Compliance to behavioural messages during crisis

Manuscript Title:
Nudge in the time of coronavirus: Compliance to behavioural
messages during crisis

Author(s) and affiliation(s)

Susannah Hume*
King’s College London

Peter John
King’s College London

Michael Sanders
King’s College London

Emma Stockdale
King’s College London

*Address correspondence to Susannah Hume at Susannah.hume@kcl.ac.uk.

Transparency: If published, the author(s) will provide the following (check all that apply):
☐ Open data
☐ Open materials
☒ Preregistration
☐ None of the above

Data will be available on request.

1
 Nudge in the time of coronavirus: Compliance to behavioural
messages during crisis

Abstract:
Successful responses to the coronavirus pandemic require those without COVID-19 and
asymptomatic individuals to comply with a range of government guidelines. As nudges have
been widely found to be effective at stimulating pro-social behaviours, how good are they for
the COVID-policy toolkit? In particular, is a reflective device or nudge plus, as an addition to
the classic nudge, able to deal with the scale of the problem? To test for the impact of nudges
and nudge plus, we implemented an online experiment with 1,481 people during a period of
the first national lockdown in the UK in April/May 2020. We show that social norms and the
portrayal of the victim do not work on their own at increasing intentions to comply with the
guidelines, but when the victim is combined with the more reflective task of writing to a
relative there is an impact. After two weeks, however, these intentions do not persist. The
implication is that there still much work to do in designing nudges in the context of COVID-
19 and other public health pandemics, yet reflection as a behavioural device can encourage
individuals to think more responsibly in a world-wide pandemic.

Keywords: (About 5 keywords) Nudge, Nudge Plus, COVID-19, social norms, compliance

2
Introduction

Compliance with public health guidelines, such as frequent hand-washing for a minimum

of twenty seconds each time, social distancing by standing two metres apart from others,

and avoiding going out where possible, remains critical for both reducing transmission of

the COVID-19 virus, and avoiding further peaks. As compliance is difficult to observe and

enforce, nudge interventions may be effective as a complement to standard policy

instruments, leading to sustained behaviour change (Gray et al., 2014). Academics and

policy-makers have argued that there is a great potential for nudge interventions given

that, for many, non-compliance may be a function of bounded cognition rather than

resistance to government public health messages (Soofi et al., 2020). This makes

interventions designed to correct such biases appropriate and potentially efficacious

(Lunn, Belton, Lavin, McGowan, Timmons, & Robertson, 2020). However, there is a need

for more research to find out if such interventions work in the context of a pandemic like

COVID-19 (West et al., 2020). Testing the effectiveness and persistence of behavioural

messages is useful for the development of messaging in current and future pandemics.

Nudge refers to any aspect of the presentation of choices (choice architecture) that

alters individual behaviour in predictable ways without reducing autonomy or changing

the underlying incentive structure (Leonard, 2008). Nudge is a key tool for today’s public

administrators, with tests being frequently reported in public administration journals (e.g.

John, 2018; Larkin et al., 2019; Vainre et al., 2020). Yet nudges are usually tested for the

routine tasks most citizens undertake, such as submitting a tax return, not in crisis

situations when both the environment and people’s choices are anything but routine. It is

a question of external validity whether these tried and tested nudge techniques work in a

crisis, when much is demanded from the citizens; when there are already significant

limitations on individual freedom; and where there is a huge amount of other information

3
citizens are receiving from governments and media outlets at the same time. Can soft

information signals break through these constraints?

Literature Review

Nudges have been tested during the pandemic in various designs, but given the

extent of official interest in this phenomenon, there is still relatively little published

experimental work using insights from behavioural public policy, with most interventions

focusing on specific public health measures, such as a default for hand washing (e.g.

National Library of Medicine, 2020). One study similar to ours, carried out in Japan, tested

interventions online, finding that only one treatment was significant and positive in effect

(Sasaki et al., 2020). Another study found behaviourally-informed text messages to be

effective in São Paulo (Boruchowicz et al., 2020), while a Danish study found that prompts

were effective in changing intentions but not actions, as in our study, though this study did

not use behaviourally-informed messages (Falco & Zaccagni, 2020). Many studies carried

out by government teams and consultancies, such as the Behavioural Insights Team

(https://www.bi.team/our-work/covid-19/), have only reported headline results, with full

study designs not yet available. In this section we review the literature that underpins the

three hypotheses tested in the present research.1

Nudge: Social norms and identifiable beneficiaries

In this paper, the focus is on two common interventions that use social context to

increase a desired behaviour: (i) providing normative feedback; and (ii) foregrounding an

identifiable beneficiary.

