Is It Safe for Us to Be Together Again? Identity,Trust,and Perceived Risk at Mass Events During the COVID-19 Pandemic

Abstract

Three years after the outbreak of COVID-19, governments are still working toward a return to “normal life.” Yet, the twin forces of ongoing disease threat and progressively relaxing restrictions raise important questions about whether, where, and when people feel safe. We analyzed data from post-event surveys of participants at live events held across Denmark between June and November 2021 (n_individuals = 4,932; n_events = 79). Consistent with the social identity model of risk-taking, identification with the audience, trust in others, and felt safety were interrelated. Multi-level modeling revealed that audiences responded to the heightened risk posed by crowds after COVID-related attendance restrictions were lifted, but also that individual differences in identification blunted the connection between crowd density at events and individual feelings of trust and safety. These findings point to a potential identity-based slippage between felt safety and actual safety in the context of collective participation and disease threat.

Keywords

COVID-19 mass gatherings crowd psychology risk trust identity

After the outbreak of COVID-19, live culture around the world suddenly stopped. Performers, their fans, and the wider industry struggled to cope with this unprecedented situation. Three years later, the world is still in the process of reopening. Yet, despite overall progress in containing COVID-19, threat persists: Infection rates rise and fall as new variants appear, and it seems likely that disease will remain a backdrop to social life for the foreseeable future. There are also worries about how extended periods of isolation might have affected the psychology of live audiences. For example, a U.K. government report expressed concerns that “increased risk-taking after lockdown will lead to a spike in drug-related deaths at festivals this summer” (Digital, Culture, Media & Sport Committee, 2021), and some media commentators pointed to “pent-up stress and anxiety from nearly two years of social isolation” (Bendix, 2021; Bennett, 2021) to explain a crowd tragedy that occurred in 2021. Although the latter analysis is unsubstantiated by evidence, it seems plausible that eager anticipation combined with a lack of recent experience of live environments could render at least some members of the audience less able to judge the line between appropriate risk and caution. If true, the present situation raises important questions about what we can expect from crowds in the “post-corona” world; how the balance between risk and safety is understood; and how this can best be managed to create experiences that are both safe and enjoyable to those who attend.

To begin the task of providing answers to these questions, we analyze data from attendees at live events held in Denmark between June and November 2021. This period affords a unique window into a changing COVID-19 landscape. Early in 2020, Denmark was swift to implement lockdown measures, including caps on all but small gatherings, mandatory working from home, closure of nonessential businesses, and cancellation of all live events that year. Following an effective program of testing, contact tracing, and eventual vaccination, in 2021 Danish authorities were progressively relaxing restrictions, allowing for larger and more varied live events to be organized across the summer of that year. Aside from the primary requirement to verify attendee’s corona status as a condition of entry via a valid corona passport, live culture had also been subject to caps on event size, spacing requirements, zoning into audience subsections, and curfews. But on September 1, 2021, all restrictions were lifted, meaning that audiences could—for the first time since early 2020—attend events just as they had before the pandemic and do so without the need for prior testing or evidence of vaccination. The significance of this date was profound, and it was celebrated by the industry and fans as the return of live culture to Denmark (e.g., Korkmaz, 2021). Because data collection spanned this period, we were able to examine the consequences of changing regulatory and audience environments for the reported experience of individual audience members. As a theoretical background to that examination, we draw on contemporary research on crowd psychology, especially as it relates to questions of perceived risk and safety.

Shared Identity and the Health Risks and Resources of the Crowd

Fears about uninhibited and risky behavior in (post-corona) crowds, as expressed in some media commentaries (Bendix, 2021; Bennett, 2021), reflect assumptions of classic crowd psychology. In his book, The Crowd, Gustav Le Bon (1895/1995) hypothesized that a loss of self, and consequently of self-regulation, occurs when individuals are immersed in crowds and cloaked by the anonymity that comes with that—psychological transformations that inevitably lead to antisocial and destructive behavior. However, cumulative research provides little evidence for Le Bon’s account (e.g., Postmes & Spears, 1998), an account that has also been criticized for failing to address the dynamic and changing nature of crowd behavior (e.g., Reicher, 1996; Warren & Power, 2015). As such, contemporary crowd scholars have largely moved away from the assumption that crowds straightforwardly compromise individual psychology.

Much contemporary research instead draws on the foundations of social identity (Tajfel & Turner, 1979) and self-categorization theories (Turner et al., 1987). The core assumption of this approach is that group membership is an important basis for individual identity (i.e., one’s social identity), and that once incorporated into the self-concept, individual thought, feeling, and action are regulated by specific social meanings, values, and norms that define the group. From this foundation, researchers have extended the scope of theorizing to consider the many ways in which group membership and collective participation structure individual thought, feeling and action across important domains (e.g., S. A. Haslam, 2014).

Applied to the health domain, this body of work documents a role for group membership in supporting individual health and well-being, thereby constituting a “social cure” (C. Haslam et al., 2018; Jetten et al., 2012). For example, when group norms value health, group-based social influences will promote individual health-protective behavior (e.g., Liu et al., 2019). Moreover, the group in-and-of-itself can be a foundation for health and well-being. Because group membership provides individuals with a framework for locating themselves in the social world and satisfies basic needs for belonging, control, meaning, and esteem, group membership supports positive psychological outcomes (Greenaway et al., 2016). Consistent with these ideas, studies show that participation in mass gatherings can support positive social relations (Hopkins et al., 2019) and contribute to the psychological health of attendees (Khan et al., 2015), at least when participation is grounded in a sense of shared identity with others present.

However, there are also risks conferred by shared identity. When group norms are oppositional to health, social influence will discourage healthy behavior (Oyserman et al., 2014; Tarrant & Butler, 2011). Moreover, precisely because shared identity fosters positive social relations (Hopkins et al., 2019), reciprocal trust between people within a group can result in the underestimation of risks, leaving individuals feeling comfortable in situations where caution might be better advised (Cruwys et al., 2021). Studies show that situationally induced or self-reported identification with a group not only amplifies trust in other group members but also attenuates disgust (Hult Khazaie & Khan, 2020; Reicher et al., 2016), contributing to comfort with interpersonal proximity (Novelli et al., 2010). In the era of social distancing, reduced sensitivity to intrusions of personal space may be concerning from a public health perspective. Social identity has also been shown to transform the appraisal and experience of environmental risks, for example, rendering people more tolerant of loud noise (Shankar et al., 2013), extreme cold (Pandey et al., 2014), and crowding (Alnabulsi & Drury, 2014; Novelli et al., 2013), when these environmental parameters are associated with a valued group membership. Again, from a health perspective, reduced sensitivity to environmental risks might be concerning.

