Social identification and risk dynamics: How perceptions of (inter)personal and collective risk impact the adoption of COVID‐19 preventative behaviors

Abstract Public adoption of preventative behaviors to reduce the transmission of COVID‐19 is crucial to managing the pandemic, and so it is vital to determine what factors influence the uptake of those behaviors. Previous studies have identified COVID‐19 risk perceptions as a key factor, but this work has typically been limited both in assuming that risk means risk to the personal self, and in being reliant on self‐reported data. Drawing on the social identity approach, we conducted two online studies in which we investigated the effects of two different types of risk on preventative measure taking: risk to the personal self and risk to the collective self (i.e., members of a group with which one identifies). Both studies involved behavioral measures using innovative interactive tasks. In Study 1 (n = 199; data collected 27 May 2021), we investigated the effects of (inter)personal and collective risk on physical distancing. In Study 2 (n = 553; data collected 20 September 2021), we investigated the effects of (inter)personal and collective risk on the speed at which tests are booked as COVID‐19 symptoms develop. In both studies, we find that perceptions of collective risk, but not perceptions of (inter)personal risk, influence the extent to which preventative measures are adopted. We discuss the implications both conceptually (as they relate to both the conceptualization of risk and social identity processes) and also practically (in terms of the implications for public health communications).

1 Study 1 additional information 1.1 Additional methodological details

Perceptions of risk and shared identity with local community
We first collected data on the participant's COVID-19 vaccination status (including number of doses, how long ago they had the latest dose, and, for the unvaccinated, willingness to have a vaccine).We then collected the following data on their perceptions of COVID-19 risk and shared identity with their local community.All items used a 7-point agreement scale from 1 ("Strongly disagree") to 7 ("Strongly agree").
Other people in my household are at risk from COVID-19.My family and close friends are at risk from COVID-19.
My local community is at risk from COVID-19.
My national community is at risk from COVID-19.
The world is at risk from COVID-19.
Local community shared identity (4 items, Cronbach's α = 0.94): I have a feeling of unity with other residents of my local community.
I have a sense of "we-ness" with other residents of my local community.Besides our differences, I share the same identity with other residents of my local community.I feel at one with the local community members around me.

Additional preventative measures
After they completed the physical distancing task, we asked participants about other preventative measures taken to reduce the spread of COVID-19, related to personal hygiene, travel, attending large gatherings, meeting up with other people, and intentions to self-isolate with symptoms.First, we asked about their intentions for the coming week.Second we asked about the preventative measures they had taken over the previous week.All items used a 7-point agreement scale from 1 ("Strongly disagree") to 7 ("Strongly agree").
Intentions to take preventative measures over the next week (9 items, α = 0.75): I will regularly wash my hands for 20 seconds.

Analysis
Our primary analyses investigated the effects of inter-personal risk (composite measure of the first three perceptions of COVID-19 risk items, above) and community risk (composite measures of the final three items) on each of standing physical distancing (SI Section 1.2.1), seated physical distancing (SI Section 1.2.2), and the additional preventative measures (SI Section 1.2.3).

Standing physical distancing
Our analysis involved linear mixed effects modelling using lme4 (Bates et al., 2013) in R (R Core Team,

2021).
First, we constructed a model to confirm that physical distancing was lower when the other individual was a friend as opposed to a stranger, and when the other individual was wearing a mask as opposed to not.We included mask wearing (sum coded: 1 if the other individual was wearing a mask; -1 if they were not) and the identity of the other individual (1 if the other individual was a friend; -1 if they were a stranger) as fixed effects.Maximal random effects structures were considered in the first instance (in line with Barr et al., 2013).Our initial model therefore included random intercepts for participant identity, trial number, vaccine status (1 if the participant had had at least one dose; -1 if they had had no doses), local community shared identity (composite measure), age, gender, and household income (converted to a linear scale and log transformed).Random intercepts were removed as necessary in the event of singular fit or non-convergence (also following Barr et al., 2013).
Our second model investigated the effects of inter-personal risk and community risk on physical distancing.Inter-personal risk and community risk (both centred) were included as fixed effects.1 Our random effects structure was the same as for the first model in the first instance, though with additional random intercepts for mask wearing and the identity of the other individual.
As an additional check of the robustness of these results, we repeated our analyses on the subsets of our data where (a) the other individual was wearing a mask, (b) the other individual was not wearing a mask, (c) the other individual was a friend, and (d) the other individual was a stranger.In all cases, and despite the smaller sample sizes involved, we get the same pattern of results: physical distance increased with community risk (b ≥ 12.012, SE ≤ 4.232, t ≥ 3.632, p < 0.001), but there was no effect of inter-personal risk (|b| ≤ 5.043, SE ≥ 2.660, |t| ≤ 1.896, p ≥ 0.060).

