Fear and the COVID-19 rally round the flag: a panel study on political trust

Abstract The onset of the COVID-19 pandemic boosted political trust in many countries. This article tests the relevance of fear of infection as the micro-level mechanism behind this rally round the flag. This study employs three-wave panel data in the Netherlands, collected days before the first lockdown (early March 2020), during that lockdown (April/May 2020), and after that lockdown (October 2020). Growth curve models isolate the rally effect and its determinants. The article reaches three main conclusions. First, fear of infection is a constituting element of the rally effect: the rise in political trust is more pronounced among people who fear infection. Second, the rise occurs in response to the direct, external threat (health concerns), not in response to the secondary threats (social isolation, economic stagnation). Third, adherents of the radical right are particularly sensitive to the external threat, but only in the short run.

incomplete for two reasons. First, earlier studies were limited by available data to testing the relationship between fear and political support only after support had sharply risen, i.e. not as an explanation of that rise (see Erhardt et al. 2021;Kritzinger et al. 2021). Second, we lack firm knowledge on the mechanisms and conditions behind this relationship.
We argue (i) that fear is one mechanism explaining the rally effect. Moreover, we specify this mechanism, arguing it is (ii) stronger among voters of the radical right, (iii) stronger during the first weeks and months of the pandemic, and (iv) stronger in response to the direct, external threat (health concerns) rather than indirect threats (social isolation, economic stagnation).
Methodologically, this article benefits from a three-wave individual-level panel data set collected in the Netherlands. Crucially, the first wave (early March 2020) was collected in the early days of the pandemic before the announcement of lockdown policies and the rise of political support (Oude Groeniger et al. 2021;Schraff 2021). The second wave took place shortly before the end of the first lockdown (April/May 2020), and the third at the onset of a second lockdown (October 2020). Growth curve models allow us to isolate the rally effect and its determinants temporally, including key determinants that by their nature could only be collected in later waves.
A better understanding of the relationship between citizens' fear and their trust in politics is particularly relevant in the light of governments' responses to the external threat. Raised levels of political trust tend to induce higher levels of acceptance of potentially far-reaching policy measures that government presents as a response to the crisis (Albertson and Gadarian 2015: 12;Hetherington and Nelson 2003). To the extent that the rally round the flag is driven by fear rather than, for instance, policy evaluation and interest alignment, this compliance may not be uniformly positive to democratic accountability.

