Rallying around the flag in times of COVID-19 Societal lockdown and trust in democratic institutions

: In times of severe international crises, such as wars and terrorist attacks, citizens tend to ‘rally around the flag’ and increase their support for political leaders. We ask if the rallying effects identified in the literature extend to the societal lockdowns in response to the COVID-19 pandemic. COVID-19-related lockdowns differ from crises studied in the existing literature because they are political crisis responses with severe and immediate negative effects on the economy. Using daily responses right before and after the announcement of the Danish lockdown on March 11, 2020, we study trust in democratic institutions among unemployed Danes over the first three weeks of a large-scale societal lockdown. OLS estimates show that trust in the Danish Prime Minister’s administration was higher immediately after the lockdown announcement. This increase lasted throughout the entire period of measurement (until the end of March). We find similarly increased trust in other institutions, most significantly the judicial system and the public sector at large, whereas findings for trust in parliament and the media are less clear. Interrupted time series estimates point to the same conclusions albeit they produce estimates with more noise. Overall, our findings are consistent with the idea that citizens tend to ‘rally around the flag’ in times of crisis and furthermore suggest that increased trust tends to spill over to institutions that are not involved in crisis management decisions.

The COVID-19 pandemic has disrupted the entire world and governments are struggling to respond.Social distancing is needed to reduce the spread of the virus and policymakers across the world have introduced measures to minimize contact, including limits to the allowed sizes of public gatherings and complete lockdowns of entire countries.One political dilemma is how to balance public health and economic considerations.Not all jobs can be reasonably performed at home (Dingel & Neiman, 2020) and thus COVID-19 in many countries threatens to replace a healthy economy with a deep recession.In the US, between March 22 and March 28, more than 6.6 million people filed claims for unemployment, beating the previous record of 3.3 million the week before and the pre-pandemic record of about 700,000 claims during one week in 1982 (Rainey & McCaskill, 2020).Many people, including the President of the United States, suggest that "the price of the cure" may have become higher than the "price of the problem" itself (Haberman & Sanger, 2020).
It is only natural that policymakers are in doubt about what is the best solution for the people they represent.It is also natural if they worry about public responses to their decisions.Some face upcoming elections, but all need the public's cooperation to combat the spread of the virus.Research suggests that high political trust is associated with a higher willingness to comply with government policies (Im et al., 2014), including social distancing (Olsen & Hjorth, 2020) and hand sanitizing recommendations (Lim et al., 2020) in response to COVID-19.Thus, if policies harm people's trust in government, it may harm the collective ability to combat the pandemic.
We investigate how trust in various democratic institutions developed in response to the announcement of a wide-ranging lockdown of one Western European society: the lockdown of Denmark announced on the evening of March 11, 2020.Research suggests that in times of international crisis, people tend to rally around the flag by increasing support for their political leaders (Brody & Shapiro, 1991;Hetherington & Nelson, 2003;Lee, 1977;Mueller, 1970;1973;Oneal & Bryan, 1995;Parker, 1995;Perrin & Smolek, 2009;Dinesen & Jaeger, 2013).To investigate rally effects in response to the Danish lockdown, we utilize data collected among unemployed Danes in the time around the lockdown announcement.Unemployed individuals serve as an interesting case as they may, as we explain in the theory section, be considered a less-likely case for rallying around the flag.We analyze data collected between March 9 (i.e., almost three days before the lockdown announcement) and March 30 (N=2,125 of which 852 responses were collected before and 1,273 were collected after the lockdown).In addition to a series of regression analyses, the timing of our data collection allows us to use an interrupted time series design, meaning that we are able to gauge the immediate response to the lockdown.We find that trust in Prime Minister Mette Frederiksen's administration increased in the aftermath of the lockdown announcement.The increase lasted throughout the entire period of measurement.Furthermore, we find similar increases in trust in other societal institutions, in particular courts and the public sector at large.Findings for trust in parliament and the media are less clear.

