Community cohesion during the first peak of the COVID-19 pandemic: A social antidote to health anxiety and stress.

Social support gained through community ties has been pivotal in dealing with stressful events. A cross-sectional community sample (N = 2,329) was gathered to assess community cohesion buffering against heath anxiety and perceived stress during the first peak of the pandemic in the UK, using structural equation modeling analyses. Community cohesion acted as a protective mechanism against both health anxiety and stress during the first national lockdown. A strong positive association was also found between health anxiety and stress. Stress and health anxiety scores peaked in the first weeks of the imposed quarantine; as the lockdown was extended, participants reported lower stress, health anxiety and community cohesion scores. The reduction of community cohesion scores was greater for those younger than 45 while the positive association between stress and health anxiety was stronger among males during the lockdown. While community cohesion effects against health anxiety were enhanced for females, community's buffering against stress were greater for males. Strengthening citizens' psychological sense of community through the publicization and support of local initiatives and mutual-aid groups and utilizing methodically green (and blue) spaces to boost neighborhood attraction might be viable strategies within which stress and health anxiety can be suppressed. Conversely, allowing community, regional and national cracks to deepen can exacerbate the impact of stressful events experienced during the COVID-19 pandemic.

INTRODUCTION

Perceptions of social support and loneliness have been pivotal in predicting depression severity. Those who perceive receiving poorer social support or experiencing greater loneliness at baseline, report more severe symptoms in follow‐ups (Wang et al., 2018). Similarly, social support is predictive of greater symptom‐remission (Jakubovski & Bloch, 2016), better quality of life (Shrestha et al., 2015) and better protection against stressful life‐events (Kingsbury et al., 2020). The above effects are partially explained by the buffering model (Cohen & Wills, 1985) proposing that social support improves wellness on a dual basis: first, through positive experiences attained by social integration in large networks where individuals adopt socially rewarding roles; second, by protecting individuals through its “buffering” against the pathogenic impact of stressful events.

Social support gained through the community has been important in disaster recovery, with individuals living in cohesive communities reporting improvements in their long‐term recovery (Bergstrand & Mayer, 2020). Importantly, community recovery is not solely driven by physical exchanges of goods and resources but also by perceptions of social support reciprocity among community members and the reassurance that one can achieve, knowing that such support will be available if needed (Bergstrand & Mayer, 2020). Cohesion aspects holding the community together are seen by disaster experts as essential to cope with community needs (Bergstrand & Mayer, 2020). Beyond its effects on individuals, community cohesion can be seen as a marker of community's ability to deal with and recover from stressful adversities (APA Health Center, 2004) and as a catalyst for community resilience supporting recovery (Ludin et al., 2019) as seen in communities exposed to natural disasters (Townshend et al., 2015). Recent research suggests that low levels of social support during the COVID‐19 pandemic were predictive of high‐stress and anxiety while social support improved sleep quality in individuals self‐isolating at home (Xiao et al., 2020). Similarly, social isolation induced through the lockdown measures has been linked to temporary disruption in socio‐cognitive ability tasks assessing emotional recognition (Bland et al., 2020). While social support's protective role against stress and anxiety has been recently revisited, less is known about the interplay between social support gained through community bonds and anxiety about health during a pandemic. A groupthink‐driven assertion may suggest that members of a community with tight social links may ultimately engage in decision‐making that does not fully consider the wider costs of group behavior in their attempt to keep the group's interests and goals intact. Examples of this may involve religious sects holding services and anti‐quarantine protests organized at the peak of the first wave (Forsyth, 2020). In such circumstances, strong community cohesion may drive health anxiety down for large groups of individuals. Alternatively, members of a community with a prototypically cautious social identity, involving full adherence to health protocols, may have encouraged other members of the community to also be cautious to fit into the group, ultimately ostracizing individuals engaging in (what they might see as) risky behavior (Tajfel et al., 1979). Such processes galvanize community responses to the pandemic and may act as a shield against anxiety about one's health.

Dealing, at a community level, with a threat such as COVID‐19 strengthen the affective ties between its members as the virus (or in some cases, the government itself imposing the restrictions) is seen as the outside enemy against which community members must be united, according to the source model of group threat (Greenaway & Cruwys, 2019). Importantly, though, the fabric keeping the community together may wear down after the initial phase of the pandemic. The Federal Emergency Management Agency (FEMA) proposes that a heroic community response manifests itself during the impact phase, followed by a “honeymoon” phase where ties between community members strengthen. Eventually, a phase of disillusionment occurs before the community is reconstructed (Townshend et al., 2015). With the FEMA predictions in mind, it was expected that community cohesion might have worn down as the lockdown‐phase of the pandemic was extended and the “honeymoon” phase neared its end.

COVID‐19 and mental‐wellness

Early empirical research originating from China has shown a decrease in well‐being among those close to the epicenters of the outbreak (Yang & Ma, 2020). Younger individuals have been affected the most, partially due to the extreme disruption of their daily routines, caused by school closures or stressors involving exam and matriculation arrangements and online education support (Cao et al., 2020; Chen et al., 2020; Wang et al., 2020a), as also shown in UAE (Saddik et al., 2020) and Spain (Ozamiz‐Etxebarria et al., 2020). Student status or young age and female gender have been associated with higher levels of stress, anxiety, or depression during the COVID‐19 pandemic (Chao et al., 2020; Qiu et al., 2020; Wang et al., 2020b). As the body of COVID‐19 related research grew out of China, the psychological vulnerability of women and young individuals was further emphasized by research involving health workers in Italy (Rossi et al., 2020), Jordanian students (Sallam et al., 2020) and the general population in Portugal (Moreira et al., 2020), Spain (González‐Sanguino et al., 2020), UAE (Saddik et al., 2020) and India (Saikarthik et al., 2020). Still, as the cases of COVID‐19 increased sharply in China, there were no corresponding clinical increases in anxiety, stress and depression. Tentative evidence supporting the downward trend in psychological impact has been presented in UK‐based‐research (Fancourt et al., 2020), although it has been highlighted that most mental health problems are still more prevalent in comparison with the pre‐pandemic population average (ONS, 2020a). Still, there is enough evidence to suggest that while the initial impact of the pandemic and the lockdown measures deteriorated stress and anxiety in particular, the direct effects may have reached a plateau, albeit stabilizing above the pre‐pandemic levels, partially due to the introduction of public health measures (Wang et al., 2020a).