Theory
The 'rally round the flag' refers to rising support for political authorities in response to external crises. These authorities encompass not only singular political leaders but also political institutions and democracy itself (Chanley 2002). The crises at the root of rallies are conventionally international and military, but also encompass other sudden threats such as terrorist attacks (Dinesen and Jaeger 2013;Wollebaek et al. 2012) and epidemics (Blair et al. 2017). What binds these threats is their external nature: they are not directly caused by (or could not reasonably be foreseen or prevented by) government (see Brody 1991).
Theoretically, fear and anxiety are likely mechanisms that relate external threat to the rally round the flag. These emotions reflect 'low self-control, low certainty, and low external agency' (Wagner and Morisi 2019). Placing trust in political leaders and institutions is a way for citizens to deal with crisis-induced anxiety (see Albertson and Gadarian 2015;Oana et al. 2021). 'When this fear is accentuated by crises, such as a war, a terrorist attack, or -in our case -a pandemic, patriotism and rallying effects follow as citizens seek psychological safety behind leaders they hope will be able to act against the threat' (Baekgaard et al. 2020: 6). Authorities then 'offer an actual and/or symbolic sense of security and safety' (Lambert et al. 2011: 344). Yet the externality of the crisis is key. If blame for the threat is on government itself, fear and anxiety are likely to drive people away from the incumbent.
Empirically, various recent studies on the COVID-19 rally paid attention to the role of anxiety and fear. The pandemic aroused higher levels of anxiety (Qiu et al. 2020;Sibley et al. 2020). The timing of the rally round the flag relates to infection rates (e.g. Baekgaard et al. 2020;De Vries et al. 2021;Schraff 2021). Erhardt et al. (2021) show that general fear and anger had a positive effect on political trust in the months after the lockdown. Similarly, Kritzinger et al. (2021) find that the perceived health threat for the population as a whole had a significant effect on changes in political trust during the lockdown in Austria, but not in France. Yet, as these studies rely on after-lockdown data, direct evidence regarding the role of fear and anxiety on the rally itself remains scarce.
H1. The rise in political trust is more pronounced among voters with more fear for the pandemic.
This general hypothesis needs further specification. We argue that the effect of fear is conditional on the subject (which groups are more sensitive to threat), on timing (when is the effect most pronounced), and on the externality of the threat (which pandemic-related threats induce a rally).
Firstly, the rally round the flag has partisan roots. Voters for opposition parties tend to experience a bigger increase in political trust than voters for government parties (Jennings et al. 2020: 26;Kritzinger et al. 2021), as external crises arouse the perception of interest alignment among common opponents (see Hegewald and Schraff 2021). Building on the logic of fear and anxiety, we can be even more specific. Voters of the radical right are likely to experience the largest boost in political trust in response to raised levels of anxiety and fear, as they are most sensitive to (feelings of) threat (Aichholzer and Zandonella 2016;Hibbing et al. 2014). The vote for the radical right is induced by a wide variety of perceived social threats, including cultural and economic ethnic threats (Tajfel and Turner 1979), decline in social status (Gidron and Hall 2020), and rural marginalisation (Harteveld et al. 2021). Generally, voters of the radical right place less trust in government and parliament in response to these perceived threats (Söderlund and Kestilä-Kekkonen 2009). The COVID-19 pandemic differs from these threats, as it was sudden and external. Rather, the pandemic positioned national political authorities as primary actors to address pandemic-related fears.
Secondly, fear of infection should have a particularly prominent effect on political trust in the early phases of the pandemic. Opinion leadership theory explains the timing and size of rally effects by the available space to raise alternative perspectives on the crisis, and the tendency of politicians and journalists to unify behind decisive policy measures to suggest widespread consensus (Baker and Oneal 2001: 668). This theory offers a particularly relevant explanation for the end of rally effects: as the crisis loses urgency, politicians begin to criticise government policy, and media broadcast alternative perspectives (Brody 1991). Extending this argument, differential partisan rally effects can die out or even invert in response to cues offered by partisan opinion leaders. In the Netherlands, the two radical right opposition parties originally called for a stringent lockdown. However, in the months after this lockdown was put in place, both parties shifted their position. Instead, they called for an end to stringent measures to reopen the economy and preserve individual rights. As their electorate was particularly likely to receive this criticism, we hypothesise that the short-term boost in political trust dissipated most strongly among voters of the radical right. Moreover, the contrarian perspective offered by the two radical right opposition parties is likely to reduce the potential impact of fear and anxiety on political support, as they offer a rival narrative to which these voters can adhere.
Combining the literature above on differentiated effects between electorates and time brings us to the following hypotheses: H2a. In the first wave of the pandemic, the rise in political trust is more pronounced among voters for the radical right-wing populist parties (RRPP) than among other groups of voters.
H2b. In the first wave of the pandemic, the effect of fear on the rise in political trust is stronger among RRPP voters than among other groups of voters.
H3a. By the onset of the second wave of the pandemic, the rise in political trust is less pronounced among voters for the RRPP than among other groups of voters.
H3b. By the onset of the second wave of the pandemic, the effect of fear on the rise in political trust is weaker among RRPP voters than among other groups of voters.
Finally, theoretically, the externality of the threat is key in the case of rally effects (Hetherington and Nelson 2003). The direct, external threat at the heart of the COVID-19 rally is the threat of infection.
Lockdown policies created secondary threats, both social (such as social isolation and loneliness) and economic (such as economic stagnation and loss of income). These secondary threats are not external but tied to government responses (see Kritzinger et al. 2021). Hence, fear for these social and economic threats is unlikely to induce a rally effect.
H4. The rise in political trust is driven by the direct external threat (fear for infection/health) rather than by the indirect secondary threats (fear for social isolation, economic performance).

The Dutch case: COVID-19 and the 'intelligent lockdown'
The timing of the waves of infection, excess mortality, and lockdown stringency in the Netherlands was comparable to other countries in Western Europe. The stringency of the measures in place by 1 April was comparable to that in Belgium, Luxemburg, and Norway, more stringent than Sweden, Germany, and the United Kingdom, but less stringent than France, Spain, and Italy (Hale et al. 2020). Statistics Netherlands estimated approximately 9000 excess deaths between week 11 and week 19 (10 May). A second wave of infections began in August; excess deaths returned in October. New lockdown policies were put in place between September and November, though not as stringent as in spring (Hale et al. 2020). The timing of events and government responses (see Table 1) is crucial to position the panel data in our study.