Empirical and theoretical background
On March 11, 2020, at 8.30PM, Danish Prime Minister Mette Frederiksen held a press conference, announcing that large parts of the Danish society would be locked down to combat the spread of COVID-19 (Danish Government, 2020).The decision came as a surprise as Italy, which was among the world's worst hit nations, was the then only European country having imposed nationwide restrictions in response to the pandemic (Baker, 2020;Wright, 2020).Denmark had not yet seen its first COVID-19-related death.
As part of the lockdown, government prohibited gatherings of more than 100 people, closed all public educational and cultural institutions, sent home all public employees in non-critical functions, and encouraged private employers to do the same.In the following days, further restrictions were introduced, for example prohibiting gatherings of more than 10 people, and closing bars, restaurants, shops, malls, etc.
Our investigation of trust in government following the lockdown announcement is informed by the literature on people's tendency to rally around the flag during times of international crisis.First proposed by Mueller (1973), the theory of rallying around the flag suggests that in times of international crisis, citizens will often react by increasing support for their political leaders.Since Mueller presented his theory, rally effects have been found in response to a variety of international crises, such as wars (Brody & Shapiro, 1991;Edwards & Swenson, 1997;Kriner, 2006;Parker, 1995) and terrorist attacks (Perrin & Smolek, 2009;Dinesen & Jaeger, 2013;Wollebaek et al., 2012).
Many studies (often from the US) have focused on presidential approval ratings during crises, where, according to the patriotism school, "Americans rally to the president as the anthropomorphic symbol of national unity -a kind of living flag" (Hetherington & Nelson, 2003, p. 37).Other studies have found rally effects beyond US presidential approval ratings.For example, in a study of reactions to 9/11, Chanley (2002) found an increased trust in government, even when controlling for presidential approval ratings.Similarly, Wollebaek and colleagues (2012) found broad increases in institutional trust in response to the 2011 Utøya attack and Dinesen and Jaeger (2013) found increases in trust, not only in political institutions but also in the justice system and the media, following the March 11, 2004 attack in Madrid.
While many studies have shown evidence of rally effects, less scholarly focus has been devoted to theorizing about the psychological mechanisms behind these effects.Like most other studies on the issue, our main contribution is empirical, but we find reason to highlight two potential, not mutually exclusive, mechanisms behind such effects.
First, according to terror management theory, leaders function as aides for citizens to manage a deepseated, primal fear of death.Accordingly, 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 (Landau et al., 2004).
An alternative mechanism is built on the insight that most people hold ambiguous (i.e., both positive and negative) views about most objects of evaluation, including government.As noted in Zaller's (1992) accessibility axiom, "the more recently a consideration has been called to mind or thought about, the less time it takes to retrieve that consideration or related considerations from memory and bring them to the top of the head for use" (Ibid.,p. 48).In other words, people's summary evaluations will always depend on the relative salience (or accessibility) of these competing considerations in people's minds.
It is possible that rally effects result from international crises reducing the salience of certain considerations, e.g., concerns about local crime and policy failures, that would else have made people critical about their leaders (Chanley et al., 2000;Chanley, 2002, p. 480).Political leaders may seek (sometimes with success) to improve public support by emphasizing threat and crisis management interventions themselves (Kaufmann, 2004;Willer, 2004), thereby shifting the political agenda away from issues that would leave them in an unfavorable light.And independent of their own communication, leaders will also tend to be helped by the fact that partisan conflict is often muted during times of international crises, meaning that people are exposed to less critique of their leaders (Brody & Shapiro, 1989) and therefore have less critical considerations accessible for use (Zaller, 1992) when making evaluations of government.This is a central argument in the opinion leadership school of rally-around-the-flag theory (Hetherington & Nelson, 2003, p. 37-38).1 The COVID-19 pandemic differs in important ways from other crises investigated in existing literature.For instance, Mueller (1973) expected rally effects to emerge in response to specific, dramatic, and sharply focused international crises.The current pandemic is certainly specific and dramatic, but it is less sharply focused than e.g.wars and terrorist attacks.It is not easy to point to a clear beginning (e.g., a bomb exploding) or end (e.g., a terrorist being killed or a peace agreement being signed) of the crisis.Moreover, when a country is locked down, beneficial public health effects are relatively invisible (the intended effect is people not contracting the virus, i.e., a non-event, and at the individual level, most people would not have experienced severe illness, even without a lockdown).What is visible to many people is the severe and immediate negative effects on the national economy, meaning that the political crisis response has negative immediate effects on people's lives.2This may very well affect their reactions.However, it is clear that during a lockdown, the pandemic becomes more salient than before, potentially accentuating fear and crowding out competing political concerns, as the accessibility-based rally mechanism suggested above would predict.Moreover, a Canadian study concludes that the COVID-19 crisis has been a "rare moment of cross-partisan consensus" (Merkley et al., 2020) and we are confident stating that the same has been the case in Denmark, at least during our data collection.