Anxiety and stress in the UK during the pandemic

Anxiety levels have soared between the end of 2019 and March 2020, especially among those experiencing a reduction of finances. Women in the UK reported higher anxiety scores than men and those older than 70 years were less anxious than younger respondents, although the latter gap has shortened during the lockdown (ONS, 2020a). Strong cross‐sectional evidence suggested that higher COVID‐19 anxiety was associated with increased somatic symptoms (Shevlin et al., 2020) with young and female populations experiencing more stress and anxiety during the first lockdown (Daly et al., 2020; Li et al., 2020; Nikčević et al., 2021).

Capturing stress during the pandemic is essential as high levels of stress can cause insulin resistance and obesity (Räikkönen et al., 1996) and moderately increase the risks of coronary heart disease (Richardson et al., 2012), all of whom raise risks of dying from coronavirus (ONS, 2020b). High stress increases susceptibility to five acute infectious respiratory diseases, including the coronavirus type 229E (Cohen et al., 1991, 1993), emphasizing the need to capture its magnitude in the community during the pandemic.

In the context of the pandemic, increased health anxiety is of concern, as it can be diagnostic of hypochondriasis (Alberts et al., 2013) leading individuals to employ catastrophizing cognitive styles to interpret body‐symptoms and sensations, ultimately inducing feelings of helplessness and inability to deal with stressful events (Rief et al., 1998), or even reduce engagement with health‐seeking and precautionary behaviors (Kellner et al., 1987; Lecci et al., 1996). Indeed, catastrophizing over one's health by misinterpreting minor bodily symptoms might lead to maladaptive behaviors (e.g., increased engagement with health services or reluctance to seek medical help to “self‐protect” against the risk of contagion) that can be detrimental as seen in previous pandemics. On the other hand, very low health anxiety may signal reluctance to seek medical advice (Asmundson & Taylor, 2020). Eventually, perceiving oneself as vulnerable toward a disease positively shapes attitudes toward key protective practices (Gilles et al., 2011).

Aims and hypotheses

The aim of this study was to assess the association between heath anxiety, community cohesion, and perceived stress within a large UK community sample during the first peak of the pandemic. It was hypothesized that community cohesion will be a protective factor reducing both stress and health anxiety while health anxiety will positively predict perceived stress. It was also proposed that as the lockdown was extended, a downward trend will be observed in stress, health anxiety, and perceptions of community cohesion.

METHOD

Data collection began just over two weeks after the World Health Organisation declared the COVID‐ 19 outbreak a pandemic and during the first week of the imposed lockdown, announced in March 23. During the enforcement of the lockdown, people were required to “stay home”, non‐essential shops and community shops were closed, and public gatherings (of more than 2 people) were prohibited (some exceptions to the above applied). Data collection was conducted in the midst of the first peak of the COVID‐19 pandemic in the UK as shown in Supplementary Figure S1. By early May 2020, the UK recorded the highest death rate in Europe and the second highest in the world (Armour et al., 2020), hence the data garnered is of noteworthy scientific and historical value.

Sample and data collection procedures

Data were collected from United Kingdom (online) community cohorts using an internet‐based self‐report survey delivered by Qualtrics. The data collection protocol and the procedure received institutional ethical approval and adhered to the Ethics Guidelines for Internet‐mediated Research (BPS, 2017). Participation was solicited by distributing the survey link in local and national social media pages (e.g., Whats on in Ramsbottom; Whats on in Rammy; Whats on in Bury; Whats on in Whitefield and Prestwich; Whats on in Tottington, Greenmount and Walshaw; 48% Preppers for families; Wirral COVID‐19 Mutual Aid Support Group; 48% preppers; SE6 SE13 Lewisham Catford Community; Camden Covid‐19 Mutual Aid; Britain for All; London advertisers group; Business connections UK, etc.) and in several student community pages. Participants received no rewards for taking part in the study. There were no incentives for participation other than contributing to a study that captures mental health and community‐cohesion indicators of the UK public during the first peak of the pandemic. Participants spent, on average, 17 min filling in the survey.

Although 2,945 respondents started filling the online survey, 416 did not complete it, and hence were removed from the final data set. Another 150 stated they were living abroad, however the country of residence was not recorded hence data were removed to allow a more selective understanding of the UK sample. The final sample comprised of 2,329 adults, reportedly living in the UK. Participants were aged 18–87 years (M = 48.08 years, SD = 13.39). Of those, 82.4% of participants (N = 1,921) were female. Most (87.3%) were White British, with the remainder clustered as “Any Other White Background” (6.7%), Irish (1.9%) and mixed/multiple ethnic background (1.1%). The remaining 3% consisted of respondents of various Asian, African and Caribbean backgrounds. A regional breakdown and an overview of the occupational characteristics of the sample are respectively shown in Tables 1 and 2.

