How Rally-Round-the-Flag Effects Shape Trust in the News Media: Evidence from Panel Waves before and during the COVID-19 Pandemic Crisis

ABSTRACT In this study, we extend the literature on the rally ‘round the flag phenomenon, that is, that international crises tend to cause an increase in citizens’ approval of political institutions. We advance this literature and highlight its relevance for political communication research in three ways: 1) by theorizing and empirically testing two arguments for why rally effects should extend to trust in the news media on the institutional level, 2) by providing empirical evidence on how rally effects on trust in the media develop over time during an international crisis, and 3) by theorizing and testing the conditions under which rally effects on media trust are more likely to occur by studying heterogeneous effects. Through a panel design with a pre-crisis baseline of Norwegian citizens’ trust in news media, we find evidence to suggest that the compound effect of the COVID-19 pandemic crisis caused a long-lasting increase in trust in the news media in Norway, and that the degree of increase varied by citizens’ education and whether they belonged to a “high-risk” group. We also provide evidence to suggest that rally effects on news media trust are contingent on how important the news media is as a source of information about the crisis and the “trust nexus” between media trust and political trust. These insights extend our current understanding of how times of crisis affect trust in the news media.


Introduction
One of the most established phenomena in the literature on how public opinion responds to political leadership and communication during acute crises is that citizens tend to "rally 'round the flag" and increase their support for their political leaders (Mueller, 1973). While prior studies on the "rally 'round the flag" effect have predominantly focused on rally effects on the approval of political leaders, this phenomenon is not exclusive to politicians. Citizens also tend to increase their trust in political institutions in general (Baekgaard et al., 2017), including the news media (Dinesen & Jaeger, 2013), amid large international crises.
However, it remains a puzzle in the political communication literature why rally effects extend to the news media. Moreover, the lack of data across time has left researchers with limited possibilities for studying the development of such rally effects on the news media during crises.
In this article, we seek to extend the literature on rally effects in three ways. First, we theorize and empirically test two arguments for why the rally effects of international emergencies extend to trust in the news media on the institutional level. Second, we further theorize the conditions under which rally effects occur by drawing on prior literature on conditional rally effects on political trust. Third, we provide empirical evidence on how rally effects on trust in the media develop over time during an international emergency. We focus on how the COVID-19 pandemic crisis -an unparalleled global crisis -influenced trust in the news media over time. In late December 2019, SARS-CoV-2 was detected in Wuhan, China, causing the coronavirus disease COVID-19 in humans. Our study takes place in Norway, one of many countries in Europe that was affected soon after the initial outbreaks in China and Italy.
We provide robust evidence to explain how, and the conditions under which, trust in the news media on the institutional level changed over time from (1) before to immediately after the onset of and (2) in the course of the COVID-19 pandemic crisis. We leverage data from five panel waves fielded in a probability-based nationally representative sample of Norwegian citizens. With these data, we can model the causal compound treatment effect (Hernán & VanderWeele, 2011) of the pandemic crisis on changes in trust in the news media, heterogeneous effects, and the development in media trust over time.

International Crises' Impact on Trust in the News Media
Exogenous shocks, such as large terror attacks and the global financial crisis of 2007-2008, have provided important insights for trust research -especially for our knowledge on trust in political institutions (e.g., Dinesen & Jaeger, 2013;Edwards & Swenson, 1997). By exogenous, we mean a phenomenon that is unrelated to the phenomenon in question, in our case meaning that the shock appeared regardless of citizens' trust. A long line of research has established that exogenous shocks such as international crises can cause a "rally 'round the flag" effect that increases citizens' approval of political institutions (Chatagnier, 2012). Such rally effects have been found in response to wars (Brody, 1991;Edwards & Swenson, 1997;Kriner, 2006;Parker, 1995), terrorist attacks (Chanley, 2013;Dinesen & Jaeger, 2013;Perrin & Smolek, 2009;Wollebaek et al., 2012), as well as after the lockdown in response to the COVID-19 pandemic crisis (e.g., Baekgaard et al., 2020;Bol et al., 2020;Devine et al., 2020;Esaiasson et al., 2020;Goldstein & Wiedemann, 2020;Schraff, 2020;De Vries et al., 2021). Rally effects tend to emerge regardless of whether political leaders succeed or fail in their response to the crisis as long as the crisis demands immediate action (Sniderman et al., 2019).
While most prior literature on rally effects focused on the approval of governments and political leaders, rally effects can also lead to increases in trust in other central institutions in society. Wollebaek et al. (2012) found broad increases in institutional trust for a range of different institutions, for example, the courts, police, public administration, and voluntary organizations, in response to the July 22, 2011, Utøya attack in Norway. Our focus is on rally effects on trust in the news media. The literature on news media trust distinguishes between four different dimensions, the source, the medium, the message, and the media as an institution (Metzger et al., 2003;Strömbäck et al., 2020), as well as between the concepts of trust and credibility (Kohring & Matthes, 2007;Van Dalen, 2019). We focus on trust in the news media on the institutional level (Strömbäck et al., 2020). Following the definition of Hanitzsch et al. (2018, p. 5), we define media trust as a form of institutional trust that involves "the willingness of the audience to be vulnerable to news content based on the expectation that the media will perform in a satisfactory matter." We view trust as a futureoriented concept, where audiences have an expectation that they can rely on the information they receive from the media, although they cannot monitor the different editorial decisions in the news-making process (Van Dalen, 2019).