The Current Research

As outlined above, the consequences of shared identity in a crowd can be both positive (e.g., mutual trust, solidarity, comfort, and well-being) and negative (underestimation of risks posed by others, tolerance for health-compromising group environments, and individual risk-taking). On the one hand, this means that some of the fears expressed about crowd behavior in the wake of the corona pandemic are probably misplaced, based as they are on an outdated understanding of crowd psychology (see also Drury et al., 2013). On the other hand, there are risks posed by the crowd. Effectively harnessing identity processes to minimize risks while simultaneously maximizing positive social and psychological processes remains a significant challenge for event organizers (Drury et al., 2021; Hopkins & Reicher, 2021).

Recent work already documents a role for social identity processes in the experience of post-corona crowds (Morton & Power, 2022; Rathbone et al., 2022; Smith & Templeton, 2022). However, this work relies exclusively on individual data and provides little insight into how subjective feelings of risk reflect the objective environments that crowds create. The present research fills this gap by examining how event-level parameters influence subjective feelings of safety after attending live events and the role of shared identity in shaping this. Following Cruwys and colleagues’ (2021) social identity model of risk-taking, we expected that identification with others in an audience would be associated with felt safety at the event (Hypothesis 1), and that this would be mediated through trust (Hypothesis 2). Following observations that identification can buffer individuals against stressful or challenging group environments (e.g., Alnabulsi & Drury, 2014; Novelli et al., 2013; Pandey et al., 2014; Shankar et al., 2013), we further expected that identification would moderate individual sensitivity to environmental risks associated with the crowd (Hypothesis 3). A conceptual model summarizing these predictions is presented in Figure 1.

Figure 1

Conceptual Model Summarizing Hypotheses About the Link Between Crowd Identification and Felt Safety (Hypothesis 1) Mediated Through Trust (Hypothesis 2), and the Role of Identification in Moderating Responses to Crowd-Relevant Environmental Parameters (Hypothesis 3)

Method

Participants and Data Collection

The data analyzed here were collected as part of the larger SAFE 2.0 project led by Dansk Live (Denmark’s interest group for festivals and venues) to evaluate the reopening of cultural and sporting events across 2021. Data on cultural events included a brief survey, developed by the Roskilde Festival Experience team (Roskilde Festival’s consulting firm) informed by previously published work on post-corona crowds (i.e., Morton & Power, 2022) and in consultation with the current authors. The survey was distributed by Dansk Live and Roskilde Festival Experience via email lists to individuals who attended a variety of live events between June 24 and October 30, 2021 (based on ticket purchase contact details). Of the total pool of 117,544 potential respondents, 5,003 attendees of 85 different events accessed the survey via the link and provided data. Separately, the Roskilde Festival Experience team collected data from event organizers about the type, timing, location, total capacity, and number of attendees at their event. Of the 85 events, 79 provided complete data. As such, the analyses reported below are based on 4,932 people who attended 79 events. No incentive or reward was offered to individuals or event organizers to participate in this research. Individual demographic and event features are summarized in Table 1. All data can be accessed via OSF: https://osf.io/hzdfj/.

Table 1

Individual and Event Characteristics

Individuals N = 4,932		Events N = 79
Age^a	M (SD) = 43.36 (13.51)	Responses per event	M (SD) = 62.43 (219.69)
	Median = 44		Median = 18
	Range = 14–81		Range: 1–1,772
Gender identity	Female: 3,053 (62%)
	Male: 1,841 (37%)	Audience size	M (SD) = 1,486.63 (5,831.21)
	Nonbinary: 12 (<1%)		Median: 179.00
	Not stated: 26 (1%)		Range = 14–49,000
Corona status	Vaccinated: 4,411 (89%)	Venue capacity	M (SD) = 1,948.19 (6,013.36)
	Immune: 115 (2%)		Median: 500.00
	Negative test: 368 (8%)		Range = 90–50,000
	Positive test: 1 (0%)
	Don’t know: 37 (1%)	Crowd density^b	M (SD) = .52 (.32)
		Crowd density^b	Median: .54
Month of attendance	June: 119 (2%)		Range = 0.02–1.03
	July: 220 (5%)
	August: 1,439 (29%)	Audience mode	Seated: 38 (48%)
	September: 3,058 (62%)		Standing: 37 (47%)
	October: 96 (2%)		Both: 4 (5%)
		Month of event	June: 4 (5%)
			July: 5 (6%)
			August: 20 (25%)
			September: 45 (57%)
			October: 5 (6%)

Two participants reported that they were born in 1870 (151 years old) and 1904 (117 years old). We assumed that these were errors and that the intended dates were 1970 and 2004, respectively. Age descriptives are reported after adjusting these two numbers. ^bCalculated as Number of Attendees/ Venue Capacity.

As noted above, data collection was conducted independently of the research team. The goal of the organizations collecting the data was simply to cover as many individuals and events as possible. Sampling was not guided by a priori power calculations based on plausible effect sizes from past research. The large number of individual respondents gives high power to detect even very small effects at that level (e.g., 95% power to detect f ² = .003 with α = .05; based on a sensitivity analysis for R ² in multiple regression with two predictors; that is, identification and trust as predictors of felt safety: Cruwys et al., 2021). However, achieving adequate power in multi-level designs is a recognized challenge (Mathieu et al., 2012), especially when the goal is to test cross-level interactions (e.g., between individual identification and event-level environmental parameters). The combination of many individuals per event, spread across a decent number of events (see Table 1), is somewhat reassuring. Yet, the power to detect all but large cross-level interactions is still limited in a sample with these properties (Arend & Schäfer, 2019). Previous field studies, based on individual rather than multi-level data, suggest medium to large effects of identification, environmental parameters (i.e., crowding), and their interaction on positive emotions and felt safety in the crowd (.13 ≤R ²≤ .55: Alnabulsi & Drury, 2014; Novelli et al., 2013). The best powered of these tests (Alnabulsi & Drury, 2014: N = 1,184) included observer estimations of crowd density (rather than individual perceptions) and produced an effect size falling between the extremes, R ² = .22, representing a medium-to-large effect (f ² = .28). Although these effect sizes come from individual-level data, there is a plausible basis for expecting medium to large effects of identification with the crowd, crowd-related environmental parameters, and their interaction on felt safety.