TABLE 1

Means, Standard Deviations (SD) and cut‐off ratios per UK region

Region	BIC	SHAI‐18	UK sample percentages scoring above the (pooled) SHAI−18 means extracted from (i) 4 pandemic samples*, (ii) non‐clinical and (iii) clinical samples (Alberts et al., 2013)			PSS‐10	UK sample percentages scoring above the normative PSS‐10 means extracted from 2009 eNation community Survey (N = 2000) (Cohen and Janicki‐Deverts, 2013)
			(i) Pandemic pseudo cut‐off point: values ≥14.48	(ii) Non‐clinical pseudo cut‐off point: values ≥12.42	(iii) Clinical pseudo cut‐off point: values ≥22.95		Pseudo cut‐off point for males: values ≥15.53	Pseudo cut‐off point for females: values ≥16.15
North West (n = 872)	59.65 (12.71)	14.73 (8.01)	43.90%	54.00%	15.70%	18.58 (7.72)	49.50%	61.10%
South East (n = 275)	57.23 (13.02)	12.97 (6.33)	31.20%	42.80%	8.70%	17.01 (7.15)	42.90%	55.90%
South West (n = 208)	59.14 (12.91)	13.16 (6.57)	34.10%	48.10%	7.70%	16.11 (7.39)	44.00%	48.10%
Greater London (n = 195)	56.59 (12.92)	14.61 (7.24)	44.60%	55.40%	11.80%	17.41 (7.55)	42.90%	65.40%
Scotland (n = 149)	58.68 (13.45)	13.70 (7.47)	38.90%	49.30%	10.07%	16.28 (7.62)	41.90%	54.20%
East of England (n = 122)	59.77 (12.63)	13.43 (6.71)	40.60%	47.50%	11.50%	15.90 (7.32)	32.40%	46.60%
West Midlands (n = 117)	56.17 (13.65)	13.85 (7.24)	42.60%	47.90%	11.10%	18.06 (7.28)	57.70%	61.50%
Yorkshire and the Hamber (n = 114)	55.25 (14.39)	13.71 (6.74)	37.70%	47.40%	8.80%	17.84 (7.40)	44.00%	57.40%
East Midlands (n = 101)	57.41 (14.55)	13.43 (6.42)	34.40%	48.50%	10.90%	17.50 (4.81)	35.50%	58.30%
Wales (n = 92)	59.60 (12.85)	13.86 (6.31)	40.02%	52.20%	13.00%	17.70 (7.77)	16.70%	60.00%
North East (n = 64)	55.22 (11.24)	16.64 (7.91)	59.40%	68.70%	21.90%	18.23 (7.73)	57.10%	64.00%
Northern Ireland (n = 20).	54.85 (14.24)	12.10 (5.41)	40.00%	50.00%	0.00%	15.85 (7.73)	0%	50.00%
United Kingdom mean (N = 2,329)	58.36 (13.08)	14.08 (7.32)	40.50%	51.00%	12.40%	17.61 (7.55)	43.60%	58.30%

Pooled means were computed for the following samples:

(i) Tull et al. (2020), US‐based, N = 500, data collected between March 27 (105.584 cases diagnosed & 2,129 COVID‐19 deaths recorded) ‐April 5 (345.345 cases diagnosed & 11.953 COVID‐19 deaths recorded). 1–4 Likert scales were used so Total scores needed to be converted as all remaining studies scored SHAI on a 0–3 scale. To do this, scores were converted by subtracting the number of questions (i.e., 18) from the original total mean score reported, as advised in Alberts et al. (2013) (SHAIMean = 14.29 SD = not available).

(ii) Jungmann and Witthöft (2020), N = 1616, Germany‐based, data collected between March 15 (5,813 cases diagnosed & March 13 COVID‐19 deaths recorded) −22 (24.813 cases diagnosed & 94 COVID‐19 deaths recorded) (SHAIMean = 14.68 SD = 6.58).

(iii) Mertens et al. (2020), N = 437, data collected between 14–17 March, International sample, 47.5% of those living in the Netherlands (SHAIMean = 13.42 SD = 6.24). No available data on cases and deaths recorded as the sample used was international.

(iv) Özdin and Bayrak Özdin (2020), Turkey‐based, data collected between April 14 (65.111 cases diagnosed & 1,403 COVID‐19 deaths recorded) ‐April 16 (74.193 cases diagnosed & 1,643 COVID‐19 deaths recorded) (SHAIMean = 15.10 SD = 7.00).

Abbreviations: BIC, Buckner's index of cohesion; PSS‐10, the perceived stress scale; SHAI‐18, The short health anxiety inventory.

TABLE 2

Means, Standard Deviations (SD) as a function of occupation and cut‐off ratios per occupation

Occupations	BIC	SHAI‐18	UK sample percentages scoring above the (pooled) SHAI‐18 means extracted from (i) 4 pandemic samples*, (ii) non‐clinical and (iii) clinical samples (Alberts et al., 2013)			PSS‐10	UK sample percentages scoring above the normative PSS‐10 means extracted from 2009 eNation community Survey (N = 2000) (Cohen and Janicki‐Deverts, 2013)
			(i) Pandemic pseudo cut‐off point: values ≥14.48	(ii) Non‐clinical pseudo cut‐off point: values ≥12.42	(iii) Clinical pseudo cut‐off point: values ≥22.95		Pseudo cut‐off point for males: values ≥15.53	Pseudo cut‐off point for females: values ≥16.15
Managers, directors and senior officials (n = 126)	58.34 (14.52)	12.23 (6.01)	26.20%	34.10%	4.80%	16.59 (6.74)	48.90%	49.10%
Professional occupations (n = 762)	58.29 (12.63)	14.16 (7.21)	33.50%	44.50%	11.80%	17.33 (7.32)	43.90%	51.90%
Associate professional and technical occupations (n = 215)	57.27 (13.24)	12.92 (6.18)	28.40%	39,10%	5.10%	17.26 (7.35)	38.20%	52.50%
Administrative and secretarial occupations (n = 224)	56.71 (13.43)	14.62 (7.24)	40.60%	49.10%	10.30%	18.21 (8.16)	41.20%	52.70%
Skilled trades occupations (n = 73)	57.62 (13.41)	13.71 (6.40)	37.00%	43.80%	6.80%	17.63 (7.52)	28.00%	58.30%
Caring, leisure and other service occupations (n = 175)	58.11 (13.34)	15.08 (7.74)	40.60%	52.00%	13.70	19.48 (7.63)	64.70%	60.80%
Sales and customer service occupations (n = 74)	57.77 (13.13)	14.00 (8.05)	33.80%	43.20%	9.50%	18.68 (7.71)	35.70%	65.00%
Process, plant and machine operatives (n = 17)	52.94 (14.35)	13.35 (7.95)	41.20%	41.20%	11.80%	17.06 (7.80)	12,50%	66.70%
Low skilled elementary occupations (n = 75)	60.32 (11.74)	14.47 (9.00)	37.30%	44.00%	13.30%	18.80 (8.40)	50.00%	59.00%
Students (n = 81)	53.07 (11.43)	17.08 (6.91)	58.00%	66.70%	13.60%	22.56 (7.07)	66.70%	76.00%
Retired (n = 369)	62.45 (12.14)	12.24 (6.45)	27.00%	36.40%	5.40%	14.48 (7.20)	26.80%	36.70%
Not working (n = 56).	53.20 (14.16)	17.96 (9.20)	55.40%	64.30%	26.80%	21.80 (8.05)	53.80%	76.70%
Home duties (n = 83)	57.87 (13.95)	18.50 (8.62)	59.00%	68.70%	24.10%	21.59 (7.11)	33.30%	76.20%

Measures

The survey comprised of measures capturing demographics (age, gender, ethnic background, occupation), health anxiety, perceived stress and community cohesion. The days living under lockdown were captured by measuring the number of days between when the “stay home” message was introduced and the survey‐completion date, the latter being automatically garnered in Qualtrics.