1 Pre-registered with the Center for Open Science, see Appendix C.

4
 Hypothesis 1: Participants who viewed either the identifiable beneficiary

norm or normative social norm message would subsequently report a greater

intention to comply with public health guidelines than participants in the

control group.

Normative feedback as an intervention is based on the notion that although people are

influenced by social norms, they are also poor at estimating actual behaviour and attitudes

(Perkins & Berkowitz, 1986). Social norms guide behaviour by representing the “rules and

standards that are understood by members of a group, and that guide or constrain social

behaviours without the force of law” (Cialdini & Trost, 1998, pg.152). People are often

strongly influenced by what others do though conformity (e.g. Goldstein et al., 2008; John

et al., 2019; Stok et al., 2016; Wiepking and Heijnen, 2011). Normative misperception can

lead to engagement in undesired behaviours due to a false belief that they are

commonplace (McAlaney et al., 2011), and therefore providing information about norms

can be very influential on behaviour. For instance, Hallsworth et al., 2016 found that

providing social norm feedback on antibiotic prescribing rates to GP practices prescribing

at a rate higher than 80% of practices in their NHS Local Area Team resulted in

significantly fewer antibiotic prescriptions than a control. In the early months of the

COVID-19 pandemic, the UK media widely reported people flouting the public health

guidelines by panic buying and not social distancing.2 Yet, highlighting instances of such

behaviour creates the perception that this is the norm, causing an increase in these

behaviours as people adjust their own behaviour in line with the perceived norm (Cialdini

et al., 2006a; Farrow et al., 2020). Previous interventions suggest that policy-makers could

address this by correcting people’s perceptions of the norm via communications with the

public or by providing normative feedback (e.g. how one’s adherence to the guidelines

2 Derbyshire Police, 2020; Ellson, 2020; Jarvis, 2020.

5
compares to the national average) (Allcott, 2011; Cialdini et al., 2006b). However, it is

unclear whether this is effective in the case of the current crisis where polls suggest that

people are generally aware of, and support, the rules (see Duffy, 2020).

Another form of social norm intervention involves highlighting a specific

individual who is affected by another individual’s behaviour. Research on the identifiable

victim (or beneficiary) has long established that people are more willing to make sacrifices

for identifiable individuals who are affected by the cause as opposed to those who are

represented as groups or statistics (Kogut & Ritov, 2005; Small & Loewenstein, 2003;

Small et al., 2007). The mechanism here is the norm of helping others, which works more

effectively when people are able to identify a specific person who will be impacted by

their actions, either to help or harm. In relation to the COVID-19 outbreak, research

suggests that highlighting a person vulnerable to infection can encourage social distancing

(Lunn, Belton, Lavin, McGowan, Timmons, & Robertson, 2020; Pfattheicher et al., 2020).

Indeed, it has been found that providing background information on why it was important

to social distance does not increase the motivation to adhere to guidelines (Pfattheicher et

al., 2020); however, when individuals felt empathy for the most vulnerable, the motivation

to distance socially increased.

Nudge plus: Self-reflection

Building on these social norm interventions, the present research also includes a

self-reflection intervention that aims to nudge individuals to think about the problem at

hand, with the goal of producing more persistent behavioural change.

Hypothesis 2: Participants who are asked to complete a self-persuasion task

will subsequently report a greater intention to comply with public health

guidelines than those who do not complete a self-persuasion task.

6
At the peak of the crisis in the UK, the majority of people knew what the guidelines were

(Geldsetzer, 2020); however, having such knowledge does not always result in adherence

(Pfattheicher et al., 2020). Self-generated arguments are often perceived by individuals as

more correct and trustworthy than those from external sources and are less likely to result

in defensive reactions (Hoch & Deighton, 1989; Liberman & Chaiken, 1992; Mussweiler

& Neumann, 2000). They may also result in a greater degree of internalisation of the social

norms (Thøgersen, 2006), which in turn creates a sense of pride from following, or shame

from not following, the guidelines. Self-persuasion has shown promise at increasing

adherence to desirable norms (e.g. Aronson, 1999; Müller et al., 2016a). Incorporation of

a reflective or self-persuasive component into a nudge has been termed ‘nudge plus’

(Banerjee & John, 2021; Kardes et al., 2001a; Müller et al., 2016b; Stoker & John, 2019).