Seated physical distancing
We analysed the seated physical distancing trials in the same way as the standing physical distancing trials.Despite there being less data for the seated trials, we see the same pattern of results.Physical distance decreased if the other individual was a friend (b = -0.564,SE = 0.023, t = -25.043,p < 0.001), or if they were wearing a mask (b = -0.181,SE = 0.023, t = -8.045,p < 0.001).Physical distance increased with community risk (b = 0.127, SE = 0.038, t = 3.321, p = 0.001), but not with inter-personal risk (b = -0.028,SE = 0.031, t = -0.887,p = 0.376).
As an additional check of the robustness of these results, we again repeated our analyses on the subsets of our data where (a) the other individual was wearing a mask, (b) the other individual was not wearing a mask, (c) the other individual was a friend, and (d) the other individual was a stranger.In all cases, and despite the smaller sample sizes involved, we again get the same pattern of results: physical distance increased with community risk (b ≥ 0.097, SE ≤ 0.048, t ≥ 2.250, p ≤ 0.026), but there was no effect of inter-personal risk (|b| ≤ 0.046, SE ≥ 0.034, |t| ≤ 1.293, p ≥ 0.198).

Additional preventative measures
To analyse the additional preventative measures, we constructed cumulative link mixed models using the ordinal package (Christensen, 2019).The fixed effects were inter-personal risk and community risk (both centred).In the first instance, we included random intercepts for participant identity, specific item, local community shared identity (composite measure), age, gender, and household income (converted to a linear scale and log transformed).
2 Study 2 additional information 2.1 Additional methodological details

Perceptions of risk and unimportance of infection
We first collected data on the participant's COVID-19 vaccination status (including number of doses and, for the unvaccinated, willingness to have a vaccine).We then collected the following data on their perceptions of COVID-19 risk and and the extent to which the participant viewed community and personal infection as unimportant.All items used a 7-point agreement scale from 1 ("Strongly disagree") to 7 ("Strongly agree").It is important for me to act in ways that stop me getting infected.(R)

Self-isolation measures
The participant was asked to imagine a scenario in which they had just tested positive for COVID-19.
We collected their intentions to assist with contact tracing, self-isolate from other households, and, if they shared their home with others, self-isolate within their household.All items here used a 7-point agreement scale from 1 ("Strongly disagree") to 7 ("Strongly agree").Finally, we presented them with a set of vignettes designed to assess the extent to which they would transgress the self-isolation requirements.These items here used a 7-point likelihood scale from 1 ("Extremely unlikely") to 7 ("Extremely likely").
Self-isolation transgression vignettes (9 items, α = 0.88): 1.There are very few people around outside at the moment.You know that you would feel a lot better if you could stretch your legs and get some fresh air by having a 10 minute walk outside.
How unlikely or likely would you be to take a walk? 2.You have no way of getting any more food delivered to you today.You would really appreciate the chocolate and snacks you could quickly get from a nearby shop.How unlikely or likely would you be to take a quick trip to the shop?3. A member of your family who lives nearby has been feeling really down recently.You'd really like to have a chat to them face-to-face to cheer them up.How unlikely or likely would you be to pop round and have a socially-distanced chat with them? 4. Your friend who lives nearby is also self-isolating.They suggest you meet up halfway between your homes to quickly swap some books and DVDs.How unlikely or likely would you be to agree to this? 5.It is early evening, and you hear that there is a spectacular meteor shower visible at the moment.
You can't see it from your home, but you could if you walked 5 minutes down the road to get a perfect view.How unlikely or likely would you be to go out and see it?6.A snack van has parked opposite your house and is selling ice creams and freshly-baked pastries.
You would really like to buy something.How unlikely or likely would you be to go out to buy something from the van? 7.You see that an elderly neighbour has received a large delivery from a garden centre.They are struggling to stack a pile of boxes by their front door.How unlikely or likely would you be to go out and help them?8.Your friend has been worried about how you have been coping since you started self-isolating, and they have come round to visit you.They are clearly expecting you to invite them in for a cup of tea.How unlikely or likely would you be to invite them in? 9.You have kept apart from the other people in your household until now, but now someone you live with is really upset.How unlikely or likely would you be to give them a hug?

Analysis
Our primary analyses investigated the effects of inter-personal risk and community risk on test-booking, supplying contact tracing information, and self-isolating (SI Section 2.2.1).We also carried out additional analyses to explore the effect of personal circumstances on the difficulties of self-isolation and testbooking (SI Section 2.2.2).Finally, we considered an alternative approach to investigating the effects of relatively (inter-)personal-level and community-level risk by investigating the effects of seeing personal and community infection as unimportant (SI Section 2.2.3).