Survey

The attendee survey was necessarily brief but covered basic information about corona status around the event, age (birth year), and gender identity. In addition, self-report measures tapped concepts of identity, trust, and felt safety adapted from prior work on would-be festival attendees (reported in Morton & Power, 2022). Respondents were presented with a series of statements to which they indicated their level of agreement (1 = not at all, 5 = very much). All individual items are reported in Table 2.

Table 2

Survey Items, Observed Scores, and Factor Loadings

Item		Observed range	M (SD)	Factor loadings^a
Item		Observed range	M (SD)	1	2	3
Based on the event you just participated in, how did you feel about participating?
	Safe	1–5	4.46 (.78)	.730	−.004	.213
	Calm	1–5	4.55 (.72)	.746	.046	.115
	Happy	1–5	4.81 (.49)	.441	.373	−.019
	Worried	1–5	1.47 (.73)	−.809	.073	.016
	Anxious	1–5	1.21 (.59)	−.733	.038	.116
I felt uncomfortable with other people’s physical proximity		1–5	1.56 (.88)	−.622	.047	−.066
I enjoyed being together with other people at the event		1–5	4.33 (.90)	.384	.397	.105
In the audience at the event, I felt like I was a separate person who was watching alone		1–5	1.27 (.69)	.071	−.722	.109
In the audience at the event, I felt like I was part of a larger gathering of people who were watching together		1–5	4.57 (.78)	.004	.836	−.040
I experienced a feeling of shared identity with other people in the audience at the event		1–5	3.87 (1.20)	−.047	.681	.169
My impression was that most people at the event cooperated with the organizers and followed any regulations or COVID-19-related restrictions		1–5	4.06 (1.08)	.010	.069	.859
My impression was that people stopped following any regulations or COVID-19-related restrictions as soon as they were inside the event		1–5	2.18 (1.23)	.063	.104	−.863
How much do you trust the organizers of this event to make the best decisions to protect the audience against unnecessary health risks?		1–5	4.21 (.94)	.192	.070	.672

Note. Factor loadings >.4 in bold.

Principal components factor analysis with oblique rotation, extraction based on eigenvalues >1.

Data Preparation

Before analyzing the survey data, individual items were subjected to a principal components factor analysis with oblique rotation. This revealed three factors with eigenvalues greater than 1, which together accounted for 57.68% of the variance. As can be seen in Table 2, all items loaded >.40 onto one of three factors corresponding to the focal concepts of felt safety, shared identity with the audience, and trust in relevant others. The only exception was the item “I enjoyed being together with other people at the event,” which displayed a primary factor loading <.40 and split across the factors representing felt safety and identification. To preserve the independence of the latter constructs, we excluded this item prior to creating composite measures by averaging the items corresponding to each factor (after appropriate reverse scoring). Variable descriptives and intercorrelations are reported in Table 3.

Table 3

Individual-Level Variable Descriptives and Intercorrelations

					rs
Variable	Observed range	M (SD)	Skew (SE)	Kurtosis (SE)	1.	2.	3.
1. Felt safety	1–5	4.60 (.51)	−1.72 (.04)	3.86 (.07)	[.81]	.50	.28
2. Trust	1–5	4.03 (.90)	−1.02 (.04)	0.70 (.07)		[.77]	.17
3. Identification	1–5	4.39 (.68)	−1.37 (.04)	1.73 (.07)			[.60]

Note. All correlations significant at p < .001; values in square brackets indicate scale reliabilities.

From the event data, we were especially interested in two environmental features: the density of the crowd (which is both relevant to COVID-19 transmission and has been the topic of previous social identity research: Alnabulsi & Drury, 2014) and the regulatory environment under which the event was held (which, as noted in the introduction, was changing in important ways across the data collection period). To create an index of crowd density, we divided the number of attendees at the event by the venue’s total capacity, subsequent values representing whether the venue (regardless of size) was at (or over) capacity versus relatively empty. With respect to regulatory environment, a significant shift occurred on September 1, 2021: Events held prior to that point were required to check the corona status of attendees as a condition of entry (e.g., via showing the official government “coronapas” app, which documented individual vaccination and testing records), whereas events held after this time point were not required to comply with any corona-related restrictions or conditions of entry. Accordingly, we created a dummy code representing whether the event was held prior to versus after September 1, 2021. Correlations among event-level parameters are reported in Table 4.

Table 4

Event-Level Parameter Intercorrelations

Parameter	2.	3.	4.	5.
1. Pre-Post September 1	−.14	.06	.06	.10
2. Crowd density		−.37***	−.55***	.31**
3. Felt safety^a			.58***	−.12
4. Trust^a				−.19
5. Identification^a

Based on event-level means.

p <.05. **p <.01. ***p < .005.

Analytic Strategy

As a first step in the analysis, to check the assumed theoretical model underlying our data (the social identity model of risk-taking: Cruwys et al., 2021), we inspected the correlations among the individual-level variables and tested the expected pattern in which trust mediated associations between identification and felt safety via the PROCESS macro (Hayes, 2017). This addresses Hypotheses 1 and 2.

Next, to address Hypothesis 3, we examined the interplay between individual identification with the audience and the environment associated with this crowd on both trust and felt safety. Because the full data set involves individuals nested within events, we analyze these data via mixed (i.e., multi-level) models using JAMOVI (The JAMOVI Project, 2021). In these analyses, individuals were clustered by event, and individual differences in identification as well as event-level parameters of crowd density and regulatory environment were included as fixed predictors. To facilitate interpretation of coefficients, continuous predictors (Identification and Crowd Density) and dependent variables (Trust and Felt Safety) were standardized via z-transformations prior to analysis. Regulatory environment was a binary factor (Pre- or Post-September 1). We tested all main effects of the predictors and all possible two- and three-way interactions. We included random intercepts for event and a random slope for identification (following recommendations for testing cross-level interactions: Heisig & Schaeffer, 2019). Parameters were assessed via restricted maximum likelihood estimation, and degrees of freedom were calculated using Satterthwaite’s approximation.