Isolating the 'rally round the flag' effect
In order to test the 'rally round the flag' effect and its relationship to fear and anxiety, cross-sectional data are insufficient (see Bernardi and Gotlib 2023). It requires individual-level, preferably panel, data that fits the timing of events as well as possible. Yet a fundamental problem in research on the 'rally round the flag' effect is that the instigating events tend to be sudden and data collection inherently is late to the party. This study makes use of long-planned survey data on political trust, that accidentally took place in the very early days of the pandemic in the Netherlands, namely between 11 March and 15 March 2020. These were the final days before the formal acknowledgement of the pandemic and the start of the lockdown in the Netherlands that induced the rise in political support (Oude Groeniger et al. 2021;Schraff 2021). The accidental timing in our data collection leaves us with an incomplete panel design (where some key variables were only introduced in later waves). We compensate for that via growth curve modelling.
Most respondents in wave 1 participated before the first measures were announced (not implemented) on 12 March; all participated before more stringent measures were announced and implemented on 15 March. 1 The second wave of data collection took place before the relaxation of the first lockdown, between 29 April and 10 May. A third wave of data collection took place between 9 and 14 October amid resurging infection rates ( Figure 1).
The sample for this study was drawn randomly from the pool of respondents available to Kantar Netherlands. This pool of respondents is representative of Dutch society on demographic characteristics. Of the 3445 respondents who participated in wave 1, 1990 also participated in wave 2, and 1636 respondents of wave 1 participated in wave 3. We eliminated respondents whose answer patterns reveal perfect straight-lining across three question batteries and ended the survey unrealistically fast (i.e. less than half the modelled time). This reduced the net sample to 1957 respondents (in wave 2) and 1611 respondents (in wave 3) (Figure 1).

Dependent variable
In order to measure political trust we rely on a question battery on trust in institutions. The question reads: On a scale of 1 to 5 where 1 indicates strong distrust in the institutions listed and 5 indicates strong trust in the institutions listed, where would you place yourself? Respondents rated six institutions, including government, the Lower House of parliament, politicians, and political parties. Political trust rises quite sharply between March and April/May, to decline slightly but significantly before October (see Online Appendix A). As these items meet the demands for a unidimensional Mokken scale, we created a single political trust scale using the unweighted average. Additional tests (see Online Appendix B) confirm the robustness of our findings (i) to separate indicators in our scale, most notably government as the most visible, responsible, and authoritative political institution in times of crisis (Hetherington and Nelson 2003), and (ii) to the related outcome measure of satisfaction with democracy (see Bol et al. 2021;Esaiasson et al. 2021).

Independent variables
Party preference is derived from the pre-lockdown survey wave. We recoded the abundance of parties in the Dutch party system to five categories: RRPP are made up by PVV and FVD (Rooduijn et al. 2019).
Because we do not model party preference dynamically across the three waves but rely on static measures from the first (pre-lockdown) wave, our analyses do not suffer from endogeneity issues.
Fear of infection with COVID-19 is measured in waves 2 and 3 as the response to the statement I am afraid that I myself or someone in my family will get infected by the Coronavirus. Response categories range from 1 (disagree completely) to 5 (agree completely). In wave 2, 52% of the respondents agreed with the statement, whereas 22% disagreed. In wave 3, 59% agreed, while 17% disagreed.
The final part of our analysis operationalises concerns with health and concerns with socio-economic issues more elaborately. A question battery in wave 3 asks: To what extent are you concerned about the consequences of the corona crisis for (your own health; the health of vulnerable groups in the Netherlands; your own financial situation; Dutch economy; your own social life; loneliness among the elderly). Answer options range from 1 (no concerns at all) to 4 (a lot of concerns). We additively combined the first two items into concerns with health, and the latter four items into socio-economic concerns.

Control variables
All models control for gender, age, level of education, and social class. 2 To isolate the effect of fear of COVID-19 in our multivariate models, we control for two other relevant COVID-19-related characteristics: (1) experience with infection of the respondent, a family member, friend, neighbour, or colleague (Bol et al. 2021;Schraff 2021), and (2) respondents' policy preference on a dilemma between reopening the economy (relaxing the lockdown measures) and protecting people's health (more stringent measures to keep infections down), measured on a seven-point scale (Oana et al. 2021).