Buckner's (1988) index of cohesion

The Buckner's index of cohesion (BIC) survey was used to capture participants' overall perceptions of community and neighborhood cohesion within the areas they lived. The BIC is an 18 item self‐report instrument that measures the psychological aspects of community cohesion, neighbor attraction and neighboring. The 18 items were scored on a 5‐point Likert scale, ranging from 0 = strongly disagree, 1 = disagree, 2 = neither agree or disagree, 3 = agree, 4 = strongly agree. Total scores range from 18–90. The BIC has shown high reliability in previous studies (Li et al., 2011; Ludin et al., 2019). To ensure that participants were referring to their overall perception of their community and sense of the neighborhood and not in how the community has become during the pandemic, as a direct result of the lockdown measures imposed by the government (e.g., visiting neighbors and physically engaging with individuals from other households was not advised during the lockdown), small changes to the wording were made. These included the addition of “before Coronavirus” on some questions involving “neighbouring” and /or physical engagements that could have been punishable by a fine during lockdown. For example, item “I rarely have neighbours over to my house to visit” changed to “Before the coronavirus I rarely had neighbours over to my house to visit”; item “I visit neighbours in their homes” changed to “Before Coronavirus I used to visit neighbours in their homes”, etc. Cronbach α was computed after the wording changes took place and was found to be very high (α = .93).

The perceived stress scale

Perceived stress was measured using The perceived stress scale (PSS‐10) (Cohen et al., 1983; Cohen & Williamson, 1988). PSS‐10 is one of the most widely used tools for measuring the perception of stress consisting of a 10‐item scale to measure the degree to which respondents consider situations in their life as stressful. Items were designed to understand how unpredictable, uncontrollable, and overloaded respondents find their lives. Items are scored on a 5‐point Likert scale ranging from 0 = never, 1 = almost never, 2 = sometimes, 3 = fairly often, to 4 = very often. Total scores ranged from 0 to 40, with higher scores indicating higher levels of perceived stress. PSS‐10 has shown high reliability in past research (Mitchell et al., 2008; Roberti et al., 2006) as was in this study (a = .90). No modifications were made to this scale.

The short health anxiety inventory

Health anxiety was measured using the short health anxiety inventory (SHAI), an eighteen‐item self‐report measure which is widely used to assess health anxiety independent of physical health status in medical and non‐medical contexts (Salkovskis et al., 2002). Each item consists of four groups of statements in which participants were asked to select the statement that best described their feelings such as 0 = I do not worry about my health, 1 = I occasionally worry about my health, 2 = I spend much of my time worrying about my health, and 3 = I spend most of my time worrying about my health. Meta‐analytical findings have been overall positive and reliability measures reported by research support its progressively frequent use (Alberts et al., 2013). Reliability of the 18‐items scale was high (a = .90). No modifications were made to this scale.

Data analysis

SPSS and AMOS software were used to process the data. Descriptive analyses of the data were followed by Principal Component analysis of the scales used. Using stringent cut‐off points, as explained below, led to a trimming of the total items used per scale in subsequent analysis. To confirm the component structure of the scales and model the data, Structural Equation Modelling (SEM) was used to test the hypothesized and alternative models. In addition to the above, the hypothesized model was tested four more times selectively using four different segments of the sample (comprising of males/females only; younger/older respondents).

RESULTS

Sample characterization

Following the reversing of scores for negatively framed items, mean scores, per UK region and occupational status were computed (Tables 1 and 2). The percentage of populations tested that scored above the tentative cut‐off points set for non‐clinical and clinical samples are also reported for health anxiety using the pooled means proposed by Alberts et al.'s (2013) meta‐analysis. Clear cut‐off points do not exist for community populations (if at all) for the SHAI so using the pooled means reported in the most comprehensive meta‐analysis, involving non‐clinical and clinical samples, was deemed appropriate. To get a sense of how the UK SHAI scores compare to those of other countries during the pandemic, pooled means of the four available studies reporting total scores for the (18‐item) SHAI (Jungmann & Witthöft, 2020; Mertens et al., 2020; Özdin & Bayrak Özdin, 2020; Tull et al., 2020) were computed and used as tentative cut‐off points.

Similarly, perceived stress thresholds were set using the normative values reported in Cohen and Janicki‐Deverts (2012) to ascertain the percentage of respondents who scored above the normative stress‐thresholds. Data were gathered in the aftermath of the 2008 economic crisis, marginally replicating the financial uncertainties experienced by the participants in this study. The authors reported separate normative values for males and females so we correspondingly broke‐down the PSS‐10 scores by gender.

Principal component analysis of the measures used

Log transformation of all data was performed before structural modeling of the data is initiated. List‐wise deletion of 8 cases took place to deal with missing values in respondents' responses and allow performing bootstrapping, hence the sample used for the final analysis included 2,321 respondents. Three PCA were performed, all with oblique (Oblimin) rotation due to high interrelation between construct factors. Kaiser's rule of maintaining Eigenvalues P1.0 was adopted for all analyses. Kaiser‐Meyer‐Olkin measure of sampling adequacy was 0.94 for health anxiety and community cohesion and 0.92 for stress with values between 0.5 and 0.7 being acceptable, 0.7+ being good to excellent (Hutcheson & Sofroniou, 1999). To ensure there were sufficient correlations between items to conduct the PCA, Bartlett's test of sphericity was also performed and found significant at the p < .0001 level for all three analyses. Using the stringent cut‐off points proposed by Guadagnoli and Velicer (1988), we retained items whose loading was equal or larger than 0.6.