Preventative measures
For each preventative measure, we ran a cumulative link mixed model (again using the ordinal package in R) with inter-personal risk and community risk (both centred) as fixed effects.In the first instance, our random effects structure included random intercepts for participant identity, trial (for the test booking measure) or item (for the other four measures), age, gender, and household income (log transformed).
For the test-booking measure, we removed the 93 trials (4% of the total) where participants booked a test with no symptoms (i.e. at Step 0).
Model outputs for the other preventative measures -contact tracing, between-household self-isolation intentions, within-household self-isolation intentions, and self-isolation transgression -are given in SI Table 1.

Effect of personal circumstances on the difficulties of self-isolation and test-booking
To investigate the effects of personal circumstances on each of the financial, practical, and other (e.g.emotional) difficulties of self-isolation, we constructed cumulative link models with the fixed effects of age (centred), household income (log transformed and centred), childcare responsibilities (centred), adult caring responsibilities (centred), number of adults in the household (log transformed), and number of children in the household (log transformed).Note that due to not all participants disclosing all this information, these models only use data from 498 of the 553 participants.
Model outputs for financial, practical, and other difficulties of self-isolation are given in SI Tables 2,   3, and 4, respectively. SI

Effect of seeing infection as unimportant
An alternative to using the perceptions of risk measures was to use the measures of seeing infection as unimportant (SI Section 2.1.1).Again, we got the same pattern of results to those we presented above.
Parallel analysis of the four non-reversed items indicated a two-factor split.Factor analysis identified a factor of community infection unimportant -comprising the items "It doesn't matter if people catch COVID-19" and "The consequences of catching for people are not very serious" -and a factor of For each preventative measure, we ran a CLMM with seeing community infection as unimportant and seeing personal infection as unimportant (both centred) as fixed effects.In the first instance, our random effects structure included random intercepts for participant identity, trial (for the test booking measure) or item (for the other four measures), age, gender, and household income (log transformed).
For the test-booking measure, seeing community infection as unimportant increased the number of steps before a participant booked a test (b = 0.243, SE = 0.100, z = 2.424, p = 0.015).There was no effect of seeing personal infection as unimportant (b = -0.125,SE = 0.082, z = -1.527,p = 0.127).

I
will touch my eyes, nose, or mouth with unwashed hands.(reversed item [R]) I will travel for nonessential reasons.(R) I will wear a facemask in public.I will avoid places where many people will gather.I will meet up with another household outdoors.(R) I will meet up with another household indoors.(R) I will attend a large gathering.(R) I will self-isolate if I develop symptoms of COVID-19.Preventative measures taken over the previous week (8 items, α = 0.65): I regularly washed my hands for 20 seconds.I touched my eyes, nose, or mouth with unwashed hands.(R) I travelled for nonessential reasons.(R) I wore a facemask in public.I avoided places where many people gathered.I met up with another household outdoors.(R) I met up with another household indoors.(R) I attended a large gathering.(R) Contact tracing (4 items, α = 0.81): I would tell my employer, school, or nursery about my result (if applicable).To the best of my ability, I would contact everyone I had recently been in close contact with to tell them about my result.If asked by NHS Test and Trace, I would provide information about where I'd been recently.If asked by NHS Test and Trace, I would provide information about everyone I'd been in close contact with.Self-isolation intentions, between households (5 items, α = 0.80): I would fully self-isolate.I would never go to work, school, or public places.I would never go out to get food or medicines.I would never have (non-essential) visitors in my home.I would never go out to exercise.Self-isolation intentions, within household (4 items, α = 0.92): I would keep fully away from the other people in my home.I would eat every one of my meals on my own in my own room.I would avoid spending any time at all in the same room as the other people.I would clean any shared rooms (e.g.bathroom or kitchen) after I have used them every time.

Table 2 :
CLM output for effect of personal circumstances on financial difficulties of selfisolation.Household income, number of adults and number of children are log transformed.Childcare and adult caring responsibilities are binary variables.Age, household income, childcare, and adult caring responsibilities are centred.

Table 3 :
CLM output for effect of personal circumstances on practical difficulties of selfisolation.Household income, number of adults and number of children are log transformed.Childcare and adult caring responsibilities are binary variables.Age, household income, childcare, and adult caring responsibilities are centred.

Table 4 :
CLM output for effect of personal circumstances on other (e.g.emotional) difficulties of self-isolation.Household income, number of adults and number of children are log transformed.Childcare and adult caring responsibilities are binary variables.Age, household income, childcare, and adult caring responsibilities are centred.

Table 5 :
CLMM output for effect of personal circumstances on test-booking.Household income, number of adults and number of children are log transformed.Childcare and adult caring responsibilities are binary variables.Age, household income, childcare and adult caring responsibilities are centred.

Table 6 :
Test booking step means (of individual participant means) by low and high personal and community infection unimportance, measured in test-booking steps.Personal/Community infection unimportance is split into low (less than or equal to mean personal/community infection unimportance) and high (greater than mean personal/community infection unimportance) for illustration purposes.Our statistical analyses treat test booking step as an ordinal variable and personal and community infection unimportance as continuous variables.