Analysis: growth curve models
We employ growth curve modelling (GCM) in Stata 15 (StataCorp 2017) to analyse the individual-level panel data. The GCM builds on a multilevel structure that nests all waves within persons. Our data cover 3914 observations (L1) nested in 1957 respondents (L2) when we compare wave 2 to wave 1; and 3222 observations (L1) nested in 1611 respondents (L2) when we compare wave 3 to wave 1. We draw two separate comparisons, because we lack data on COVID-19-related attitudes (fear, experiences, policy position) in wave 1, i.e. before the pandemic became a salient issue. Hence, as we are unable to model these determinants as dynamic (time-variant) factors at L1, we model them as static (time-invariant) factors at L2.
The inclusion of a variable 'wave' at level 1 models the difference between the three survey waves. By allowing the slope of the wave effect to vary over respondent characteristics, we can separate static correlates of political trust from dynamic explanations (time-variant, within-person changes). The differential effect of independent variables (IVs) between wave 1 (before the crisis hit hard) and wave 2/3 (after it did) allows for the crucial, dynamic interpretation. The GCM enables us to isolate both the fixed effect of fear (see Bernardi and Gotlib 2023) and its dynamic effect (see Kritzinger et al. 2021). Our models incorporate such cross-level interaction effects both for the IVs and for the control variables. All explanatory models are random-intercept, random-slope models. The GCM is a somewhat conservative estimator as each model has only two waves (Singer and Willett 2003). Table 2 shows the outcomes of GCM on political trust. The cross-level interaction effects with wave assess the conditioning effect of fear on the boost in political trust.

The short-term boost in political trust
First, we assess the short-term effects (between waves 1 and 2). Model 1 shows that fear of infection significantly and positively conditions the effect of survey wave on political trust (b = 0.07). In line with hypothesis 1, the boost in political trust is stronger among people who fear infection more. Additional analyses suggest that the boost in political trust is twice as large among the most than among the least fearful. Political trust Table 2. Growth curve models of trust in political institutions.  increased among all groups of voters (see Online Appendix A). But in line with hypothesis 2a, RRPP voters experienced a significantly stronger boost in political trust between waves 1 and 2 than both government party voters and left-wing opposition voters. 3 The three-way interaction in Model 2 shows that the conditioning effect of fear is not systematically stronger among RRPP voters. However, that is not the full story. The marginal effects plots (see Figure 2) show the net effects of fear in Model 2 in each of the two waves. The net effect in wave 1 represents some confounding effects related to fear of infection. Fear of infection itself could hardly have a causal effect on political support in wave 1, as the salience of the pandemic was still low and fear of infection was only measured in wave 2. Rather, we are interested in the net effects in wave 2 and particularly the difference in the net effect compared to wave 1. The left panel in Figure 2 shows that fear of infection has a significantly more positive effect on political support in wave 2 than in wave 1 among RRPP voters. By contrast, among other groups of voters this differential effect is generally not significant. This supports hypothesis 2b.

The longer-term development of political trust
Model 3 in Table 2 shows that that fear of infection also stimulates the boost in political trust from wave 1 (March) to wave 3 (October), affecting trust in government (b = 0.06). The effect is approximately three times larger among the most fearful than among the least fearful respondents. This is in line with hypothesis 1. However, by October, the boost in political trust is not significantly stronger but also not significantly weaker among RRPP voters than among any of the other groups of voters (for an expanded analysis of the partisan nature of the rally effect, see Online Appendix A). We thus reject hypothesis 3a.
Model 4 reveals that the boost is not significantly stronger among RRPP voters than among other groups of voters: none of the three-way cross-level interaction effects in Table 2 is significant. Moreover, the right panel in Figure 2 visualises the general lack of significant marginal effects among various groups of voters. The direct and conditional differences between RRPP voters and the other voters that were so pronounced in wave 2 (April/May 2020) are absent in wave 3 (October 2020). We thus reject hypothesis 3b.
RRPP voters responded more strongly (in level and effect) to the threat during the first weeks of the pandemic in spring, but that difference had disappeared by the advent of the second wave of the pandemic in autumn 2020. The opinion leadership model offers a partial explanation for this change. Leading politicians of the two RRP parties moved from a position of outspoken concern with the pandemic in the first months of 2020, to positions of scepticism about government policies (as formulated by the PVV) and even about the underlying science on the threat of the virus and the efficacy of the lockdown (as formulated by Forum). This is likely to have affected the attitudes of their adherents.