Preliminary Checks

Preliminary checks revealed that relative to the reference category of female, male respondents felt safer at events (p < .001), whereas nonbinary/unidentified respondents felt less safe (p < .001) and reported lower levels of trust (p = .017) and identification (p < .001). Older respondents reported higher levels of trust and felt safety (ps < .001), and those who were vaccinated (binary-coded against all other corona status categories) were lower in trust (p = .004) but higher in identification (p = .001). Tables 5 –7 report results of analyses with and without these covariates included.

Table 5

Regression Models Testing Links Between Identification, Trust, and Felt Safety

	Trust										Felt safety
Parameter	B	SE	β	t	p	B	SE	β	t	p	B	SE	β	t	p	B	SE	β	t	p
Identification	.22	.02	.16	11.72	<.0001	.22	.02	.17	12.29	<.0001	.15	.009	.20	16.49	<.0001	.15	.009	.20	16.50	<.0001
Trust											.26	.007	.47	38.47	<.0001	.26	.007	.46	37.09	<.0001
Male^a						−.04	.03	−.02	−1.66	.10						.06	.013	.06	4.90	<.0001
Nonbinary/Other^a						−.24	.14	−.02	−1.70	.089						−.14	.07	−.02	−1.98	.048
Age						.01	.0009	.20	14.27	<.0001						.001	.0005	.04	2.93	.003
Vaccinated^b						−.28	.04	−.09	−6.69	<.0001						−.04	.02	−.02	−1.89	.059
Model summary:	R ² = .03					R ² = .07					R ²2 = .29					R ² = .30
Model summary:	F(1, 4930) = 137.26					F(5, 4926) = 73.51		73.51			F(2, 4929) = 1,007.74					F(5, 4925) = 345.59
	p <.0001					p <.0001					p <.0001					p <.0001
Pathways:											Without covariates					With covariates
											Effect	SE	95% CI			Effect	SE	95% CI
					Direct effect (Identification to Safety)						.15	.009	[.1315, .1669]			.15	.009	[.1320, .1676]
					Indirect effect (via Trust)						.06	.006	[.0464, .0692]			.06	.006	[.0470, .0693]
					Standardized indirect effect (via Trust)						.08	.007	[.0626, .0911]			.08	.007	[.0639, .0922]

Note. CI = confidence interval.

Reference category = Female. ^b Reference category = Unvaccinated.

Table 6

Mixed-Models Predicting Trust in Organizers and Fans

	Model 1					Model 2					Model 3
Parameter	Est (SE)	[95% CI]	df	t	p	Est (SE)	[95% CI]	df	t	p	Est (SE)	[95% CI]	df	t	p
Intercept	0.34 (.04)	[0.26, 0.42]	82.69	8.48	<.001	0.26 (.05)	[0.17, 0.35]	57.71	5.63	< .001	−0.22 (0.06)	[0.11, 0.33]	157.82	3.77	<.001
Identification						0.20 (.02)	[0.17, 0.23]	11.27	12.17	< .001	0.19 (0.02)	[0.16, 0.23]	10.46	11.95	< .001
Crowd density						−0.09 (.03)	[−0.15, −0.04]	96.88	−3.32	0.001	−0.08 (0.03)	[−0.14, −0.03]	95.65	−3.05	0.003
Pre-Post September 1						−0.25 (.09)	[−0.43, −0.07]	57.71	−2.71	0.009	−0.28 (0.09)	[−0.46, −0.11]	56.62	−3.14	0.003
Id × Pre-Post						0.07 (.03)	[0.01, 0.14]	11.27	2.13	0.056	0.07 (0.03)	[0.01, 0.14]	10.46	2.32	0.042
Pre-Post × Crowd						−0.16 (.06)	[−0.27, −0.05]	96.88	−2.78	0.006	−0.16 (0.06)	[−0.27, −0.05]	94.65	−2.93	0.004
Id × Crowd						0.03 (.01)	[0.00, 0.06]	44.08	2.33	0.024	0.03 (0.01)	[0.00, 0.05]	43.42	2.20	0.033
Id × Crowd × Pre-Post						−0.01 (.03)	[−0.06, 0.04]	44.08	−0.49	0.626	−0.01 (0.03)	[−0.06, 0.04]	43.20	−0.42	0.674
Male^a											0.02 (0.03)	[−0.04, 0.07]	4,823.15	0.63	0.527
Nonbinary/Other^a											−0.39 (0.15)	[−0.67, −.011]	4,909.97	−2.68	0.007
Age^b											0.00 (0.00)	[0.00, 0.01]	3,765.79	3.31	< .001
Vaccinated^c											0.04 (0.05)	[−0.06, 0.13]	4350.90	0.77	0.439
Model summary:	Intercept variance: 0.08					Intercept variance: 0.06					Intercept variance: 0.05
	Slope variance: —					Slope variance: 0.00					Slope variance: 0.00
	Residual variance: 0.83					Residual variance: 0.77					Residual variance: 0.76
	ICC: .08					ICC: .07					ICC: .06
	AIC: 13,178.91					AIC: 12,785.42					AIC: 12,771.24
	Marginal R ²: .00					Marginal R ²: .08					Marginal R ²: .09
	Conditional R ²: .08					Conditional R ²: .14					Conditional R ²: .15

Note. Trust, Identification, and Crowd Density standardized (z-transformed) prior to analysis. CI = confidence interval; ICC = interclass correlation; AIC = Akaike information criteria.

Reference category = Female. ^b Mean centered. ^c Reference category = Unvaccinated.