Based on the literature on rallying around the flag, we therefore expect an increase in trust in government following the announcement of the lockdown.
Our expectations are strengthened by a few existing studies having investigated reactions to COVID-19 as well.One study analyzed votes at the 2020 election in the German state of Bavaria and found that the dominant party benefited in counties with many COVID-19 cases (Leininger & Shaub, 2020).
Moreover, a study has documented increases in political trust in reaction to lockdowns in 15 Western European countries (Bol et al., 2020; see also Harell, 2020).
These studies, however, focused on average reactions to the pandemic.Our investigation concerns reactions among a group of people who had a low pre-crisis trust in government and who are vulnerable to the economic consequences of the lockdown.3Our respondents are therefore likely to have become more economically pessimistic due to the lockdown and economic pessimism has been found to reduce trust in government (Chanley et al., 2001;Chanley, 2002), meaning that rally effects may have to compete with an opposite negative effect.Moreover, Edwards and Swenson (1997) argue that rally effects are mainly driven by people who already support their political leaders.Similarly, in Perrin and Smolek's (2009) study of 9/11, rally effects were mostly driven by white respondents.
African Americans had a lower pre-trust in government and their trust was less affected by the terrorist attack.In general, literature suggests that people's pre-crisis trust will be positively associated with the size of rally effects of international crises (Chatagnier, 2012).Based on this, our sample's low trust in government prior to the pandemic gives us reason to consider unemployed individuals a less-likely case for rallying around the flag.

Data and analysis
We utilize data, which we collected among unemployed members of 3FA, a Danish unemployment insurance fund servicing primarily blue-collar workers.The data were collected for the purpose of a research question not related to the COVID-19 crisis but happened, due to the timing of the data collection, to provide good conditions for an investigation of rally effects.Thus, answers were collected between March 9 (i.e., almost three days before the lockdown announcement) and March 30, 2020, when the lockdown was still in place.Data were collected using an online survey, which was sent to a sample of 10,000 individuals.In total, we received 852 responses prior to the announcement of the lockdown and 1,273 responses afterwards (see Supplementary Information Section 3 for descriptive statistics and item response rates).A first survey invitation was sent to individuals on March 9 and reminders were sent on March 13 and 25.
Importantly, we are able to track incoming responses on an hour-by-hour basis around the lockdown, which is an advantage compared to other rally studies that have most often relied on separate data collections more or less shortly before and after the events being studied.Our data enable us to employ an interrupted time series (ITS) design to plausibly improve causal traction.Our analysis of the ITS resembles analyses of regression discontinuity designs (RDD).In a standard RDD, a naturally occurring threshold is utilized around which respondents are as-if randomly assigned, making standard RDDs theoretically clean natural experiments.The strength of this as-if random assumption is increased by basing conclusions on data in a narrow bandwidth around the threshold (Dunning, 2012).
In our setup, the discontinuity occurs in time: If we consider the abrupt lockdown announcement on March 11 as a threshold, and hence assume that our respondents sort as-if randomly around answering the survey before and after the lockdown announcement, we can use the announcement as an exogenous treatment, particularly if we narrow our empirical focus to the days immediately preceding and following the announcement.In effect, on this assumption, respondents before the lockdown announcement and respondents after the lockdown announcement will not differ systematically on observable and unobservable characteristics, and causal estimates of the effect of the lockdown can be obtained.