In terms of rally effects on news media trust, Dinesen and Jaeger (2013) found increases in trust in the news media, political institutions, and justice system after the March 11, 2004, attack in Madrid. In addition, using a longitudinal design which allowed the authors to follow daily new responses to a survey right before and after the announcement of the Danish COVID-19 pandemic crisis lockdown, Baekgaard et al. (2020) found evidence to suggest that institutional trust in political institutions and in the Danish news media increased among unemployed Danes over the first three weeks following the lockdown in Denmark. These studies demonstrate that rally effects are far from exclusive to the approval of political leaders and the government. Taken together, the literature reviewed above suggests that large international crises, such as the COVID-19 pandemic crisis, increase trust in the news media on the institutional level. Accordingly, we formulate the following hypothesis: [H1] Citizens' trust in the news media will increase from before the crisis to shortly after the outbreak of the COVID-19 pandemic crisis.

Theorizing Reasons for Rally Effects on Institutional News Media Trust
Why has prior research found that rally effects occur not only for the governing parties and political leaders but also for the news media? Dinesen and Jaeger (2013) assumed that the media can work to mitigate national trauma of a crisis by creating a sense of "togetherness." We propose two additional possible explanations. First, prior studies have consistently shown a relationship between news media use and news media trust on the institutional level (Strömbäck et al., 2020) although audiences also use news media outlets that they do not trust (e.g., Tsfati & Cappella, 2003;Tsfati & Cohen, 2005). During a crisis, citizens have strong incentives to update themselves on the development of the crisis and the government's response to the crisis . Based on a panel study of news use before and during the COVID-19 pandemic in 17 countries, Van Aelst et al. (2021) showed that TV news use and internet news use increased, while radio and newspaper use decreased. Van Aelst et al. (2021) also demonstrated a relationship between increased news use and news media trust: the increase in news use was higher for citizens with higher pre-pandemic levels of trust in legacy news media compared to citizens with lower levels of media trust in times of non-crisis. In addition, the news media is often among the most important, if not the most important, arena for political leaders to disseminate their proposed solutions and measures to counter crises through press conferences and interviews. If our first theoretical model of why rally effects tend to extend to the news media is correct, we should observe a stronger rally effect on trust in the news media among those who perceive the news media as an important source of information about the crisis compared to those who do so to a lesser extent. A hypothesis on this conditional effect follows.
[H2] The more the news media is perceived to be an important source of information about the COVID-19 pandemic crisis, the more citizens' trust in the news media will increase from before the crisis to shortly after the outbreak of the COVID-19 pandemic crisis.
Second, we theorize that increased trust in the government is linked to increased institutional trust in the news, and vice versa. Specifically, we argue that the rally effect on trust in the news media is linked to rally effects on the government and other political institutions (Hanitzsch et al., 2018). For instance, building their argument around the idea that the sharp decline in trust in the news media is closely related to the decline of public confidence in political institutions (S. E. Bennett et al., 1999), Hanitzsch et al. (2018) presented evidence for what they called a trust nexus. According to this concept, citizens' trust in the news media is linked to the way they evaluate political institutions. The idea of a trust nexus is supported by studies that have found that the decrease in trust in the news media observed in some countries seems to be connected to citizens' disdain for social institutions more broadly, and particularly political institutions (e.g., Ariely, 2015;Cappella & Jamieson, 1997;Hanitzsch et al., 2018). While Hanitzsch et al. (2018) mainly used the trust nexus to describe a decline in trust, it is reasonable to assume that this link between political trust and media trust also exists when we expect trust to increase.
We theorize that a second explanation for the rally effect on news media trust is that trust in the news media and political trust is connected through the "trust nexus." If this model is correct, citizens should be more likely to increase their trust in the news due to the pandemic crisis the more their trust in political institutions increases due to the pandemic crisis. To test this assumption, we suggest the following hypothesis: [H3] The increase in citizens' trust in the news media from before the crisis to shortly after the outbreak of the COVID-19 pandemic crisis is conditional on a parallel increase in trust in political institutions.
In addition to the importance of news media as source of information and the trust nexus, we assume that rally effects on the news media vary based on many of the same predictors of rally effects on political trust and approval. More specifically, we focus on the possible conditional effects of education, government party support, and worry. Regarding education, Baum (2002) showed through time series analyses that highly educated US citizens were the least likely to show increased support for the president following the use of force abroad. Building on the work of Zaller (1992) on how the middle category (i.e., neither the least nor the most) of politically aware citizens are most likely to change their opinions, Baum argued that citizens in the middle category of education are more likely to rally than citizens with low or high education (assuming that education is a proxy for political awareness). The possible heterogeneous effect of education was also noted by Esaiasson et al. (2020), who found that the increase in satisfaction with the Swedish government in responding to the COVID-19 pandemic crisis varied by level of education. While this conditional effect of education has not been tested for rally effects on trust in the news media, it is reasonable to assume that the increase in trust in the news media is conditional on citizens' education, given that media trust and satisfaction with the government are likely connected through the trust nexus (Hanitzsch et al., 2018). As the results of Esaiasson et al. (2020) and Baum (2002) provide conflicting predictions in terms of how the education effect plays out, we specify two conflicting hypotheses: [H4a] The effect of the increase in citizens' trust in the news media from before the crisis to shortly after the outbreak of the COVID-19 pandemic crisis is stronger among the higher educated than the lower educated.