Using both scree plot and Eigenvalues >1 to determine the underlying community‐cohesion components, yielded two factors: factor 1 consisted of 7 items capturing psychological sense of community and factor 2 consisted of 4 items capturing neighborhood attraction. The two component structure revealed differed from the three‐construct structure proposed elsewhere (Li et al., 2011; Ludin et al., 2019); specifically, while items related to neighborhood attraction and psychological sense of community remained relatively unaffected, items originally linked to neighboring were loaded into the second factor (i.e., sense of community) in the current study. Moreover, since the loading thresholds set were high to allow robust modeling, some items originally used to describe the psychological sense of community felt by respondents were completely removed as they did not load sufficiently to any factor. This finding highlights that the factor‐structure and items loaded in each factor are context and sample‐dependent for BIC as also seen in studies using multiple samples (Li et al., 2011). So, bearing in mind that this is, to the knowledge of the authors, the largest sample used to test component structure of the BIC scale in the UK and the first to do so during the pandemic, such component fluctuations should not be surprising.

Following the same procedure, a second PCA yielded 2 stress factors: factor 1 consisted of 6 items capturing helplessness and factor 2 consisted of 3 items capturing self‐efficacy. The component structure resembled closely the component structure reported elsewhere (Schneider et al., 2020).

A third PCA yielded 3 health anxiety factors: factor 1 was represented by 6 items relating to illness likelihood, factor 2 was represented by 4 items relating to illness severity and factor 3 was represented by 3 items relating to body vigilance. Further analyses displayed an acceptable to high internal consistency for illness severity and illness likelihood (Cronbach's Alpha > .70) but fair internal consistency for body vigilance (Cronbach's Alpha = .64) hence the latter subscale was not retained in structural modeling. The two‐factor component structure extracted resembled the two‐factor structure reported in Wheaton et al. (2010); however, while the four items loaded in the illness severity subscale were identical in both studies, Wheaton et al. (2010) showed that 13 items were loaded in the illness likelihood factor while 6 were loaded in this study. This could be partially attributed to the different samples (i.e., Wheaton, et al, similarly to most studies administering SHAI in non‐clinical samples, used a student sample consisting of 636 individuals) and factor thresholds used (i.e., Wheaton et al. accepted loadings ≥ 0.4 while we accepted loadings ≥ 0.6). Reliability measures for the final sets of items comprising each construct and confirmatory factor loadings for all items used in subsequent analysis are shown in online Table S1.

Model description and associations between variables

Perceived stress was the dependent variable for the hypothesized model. Initially, SEM analysis assessed the direct effects of health anxiety, community cohesion, and days living under lockdown on perceived stress. The model also assessed whether the latter two predictors have indirect effects on perceived stress. Health anxiety, community cohesion and perceived stress were operationalized as second order latent variables (see Figures 1 and 2). Unstandardized regression coefficients are reported in text to explain the associations between the model variables. Unstandardized positive and negative regression coefficients, respectively, indicated the increase and decrease in stress scores per unit of change in community cohesion, health anxiety and days living under lockdown. Bias corrected bootstrapping was also operated to extract 95% confidence intervals (CI) and significance p‐values for all unstandardized coefficients and bias‐corrected CI for the total and indirect effects found in the model. Two error covariances were drawn between the (only) two negatively framed items from the community cohesion scale and their positively framed equivalents, as their wording was almost identical leading to high error covariance (i.e., Remain in the area for a number of years/ Given the opportunity, I would move out of the area; Before the coronavirus, I used to visit neighbors in their homes/ Before the coronavirus, I rarely had neighbors over to my house to visit). Model fit indices for the main model are reported before and after covariances were drawn.

FIGURE 1

Standardized path coefficients among constructs: A, neighbourhood attraction; H, helplessness; IL, illness likelihood; IS, illness severity; PSYC, psychological sense of community; SE, self‐efficacy; **p < .001

FIGURE 2

Standardized path coefficients among constructs for the four subsamples: A, neighbourhood attraction; H, helplessness; IL, illness likelihood; IS, illness severity; PSYC, psychological sense of community; SE, self‐efficacy; *p < .01; **p < .001; ns = non‐significant; ↑/↓: path coefficient larger/lower than the equivalent reported in the full hypothesized model displayed in Figure 1

Structural equation model fit

Absolute (χ 2/df [1,755.214/763] = 4.169; RMSEA = 0.037; SRMR = 0.036) and incremental (NFI = 0.942; TLI = 0.951; CFI = 0.955) fit indices overall showed a good model fit. A good to acceptable model fit was also found before the two error covariances were drawn: (χ 2/df [2,474.587/423] = 5.850; RMSEA = 0.046; SRMR = 0.039; NFI = 0.919; TLI = 0.925; CFI = 0.932). Using unstandardized beta values, indicated direct negative association between days living under lockdown and community cohesion (B = −0.04, p < .01, 95% CI: −0.07 to −0.01); health anxiety (B = −0.04, p < .001, 95% CI: −0.05 to −0.02) and stress (B = −0.06, p < .001, 95% CI: −0.08 to −0.02). Community cohesion was directly and negatively associated with stress (B = −0.11, p < .001, 95% CI: −0.17 to −0.06) and health anxiety (B = −0.18, p < .001, 95% CI: −0.22 to −0.13). Health anxiety was the sole factor to be positively associated with stress (B = 1.07, p < .001, 95% CI: 0.89 to 1.21). Community cohesion indirectly suppressed stress through its effects on health anxiety (B = −0.179, p < .001, 95% CI: −0.239 to −0.128). Days living under lockdown did indirectly and negatively reduced stress (B = −0.03, p < .001, 95% CI: −0.04 to −0.01) but indirectly increased health anxiety (B = −0.01, p < .001, 95% CI: 0.001 to 0.01; indirect and total effects for all models are shown in Table 3).