In our data, this assumption is somewhat problematic.While some respondents provided their answer after the lockdown announcement, all were unemployed prior to it, mitigating concerns that characteristics of the average unemployed blue collar worker change as many lose jobs to the pandemic.Moreover, in an examination of observed characteristics, we do not identify any significant differences in gender ( − = 0.144), age ( − = 0.371), education ( − = 0.185), full-time unemployment status ( − = 0.565), or duration ( − = 0.220) between people who responded before and after the announcement.We do find imbalances on number of unemployment periods ( − = 0.012) and of children ( − = 0.001).
However, these do not maintain significance inside the narrower bandwidth used for our ITS analyses below (Supplementary Information Section 5).
More generally, ITS suffers from a tradeoff between exogeneity (the plausibility of the as-if random assignment assumption) and bandwidth, meaning the width of the timespan in which exogeneity is assumed to hold (Hausman and Rapson, 2018).This potential issue is accentuated in our data since data are relatively sparse close to the threshold (see Figure 1).To gain sufficient statistical power, we rely on responses up to one week from the lockdown announcement.This weakens our claim to causal identification.Consequently, to mitigate the potential biases from covariate imbalances along our threshold, we report ordinary least squares (OLS) estimates from models including a series of controls alongside our ITS estimates.All ITS estimates are produced using Dimmery's (2016) RDD package for the R environment.Estimates are local estimates of the impact of the lockdown estimated with robust standard errors and the triangular kernel recommended in the literature (Lee & Limieux, 2010).
We noted above that literature has linked economic pessimism to lower trust in government (Chanley et al., 2001;Chanley, 2002), meaning that trust after the lockdown announcement may be negatively affected by the economic consequences of closing down the country.
In our survey, we asked about respondents' expectations regarding their employment perspectives and unsurprisingly, on average, people became significantly more pessimistic following the lockdown ( − = 0.009).Before the lockdown, 52 percent expected to get a job within two months but after the lockdown, the share of respondents who expected this to happen had dropped to 46 percent.Also, as will be visible from Table 1, more pessimistic unemployment expectations are associated with a lower trust in government.Turning to rally effects on trust in government, Figure 1 shows that people's trust in the Prime Minister's administration increased significantly in response to the announcement of the lockdown.
From before to after the announcement, average trust in the administration increased from 4.16 to 5.57 on a scale from 0 (no trust at all) to 10 (complete trust), corresponding to an increase of 34 percent ( − < 0.001); an increase, which lasted throughout the entire data collection.
However, due to the weaknesses noted above, Figure 1 should be read as indicative and not as evidence of a causal effect since it reflects a simple difference in average trust along two time periods, i.e., without controlling for any background characteristics.
Table 1 addresses this using a regression framework.Model 1 shows the raw difference in means estimate before and after the lockdown.Model 2 includes demographic background variables for which we have administrative data for all respondents.Model 3 adds survey data on respondents' education, number of children, full-time unemployment (as opposed to e.g.receiving part time education), duration of unemployment, and number of unemployment periods in the past five years.
Model 4 includes a dummy for respondent expectations to remain unemployed in two months to control for effects of the lockdown on people's employment expectations.Lockdown estimates remain substantial and statistically significant in all models.
We advance the analyses above in two ways.First, we provide ITS estimates of the effect of the lockdown on people's trust in Mette Frederiksen's administration.In an attempt to balance power and plausibility of the 'natural experiment', we opt for a bandwidth of one week.4Second, the observed increase in trust may either be exclusively directed at the Prime Minister and her administration as leaders of the crisis management efforts.Or, as suggested by findings from Chanley (2002), Wollebaek and colleagues (2012), and Dinesen and Jaeger (2013), it could be directed at democratic institutions more broadly.Our data allow for an examination of this.In addition to trust in Mette Frederiksen's administration, our survey included items measuring respondents' trust in the Danish parliament, the public sector at large, the courts, and the media.