Self-persuasion strategies can increase the internalisation, and hence persistence, of

nudge-style messages. It may be the case that with the large amount of information that

people are receiving encouraging them to comply, as with public information messages,

an activity that asks people to stop and reflect might be more suitable for the COVID

pandemic than standard nudge messages, or at least enhance a nudge message. Also, when

so many regulations have been introduced at such great speed and without much

deliberation by legislators, by executive action and delegation, the nudge plus does seek

to introduce a degree of citizen autonomy and respect into the proceeding through the

process of reflection on the argument for these regulations.

Interaction effects: Nudge and nudge plus

It is not unreasonable to assume that self-persuasion will be more effective at increasing

intention to adhere to guidelines after seeing certain types of social messages than others, or

7
than self-persuasion alone. This is core to the nudge plus idea that the plus (the think) is

presented alongside a classic nudge, so enhancing the nudge with the plus (see Banerjee and

John 2021). This is similar to the argument in the boost literature (Hertwig and Grüne-Yanoff

2017) that the individual needs extra capacity and resources conveyed by the boost to respond

to an information signal effectively.

Hypothesis 3: There may also be a complementary effect of the nudge

messages and the self-persuasion/nudge plus task.

The experiment

Although promising, much of the research on self-persuasion and nudge plus, has

been conducted in the lab environment (for exceptions see, for example, Damen et al.,

2015; Kardes et al., 2001b; Müller et al., 2016b) making generalisability to the current

context hard to ascertain. Further, most existing nudge research to date is not conducted

in crisis contexts, and there is reason to be cautious about extrapolating effects. In the

present situation, fear and panic can dominate (Association, 2020; McKeever, 2020)

making behaviour more difficult to anticipate and influence.

The present study tested whether intentions of compliance with public health

guidelines related to COVID-19 (including, but not limited to, hand washing, social

distancing, and self-isolating) can be improved using behavioural communication

strategies. The research was carried out in the UK the middle of the first national lockdown

designed to suppress the spread of the virus. From March 23 the UK government confined

everyone (bar essential workers) to their homes, only with permission to exercise outside

one hour per day, rules that gradually eased from 16 May.3 The first wave of the main

3 There were variations in these rules across the four nations of the UK, though these were generally not
large at this period of time.

8
experiment ran from 23 April 2020 at 2pm to 24 April 2020 at 8am; the second wave took

place from 7 May 2020 at 11am to 9 May 2020 at 4pm. These dates show that the

experiment occurred right in the middle of the period when citizens were required to

follow the most stringent set of rules from government. However, owing to reports in the

media around 7 May, wave 2 participants may have been expecting that some of the

restrictions might be about to be lifted - this announcement occurred on 10 May.4 We

provide a full timeline of announcements in the months surrounding the trial in Appendix

D.

There is always a challenge, especially with behavioural interventions, as to

whether people who are mobilised with good intentions turn them into action (Brandon et

al., 2017; Gaudeul & Kaczmarek, 2017). If the good intentions are not accompanied by a

plan to put them into action, or are triggered at a point when they cannot be acted on, they

can be forgotten. Pro-social behavioural intentions can also fade as the message becomes

more temporally distant (Västfjäll et al., 2014). On the basis that good intentions do not

help anyone, we wished to test whether these intentions translated to reported changes in

behaviour, at a two-week follow up.

Method

Design and materials

After consenting and answering introductory questions, participants were randomly

allocated to see one of three scenarios: a control, a normative (‘classic’) social norm, or

the identifiable beneficiary of an elderly person. The text of all three interventions is given

4Note that the most important event that changed people’s response to government during the lockdown
was the furore over the breach of guidelines by the Prime Minister’s Chief of Staff, Dominic Cummings. This
occurred after our experiment, from 22 May.

9
below. The interventions were mocked-up to look like online news articles and were

identical in content - bar the title and opening paragraphs outlined below - with each

having three further paragraphs detailing factually correct information about the lockdown

announced by the prime minister, Boris Johnson, on 23 March.