TABLE 3

Total and indirect effects for structural equation models

Sample assessed	Full‐sample (N = 2,321)				≤45 years old (n = 1,370)				>45 years old (n = 951)				Females only (n = 1915)				Males only (n = 406)
	95% CI				95% CI				95% CI				95% CI				95% CI
	B	β	LL	UL	B	β	LL	UL	B	β	LL	UL	B	β	LL	UL	B	β	LL	UL
Total effects
Days living under lockdown → Stress	−0.086	−0.165**	−0.208	−0.128	−0.092	−0.184**	−0.236	−0.130	−0.053	−0.107**	−0.170	−0.041	−0.053	−0.133**	−0.178	−0.085	−0.053	−0.082	−0.175	0.130
Days living under lockdown → Health anxiety	0.007	−0.085**	−0.131	−0.033	−0.042	−0.115**	−0.193	−0.048	−0.009	−0.028	−0.106	0.058	−0.009	−0.059	−0.118	−0.009	−0.002	−0.007	−0.141	0.029
Community cohesion → Stress	−0.288	−0.256**	−0.304	−0.206	−0.262	−0.198**	−0.283	−0.113	−2.16	−0.210**	−0.297	−0.090	−0.216	−0.263**	−0.327	−0.202	−0.478	−0.294**	−0.423	−0.140
Indirect effects
Days living under lockdown → Community cohesion, Health anxiety → Stress	−0.025	−0.048**	−0.084	−0.013	−0.036	−0.072**	−0.127	−0.027	−0.005	−0.009	−0.064	0.048	−0.005	−0.030	−0.071	0.004	0.000	0.001	−0.103	0.095
Days living under lockdown → Community cohesion → Health anxiety	−0.007	0.020**	0.005	0.037	0.002	0.005	−0.004	0.025	0.013	0.042**	0.015	0.077	0.013	0.022**	0.009	0.046	0.002	−0.006	−0.012	0.060
Community cohesion → Health anxiety → Stress	−0.179	−0.159**	−0.214	−0.109	−0.135	−0.102**	−0.168	−0.043	−0.164	−0.160**	−0.246	−0.083	−0.164	−0.160**	−0.230	−0.119	−0.240	−0.148**	−0.312	−0.043

Abbreviations: % CI, 95% confidence interval of standardized path coefficient;B, unstandardized path coefficient; UL, upper limit of 95% CI; β, path coefficient LL, lower limit of 95% CI.

p < .001.

In addition to the full hypothesized model displayed in Figure 1, we adhered by Kelloway's (1998) recommendation to access model fit against alternative models. Following contemporary SEM guidance (Zhao et al., 2010), we opted for a nonmediated (direct effects only) alternative model whereby all variables (i.e., community cohesion, days living under lockdown and health anxiety) are positioned as exogenous factors directly effecting stress and no indirect effects are assessed. Furthermore, we also examined an indirect‐effects model whereby only the indirect effects of community cohesion and days living under lockdown, mediated by health anxiety, on stress were assessed but the equivalent direct paths (i.e., direct effects of community cohesion and days living under lockdown on stress) were not captured. Although the nonmediated and indirect effects models also displayed good fit, all six indices of model fit were superior in the hypothesized model. While some of fluctuations in model fit indices might be due to changes in model‐complexity and equivalent reduction in paths examined, the hypothesized model assessing the direct and indirect effects of the tested factors on stress fitted better our data (see Table 4 for an overview of fit indices).

TABLE 4

Fit indices for all models

Model	χ 2/df	SRMR	NFI	TLI	CFI	RMSEA
Full hypothesized model (N = 2,321)	4.169	0.036	0.942	0.951	0.955	0.036
Indirect effects model (N = 2,321)	4.240	0.037	0.941	0.950	0.954	0.037
Nonmediated Model (N = 2,321)	4.326	0.048	0.940	0.948	0.953	0.038
Males only (n = 406)	1.597	0.051	0.867	0.940	0.945	0.038
Females only (n = 1,915)	3.725	0.035	0.938	0.949	0.954	0.038
≤45 years old (n = 1,370)	2.798	0.037	0.930	0.949	0.954	0.036
>45 years old (n = 951)	2.269	0.040	0.926	0.953	0.957	0.037

Abbreviations: CFI, comparative fit index; NFI, normed fit index; RMSEA, root mean square error of approximation; SRMR, standardized root mean residual; TFI, Tucker‐Lewis fit index.

Four more SEM analyses, using the path‐structure of the hypothesized model, were conducted to test the diagnostic value of the proposed model and capacity to fit four subpopulations: males only, females only, those aged 45 years old or above, and those younger than 45 years old (see Table 4 for model fit indices). The use of 45 years as a threshold to divide the population was devised to split the original data set while matching UK's ONS (2020c) social impact report (data files) using similar age bracketing to highlight a jump in deaths (i.e., deaths in March & April 2020, doubled from 167 in ages 40–44 to 334 in ages 45–49). Standardized direct effects using Beta values, for the four additional models are outlined in Figure 2. Factor loadings for first‐to‐second order latent variables were good with one exception; namely, (standardized) loadings of neighborhood attraction on community cohesion superseded 1 among males and those older than 44 years. This finding mirrors previous applications of the BIC showing factor‐loading fluctuations as a function of the specific population used (Li et al., 2011). In this study, higher loadings do reflect multicollinearity issues, partially occurring due to the high fluctuation of correlations among the first‐order factors further induced by repeated assessment of loadings using four idiosyncratically diverse subpopulations (for more information on the occurrence of standardized Betas above 1, see Deegan, 1978; Jöreskog, 1999).