[H4b] The effect of the increase in citizens' trust in the news media from before the crisis to shortly after the outbreak of the COVID-19 pandemic crisis is stronger for individuals in the middle category of education than those who are lower and higher educated. Baum (2002) also argued that those who do not support the parties in power are most likely to rally in terms of increased support for the government, given that the levels of trust between voters of the opposition parties and the incumbent parties were different before the crisis hits but indistinguishable from chance during the crisis. Baum contended that, because there is little to gain in presidential approval among partisans who already support the president, it is voters of the opposition party that are most likely to rally. However, Baum only found partial support for this argument (i.e., only for Republican administrations). Esaiasson et al. (2020) also found some indications of heterogeneous effects of party preference, but this finding was not robust across all samples in their study. However, Esaiasson et al.' study on the effects on COVID-19 in Sweden found conditional effects of party preference in the opposite direction compared to what Baum (2002) hypothesized. Phrased differently, opposition party supporters was statistically significantly less likely to rally in the Swedish COVID-19 study by Esaiasson et al. (2020). As we expect that rally effects on news media trust are connected to rally effects on trust in politicians and approval for the government (i.e., H3), we aim to study whether there are heterogeneous differences in rally effects on news media trust in terms of government party support. Again, as the results by Esaiasson et al. (2020) and Baum (2002) provide conflicting predictions in terms of how the government support effect plays out, we specify two conflicting hypotheses: [H5a] The increase in citizens' trust in the news media from before the crisis to shortly after the outbreak of the COVID-19 pandemic crisis is stronger among government party supporters.
[H5b] The increase in citizens' trust in the news media from before the crisis to shortly after the outbreak of the COVID-19 pandemic crisis is stronger among opposition party supporters. Schraff (2020) argued that an anxiety effect has driven the rally effect of the COVID-19 pandemic crisis on political support and that citizens accordingly rally around their political institutions as a lifebuoy. This argument is supported by a study showing that concern for the COVID-19 pandemic increased after lockdown measures were initiated in the UK (Harding & Eggers, 2021). Moreover, Schraff's argument is in line with several social psychological models that offer a useful lens for understanding the causes of rally effect, such as uncertainty management theory (Van den Bos et al., 2005) and anxiety-based formulations of authoritarianism (Doty et al., 1991). As noted by Lambert et al. (2011), such models emphasize the idea that people are motivated to see the world as a secure place. That said, Lambert et al. (2011) argued against anxiety as a cause of rally effects in terms of reactions to war, instead highlighting that anger is a key motivator for increased political support. However, while anger might be a relevant factor in terms of terrorist attacks and war, we argue that in the case of a pandemic crisis, where there is no concrete enemy or out group other than the virus, worry and anxiety are more relevant factors. Importantly, one of the central effects of international pandemic crises in the public is the feeling of threat, such as the threat of a pandemic getting out of control and infecting one's relatives and friends. Following the argument of Schraff (2020), we assume that a pandemic threat fosters worry and anxiety in many citizens and that the news media represent one potential bulwark against such threats. If this extends to trust in the news media, trust in the news media should be more likely to increase among worried citizens compared to citizens who are less worried. Accordingly, we present the following hypothesis: [H6] The increase in citizens' trust in the news media from before the crisis to shortly after the outbreak of the COVID-19 pandemic crisis is stronger among worried citizens than among citizens who are less worried.
Relatedly, Esaiasson et al. (2020) suggested that vulnerable groups are inclined to react differently in terms of rally effects and tested this indirectly (represented by elderly and the lower educated). We argue that vulnerable groups are more inclined to increase their news media trust after a crisis because they have a significant incentive to be more reliant on the news media's reporting of a crisis. For instance, citizens in a high-risk group, such as people who have an increased chance of becoming seriously ill or dying due to COVID-19, should have an increased incentive to rely on both the news media and political institutions for information and interpretation for crisis-related issues. Thus, one could assume that citizens in the risk group experienced a larger increase in trust than citizens who are not in the risk group. Another example is citizens who are harmed directly by a crisis. Using the COVID-19 pandemic crisis as an illustration, citizens can have a direct experience with the crisis as a result of the lockdown measures' influence on one's personal economy. For instance, De Simone et al. (2022) showed that during the COVID-19 pandemic crisis, a pessimistic prospect of ones' future financial situation was correlated with lower satisfaction with democracy. Moreover, Mann et al. (2020) and Fetzer et al. (2021) found that the economic impact of COVID-19 led to an increase in economic anxiety. As we do not build these two arguments for conditional effects on concrete evidence from prior literature on rally effects, we state the following research questions rather than formal hypotheses: [RQ1a] Does the rally effect on trust in the news media depend on belonging to (or being close to someone who belongs to) a risk group?