Control

What are the public health guidelines in the UK?

When staying at home in order to social distance or self-isolate for an extended

period of time it is important to maintain your wellbeing by eating a balanced diet,

keeping mobile and staying in contact with family online or via mobile.

You should also keep your mind busy with activities such as cooking, sewing or

painting and creating a routine can also help maintain a sense of normality.

Classic social norm

Continue to adhere to the public health guidelines

Recent surveys suggest that, in the UK, 94% of people are complying with a

minimum of one public health guideline. Overall:

• More than 80% of people are avoiding public places.

• 82% of people are washing or sanitising their hands more frequently than

usual.

• 61% of people are avoiding all travel by car, train or bus

• In London, movement around the city has dropped 91% compared to usual.

Identifiable beneficiary

Follow the public health guidelines to protect the elderly

10
 “We feel completely powerless with this situation. We actually have less freedom

than the younger generation because we’re faced with this virus while in the last

chapter of our life.” By adhering to the guidelines you protect Sam, a 72 year-old

retiree with three grandchildren from developing COVID-19, but also from the

loneliness and uncertainty of being isolated indefinitely.5

Following the article, participants were randomised to see a nudge plus self-

reflection intervention or not. If they did not receive the nudge plus, they moved straight

to the next stage of the survey; if they received the intervention they saw a message which

read:

The chances are that everyone knows someone who is a key worker - helping keep

the country running - or is either elderly, or has an underlying health condition, and

so would be substantially more impacted by the virus if they caught it.

Reflecting on this person you know, how would you explain what steps you are

taking to reduce the spread of coronavirus, and why you consider that this is

important? Please use the box below.

The median response length for those who saw the self-reflection prompt was 36.5

words; the first quartile was 21 words. Only 5 respondents (out of 736) in the plus

condition provided no text in response.

All participants then indicated their intention to adhere with public health

guidelines on a 7-point scale from (1) ‘never’ to (7) ‘always’ with an option to select ‘I

5This beneficiary was chosen via a pilot study (N=300, on 6 April in the afternoon, followed by a booster
sample on 11 April also in the afternoon), which aimed to identify which of three potential beneficiaries of
participants’ good behaviour (or victims of poor behaviour) participants found the most relatable, based on
scores on measures of emotional valence. The vignettes featured a medical professional, an elderly person,
and a younger immunocompromised person. After reading the vignettes, participants were asked a series
of questions about how much they related to the character described as well as some factual questions
about the government guidance. The elderly vignette provided the highest relatability/emotional valence
scores (p=0.254, Cohen’s d =0.2), as well as a higher proportion of correct answers on the factual questions.

11
don’t know / not applicable’. Items on this scale were based on Weinberg (2020) and were

amended to cover official and unofficial guidelines in place in the UK at the time of the

study, and to include seven response levels rather than five. Items included ‘Working from

home unless essential to do otherwise’ and ‘Social distancing from others apart from those

in your household’. Participants were also asked to indicate their intention to leave their

home in the next five days: ‘Yes’, ‘No’, ‘I’m not sure’. If they answered ‘Yes’ or ’not

sure’ they were then asked how frequently they intended to leave their house across a

range of different activities. This was originally used by Barari et al. (2020), who found

62 percent of respondents stated a need to leave the house. Appendix A gives the full set

of activities for both outcomes.

Sample

After obtaining ethical approval, 6 we recruited a total sample of 1,481 participants on

Prolific with an average age of 46.1 and a standard deviation of 15.4 years. The sample

was recruited to be representative of the UK population with regards to age, gender, and

ethnicity. Sample size was mainly determined by financial constraints, but we calculated

that a sample of 250 participants per cell was powered to detect effects greater than

Cohen’s d of 0.25, a fairly modest effect. This achieved cell size is in keeping with the

guidance of Simmons et al (2018).

Respondents were re-sampled two weeks later. 1,211 (79%) participants completed

the follow-up survey. There was no imbalance in retention by intended compliance, but

6University identity is blinded, GGC Research Ethics Panel, reference: LRS-19/20-19268, 19 April 2020,
amended 26 May 2020.

12
some variation in retention on ethnicity, and those seeing the beneficiary plus interventions

had higher rates (5 percentage points, p = 0.1) of responding to wave 2.