Table 7

Mixed-Models Predicting Felt Safety

	Model 1					Model 2					Model 3
Parameter	Est (SE)	[95% CI]	df	t	p	Est (SE)	[95% CI]	df	t	p	Est (SE)	[95% CI]	df	t	p
Intercept	0.15 (.04)	[0.08, 0.22]	67.06	4.35	<.001	0.10 (.03)	[0.03, 0.17]	39.85	2.95	0.005	0.07 (.05)	[−0.03, 0.17]	198.85	1.37	.173
Identification						0.27 (.03)	[0.21, 0.33]	39.01	9.10	< .001	0.26 (.03)	[0.21, 0.32]	37.75	9.11	< .001
Crowd density						−0.08 (.02)	[−0.13, −0.04]	74.30	−3.51	< .001	−0.06 (.02)	[−0.11, −0.02]	70.71	−2.82	.006
Pre-Post September 1						−0.12 (.07)	[−0.25, 0.02]	39.85	−1.68	0.101	−0.12 (.07)	[−0.25, 0.02]	37.36	−1.67	.103
Id × Pre-Post						0.03 (.06)	[−0.09, 0.15]	39.01	0.51	0.613	0.03 (.06)	[−0.08, 0.15]	37.77	0.59	.557
Pre-Post × Crowd						−0.11 (.05)	[−0.20, −0.02]	74.30	−2.49	0.015	−0.11 (.05)	[−0.20, −0.03]	69.14	−2.52	.014
Id × Crowd						0.04 (.02)	[0.00, 0.08]	72.90	2.04	0.045	0.04 (.02)	[0.00, 0.08]	71.32	2.08	.041
Id × Crowd × Pre-Post						−0.00 (.04)	[−0.08, 0.07]	72.90	−0.02	0.982	0.00 (.04)	[−0.07, 0.08]	70.88	0.07	.943
Male^a											0.15 (.03)	[0.10, 0.21]	4,204.34	5.37	< .001
Nonbinary/Other^a											−0.38 (.15)	[−0.68, −0.08]	4,918.38	−2.51	.012
Age^b											0.00 (0.00)	[0.00, 0.01]	1,800.44	4.10	< .001
Vaccinated^c											−0.04 (0.05)	[−0.13, 0.05]	2,778.17	−0.81	.418
Model summary:	Intercept variance: 0.05					Intercept variance: 0.03					Intercept variance: 0.02
	Slope variance: —					Slope variance: 0.02					Slope variance: 0.02
	Residual variance: 0.95					Residual variance: 0.83					Residual variance: 0.82
	ICC: .05					ICC: .03					ICC: .03
	AIC: 13,827.37					AIC: 13,194.42					AIC: 13,141.28
	Marginal R ²: .00					Marginal R ²: .09					Marginal R ²: .10
	Conditional R ²: .05					Conditional R ²: .13					Conditional R: .14

Note. Felt Safety, Identification, and Crowd Density standardized (z-transformed) prior to analysis. CI = confidence interval; ICC = interclass correlation; AIC = Akaike information criteria.

Reference category = Female. ^b Mean centered. ^c Reference category = Unvaccinated.

Results

Individual-Level Patterns

Looking first at the individual-level data (summarized in Table 3), two things can be seen. First, overall levels of felt safety, trust, and identification with the audience were all high. Despite the wider pandemic context, people who chose to attend live events in 2021 seem comfortable with their decision, connected to others in the audience, and trusting of both others at events and the organizers to behave properly and make decisions in their interests. The high levels of trust should perhaps be understood as reflective of the wider culture of trust in Denmark (see Lindholt et al., 2021; Petersen et al., 2021; Power et al., 2023). Although the high endorsement resulted in a negative skew for all variables, there was still variation and individual responses covered the full possible scale range.

Second, and as expected based on past work, identification, trust, and felt safety were significantly intercorrelated. Guided by the social identity model of risk-taking (Cruwys et al., 2021), we tested the indirect pathway between identification and felt safety via trust using PROCESS (Model 4; Hayes, 2017). Regression models and tests for mediation can be found in Table 5. The predicted indirect pathway connecting the focal variables was significant. The data are therefore consistent with the underlying assumptions of the theoretical model (Hypotheses 1 and 2). However, these being correlational data, the causality assumed under that model cannot be directly confirmed and alternative sequences remain equally plausible.

Cross-Level Patterns

As can be seen in Table 4, the event-level predictors of regulatory environment and crowd density were unrelated. On the whole, however, events that were more crowded were experienced as more risky (i.e., lower average trust and felt safety), although these events were also associated with higher average levels of identification. To explore the interplay between event and individual parameters further, we ran multi-level models according to three steps. First, we specified an empty model predicting trust including only the random intercept for event (to provide an estimate of the interclass correlation), then we included the three predictors and their interactions, and then we included covariates. The results of these analyses are summarized in Table 6 (trust as outcome) and Table 7 (felt safety as outcome).

There were significant main effects of individual identification and event-level crowding on trust and felt safety. People who identified more strongly with the audience around them, and who were at events characterized by lower crowd density, were more trusting and felt safer. In addition, after accounting for the other factors, trust was higher at events held before COVID-related restrictions were lifted on September 1, 2021, than after this date, although this shift did not significantly affect felt safety at the events.

Beyond these main effects, two 2-way interactions consistently emerged: Crowd Density × Regulatory Environment and Identification × Crowd Density.¹ Simple effects relating to each of these interactions are reported in Table 8. For completeness, the full pattern of data combining all variables is plotted in Figures 2 and 3. Higher levels of crowding were associated with both less trust and less felt safety only after COVID-19 restrictions were lifted, but not before (i.e., the Crowd Density × Regulatory Environment interaction; seen in the contrast between left and right panels of Figures 2 and 3). Regardless of the regulatory period, higher levels of identification attenuated the associations between crowding and each outcome (i.e., the Identification × Crowd Density interaction; seen in the consistently attenuated slope for high and moderate identifiers relative to low identifiers). The latter pattern provides support for Hypothesis 3, at least with respect to crowding, if not the wider regulatory environment surrounding events.

Table 8

Simple Effects of Crowd Density According to Regulatory Period and Identification Levels

Outcome	Moderator	Level	Estimate (SE)	[95% CI]	df	t	p
Trust	Period	Before September 1	−0.00 (.05)	[−0.09, 0.09]	93.20	−0.08	.94
		After September 1	−0.17 (.03)	[−0.23, −0.10]	99.45	−5.32	<.001
	Identification^a	16%	−0.11 (.03)	[−0.18, −0.05]	92.98	−3.43	<.001
		50%	−0.07 (.03)	[−0.13, −0.02]	97.79	−2.69	.008
		84%	−0.06 (.03)	[−0.11, −0.00]	99.88	−2.13	.035
Felt safety	Period	Before September 1	−0.01 (.04)	[−0.08, 0.07]	63.72	−0.19	.853
		After September 1	−0.12 (.03)	[−0.17, −0.07]	85.28	−4.66	<.001
	Identification^a	16%	−0.11 (.03)	[−0.17, −0.04]	84.25	−3.03	<.003
		50%	−0.05 (.02)	[−0.09, −0.00]	51.71	−2.24	.030
		84%	−0.03 (.02)	[−0.08, 0.02]	40.46	−1.23	.226

Note. Identification, Crowd Density, Trust, and Felt Safety standardized (z-transformed) prior to analysis; simple effects were tested with covariates included in the model. CI = confidence interval.