[RQ1b] Does the rally effect on trust in the news media depend on one's perceived economic situation after the crisis hits? Sniderman et al. (2019) argued that although crises tend to lead to changes in public opinion shortly after the crisis event, public opinion tends to return to baseline over time and thus be stable in the long term. For instance, Dinesen and Jaeger (2013), who studied rally effects after the 3/11 terror attack in Madrid, found that trust in what they called "media institutions" (i.e., the press, radio, and television) clearly increased immediately after the 3/11 attack but then decreased to below pre-attack levels seven months after the terror attack. They argued that the rally effect on trust in media institutions is more short term and volatile than the rally effect on political institutions. A parallel finding is the short-lived increase in trust in local news in the US after the 9/11 terror attacks (Putnam, 2002;Skocpol, 2002). According to the existing literature on how crises influence trust in political institutions and the news media over time, citizens' trust in the news media is likely to decrease toward baseline levels a few months after the government's response to the crisis.

How Long Do Rally Effects on Institutional Media Trust Last?
The COVID-19 pandemic crisis is, however, a very different crisis than terror attacks and other event-based crises. The pandemic and the government response to it was, at the time of the last survey wave in our data, still ongoing two years after the initial outbreak. As such, one might expect the rally effects to last longer. In terms of lasting events, a more comparable crisis event is the refugee crisis in Europe in 2015/2016. In a study of public reactions to the refugee crisis, Nordø and Ivarsflaten (2022) found a marked exclusionary reaction to the sudden influx of refugees, and only two years after the influx of refugees was under control, opinions about refugee rights and immigration had reverted back to pre-crisis baseline levels. Although Sniderman et al.'s (2019) argument would lead us to expect that rally effects on news media trust should gradually decrease and return to its baseline values, the longlasting COVID-19 crisis could result in a different outcome, such as stable changes over time or a gradual decrease in trust toward baseline levels. This leads us to formulate a research question rather than a formal hypothesis: [RQ2] Does trust in the news media continue to increase, stabilize, or decrease over time during a long-lasting international crisis?

Data and Methods
We leverage panel data from five different waves of the Norwegian Citizen Panel (NCP)which ran in January 2020, March 2020, June 2020, January 2021, and January 2022. This leaves us with a pre-crisis baseline (January 2020), a measure during the initial stage of the Corona pandemic (March 2020), and three measures at later stages of the crisis (June 2020, January 2021, and January 2022). The NCP is a probability-based, nationally representative online survey panel that began in 2013, and its participants have been recruited regularly in seven waves of postal recruitment. Participants were randomly selected for recruitment from Norway's National Registry -a list of all individuals who either are or have been a resident in Norway, maintained by the Norwegian Tax Administration. This means that the entire Norwegian population has an equal and known probability of being invited. More details about response rates or other methodological matters can be found in the NCP methodology reports (Skjervheim et al., 2022).

Dependent and Independent Measures
Our dependent variable, trust in the news media (on the institutional level), was measured with the question "Generally speaking, to what degree do you trust the Norwegian news media?," which was answered on a five point scale from one (to a very great extent) to five (not at all). For analytical purposes, the scale was reversed so that five represented the highest degree of news media trust.
To test our hypotheses and answer our research questions about the effect of the COVID-19 crisis on trust in the news media, we performed three sets of analyses. First, we tested H1 by studying the short-term effect from before the crisis (January 2020) to shortly after the outbreak (March 2020). For all analyses of short-term effects, time was measured through the survey waves, with the January 2020 baseline coded as 0 and the March 2020 crisis wave coded as 1. To test H2 and H3, we included an interaction term between our measure of the short-term effect (from January 2020 to March 2020) and the variables that we argue are theoretically relevant for explaining the rally effect on news media trust: a) importance of the news media as a source of information about COVID-19 t 1 and b) change in political trust, that is, the change in trust in politicians from January to March 2020. Regarding a), we use the measure "Many people are seeking information about the corona virus these days. How important are the following sources to you as sources of information about COVID-19 . . . the news media's coverage (e.g., the NRK, TV2, VG)," anchored on a scale from 1 (not important at all) to 5 (very important). Regarding b), we used the measure "Generally speaking, to what degree do you trust politicians in Norway?" anchored on a five point scale from 1 (to a very great extent) to 5 (not at all), measured in January 2020 and March 2020. We created a change variable by subtracting respondents' trust in politicians in March from that in January 2020.
Second, we analyzed the heterogeneous effects (H4-H6 and RQ1) of the observed change in media trust in the initial stage of the crisis (focusing only on the short-term change from January 2020 to March 2020). We interacted the time variable (January and March 2020) with respondents' level of education, belonging to a risk group, perceived state of personal economy during the lockdown, intention to vote for one of the government parties in January 2020, and feeling worried during the last week in the March 2020 wave. Education was measured based on the respondents' highest completed level of education, divided into the following categories: 1 "No education/elementary school," 2 "Upper secondary education," and 3 "University/University college." Belonging to a high-risk group was measured by combining two items asking whether the respondent or anyone in the same household was medically vulnerable to COVID-19 (risk group). If the respondent confirmed belonging to a risk group themselves or living with someone in a risk group, they were coded as 1; otherwise they were coded as 0. The dummy variable for voting for the government parties (1) or opposition (0) was measured through a question asking, "If there was an election to the Norwegian Storting tomorrow, which party would you vote for?" Participants who indicated they would vote for the governing parties (i.e., the Conservative Party, The Liberal Party, or the Christian Democratic Party) were all coded 1, while support for all other parties was coded 0. We measured perceived state of personal economy by asking "How would you assess your own economic situation today?," anchored on a five point scale from "very bad" to "very good." Worry in the last week was measured with the following questions: "Think about how you have felt during the last seven days, to what extent were you . . . " (1)" . . . worried?" and (2)" . . . anxious?" Considering these two feelings as part of the same dimension, we recoded these two measures into an index scale (α = .861) labeled "Worried last seven days" in our analyses. The scale ranged from 0 "Not at all" to 10 "To a great extent." Third, we studied the durability of the crisis effect (H1 and RQ2). Here, we focused on whether the hypothesized (H1) effect of the COVID-19 crisis on trust in the news media from January 2020 to March 2020 was short-lived and moved back to a baseline level once the initial crisis was over, or whether this change in trust was more of a long-lasting effect (RQ2). For these analyses, time was operationalized as the five survey waves of the NCP, running from 1 (pre crisis wave of January 2020) to 5 (fourth crisis wave of January 2022), comparing the pre-crisis baseline trust level with the trust levels in the four crisis waves (March 2020, June 2020, January 2021, and January 2022).