Analysis

The analysis proceeded using an Ordinary Least Squares (OLS) regression with ‘HC2’

robust standard errors (Hayes and Cai, 2007).

The analysis specified the treatment indicator as a factor with six levels,

corresponding to the six combinations of the nudge and plus interventions, with those

allocated to control in both randomisations as the reference category. This specification

was chosen as it is the most straightforward to interpret and has the most meaning from a

policy standpoint. Our pre-registration used a 3 x 2 factorial design, which is provided in

Appendix B. In addition, we have analysed wave 1 and wave 2 separately, again for

reasons of interpretability. However, we present a panel specification, with standard errors

clustered at the level of the individual respondent, in the Appendix. Our pre-registration

plan included the use of a baseline measure of the outcome, but ultimately we did not

collect this as we did not think it sensible to administer the same instrument twice within

a seven-minute survey. Instead, we included gender, age, and ethnicity as covariates, and

present specifications with and without these covariates in the Appendix. Our outcome

measures were the proportion of activities where a respondent indicated an affirmative

intention to comply with regulations, and how often they intended to leave their house

over the coming week. Our pre-registration assumed we would be able to code this

outcome as the average of the responses, from 1 (‘never’) to 7 (‘always’), but did not pre-

specify an imputation strategy for ‘don’t know’ responses. However, likelihood of

answering ‘don’t know’ was close-to-significantly affected by treatment assignment (p =

13
0.1), so we determined that the most parsimonious way to code this would be to treat the

outcome as the proportion of instances where participants indicated a strong intention to

comply with the guidelines (i.e. if they answered ‘almost always’/‘always’ - or ‘almost

never’/‘never’ in some cases). We considered this the approach that assigned ‘don’t know’

most appropriately. The outcome measure was therefore a continuous variable from 0 to

1, representing the proportion of the activities presented where the respondent intended to

comply with the regulations (see Appendix A for list of activities). We present the analysis

as pre-registered, with ‘don’t know’ responses conservatively imputed with the item

sample mean, in Appendix B.

This outcome was re-collected in the follow-up survey two weeks later. As there was

a close-to- significant imbalance on response to the follow-up, there was a risk of bias in

complete case analysis, or when using last observation carried forward to impute wave 2

outcomes for these respondents. As such, we opted to impute the sample mean for each

item in analysing the follow-up.7 We also conducted complete case analysis, but the results

change very little, so we present only the analysis with imputed values in the paper.

The second outcome related to respondents’ intentions to leave the house.

Respondents were first asked whether they needed to leave the house in the next week.

Respondents who answered ‘no’ to this initial question were assigned an outcome value

of 1. Respondents who answered ‘yes’ or ‘I’m not sure’ were then shown a list of reasons

for leaving the house and asked for each how frequently they would leave for that reason.

Each item was coded from 1 (‘never’) to 6 (‘2-3 times a day’) and averaged across the 13

7This imputation approach carries two risks. Compared to multiple imputation, it tends to under-estimate
the variance in the data, and lead to spurious rejection of the null hypothesis. By contrast, by imputing the
sample mean, rather than the within-treatment mean, risks attenuating the treatment effect, leading to
spurious acceptance of the null hypothesis. Given our findings, and the scarcity of our data, multiple
imputation would not change the interpretation of our findings, while these findings are also robust to the
use of alternative imputation strategies.

14
activities. There was no imbalance on ‘don’t know’ item responses by treatment, which

were imputed with the item sample mean.

Results

This section presents a summary of the results of the experiment. Full regression tables,

including alternative specifications, may be found in Appendix B.

Compliance with guidelines

Figure 1 gives the proportion of the guidelines shown to respondents that they intended to

comply with either ‘always’ or ‘almost always’, broken down by treatment condition

(coded as a six-level factor, with those who were in the control for both the ‘nudge’ and

‘plus’ components as the reference group), at both wave 1 (directly post-treatment) and

wave 2 (two weeks later).

From this, we can see that overall most of the treatment conditions had no effect, with

the exception of those who saw the beneficiary nudge, followed by the plus prompt. This

group intended to comply with a significantly higher proportion of the guidelines

applicable in the UK at the time of the experiment. However, by the two-week follow up,

there were no significant differences in compliance intentions, for any condition. Note

there are differences in baselines across wave 1 and 2 with the intercepts in the regression

decreasing from .77 to .32. This indicates a change in population behaviour between the

waves, possibly in response to reports about easing of lockdown.