Figure 2: Results across institutions
Notes: Results from ordinary least squares (OLS) and ITS models with 95 and 90 (bolded) percent confidence intervals."OLS" provides OLS estimates of the full sample with no controls; "ITS" provides RDD package estimates use a seven-day bandwidth and no controls; and "OLS w. controls" provides OLS estimates from models replicating model 4 in table 1 above.
Figure 2 shows the estimated changes in trust in all these institutions from before to after the March 11 lockdown announcement, using both OLS and the ITS framework.Consistent with the findings of Chanley (2002), Wollebaek and colleagues (2012), and Dinesen and Jaeger (2013), all institutions enjoyed an increase in trust following the announcement.Whether differences are statistically significant to some extent depends on the model employed.All differences are highly significant in OLS models with and without the full battery of controls (replicating models 1 and 4 in Table 1).In the ITS models, however, trust in the Prime Minister's administration only reaches significance at the 0.10 level, while trust in the media and parliament are not significant at conventional levels.The ITS model results are presented in table form in Table 2.
It is worth noting that our OLS estimates for trust in the PM's administration retain statistical significance at conventional levels using the 7-day bandwidth, or any of a number of other bandwidths the data permit (Supplementary Information Section 6).This may indicate that our mixed findings in the ITS framework are the result of the relatively inefficient local estimation used in RDD style models and the sparsity of data very close to the threshold.

Discussion
In the section above, we have investigated changes in institutional trust among unemployed Danes following the Danish COVID-19 lockdown announcement.While some institutions enjoyed bigger increases in trust than others, and while interrupted times series estimates are more noisy than OLS estimates, our findings overall indicate that people reacted to the announcement by increasing their trust in both the Prime Minister's administration, the public sector, the courts, as well as possibly the media and parliament.Below, we discuss alternative explanations of the findings.
A limitation is that we have to compare levels of trust for different individuals at different points in time.Hence, a concern is if differences in trust are due to differences in respondent characteristics.
As noted, balance tests along the lockdown announcement show that early responders do not differ significantly from late responders on observable background variables, at least within a seven-day bandwidth.However, for obvious reasons, we do not have (and cannot create) experimental variation regarding interventions like the one we study and thus, we cannot entirely reject the possibility that early and late responders differ on unobserved characteristics.
Moreover, the lockdown was not the only event during the days around March 11, 2020, with a potential to affect unemployed people's trust in the Danish government.The lockdown announcement was accompanied by a series of other public announcements, meaning we cannot fully reject the possibility that our findings are, at least in part, due to events other than the lockdown itself.
In placebo tests (reported in the Supplementary Information Section 9), in which we test whether estimates of the lockdown change if we use another cutoff date than March 11, we find even stronger effects in the ITS framework with a cutoff set on March 12 to 13, whereas the lockdown coefficient becomes statistically insignificant if the cutoff is set earlier than March 11 or later than March 13.This raises concerns about the impact of events after March 11 (a possibility we discuss shortly).It should, however, be noted that the placebo tests are somewhat susceptible to a lack of power due to the relatively low number of responses that we obtained in the days right after March 11.Moreover, in an OLS framework, a similar exercise shows consistently significant differences across the date threshold -which is not surprising given the much more efficient estimator -but without the increase in estimate size for thresholds immediately after the actual lockdown.This increases our confidence in at least until trust had reached pre-9/11 levels around six months later (Ibid.,p. 141).We find reason to believe that the durability of the rally effects in response to the COVID-19 crisis and lockdown will depend on at least two factors.
First, for how long will the ongoing pandemic continue to put heavy restrictions on our lives?We noted in the theory section that the COVID-19 crisis differs from other crises, such as terrorist attacks, by being much less focused in time.The sense of acute crisis, which emerged as the lockdown was announced, continued throughout the entire data collection and as we write, it seems that our lives will continue to be influenced by the pandemic, at least until a vaccine becomes available or an effective treatment is identified.However, as solutions are found and lives begin to normalize, the COVID-19 crisis will also become less salient.Rally effects will then probably fade out if they have not already done so since the end of our data collection.