For all analyses reported in the manuscript, we include the standard demographic variables age, gender, and education to control for biases in the sample. The substantive results from the main models are also reproduced when we use survey weights instead of demographic controls as well as with different estimation techniques (please see the Online Appendix for all robustness analyses).

Treatment and Analysis
We consider the initial treatment effect to be the difference between the pre-crisis baseline (January 2020) and the initial crisis wave (March 2020). Importantly, we did not attempt to isolate effects of, for example, specific press conferences by political leaders, the content in the news media coverage, or the lockdown measures. Rather, we followed the literature on rally effects and focused on the compound treatment (Hernán & VanderWeele, 2011), considering the combined experiences that people had during the initial phase of the crisis.
We conducted panel analyses of our collected data at two and five different points in time. In all panel waves, respondents were asked the exact same question about their trust in the news media. In terms of modeling strategy, we conducted multilevel growth models, conceiving of Y as a function of time itself. The growth models allowed each respondent to have an individual trust level at the outset and to change opinions at a different pace over time while taking into account the nested structure of waves of observations (level 1) within individuals (level 2.) Thus, we studied change in news media trust over time by modeling the outcome for each individual as a positive or negative function of time. Moreover, the growth models allowed us to take unobserved heterogeneity, like measurement error as well as unobserved explanatory variables, into account through random effect components. Further, we allowed for unexplained random variation both at the intra-individual level and at the inter-individual level. Last, the growth models made it possible to include both time-varying and time-invariant variables in the same models, ensuring more effective use of our data material. All growth models were run on a balanced panel, meaning they only included the respondents who answered in all survey waves in the study (Singer & Willett, 2003). Thus, we estimated the effect of the COVID-19 crisis on change in news media trust with the following econometric specification: where i indexes each respondent and t each survey wave, γ estimates the change in trust in news media, and Time is a categorical indicator operationalized as the survey waves (January 2020, March 2020, June 2020, January 2021, and January 2022). X i φ is a row vector of time-invariant independent variables and controls and an intercept. The parentheses include the random part of the model to account for unobserved heterogeneity. ζ 0 i is the random portion of the individual-level intercept, ζ 1 i is the random portion of the individual-level slope for Time, and e it is the idiosyncratic error term for a given individual at a given time. Robust standard errors clustered at the respondent level are included to account for potential serial correlation and heteroskedasticity.
To test H1-H6, and answer RQ1a and RQ1b, we focused on two of the five survey waves. Specifically, we focused on the difference from before the outbreak (January 2020) to after the initial outbreak (March 2020) (n =3155, N =6310). To answer RQ2, we focused on the long-term effects of the crisis and included all five survey waves from January 2020 to January 2022 (n =1636, N =8180).

The Rally Effect on News Media Trust
We initially examined our first formal hypothesis, H1, that trust in the news media increased from before (January 2020) to shortly after (March 2020) the outbreak of the COVID-19 pandemic crisis in Norway. A paired t-test among respondents (n = 3,155) who answered both the pre-crisis January 2020 and the initial crisis March 2020 waves confirmed that there indeed was an increase in trust in the news media in this period of 6.4% (mean difference = .215, 95% CI [.255, .175], t 6308 = 10.566, p < .0001). In the next step, we studied the effect of the COVID-19 crisis based on a multivariate growth model, taking the nested structure of the data into account and controlling for the degree to which the news media was an important source of news about COVID-19 and the rally effect on trust in politicians. As shown in the leftmost model in Table 1, the rally effect on trust in the news media remained robust. In sum, the findings support H1: The compound treatment effect of the COVID-19 pandemic crisis increased citizens' trust in the news media. Note: Entries are maximum-likelihood estimates of growth models, represented as coefficients and robust standard errors. The dependent variable ranges from 1 (least trust) to 5 (most trust). � p < 0:05, �� p < 0:01, ��� p < 0:001

Heterogeneous Effects
To test our two theoretical arguments for why trust in the news media experienced a rally effect (H2 and H3), we first interacted the difference in trust in the news media between the January 2020 and March 2020 waves with our measure of how important the news media were perceived to be in terms of information about the pandemic crisis (measured in March 2020). The middle model in Table 1 shows a positive and statistically significant interaction effect between the increase in trust between the January 2020 pre-crisis wave and the March 2020 initial crisis wave and how important the news media was as a source of news. This supports our first argument (H2), that the rally effect on news media trust was conditional on the news media being perceived as an important source of information about the pandemic. This finding was robust when we also included the following possible confounding variables for the perceived importance of the news media as controls: how important (a) social media and (b) press conferences were perceived to be as a source of information about the pandemic (please see Table  A13 in the Online Appendix for details). The rightmost model in Table 1 analyzes the interaction effect between the change in trust in news media and the change in trust in politicians from the pre-crisis wave to the initial crisis wave. Again, we found a positive and statistically significant interaction interaction effect, thus supporting our second argument (H3) that the rally effect on news media trust is conditional on the increase in trust on other political institutions (i.e., the "trust nexus").