15
 Figure 2: Proportion of the guidelines that respondents indented to comply with
always or almost always, by treatment allocation.

Intended frequency of leaving the house

Figure 2 below gives the effect of treatment condition on participants’ intention to leave

the house coded two ways: (1) as a binary set to 0 if they stated they did not intend to leave

the house at all in the coming week, and 1 if they said they did intend to leave the house

or weren’t sure; and (1) as a continuous variable representing the average frequency with

which they stated they intended to leave the house for a range of purposes, with 0

representing no intention to leave the house and 1 representing the intention to leave the

house 2-3 times a day across all the listed purposes.

16
 Again, we see limited effectiveness of all conditions, with the beneficiary nudge

condition overall as the most promising, and - consistent with our other outcome - the

beneficiary condition with the self-reflection ‘plus’ prompt being the only condition to

have a significant impact on intention to leave the house, compared to the control.8

Figure 2: Intended frequency of leaving the house, by treatment allocation.

Discussion

Our study tested the efficacy of nudge treatments to attract the attention of respondents in

an information-heavy environment of COVID-19. We find that the addition of nudge plus

increased intentions to comply directly following treatment. But when followed up two

8Note that the beneficiary plus result is significantly different (p <0.05, t >1.96), from the beneficiary only
result in two out of the four specifications in Table 1, with a third yielding a result close to significance (
<0.1, t >1.69), and three out of the four specifications in Appendix B, Table 2.

17
weeks later, participants reported no changed intentions going forward. The importance of

these findings is the lack of impact of classic nudges, except when combined with nudge

plus, and the lack of any persistence in changed intentions.

The key question from this research is why is there no treatment effect on intention

to change and actual behaviours when stimulating people to wish to change their behaviour

for the classic nudges. It seems that classic nudges are not well adjusted for the COVID

environment. There is much more promise for when a nudge is combined with a reflective

device, the nudge plus. Yet even this did not transfer to compliance. This is a familiar

problem with COVID interventions (e.g. Favero and Pedersen, 2020).

Information effects need to be considered. Our sample was recruited via Prolific.

Conducting the study online allowed for the research to be conducted where face-to-face

contact was not possible and as such was an appropriate method. However, it is possible

that participants may have been involved in other COVID-19 related studies available on

the platform around the time of our research. They may therefore have been exposed to

greater levels of information on COVID-19 and interventions which may have

inadvertently countered the aims of the messages in the present study.

Although this may have influenced the ability of the messages in our study to cut

through, this is also precisely the challenge faced by policy-makers hoping to nudge

people’s compliance. At the beginning of the pandemic, the government relied heavily on

nudge strategies (Sibony, 2020; Yates, 2020), amidst a multitude of messages being

directed at the public. Our study suggests, with all the other pressures on people’s

behaviour during a crisis, nudges may just be crowded out. The invocation to reflect might

be the one way governments can break through this information overload. Agencies might

18
be able to use reflection devices in the roll out of messaging and in encouragements for

citizens to think more slowly.

19
Acknowledgement

We acknowledge and thank the funder (blinded).

Author Contributions

(blinded)

Statement of Pre-registration

The study was pre-registered with the Center for Open Science (registration reference

withheld for blinding). Neither the data nor the materials have been made available on a

permanent third-party archive; requests for the data or materials can be sent via email to

the lead author.

Competing Interests Statement and Ethical Consideration

There are no competing interests. All studies were approved by the (blinded) Ethics

Committee.

20
 References

Allcott, H. (2011). Social norms and energy conservation. Journal of Public Economics,

95 (9-10), 1082– 1095.

Aronson, E. (1999). The power of self-persuasion. American Psychologist, 54 (11), 875.

American Psychological Association. (2020). Speaking of psychology: Coronavirus

anxiety (part 1).

Banerjee, S., & John, P. (2021). Nudge plus: Incorporating reflection into behavioural

public policy. Behavioural Public Policy, 1-16.

Barari, S., Caria, S., Davola, A., Falco, P., Fetzer, T., Fiorin, S., Hensel, L., Ivchenko,

A., Jachimowicz, J., King, G., et al. (2020). Evaluating covid-19 public health

messaging in Italy: Self-reported compliance and growing mental health

concerns. MedRxiv.