Second, as noted in the theory section, the COVID-19 crisis response has been characterized by a high degree of interparty cooperation, meaning that, at least during our data collection, there was a very low level of political conflict.According to the opinion leadership school of rally-around-the=flag theory, lack of partisan conflict is a central ingredient in creating rally effects, and literature has pointed to partisan consensus as a central factor in predicting the durability of rally effects (Hetherington & Nelson, 2003).Since the end of our data collection, political discussions in Denmark have begun focusing on reopening society, and these discussions have been characterized by more partisan conflict.It is possible that, as the political climate becomes more polarized and critiques of the government become more accessible, this will accelerate the process towards pre-COVID-19 trust in government.Unfortunately, we do not have the data to test this proposition.
A final issue, which needs to be addressed, is the external validity of our findings.We have investigated rally effects in the context of one specific group of people (unemployed Danes) and it is relevant to ask if findings generalize to other parts of the Danish society as well as to other countries.Starting with the question of generalizability to the rest of the Danish society, we have noted that because of their low level of pre-crisis trust in government and economic vulnerability, we consider unemployed people a less-likely case when it comes to rally effects on trust in government.There is, however, a nuance to this, stemming from the fact that our respondents are all unemployed blue-collar workers.
A sizable proportion of Denmark's blue-collar workers tend to vote for the Social Democratic party, i.e., the party of the Prime Minister during the COVID-19 crisis (Holst, 2019).We have referred to Chatagnier's (2012) finding of pre-crisis trust in government being positively related to rally effects, but these effects are mainly driven by people who did not vote for their political leaders (in the case of Chatagnier's study, people who had not voted for the president).Pre-crisis trust in government does matter for people who voted for the leaders as well, but to a lesser degree (Ibid., p. 642).In that sense, the blue-collar background may have led to stronger rally effects than what would be seen among other groups of unemployed Danes.Unfortunately, our data do not include measures of partisanship and thus, we are not able to test whether Social Democrats reacted differently from opposition party voters.It is, however, hard to believe that no rallying has happened throughout the rest of the society as well, which is also reflected in media polls reporting a general, high support for the way government has responded to the pandemic (Toft 2020).
Another question regards generalizability to societies other than Denmark.Different countries (and jurisdictions within countries, e.g., states and municipalities in the USA) have differed in their COVID-19 crisis response and even when looking at countries where lockdowns have been in place, differences, e.g. in the political culture, may affect how citizens respond.While our findings are consistent with other COVID-19 related investigations in Western Europe and Canada (Bol et al., 2020;Harell, 2020;Leininger & Shaub, 2020), where rally effects have also been found, there are countries, including the USA, where the pandemic response has been more polarized along partisan lines (Motta et al., 2020).It is likely that weaker rally effects may be found in more partisan-polarized settings (Brody & Shapiro, 1989;Chanley, 2002;Hetherington & Nelson, 2003), which may also explain why President Donald Trump has not enjoyed the same rally tendencies as other Western governments (Enten, 2020).

Notes
1: Note, however, that Dinesen and Jaeger (2013) found an increased trust in government following the 2004 terrorist attack in Madrid despite a hostile political environment (Ibid., p. 919), suggesting that although lack of partisan conflict is probably helpful in terms of generating rally effects, it is not a necessary condition.
2: During the weeks after the lockdown announcement, the lockdown was accompanied by a series of economic rescue packages to curb the economic effects of the lockdown and prevent mass layoffs.
Despite these measures, however, it became immediately clear that the lockdown would have dramatic effects on the Danish economy.According to prognoses from The Danish Economic Council, between 110,000 and 160,000 Danes will lose their jobs because of the current crisis (Bendtsen, 2020), which should be seen in light of the fact that only 131,742 Danes were unemployed before the lockdown (The Danish Agency for Labour Market and Recruitment, 2020).

Figure 1 :
Figure 1: Trust in the Prime Minister's administration before and after the March 11 estimates of the effect of lockdown within a 7-day bandwidth and robust standard errors in parentheses.* p < 0.10, ** p < 0.05, *** p < 0.01.

Table 2 :
ITS Estimates for trust in institutions