To test the hypothesized heterogeneous effects of education (H4), government party support (H5), experiencing worry/anxiety (H6), belonging to a risk group (RQ1a), and economic situation (RQ1b), we included these variables as interactions with time across models in the growth model reported in Table 2.
Using the leftmost model in Table 2, we tested whether the change in news media trust from January 2020 to March 2020 was contingent upon the respondents' level of education. The results showed a statistically significant effect of the interaction among respondents with medium education, indicating that respondents whose highest level of education was high school/upper secondary school became significantly more trusting of news media after the outbreak of the COVID-19 crisis compared to those with lower levels of education, supporting H4.
The model in the next column was used to test the interaction effect between the change from January 2020 to March 2020 in trust in the news media and intention to vote for the governing parties (measured in January 2020). We found a positive yet statistically insignificant effect of government party support on change in news media trust. Thus, we did not find support H5.
Turning to RQ1a and RQ1b, the next two columns show the interaction effect of evaluation of one's economic situation and belonging to a risk group and change in news media trust from January 2020 to March 2020. While we did not identify a statistically significant interaction effect with evaluation of one's economic situation, we did find a positive and statistically significant interaction effect of belonging to a risk group and change in news media trust. Phrased differently, the interaction suggests that respondents belonging to (or living with someone who belongs to) a risk group were more inclined to increase their trust in the news media from January 2020 to March 2020 compared to those who did not belong to a risk group.
The rightmost model in Table 2 shows that although the interaction effect between the change from January 2020 to March 2020 in trust in news media and feeling worried the last seven days goes in the expected direction, the effect size is small and is not statistically significant. Thus, we do not find support for H6.

The Change in Trust in the News Media over Time
Until now, we have concentrated on the change from before (January 2020) to immediately after the outbreak of the COVID-19 crisis in Norway (March 2020). To answer RQ2, whether the effect on trust lasts over time or quickly returns to its baseline value, we added the three "later crisis" waves of June 2020, January 2021, and January 2022 to our analysis. Figure 1 shows the predicted change in trust in news media before, during the initial outbreak, and as the COVID-19 pandemic crisis in Norway developed. Note: Entries are maximum-likelihood estimates of growth models, represented as coefficients and robust standard errors. The dependent variable ranges from 1 (least trust) to 5 (most trust). � p < 0:05, �� p < 0:01, ��� p < 0:001 The results indicate that the heightened level of media trust found in response to the COVID-19 crisis lasted for a considerable period of time. Although trust was at its highest during the initial phase of the pandemic (March 2020), the mean trust levels had not yet returned to precrisis levels two years later (January 2022). This is shown through the positively signed and statistically significant coefficients found across the waves succeeding the initial stage of the COVID-19 pandemic. That said, the size of the coefficient decreases over time, indicating that there might be a movement toward the pre-crisis levels of trust in the longer term.
In the Online Appendix (see Table A15), we present supplementary analyses of heterogeneous differences in how trust in the news media developed over time from the outbreak of the pandemic crisis, showing that citizens tended to follow the pattern described in Figure 1 across subgroups. We found no subgroup heterogeneity in the decrease in trust from March 2020 in terms of education, economic situation, and worry. However, individuals belonging to a risk group in the initial crisis phase (March 2020) were significantly more likely to decrease their trust in the news media from March 2020 to January 2021 compared to those who did not report belonging to a risk group. In addition, government party voters were less likely to decrease their trust in the news media from March 2020 to January 2022 compared to opposition party voters. Note that government party voters here refers to people who voted for the parties in government prior to the COVID-19 pandemic (January 2020), and that there was an election in September 2021 resulting in a change in government prior to the January 2022 wave.

Discussion
Based on high-quality panel data collected before, immediately after, and at three more mature stages of an unfolding pandemic crisis, our study demonstrates clear and long- lasting rally effects on trust in the news media. While prior studies on rally effects have focused on how political institutions are affected by international crises (Dinesen & Jaeger, 2013;Esaiasson et al., 2020;Wollebaek et al., 2012), we contribute with original empirical and theoretical knowledge on how trust in the news media is affected by an international crisis. Our findings show a robust and statistically significant increase in citizens' trust in the news media from before the pandemic to immediately after the lockdown was imposed in Norway. This is in line with the literature on rally effects with respect to international crises (Dinesen & Jaeger, 2013;Mueller, 1973), including the COVID-19 pandemic crisis (Baekgaard et al., 2020;Esaiasson et al., 2020).