We test simple effects at high and low values of identification defined by percentiles rather than ±1 SD to ensure that values were within range of the observed data given variable skew. Similar patterns are observed when ±1 SD are instead used.

Figure 2

Relationship Between Crowd Density and Trust at Different Levels of Identification Before (Left Panel) and After (Right Panel) Regulations Changed on September 1, 2021

Figure 3

Relationship Between Crowd Density and Felt Safety at Different Levels of Identification Before (Left Panel) and After (Right Panel) Regulations Changed on September 1, 2021

Discussion

Research already points to the importance of shared identity (e.g., van Bavel et al., 2022; Vignoles et al., 2021) and trust (e.g., Bargain & Aminjonov, 2020; Bicchieri et al., 2021) in shaping individual and collective coping with the global threat of COVID-19. Here, we show the relevance of these same concepts for understanding how individuals cope with the “return to normal” as threat subsides. Higher identification with audiences was generally connected to higher levels of trust and felt safety at events held in Denmark across the summer of 2021. Although audience members seemed attuned to the altered meaning of crowded spaces after corona-related restrictions were lifted on September 1, 2021, higher levels of identification blunted individuals’ appraisal of the risks posed by such crowding.

Implications

Reflecting the social identity approach, we started from the assumption that collective participation is a meaningful (rather than meaningless) human activity that can contribute to positive (not just negative) individual and collective outcomes (cf. Le Bon, 1895/1995). Our data dovetail with this meta-theoretical standpoint: We observe people feeling safe, trusting, and connected to the others at what were sometimes quite large and crowded events held at a time of varying degrees of threat and uncertainty. The data also broadly suggest that people were not unthinking about their engagement with strangers at that time. Instead, perceptions of safety varied as a function of objective circumstances and the risks inherent in these: Environments that were objectively more risky (e.g., by being both more crowded and less regulated) were subjectively experienced as less safe (i.e., the consistent Regulatory Environment × Crowding interaction).

However, we also find that identification with others in a crowd can moderate how individuals respond to the environment the crowd creates. In particular, higher levels of identification attenuated the negative effects of crowding on trust in others and feelings of safety at live events (Cruwys et al., 2021). Although others have observed a role for identification in transforming responses to group-based environments, the current data are unique by drawing on real-world event data collected independently of individual responses rather than, for example, experimental manipulations of environmental stimuli (e.g., Shankar et al., 2013) or self- (Novelli et al., 2013) and other reported (Alnabulsi & Drury, 2014) perceptions of environmental properties.

The moderating role of identification emerged most consistently in response to the environmental parameter of crowding but was less clear with respect to changing regulatory environments. Speculatively, of these factors, crowding is most strongly connected to identity, being literally composed of the others with which one identifies (or not). Indeed, our findings replicate previous research that has focused on crowding as a highly relevant environmental parameter (Alnabulsi & Drury, 2014). By comparison, regulatory environments are disembodied and more diffuse. Yet, our respondents were also not oblivious to shifts in the regulatory environment and the implications of these: The meaning of crowding was reappraised after the corona status of others in the crowd could no longer be assumed. From a theoretical standpoint, one could expect social-relational factors, like social identity, would be especially important for resolving questions of risk in the context of uncertainty (e.g., Smith et al., 2007). This possibility would imply further interactions involving identification and regulatory environment in addition to crowding. Such a pattern was hinted at in our data (Note 1) but was less consistent and therefore awaits further testing.

Overall, the observed patterns involving identification imply that this factor not only contributes to felt safety in general (Cruwys et al., 2021), but also that it can disrupt the link between objective and subjective risk, something that might be concerning from a public health perspective. Although social identity processes are acknowledged to be both a resource and a risk when it comes to crowd safety and event management (Drury et al., 2021), this finding raises more precise questions about what event organizers should do to cultivate the former while minimizing the latter. Answering such questions is difficult based on the current data. Yet, alongside the theoretical framework that informed their analysis, these data again point to the centrality of trust in discussions of risk and safety (Cruwys et al., 2021). The high levels of trust observed in the current data are perhaps testament to the previous work of organizers to build up trusting relations with their audience (and of the wider society within which these events were held; Lindholt et al., 2021; Petersen et al., 2021; Power et al., 2023), but they are also a reminder of organizers’ (and societal decision-makers’) responsibility to act in ways that continue to demonstrate their trustworthiness. Along these lines, as society continues to open, it would seem most productive to work with (rather than against) audiences to create events premised on shared concerns for the safety of others (see also Reicher & Bauld, 2021). Were organizers to instead communicate distrust to their audience, this could undermine shared identity and weaken willingness to comply with health and safety regulations (e.g., Drury et al., 2020; Petersen et al., 2021).

Limitations

The data we draw on are unique, covering a wide variety of events against the backdrop of a dynamic and changing health threat. Yet, these data are limited by their coarseness. Individual data provide only a broad impression of how people were thinking and feeling at that time. Observational and qualitative data would complement the current analysis, potentially revealing how individuals within crowds express their identity with others, demonstrate their trust, and come to experience environments as safe versus risky (Power et al., 2018). In addition, while we have focused only on variations in crowding and the major regulatory shift that occurred on September 1, 2021, regulations and audience environments were continually changing in other, more gradual, ways across this timeline. There is also likely to be undocumented variation in how regulations were enacted on the spot at individual events. Data at this level of granularity are difficult to obtain, but it would be very interesting for future research to try to link individual experiences and perceptions of safety to the more dynamic micro-environments that occur within larger crowds and events.