DISCUSSION

Structural equation modeling showed that community cohesion negatively predicted perceived stress and health anxiety for the entire population but also for the separate groups tested. A strong positive association was found between health anxiety and perceived stress. As the lockdown was extended, self‐reported scores of perceived stress, anxiety and community cohesion were reduced for all groups however the associations (i.e., between days living under lockdown and health anxiety, perceived stress and community cohesion) were not significant for males. Assessing the indirect effects showed that community cohesion functioned as a double stress‐barrier, first through its direct effects on perceived stress but also indirectly, through its direct suppression of health anxiety that induces perceived stress. Although the days living under lockdown directly reduced health anxiety scores, they have indirectly increased health anxiety through lockdown's suppressing effects on community cohesion. While the days living under lockdown coincided with a greater reduction of community cohesion for those younger than 45, the positive association between perceived stress and health anxiety was stronger among males. Community cohesion effects against health anxiety were enhanced for females and community's buffering against perceived stress were greater for males (a trend supported by the standardized total effects). Although all tested models displayed good to acceptable model fit, the full hypothesized model fitted better our data set. Even though the occupational status of the participants was not processed through inferential analysis, it was evident that some groups were put under greater psychological strain, partially due to their age and gender composition. For example, students reported some of the highest stress and health anxiety scores while those retired overall displayed the lowest scores in stress and health anxiety. On the other hand, there were trends observed amongst both males and females in specific professional occupations hinting that the psychological burden experienced may have also been profession specific, possibly reflecting existing inequalities that were further exacerbated during the pandemic. For example, while nearly 14% of those in caring, leisure and other service occupations scored above the clinical cut‐off points for health anxiety, less than 5% of those working as managers, directors and senior officials scored as highly, potentially reflecting the disparities in the available opportunities to work from home and protect one's health during the pandemic. Highly anxious individuals were more likely to experience higher stress levels during the pandemic mirroring the findings extracted during the H1N1 influenza outbreak (Wheaton et al., 2012) and supporting the wide‐ranging effects of health anxiety on health observed in pandemic samples (Jungmann et al., 2020; Nikčević et al., 2021). Highly stressed individuals have been struck by a double stressor also experiencing greater anxiety during a period where there has been an abundance of (other) psychological challenges such as worrying for essentials including availability of groceries and lack of ability to make long‐term life plans (ONS, 2020d). While low SHAI scores were indicative of low perceived stress scores, arguably a positive health outcome during a pandemic, they might have also constituted markers of maladaptive behavioral reactions that can negatively affect public health safety in the long term. For example, lower levels of health anxiety induced carelessness during the SARS pandemic and limited precautionary behaviors that could have reduced its spread. Indeed, perceived susceptibility to SARS, avian flu, swine flu, or pandemic influenza have previously shaped behaviors with those feeling unsusceptible being less likely to engage with precautionary measures (Bish & Michie, 2010).

The psychologically protective role of community cohesion was empirically supported, for the first time, during the COVID‐19 pandemic. Community's protective role against adversity may be partially explained through community activities set up to help those in need by mutual‐aid groups (Mutual Aid UK, 2020). In line with Kropotkin's (1902) socio‐biological, quasi‐Darwinian assertions, community members may naturally reassert their mutual obligations during a crisis and suppress individualism to embrace collectivist values and reciprocity during the (early days of the) pandemic (Springer, 2020). Still, a negative association was found between community cohesion and days lived under lockdown, perhaps hinting that the buffering of community cohesion might not be long‐lasting should the lockdowns and/or the extensive disruptions continue. It is notable, in that respect, that community (and regional) divisions have been reported in the UK in recent months following the lockdown easing (Duffy & Allington, 2020; ONS, 2020e) hinting that community cracks may have already appeared. According to our models, as community cohesion weakens, both perceived stress and health anxiety will increase, ultimately making the population more susceptible to infection and mental health difficulties. So, although the beneficial role of social support gained through strong community bonds and, conversely, the damaging effects of social isolation on stress and anxiety are well supported in the literature (Cohen & Wills, 1985; Kingsbury et al., 2020; Wang et al., 2018) the soothing effects of community cohesion on health anxiety are harder to fully understand. While it is assumed that community cohesion “protects” against health anxiety, one might claim that strong community bonds can equally induce poor adherence to the recommended health protocols and carelessness in communities where the prototypical attitudes and behaviors are antithetical to the health‐measures imposed. So, even though the buffering effects of community cohesion are assessed positively within the context of this study, it is important to further understand the exact mechanisms suppressing health anxiety. For example, it is likely that the lower health anxiety scores in cohesive communities are due to a harmonizing implementation of the proposed health protocols by community members that ultimately reduce worries about health. Alternatively, it is equally likely that a collective underestimation of health‐risks by groups or a shared perception that the virus is harmless can lead some cohesive and similar‐minded communities to suppress health‐related concerns. Crucially, since both processes can occur simultaneously, the tentative community cohesion reductions, observed during lockdown, might partially reflect the conflicting but salient social identities emerging in response to the pandemic.

It is noteworthy that the negative association between days living under lockdown and community cohesion was enhanced among those under 45 years of age. Coupled with the very high perceived stress scores among students and young individuals, it is highly likely that the younger population has been disproportionately affected by the disruptive lockdown measures as their routine has been most severely affected. Furthermore, younger individuals have spent their time during the lockdown in accommodations that possessed less (interior and exterior) space, were damper and situated in crowded neighborhoods (Judge & Rahman, 2020) and are generally less likely to own a home (Gov.UK, 2020). Neighborhood conditions may have accordingly challenged younger individuals to get support from others (Cramm & Nieboer, 2015). Renting properties (vs. owning) may have further suppressed neighborhood attachment and belongingness (Stone & Hulse, 2007) ultimately inducing helplessness and low self‐efficacy. Yet, caution must be exercised in interpreting community cohesion, perceived stress and health anxiety reductions during lockdown in our sample. The items capturing community cohesion encompassed global notions of psychological belongingness and community attraction that are unlikely to have changed dramatically within a short time. Similarly, four items in the psychological sense of community subscale captured neighboring activities that inevitably took place prior to the lockdown, as social engagements were not allowed during data collection. Likewise, SHAI‐18 and PSS‐10 garnered, by design, information about how respondents have felt during a period that includes but is not limited to the date of data collection, also covering a period that precedes it. It is hence proposed that while participants' perceptions of community cohesion, stress and health anxiety were instantly captured, reflecting how respondents felt at the time, dynamic changes should not be fully assumed on the sole basis of cross‐sectional evidence. While recent and strong longitudinal evidence supports the declining trends of community bonds in England (Borkowska & Laurence, 2020) the assessments entailed pre‐pandemic (and pre‐Brexit‐referendum) to pandemic comparisons, hence it is advised that longitudinal evidence is extracted during the pandemic to understand better the proposed decline of community cohesion during the lockdown. Still, one must observe that preliminary and recent cross‐sectional evidence supports the empirical assertion that community divisions gradually emerge in the UK. Indeed, Duffy and Allington (2020) proposed that divisions have given rise to three main groups: those who while worried about the virus, are trusting the government and the measures it implements; those who are most worried about the health effects of the pandemic and express greater criticism for the government and those who are least anxious about the health effects of the pandemic, hence wishing lockdown measures to be lifted while feeling ambivalent about the government's response.