Concern with health, or concern with socio-economic issues
Finally, wave 3 allows us to delve more deeply into the nature of citizens' concerns in response to the pandemic and the measures that government issued in response. To that purpose, we estimated a growth curve model which included the rival (two-and three-way) interaction effects of primary threats (health concerns) and secondary threats (socio-economic concerns) simultaneously. Based on these models, we estimated the marginal effects.
The left panel in Figure 3 shows the marginal effects of health concerns. In line with our earlier findings, the top panel ('full sample') confirms that the boost in political trust is stronger among people who are concerned with the health consequences of the pandemic. The right panel in Figure 3 shows the marginal effects of socio-economic concerns (such as social isolation and the state of the economy). These concerns are not significantly related to the rise in political trust. Although socio-economic concerns arise partly in response to government measures, they relate neither to a rise nor to a decline in political trust. All in all, we find support for hypothesis 4.

Conclusions
This study aimed to test one of the oft-proposed drivers of the rally effect, namely that of fear. Three-wave panel data in the Netherlands with a pre-lockdown wave allowed us to better isolate the 'rally round the flag' effect temporally. Specifically designed measures of fear for primary external threats (infection) and secondary threats (social isolation, economic stagnation) allowed us to isolate the mechanisms.
We reach three main conclusions. First, fear of infection explains a large part of the boost in political trust, i.e. the rally effect. This lines up with findings by Erhardt et al. (2021) and Kritzinger et al. (2021) for the subsequent period. The rise in political trust is significantly and substantially more pronounced among people who fear infection; the rally effect is estimated to be more than twice as large among the most than the least fearful. Second, while fear about external threat (infection) drives the rally effect, anxieties about secondary threats (such as social isolation and economic stagnation) do not. Third, the effect of fear of infection is stronger among adherents of the radical right, but only in the short run. Opinion leadership (Baker and Oneal 2001;Brody 1991)  offers a likely explanation for why that effect dies out: once radical right politicians opened up the debate to rival, critical perspectives, their voters were the first to begin losing some of the original gains in political trust.
An optimistic reading of our findings suggests that fear and anxiety about the external threat provide a beneficial boost in political trust that stimulates support for and compliance with government policies to combat that external threat (Albertson and Gadarian 2015, 12;Blair et al. 2017;Dinesen and Jaeger 2013;Hetherington and Nelson 2003;Olson and Hjorth 2020). By their very nature, external threats induce fear and anxiety that create a brief window of opportunity in which policy makers can rely on higher than normal levels of support. For this window of opportunity to arise, it is crucial that the threat is truly external, and not perceived to be induced, preventable, or exacerbated by political authorities.
However, there is also a more sceptical implication of the findings in this study. The larger reservoir of political trust is instigated by fear of the external threat, not evidently by an increased trustworthiness of political authorities. Much scholarly research suggests that political trust is conventionally an evaluation according to which one's government not only acts competently, but will also act with integrity when left unattended (Levi and Stoker 2000). Political trust thus tends to be an evaluation of procedural fairness and the quality of policy output (Citrin and Stoker 2018;Van der Meer 2017). During the rally round the flag, one can argue, levels of trust are higher than the procedural and output performance of political authorities warrant. The rally thus places a burden on the responsibility of governments to preserve procedural fairness and meet the heightened hopes and expectations of the population even beyond the external crisis.

Data
The panel data set will be deposited at data archive DANS before March 2023. Before deposition, the data are available for replication purposes upon request.

Notes
1. The findings are robust by limiting the sample to respondents who participated before the Thursday announcement. 2. Because we found no differential effects of these control variables over the three survey waves, we do not control for their interaction with survey wave to preserve parsimony. We considered two more control variables: subjective evaluations of government policy, and emotions such as societal unease and anomie. We left them out of the analyses out of concern for endogeneity. Nevertheless, additional robustness checks reveal that our findings are robust to their inclusion. 3. Given the distribution, bottom and ceiling effects are unlikely causes for this differential effect.

Disclosure
The authors declare that the research was conducted in the absence of any potential conflict of interest.

Funding
This work was supported by the Dutch Science Foundation NWO under Grant 452-16-001, 2016.