Although our data set contains many individual observations and covers a variety of events, the power to properly detect and explore cross-level interacts was still restricted (Arend & Schäfer, 2019). This is a common issue with multi-level designs, especially when researchers do not have control over sampling (Mathieu et al., 2012), as was the case here. Future research should therefore treat the patterns we have observed as indicative, rather than conclusive, and continue to explore the robustness of Identification × Crowding interactions and their meaning for individual experiences within crowds. Future research might also like to further explore whether and how social-relational processes (like identification, not just with crowds but also with regulatory authorities) shape the interpretation of crowd management procedures and practices. Although we did not find consistent evidence of a role for identification in moderating responses to changing regulatory environments, we believe this remains a plausible focus for hypotheses based on prior social identity work in other crowd management settings (e.g., Stott & Radburn, 2020).

Conclusion

After COVID-19 pressed the pause button on cultural life, new questions arise over how individuals feel about the prospect of being together again and where to draw the line between risk minimization and the joy of shared experience. Data from individuals who attended a variety of different events held in Denmark over the summer of 2021 suggest that individuals were attuned to the risks posed by their return to less regulated and more crowded environments, but that shared identity with others in the audience offset concerns about the risks posed by the crowd. As such, these findings also highlight how shared identity can open up a discrepancy between what feels safe and what is safe.

Footnotes

Acknowledgements

Thanks to Roskilde Festival Experience for providing access to data and facilitating the research, especially Nethe Katrine Jørgensen, Nils Nagel, and Morten Therkildsen. Thanks also to Merlin Schaeffer for statistical advice.

Handling Editor: Miyamoto Yuri.

Declaration of Interest

The data reported here were collected as part of the SAFE 2.0 research project led by Dansk Live and supported by the Danish Ministry of Culture. Key findings from the overall project are summarized in the report “Kultur og Idræt under en Pandemi” (“Culture and Sport under a Pandemic”), which can be accessed here (in Danish): . The survey was developed with input from the authors, but data collection and compilation were conducted independently by Roskilde Festival Experience (Roskilde Festival consulting firm) on behalf of Dansk Live. The analyses conducted in this article are independent of the commercial interests of Dansk Live or any of the participating organizations.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The data reported here were collected as part of the SAFE 2.0 research project funded by the Danish Ministry of Culture.

ORCID iDs

Thomas A. Morton

Séamus A. Power

Notes

Author Biographies

Thomas A. Morton is Professor of Social Psychology at the University of Copenhagen. His current research interests center on the ways in which identity is experienced and expressed in response to the symbolic and material properties of social and physical environments.

Séamus A. Power is an Associate Professor at the Department of Psychology at the University of Copenhagen. He is a cultural psychologist with interests in social and political psychology and in mixed qualitative and quantitative research methods.

References

Alnabulsi

Drury

(2014). Social identification moderates the effect of crowd density on safety at the Hajj. Proceedings of the National Academy of Sciences, 111(25), 9091–9096.

Arend

M. G.

Schäfer

(2019). Statistical power in two-level models: A tutorial based on Monte Carlo simulation. Psychological Methods, 24, 1–19.

Bargain

Aminjonov

(2020). Trust and compliance to public health policies in times of COVID-19. Journal of Public Economics, 192, Article 104316.

Bendix

(2021, November 10). Pandemic-linked stress may have fuelled the deadly crowd surge at Astroworld, a sociologist says. Business Insider. https://www.businessinsider.com/why-did-crowd-surge-astroworld-psychology-2021-11?r=US&IR=T

Bennett

(2021, November 8). Death at Astroworld: 5 ways crowds can quickly go bad. Psychology Today. https://www.psychologytoday.com/us/blog/modern-minds/202111/death-astroworld-5-ways-crowds-can-quickly-go-bad

Bicchieri

Fatas

Aldama

Casas

Deshpande

Lauro

. . .Wen

(2021). In science we (should) trust: Expectations and compliance across nine countries during the COVID-19 pandemic. PLOS ONE, 16(6), Article e0252892.

Cruwys

Greenaway

K. H.

Ferris

L. J.

Rathbone

J. A.

Saeri

A. K.

Williams

. . .Grace

(2021). When trust goes wrong: A social identity model of risk taking. Journal of Personality and Social Psychology, 120(1), 57–83.

Digital, Culture, Media and Sport Committee. (2021). The future of UK music festivals (HC 49). https://committees.parliament.uk/publications/6136/documents/68377/default/

Drury

Novelli

Stott

(2013). Psychological disaster myths in the perception and management of mass emergencies. Journal of Applied Social Psychology, 43(11), 2259–2270.

10.

Drury

Reicher

Stott

(2020). COVID-19 in context: Why do people die in emergencies? It’s probably not because of collective psychology. British Journal of Social Psychology, 59(3), 686–693. https://doi.org/10.1111/bjso.12393

11.

Drury

Rogers

M. B.

Marteau

T. M.

Yardley

Reicher

Stott

(2021). Re-opening live events and large venues after Covid-19 ‘lockdown’: Behavioural risks and their mitigations. Safety Science, 139, Article 105243.

12.

Greenaway

K. H.

Cruwys

Haslam

S. A.

Jetten

(2016). Social identities promote well-being because they satisfy global psychological needs. European Journal of Social Psychology, 46, 294–307.

13.

Haslam

Jetten

Cruwys

Dingle

G. A.

Haslam

S. A.

(2018). The new psychology of health: Unlocking the social cure. Routledge.

14.

Haslam

S. A.

(2014). Making good theory practical: Five lessons for an applied social identity approach to challenges of organizational, health, and clinical psychology. British Journal of Social Psychology, 53(1), 1–20.

15.

Hayes

A. F.

(2017). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. Guilford.

16.

Heisig

J. P.

Schaeffer

(2019). Why you should always include a random slope for the lower-level variable involved in a cross-level interaction. European Sociological Review, 35(2), 258–279.

17.

Hopkins

Reicher

(2021). Mass gatherings, health, and well-being: From risk mitigation to health promotion. Social Issues and Policy Review, 15(1), 114–145.

18.

Hopkins

Reicher

Stevenson

Pandey

Shankar

Tewari

(2019). Social relations in crowds: Recognition, validation and solidarity. European Journal of Social Psychology, 49(6), 1283–1297.

19.

Hult Khazaie

Khan

S. S.

(2020). Shared social identification in mass gatherings lowers health risk perceptions via lowered disgust. British Journal of Social Psychology, 59(4), 839–856.

20.