We have theorized and empirically tested two arguments for why rally effects should extend to the news media. First, we argued that an increase in trust in news media is likely due to the importance of the news media as a source of information, as the news media are, for many people, the most important arena for learning about the government's handling of the crisis and the development of the crisis. Note, however, that we only measured the perceived importance of the news media as a source of information about the pandemic (measured in March 2020 -after the pandemic outbreak in Norway). Thus, our data did not allow us to model the change in news use or exposure in general. To provide a stronger test of the argument that trust in the news increases partly due to the news media becoming a more important source of information, it would be preferable to also measure the development of news use and exposure from before the pandemic (January 2020) to the initial crisis phase (March 2020). That said, our argument is in line with a study by Fletcher et al. (2020), who presented evidence suggesting that both news media trust and news use increased during the outbreak of the COVID-19 crisis in the United Kingdom. While Van Aelst et al. (2021) did not measure change in media trust over time, they provided evidence that as the COVID-19 crisis emerged, news use increased for some types of media. It is reasonable to assume that the observed increase in trust in the media in our study would correlate with citizens' increased use of news media.
Second, we found support for our expectation that the trust nexus -the relationship between trust in the news media and political trust (Hanitzsch et al., 2018) -partly explains the increase in trust in the news media on the institutional level. However, these analyses did not allow us to empirically disentangle whether the increase in trust in politicians caused an increase in trust in the news media (or vice versa). To address that possibility, we ran cross-lagged panel models, reported in Figure A1 in the Online Appendix. We utilized observations from all five survey waves to assess whether the effects run from media trust to political trust, from political trust to media trust, or whether there is a reciprocal effect between media trust and political trust (Acock, 2013). The analysis indicate reciprocal effects, as we found positive and statistically significant effects for all waves, running both ways, with substantively similar effect sizes. A possible explanation for these reciprocal effects is that the surge in trust in both the media and the politicians likely occurred almost simultaneously. On the one hand, increased institutional trust in political institutions and the news media is likely to develop shortly after the public learns how the government responds to the threat. On the other hand, the government's response is often presented to the public through press conferences which are conveyed by news media outlets.
We also found support for other hypothesized heterogeneous rally effects that have been found in previous studies, as the level of increase varied by citizens' education. This heterogeneous effect echoes both the findings by Baum (2002) and Esaiasson et al. (2020) in terms of a general effect of education. In line with Baum (2002), we found that the middle category of education experienced a larger rally effect. Esaiasson et al. (2020), however, found a linear effect of education, using a different and more fine-grained measure.
The lack of a significant effect of government party vote on increased trust in the news media runs counter to the findings and theoretical arguments by Baum (2002) and Esaiasson et al. (2020). Our analyses of heterogeneous differences over time indicated that trust in the news media declined less among the government party voters than among the opposition party voters from March 2020 to January 2022. Even so, this result is not comparable to those of Baum (2002) and Esaiasson et al. (2020), as there was an election between January 2021 and January 2022, resulting in a new government. This means that the government party voters in March 2020 were actually opposition party voters in January 2022.
We also hypothesized that worry would foster increased trust in the news media. Contrary to the argument by Schraff (2020), we did not find any signs of such evidence. An important limitation in this regard is that we relied on a measure combining a feeling of worry and a feeling of anxiousness in the last seven days. It is possible that a measure of anxiety for getting infected, or for someone in one's close network getting infected, would have produced different results. This is likely since we found statistically significant interaction effects of belonging to, or having a close relative that belongs to, a risk group. In other words, while generalized anxiety might not influence trust, the experience of a direct threat toward oneself or close ones could prompt an increase in trust due to heightened attentiveness in a threatening situation leading to increased trust. Although prior studies (e.g., Fetzer et al., 2021) have found that COVID-19 influenced economic anxiety, we did not find evidence of substantive or significant effects of the respondents' economic situation on media trust. We encourage future studies to test whether and to what extent trust in the news media increases as a result of increased dependency on the news media in order to protect oneself or close ones during an international crisis.
Extending the literature on the duration of rally effects (e.g., Dinesen & Jaeger, 2013;Putnam, 2002), we also found that the levels of trust in the news media were still substantially higher two years after the lockdown, yet statistically significantly lower than the peak in trust immediately after the outbreak of the COVID-19 crisis in Norway. This drop in trust echoes the results from a panel study during the COVID-19 pandemic in the UK, showing a clear decline in trust in COVID-19 information from the news media (and the UK government) from April to May 2020 Nielsen et al., 2020). However, we did not find support for the argument by Sniderman et al. (2019) in terms of how public opinion tends to return to baseline values over time as the crisis fades. Compared to abrupt events like terrorist attacks, the COVID-19 pandemic was still an ongoing crisis in Norway at the time of our last data point. After the pandemic has ended, future studies should explore whether the COVID-19 pandemic permanently altered people's trust in media.
In sum, these insights extend our current understanding of how times of crises affect trust in the news media and provide a more nuanced picture of the nature and extent of rally effects on news media trust. We empirically demonstrated that while trust in the news media on the institutional level seemed to enjoy a close to across-the-board increase, there were important individual-level differences in the level and degree of this increase in terms of education and belonging to a risk group. Importantly, we found evidence suggesting that some of the same heterogeneous effects that have been found in research on rally effects on political trust are also important for rally effects on trust in the news media.
