T3 among Democrats, but, again, not Republicans. SNS/mobile media also increased Democrats’ negative emotions at T3. The effect among Democrats might stem from the still‐rising rates of infection, trends toward “reopening” society, and the perceived failure of the Trump administration's response to the pandemic at that time (May, 2020). Overall, it seems that media use tended to have stronger emotional consequences for Democrats than for Republicans. To the extent that Republicans were already anticipating or enacting the return to “normal life” in May 2020, and Democrats were not, Democrats’ emotional state may have depended more on their media consumption.

4.3. Limitations and implications

As with all research, the present study is not without limitations. First, although Qualtrics provides national U.S. samples that tend to be representative demographically, our sample was not random. To reduce questionnaire length and reduce cost, we measured media use at Time 2 only, which constrains our ability to assess or control for the effect of media use at earlier times. Given the highly dynamic nature of public health crises such as COVID‐19, future research would benefit from collecting longitudinal data on media use, which will allow not only a more rigorous test of the link between media use and psychological responses but also observations of potential changes in effects over time. Further, we focused on the consumption of different types of media and did not assess use of specific media content (e.g., Fox News, CNN's Facebook page). Consistent with prior research that employed similar measures (e.g., Nabi et al., 2022), findings of the study demonstrated the critical and meaningful role played by various forms of media use in understanding stress responses during a crisis. However, it should be acknowledged that the content people consume on media platforms varies substantially and can lead to diverse consequences (Chao et al., 2020). Therefore, future research that includes a more fine‐grained assessment of the specific content of consumption is needed to deepen our knowledge of the relationships examined in this study. Lastly, the data were collected during April and May 2020 regarding the period from March 2020 to May 2020; we do not know if these relationships were sustained as the pandemic proceeded.

Despite these limitations, this study employed a rigorous, three‐wave panel design to assess the over‐time interplay of risk perceptions, emotions, stress, and media use, as well as the moderating influence of political party identification. Generalizing across our findings, we observe that use of different types of media are driven by risk perceptions, emotions, and stress in varying ways; media use influences emotional and stress responses, which in turn affect risk perceptions. Political affiliation alters the factors driving media use, which subsequently alter the experience of risk, stress, and emotion. Theoretically, these findings underscore the importance of both risk appraisals and media use for explaining behavioral responses to the COVID‐19 pandemic and future public health crises. Practically speaking, partisan differences in perceived risk, anger, and social media use help to explain higher rates of infection in Republican‐leaning areas (Bump, 2020).

To reach partisan audiences for future pandemics, public health advocates need to craft messages utilizing trusted voices and reflecting partisan concerns, and place them strategically where they can be heard. Additionally, our data suggest that health advocates do not necessarily need to warn Americans against consuming news in order to avoid stress. As an alternative, they could encourage functional stress appraisal and coping strategies, including limits on using media purely as a distraction (Hagger et al., 2020). A more nuanced strategy for making stress‐related recommendations against media use could also help public health officials avoid eliciting reactance from audiences who are likely to want to use media to help them cope during the stressful, and risky, times of a viral pandemic.

1.

See appendix Table A1, Table A2, Table A3, Table A4

TABLE A1.

Measurement instruments

Variable	Item	Scale
Risk perceptions*	How likely is it that the COVID‐19 pandemic will harm your health?	0 = Will certainly not harm my health,10 = certain to harm my health
	If you develop COVID‐19, how serious do you think the harm would be?	0 = Not serious at all, 10 = as serious as it could possibly be
Emotions	To what extent do you have the following feelings about the COVID‐19 pandemic? Fearful, afraid, scared (fear) Irritated, angry, annoyed (anger) Sad, dreary, dismal (sadness) Anxious, worries, nervous (worry) Hopeful, optimistic, positive (hope) Contented, peaceful, calm (contentment)	0 = Not at all, 6 = very much so
Stress responses	Please indicate to what extent the following statements apply to you. I found it hard to wind down I tended to overreact to situations I felt that I was using a lot of nervous energy I found myself getting agitated I found it difficult to relax I was intolerant of anything that kept me from getting on with what I was doing I felt that I was rather touchy	0 = Did not apply to me at all, 3 = applied to me very much or most of the time
Frequency of media use	In the past 2 weeks, how often do you consume the following types of media? News media (e.g., newspapers, news websites, radio or TV news) Social media (e.g., Facebook, Twitter, Snapchat, Instagram, TikTok) Phone‐based texting (e.g., iMessages, SMS messages, WhatsApp, or other texting‐only apps) Entertainment television (e.g., comedy or drama shows, on a mobile device, computer, or traditional television) Daytime talk shows (e.g., The View, Kelly & Ryan) Late night comedy (e.g., Fallon, Colbert) Books (paper or e‐reader) Magazines (paper or e‐reader) Fictional movies Documentaries Podcasts Talk radio shows (e.g., Rush Limbaugh, Howard Stern)	1 = Never, 5 = constantly, except when sleeping
Party identification	Generally speaking, do you think of yourself as a Democrat, a Republican, or something else?	1 = Democrat 2 = Republican 3 = Independent/no party 4 = Member of a third party 5 = Decline to indicate

^*The variable was calculated as the product of the two items.