The JAMOVI Project. (2021) JAMOVI (Version 1.6) [Computer software]. https://www.jamovi.org

21.

Jetten

Haslam

S. A.

(2012). The social cure: Identity, health and well-being. Psychology Press.

22.

Khan

S. S.

Hopkins

Reicher

Tewari

Srinivasan

Stevenson

(2015). Shared identity predicts enhanced health at a mass gathering. Group Processes & Intergroup Relations, 18(4), 504–522.

23.

Korkmaz

N. F.

(2021, September 5). Se 15.000 mennesker fejre, at livemusikken for alvor er tilbage [Watch 15,000 people celebrate that live music is back for real]. DR. https://www.dr.dk/musik/festival/se-15000-mennesker-fejre-livemusikken-alvor-er-tilbage-det-er-meget-mere-roskilde

24.

Le Bon

(1995). The crowd: A study of the popular mind. London: Transaction. (Originally published: Psychologie des foules, 1895).

25.

Lindholt

M. F.

Jørgensen

Bor

Petersen

M. B.

(2021). Public acceptance of COVID-19 vaccines: Cross-national evidence on levels and individual-level predictors using observational data. BMJ Open, 11(6), Article e048172.

26.

Liu

Thomas

J. M.

Higgs

(2019). The relationship between social identity, descriptive social norms and eating intentions and behaviors. Journal of Experimental Social Psychology, 82, 217–230.

27.

Mathieu

J. E.

Aguinis

Culpepper

S. A.

Chen

(2012). Understanding and estimating the power to detect cross-level interaction effects in multilevel modeling. Journal of Applied Psychology, 97, 951–966.

28.

Morton

T. A.

Power

S. A.

(2022). Coming together after standing apart: What predicts felt safety in the post-coronavirus crowd? Social Science & Medicine, 293, Article 114649.

29.

Novelli

Drury

Reicher

(2010). Come together: Two studies concerning the impact of group relations on personal space. British Journal of Social Psychology, 49(2), 223–236.

30.

Novelli

Drury

Reicher

Stott

(2013). Crowdedness mediates the effect of social identification on positive emotion in a crowd: A survey of two crowd events. PLOS ONE, 8(11), Article e78983.

31.

Oyserman

Smith

G. C.

Elmore

(2014). Identity-based motivation: Implications for health and health disparities. Journal of Social Issues, 70(2), 206–225.

32.

Pandey

Stevenson

Shankar

Hopkins

N. P.

Reicher

S. D.

(2014). Cold comfort at the Magh Mela: Social identity processes and physical hardship. British Journal of Social Psychology, 53(4), 675–690.

33.

Petersen

M. B.

Bor

Jørgensen

Lindholt

M. F.

(2021). Transparent communication about negative features of COVID-19 vaccines decreases acceptance but increases trust. Proceedings of the National Academy of Sciences, 118(29), Article e2024597118.

34.

Postmes

Spears

(1998). Deindividuation and antinormative behavior: A meta-analysis. Psychological Bulletin, 123(3), 238–259.

35.

Power

S. A.

Schaeffer

Heisig

J. P.

Udsen

Morton

(2023). Why trust? A mixed-method investigation of the origin and meaning of trust during the Covid-19 lockdown in Denmark. British Journal of Social Psychology. Advance online publication. https://doi.org/10.1111/bjso.12637

36.

Power

S. A.

Velez

Qadafi

Tennant

(2018). The SAGE model of social psychological research. Perspectives on Psychological Science, 13(3), 359–372.

37.

Rathbone

J. A.

Stevens

Cruwys

Ferris

L. J.

(2022). COVID-safe behaviour before, during and after a youth mass gathering event: A longitudinal cohort study. BMJ Open, 12(7), Article e058239.

38.

Reicher

(1996). The crowd century: Reconciling practical success with theoretical failure. British Journal of Social Psychology, 35, 535–553.

39.

Reicher

Bauld

(2021). From the “fragile rationalist” to ‘collective resilience’: What human psychology has taught us about the COVID-19 pandemic and what the COVID-19 pandemic has taught us about human psychology. Journal of the Royal College of Physicians of Edinburgh, 51(Suppl.1), 12–19.

40.

Reicher

S. D.

Templeton

Neville

Ferrari

Drury

(2016). Core disgust is attenuated by ingroup relations. Proceedings of the National Academy of Sciences, 113(10), 2631–2635.

41.

Shankar

Stevenson

Pandey

Tewari

Hopkins

N. P.

Reicher

S. D.

(2013). A calming cacophony: Social identity can shape the experience of loud noise. Journal of Environmental Psychology, 36, 87–95.

42.

Smith

J. R.

Hogg

M. A.

Martin

Terry

D. J.

(2007). Uncertainty and the influence of group norms in the attitude–behaviour relationship. British Journal of Social Psychology, 46(4), 769–792.

43.

Smith

Templeton

(2022). Social identity processes associated with perceived risk at pilot sporting events during COVID-19. British Journal of Social Psychology, 61, 1376–1399.

44.

Stott

Radburn

(2020). Understanding crowd conflict: Social context, psychology and policing. Current Opinion in Psychology, 35, 76–80.

45.

Tajfel

Turner

J. C.

(1979). An integrative theory of intergroup conflict. In Austin

W. G.

Worchel

(Eds.), The social psychology of intergroup relations (pp. 33–47). Brooks-Cole.

46.

Tarrant

Butler

(2011). Effects of self-categorization on orientation towards health. British Journal of Social Psychology, 50(1), 121–139.

47.

Turner

J. C.

Hogg

M. A.

Oakes

P. J.

Reicher

S. D.

Wetherell

M. S.

(1987). Rediscovering the social group. Blackwell.

48.

Van Bavel

J. J.

Cichocka

Capraro

Sjåstad

Nezlek

J. B.

Pavlović

. . . Jørgensen

F. J.

(2022). National identity predicts public health support during a global pandemic. Nature Communications, 13(1), 1–14.

49.

Vignoles

V. L.

Jaser

Taylor

Ntontis

(2021). Harnessing shared identities to mobilize resilient responses to the COVID-19 pandemic. Political Psychology, 42(5), 817–826.

50.

Warren

Power

S. A.

(2015). It’s Contagious: Rethinking a Metaphor Dialogically. Culture & Psychology, 21(3), 359–379.