Perceived stress disparities between males and females were found across geographical regions and occupations despite the higher thresholds set for females. While sex hormone fluctuations, particularly those involving high levels of progesterone, can partially enhance female susceptibility to anxiety and stress following a traumatic event (Li & Graham, 2017), one must also look at societal markers of stress to understand the psychological impact of the pandemic. For example, men have traditionally better occupational backgrounds and lower likelihood to be single at an older age (Van der Meer, 2006), while women are more likely to be heavily engaged with stress‐inducing unpaid care work than men (Foster & Elntib, 2020), as vividly manifested in the gender stress‐disparity among those taking over home‐duties in this sample. Evidence from past pandemics has also shown women are at much higher risk of domestic violence (Peterman et al., 2020) and are more heavily engaged with care responsibilities such as child‐care following school closures and caring for ill family members (Bandiera et al., 2019; Wurth et al., 2017).

Implications and limitations

The study added unique empirical evidence regarding the perceived stress and health anxiety levels experienced by different sections of the British public. It also highlighted the protective role of community cohesion proposing a mechanism through which perceived stress is reduced. The study also highlighted the risks involved in a situation where community cohesion is further weakened but also potentially underlined aspects of a healing process that might have started following the initial impact as seen in the slowing trends in perceived stress and health anxiety scores during the lockdown. It is advised that scholars agree on clear upper and lower health anxiety threshold‐points during the pandemic, considering their high relevance in the fight against COVID‐19 (Asmundson & Taylor, 2020). It is also advised that the links between health anxiety and engagement with specific precautionary and protective measures are understood. For example, if very low health anxiety scores lead to reckless social behaviors harming public safety, then it is important to investigate the cognitive causes facilitating such behaviors and their magnitude and prevalence in the community. For example, recent UK‐based empirical research, has shown that those who hold COVID‐19 conspiracy beliefs (e.g., suggesting that the virus does not exist, or challenging the viral origin of its symptoms) are less likely to support or adhere to health‐protective behaviors and more likely to use social media platforms, especially You Tube, to get informed about the pandemic (Allington et al., 2020). Similarly, lack of trust toward official government officials, and leadership that conveys that it is not prepared to share the burden of the crisis, or lead by example, might shape follower indifference, reducing willingness to abide by health and social distancing policies (Van Bavel et al., 2020).

The study misrepresented UK's male population with a disproportionally large female representation. Although a stratified sampling approach was not fully achieved, it was still surprising that male participation was comparatively as low as both genders were offered equal opportunities to participate, resembling a trend found in similar studies (González‐Sanguino et al., 2020; Jungmann & Witthöft, 2020; Mertens et al., 2020; Ozamiz‐Etxebarria et al., 2020; Rossi et al., 2020). Such imbalance can be partially explained by women's higher engagement with mediated technology (Kimbrough et al., 2013) and more intensive use of social networking sites (Muscanell & Guadagno, 2012). Nevertheless, it might be fruitful to empirically investigate whether men are more reluctant than women to share their concerns about their health during the pandemic. The study has similarly overrepresented the North‐West region, potentially skewing the average national opinions held by the British public. Still, the North West region has been in the midst of the pandemic crisis and is one of hardest‐hit areas in the country, hence valuable insights can be extracted by looking at markers of wellness in this county. While this study was representative of older UK population, it was conducted online so it restricted those without access to or use of internet, potentially failing to give a voice to those living in care homes, one of the main epicenters of the pandemic (Holt & Butcher, 2020). By conducting the study online, we could not allow the computer or smart‐phone illiterates to take part. The study survey did not gather information about pre‐existing health status, including whether participants were diagnosed or have had a relative diagnosed with COVID‐19, so it is likely the findings are skewed by responses driven by health status (of loved ones) rather than anxiety and stress per se (Mertens et al., 2020). Similarly, no information about the finances of the participants was gathered so while stress can be a by‐product of the financial or health‐related pressures, the study was limited to the narrow set of factors tested.

Conclusion

This was, to the authors' knowledge, the first UK‐based study that empirically highlighted the buffering role of community cohesion against perceived stress and health anxiety during the first peak of the pandemic, re‐asserting the protective role of community factors emphasized in pre‐pandemic research. Since the emergence of the virus unprecedented emphasis has been put in the implementation of draconian social distancing measures leading to lockdown measures imposed to half of earth's population by early April, 2020 (Sandford, 2020). Still, as of May 2021, no corresponding and systematic guidelines have been developed to promote safe socializing within neighborhoods and community hubs with an aim to boost neighborhood belongingness and attraction among its members.

Therefore, it is advised that government planning also focusses on implementing measures that ensure that green (and blue) spaces are openly available for and safely used by community members. For example, instead of closing down large areas of green spaces to “secure” social distance between members, leading to greater congestion in densely populated areas, parks and other green (and blue) spaces should be accessible areas freely booked by members of the community who can check (perhaps by using an appropriate app) the availability and real‐time capacity of open‐air community spots. According to our research, increasing community attraction through easing access to safe community hubs can partially mediate the harmful effects of anxiety and stress so if meeting others outside can be promoted sensibly, even for the professional and age groups most severely affected (e.g., younger population, health‐service personnel, etc), then public guidance and strategies adopted can be accordingly reformed.

Similarly, to boost the psychological sense of community, local mutual‐aid initiatives already supporting marginalized community members (see Mutual Aid UK, 2021 for an indicative list of over 4,000 groups based in the UK) must be publicized so that their contributions are appreciated and understood. Such mutual aid groups often fill gaps in services that the state mismanages so making the public aware of such initiatives may amalgamize community members' understanding of and involvement in mutual aid community work and support community resilience and reciprocity amongst its members. Encouraging the engagement of members in community activities can boost the psychological sense of belongingness through member interactions and the mere realization that neighbors and community members will help in an emergency, especially when the state‐response is inadequate. Such autonomous community initiatives were enacted successfully by North‐London mutual aid groups, during the lockdown (Chevée, 2021) distributing and securing essentials (e.g., food and medication) to vulnerable and neglected populations during the lockdown.

Ultimately, this research highlighted the positive role of community cohesion during the pandemic and the potential health risks involved in allowing community cracks to grow. The protective role of community cohesion against stress and health anxiety hints toward the need to also consider and develop non‐medical interventions, involving community mobilization, to improve the mental‐health and general wellness of the public during the COVID‐19 pandemic.

Fig S1

Click here for additional data file.

Table S1

Click here for additional data file.