Boruchowicz, C. G., Bóo, F. L., Pfeifer, F. F., Russo, G. A., & Pacheco, T. S. (2020).

Are behaviorally informed text messages effective in promoting compliance with

covid-19 preventive measures? Evidence from an RCT in the city of São Paulo.

Brandon, A., Ferraro, P. J., List, J. A., Metcalfe, R. D., Price, M. K., & Rundhammer, F.

(2017). Do the effects of social nudges persist? theory and evidence from 38

natural field experiments (NBER Working Papers No. 23277). National Bureau

of Economic Research, Inc.

Cialdini, R. B., Demaine, L. J., Sagarin, B. J., Barrett, D. W., Rhoads, K., & Winter, P.

L. (2006a). Managing social norms for persuasive impact. Social Influence, 1(1),

3–15.

21
Cialdini, R. B., & Trost, M. R. (1998). Social influence: Social norms, conformity and

compliance. McGraw-Hill.

Damen, T. G., Müller, B. C., van Baaren, R. B., & Dijksterhuis, A. (2015). Re-

examining the agentic shift: The sense of agency influences the effectiveness of

(self) persuasion [pmid:26053303]. PloS one, 10 (6).

Derbyshire Police. (2020). Despite posts yesterday highlighting issues of people still

visiting the #peakdistrict despite government guidance, the message is still not

getting through. @derpoldroneunit have been out at beauty spots across the

county, and this footage was captured at #curbaredge last night. [Tweet].

https://twitter.com/derbyspolice/status/1243168931503882241

Duffy, B. (2020). UK still staying home and even coming to terms with lockdown.

Retrieved from: https://www.kcl.ac.uk/news/uk-still-staying-home-and-even-

coming-to-terms-with-lockdown

Ellson, A. (2020). Panic buyers fill trolleys with rice, pasta and hand gel. Retrieved from:

https://www.thetimes.co.uk/ article/shoppers-stockpiling-pasta-and-hand-gel-

urged-to-stop-panic-buying-z0tdpkvqz

Falco, P., & Zaccagni, S. (2020). Promoting social distancing in a pandemic: Beyond the

good intentions (OSF Preprints). Center for Open Science.

https://doi.org/10.31219/osf.io/a2nys

Farrow, K., Grolleau, G., & Mzoughi, N. (2020). When more is not better: Three

common mistakes in health messaging interventions. Journal of Health Politics,

Policy and Law, 45 (1), 143–152.

22
Favero, N., & Pedersen, M. J. (2020). How to encourage “togetherness by keeping apart”

amid COVID- 19? the ineffectiveness of prosocial and empathy appeals. Journal

of Behavioral Public Administration, 3 (2).

Gaudeul, A., & Kaczmarek, M. C. (2017). Does nudging intentions translate into action?

why nudging pledges to charities does not result in increased donations (Center

for European, Governance and Economic Development Research Discussion

Papers No. 318). University of Goettingen, Department of Economics.

Geldsetzer, P. (2020). Knowledge and perceptions of Covid-19 among the general public

in the United States and the United Kingdom: A cross-sectional online survey.

Annals of Internal Medicine.

Goldstein, N. J., Cialdini, R. B., & Griskevicius, V. (2008). A room with a viewpoint:

Using social norms to motivate environmental conservation in hotels. Journal of

Consumer Research, 35 (3), 472–482.

Gray, K., Ward, A. F., & Norton, M. I. (2014). Paying it forward: Generalized

reciprocity and the limits of generosity. Journal of Experimental Psychology:

General, 143 (1), 247.

Hallsworth, M., Chadborn, T., Sallis, A., Sanders, M., Berry, D., Greaves, F., Clements,

L., & Davies, S. C. (2016). Provision of social norm feedback to high prescribers

of antibiotics in general practice: A pragmatic national randomised controlled

trial. The Lancet, 387 (10029), 1743–1752.

Hayes, A. F., & Cai, L. (2007). Using heteroskedasticity-consistent standard error

estimators in OLS regression: An introduction and software implementation.

Behavior Research Methods, 39 (4), 709–722.

23
Hertwig R, Grüne-Yanoff T. (2017). Nudging and boosting: Steering or empowering

good decisions. Perspectives on Psychological Science, 12(6), 973-986.