TABLE A2.

Other significant paths in the model

Predicting risk perceptions at T2	β	Within‐wave paths	β
NegativeT1→RiskT2	0.24***	RiskT1←→NegativeT1	0.56***
AngerT1→RiskT2	−0.15***	RiskT1←→PositiveT1	−0.09**
Predicting emotions at T2		RiskT1←→AngerT1	0.21***
StressT1→NegativeT2	0.12**	RiskT1←→StressT1	0.28***
StressT1→AngerT2	0.12*	StressT1←→NegativeT1	0.42***
Predicting stress at T2		StressT1←→AngerT1	0.35***
NegativeT1→StressT2	0.12**	RiskT2←→NegativeT2	0.21***
Predicting risk perceptions at T3		RiskT2←→AngerT2	0.19***
NegativeT2→RiskT3	0.17**	StressT2←→NegativeT2	0.26***
AngerT2→RiskT3	−0.12**	StressT2←→AngerT2	0.28***
StressT2→RiskT3	0.11*	RiskT3←→NegativeT3	0.30***
Predicting emotions at T3		RiskT3←→AngerT3	0.18**
RiskT2→NegativeT3	0.09**	StressT3←→NegativeT3	0.16*
StressT2→NegativeT3	0.11*	StressT3←→AngerT3	0.19*

Note: All autoregressive paths were significant at the level of α < 0.001 (not shown in the table).

^* p < 0.05;

^** p < 0.01;

^*** p < 0.001.

TABLE A3.

Multigroup analysis comparing republicans and non‐Republicans

	Model for non‐Republicans	Model for Republicans
	(n = 690)	(n = 307)
RiskT1 →Entertainment media	0.13*	0.25**
PositiveT1→Entertainment media	0.32***	0.24**
StressT1→Entertainment media	0.42***	0.45***
RiskT1 →News	0.14*	0.08
NegativeT1→News	0.08	0.28*
AngerT1→SSN/mobile media	0.30**	−0.14
NegativeT1→ SNS/mobile media	0.03	0.51*
StressT1→SNS/mobile media	0.23**	0.56***
NegativeT1→RiskT2	0.23***	−0.003
PositiveT1→RiskT2	0.10*	−0.06
News→RiskT2	0.01	0.12*
Entertainment media→PositiveT2	0.21***	0.05
News→PositiveT2	−0.08*	−0.002
RiskT1→StressT2	−0.04	0.22*
NegativeT1→StressT2	0.16**	−0.25
AngerT1→StressT2	0.03	0.21**
Entertainment media→StressT2	0.14**	0.07
SSN/mobile media→StressT2	−0.08	0.48*
News→StressT2	−0.10*	0.05
NegativeT2→RiskT3	0.16**	0.33
StressT2→RiskT3	0.09	0.74*
SSN/mobile media→RiskT3	−0.01	−0.77*
News→PositiveT3	−0.08*	−0.15*
News→AngerT3	0.09*	0.07
SNS/mobile→AngerT3	0.17*	−0.10
Entertainment media→StressT3	0.13*	0.23*
SNS/mobile→StressT3	0.17*	0.21

Note: Fit of the unconstrained model: χ ²(7158) = 12,536.73, p < 0.001. Fit of the constrained model: χ ²(7253) = 12,693.90, p < 0.001. The constrained model fit the data significantly worse: Δχ ²(95) = 157.17, p < 0.001. Only paths that were significant for at least one of the groups are presented in the table. All autoregressive paths were significant across the two groups (not shown in the table). Path coefficients were standardized.

^* p < 0.05;

^** p < 0.01;

^*** p < 0.001.

TABLE A4.

Other paths that differ for Democrats and Republicans

	Model for Democrats	Model for Republicans
	(n = 424)	(n = 307)
RiskT1→StressT2	−0.19**	0.22*
NegativeT1→StressT2	0.23**	−0.25
AngerT1→StressT2	0.08	0.21**
StressT2→RiskT3	0.07	0.74*
RiskT2→NegativeT3	0.18*	0.14
AngerT2→StressT3	−0.36*	0.06

Note: Only paths that were significant for one group but not the other or had different signs across the two groups are presented. Path coefficients were standardized.

^* p < 0.05;

^** p < 0.01;

^*** p < 0.001.

Zhou, Y. , Myrick, J. G. , Farrell, E. L. , & Cohen, O. (2022). Perceived risk, emotions, and stress in response to COVID‐19: The interplay of media use and partisanship. Risk Analysis, 00, 1–15. 10.1111/risa.14044