Our findings warrant further explorations of how political trust and news media trust might also be connected through the way in which the news media cover the crisis, the political handling of the crisis, and the absence of political conflict during a crisis. As argued by Dinesen and Jaeger (2013), the news media can foster a form of "togetherness," emphasizing how citizens can stand united and maintain hope and morale during a crisis. This argument goes hand in hand with one of the most accepted explanations for rally effects on political leaders (Groeling & Baum, 2008): consistent elite messages. In times of global crises that demand an immediate response, there is little time for criticism and disagreement to develop (Brody & Shapiro, 1989). When there is an immediate threat, it is more important that the government formulates a response that can deal with the threat rather than to criticize the government (Sniderman et al., 2019). Thus, members of the political opposition "have substantial incentive to remain silent or to be vaguely supportive -support that later can be, and usually is, withdrawn -and almost no incentive to criticize" (Brody & Shapiro, 1989, p. 90). As long as there is little or no political conflict, the news media tend to convey a picture of cross-party consensus regarding the government's handling of a crisis (Sniderman et al., 2019). Following Indexing theory (W. L. Bennett, 1990; W. L. Bennett et al., 2006), the lack of political disagreement during a crisis should lead to less conflict-driven news stories. As summarized by Robinson (2008), when there is elite consensus, news media operate within the sphere of that consensus, assuming more of a megaphone role, as opposed to a mirror role in which they reflect elite dissensus.
In conclusion, we highlight some important limitations concerning our measure of trust in the news media. Our study is subject to the classic trade-off between the availability of highquality panel data on the one hand and depth and nuance on the other hand. We used highquality panel data from a well-established social science panel infrastructure in Norway, which allowed us to establish a pre-crisis baseline measure of news media trust and political trust before the outbreak of the COVID-19 pandemic crisis. Yet, this also means that there was limited space for panel survey questions. Thus, we opted for a panel item measuring news media trust by asking directly about trust in the news media as an institution rather than measuring specific news brands, differing between platforms, and so on. The use of a single item as a measure of trust in the news raises issues, as trust is not such a unidimensional construct as conceptualized here (Daniller et al., 2017;Hanitzsch et al., 2018). Further, by focusing on trust in the media on the institutional level, we could not capture the rich nuances of difference in trust between individual news brands (Newman et al., 2020).
As these concerns are important for the validity of our findings, we provide supplemental analyses based on the Reuters Digital News Report datasets for Norway in 2020, 2021, and 2022 in the Online Appendix (i.e., Table A16 -Table A19) to further explore nuances in the changes in trust in the media in Norway. While it is important to stress that the Reuters data are not panel data and do not provide measures of news media trust immediately after the lockdown in Norway, they do provide data on trust in specific news brands at the same time as our pre-crisis baseline in January 2020, our fourth panel wave in January 2021, and our fifth panel wave in January 2022. Similar to our results, the results from the analyses of the Reuters data show a statistically significant increase in trust in the news media on the institutional level from January 2020 to January 2021 and a statistically significantly higher level of trust in 2022 compared to the January 2020 baseline. Importantly, we did not identify a statistically significant increase in trust in news on social media and in news from search engines from 2020 -2021. This suggests that our measure of trust in the Norwegian news media on the institutional level is not a measure of increased trust in news from social media platforms and search engines. Among the 11 mainstream brands included in the Reuters survey for Norway in 2020, 2021, and 2022, six brands enjoyed a statistically significant increase in trust from January 2020 to January 2021, and all mainstream brands enjoyed an increase from January 2020 to January 2022. Two important exceptions are so-called right-wing alternative news sites, which stand in opposition to the Norwegian mainstream news media. These analyses of the Reuters data suggest that the increase in trust that we identified with the NCP data is not only restricted to the institutional level but is also observable on the individual brand level.
It is also fair to ask what conclusions can be drawn from the Norwegian case for other media systems. For instance, an analysis covering rally effects in 17 mostly European countries found clear differences in rally effects across countries, showing that the change in trust in Norway was above the mean . One expectation could be that the level of affective polarization in a country could lead to partisan asymmetry in the increase in trust, and Norway is seemingly less polarized compared to for example, the US (e.g., Boxell et al. (2020), but see, Knudsen (2021), who argues that such conclusions depend on how affective polarization is measured). In terms of our general argument that the news media are likely to experience rally effects due to large international crises, we find strong support from the Reuters Digital News report in 2021, showing that 35 out of 40 countries included in both their 2020 and 2021 surveys experienced an increase in media trust on the institutional level. In the 2020 and 2021 Reuters Digital News Reports (Newman et al., 2020(Newman et al., , 2021, there was a decrease in overall trust in Turkey, Mexico, and Bulgaria and a stable (i.e., no change) development from 2020 to 2021 in the US and Spain. However, that is not to say that the degree of increase we identified in our main study, or our results for the heterogeneous effects, translate to other media systems. We need more comparative research to tease out possible similarities and differences across countries regarding who rallies and why some countries see rally effects while some do not.
Considering these limitations, we encourage more research on rally effects on news media trust in order to gain a richer understanding of how international crises, such as the COVID-19 pandemic, affect media trust. First and foremost, we would argue that we need studies on rally effects that connect media use and exposure, media content, and trust on the outlet level. In this way, we could learn whether people who do not increase their news consumption during times of crises increase their levels of trust in news and the extent to which the content that the different news media publish affects citizens' trust. By linking fine-grained measures of news media trust and use on the brand level with content analysis of crisis coverage (De Vreese et al., 2017), future studies could also explore the extent to which the news media enjoy increased trust due to the fact that they are conveying consensus rather than conflict messages in responses to crises.