Hoch, S. J., & Deighton, J. (1989). Managing what consumers learn from experience.

Journal of Marketing, 53 (2), 1–20.

Jarvis, J. (2020). Panic buyers swarm London supermarkets the minute they open during

coronavirus crisis. Retrieved from:

https://www.standard.co.uk/news/london/london-supermarkets- panic- buyers-

food- shop-first-thing-a4390321.html

John, P., Sanders, M., & Wang, J. (2019). A panacea for improving citizen behaviors?

introduction to the symposium on the use of social norms in public

administration. Journal of Behavioral Public Administration, 2 (2).

John, P. (2018). How best to nudge taxpayers? The impact of message simplification and

descriptive social norms on payment rates in a central London local authority.

Journal of Behavioral Public Administration, 1 (1), 1–11.

https://doi.org/10.30636/jbpa.11.10

Kardes, F. R., Cronley, M. L., Pontes, M. C., & Houghton, D. C. (2001a). Down the

garden path: The role of conditional inference processes in self-persuasion.

Journal of Consumer Psychology, 11 (3), 159–168.

Kardes, F. R., Cronley, M. L., Pontes, M. C., & Houghton, D. C. (2001b). Down the

garden path: The role of conditional inference processes in self-persuasion.

Journal of Consumer Psychology, 11 (3), 159–168.

Kogut, T., & Ritov, I. (2005). The “identified victim” effect: An identified group, or just

a single individual? Journal of Behavioral Decision Making, 18 (3), 157–167.

24
Larkin, C., Sanders, M., Andresen, I., & Algate, F. (2019). Testing local descriptive

norms and salience of enforcement action: A field experiment to increase tax

collection. Journal of Behavioral Public Administration, 2 (1).

https://doi.org/10.30636/jbpa.21.54

Leonard, T. C. (2008). Richard H. Thaler, Cass R. Sunstein, Nudge: Improving decisions

about health, wealth, and happiness. Yale University Press. New Haven, CT.

Liberman, A., & Chaiken, S. (1992). Defensive processing of personally relevant health

messages. Personality and Social Psychology Bulletin, 18 (6), 669–679.

Lunn, P. D., Belton, C. A., Lavin, C., McGowan, F. P., Timmons, S., & Robertson, D.

(2020). Using behavioural science to help fight the coronavirus: A rapid,

narrative review. Journal of Behavioral Public Administration, 3 (1).

Lunn, P. D., Belton, C. A., Lavin, C., McGowan, F. P., Timmons, S., & Robertson, D. A.

(2020). Using behavioral science to help fight the coronavirus. Journal of

Behavioral Public Administration, 3 (1).

McAlaney, J., Bewick, B., & Hughes, C. (2011). The international development of the

‘social norms’ approach to drug education and prevention. Drugs: Education,

Prevention and Policy, 18 (2), 81– 89.

McKeever, A. (2020). Panic is spreading like the coronavirus. here’s why we evolved to

feel anxiety. Retrieved from: https://www. nationalgeographic.co.uk/history-and-

civilisation/2020/03 /panic-spreading-coronavirus-heres-why-we-evolved-feel-

anxiety

Müller, B. C., Ritter, S. M., Glock, S., Dijksterhuis, A., Engels, R. C., & van Baaren, R.

B. (2016a). Smoking-related warning messages formulated as questions

25
 positively influence short-term smoking behaviour. Journal of Health

Psychology, 21 (1), 60–68.

Mussweiler, T., & Neumann, R. (2000). Sources of mental contamination: Comparing

the effects of self- generated versus externally provided primes. Journal of

Experimental Social Psychology, 36 (2), 194–206.

National Library of Medicine. (2020). Nudges to improve health behaviors that limit

covid-19 spread. Retrieved from:

https://clinicaltrials.gov/ct2/show/NCT04379375

Palan, S., & Schitter, C. (2018). Prolific Academic —a subject pool for online

experiments. Journal of Behavioral and Experimental Finance, 17, 22–27.

Peer, E., Brandimarte, L., Samat, S., & Acquisti, A. (2017). Beyond the Turk:

Alternative platforms for crowdsourcing behavioral research. Journal of

Experimental Social Psychology, 70, 153–163.

Perkins, H. W., & Berkowitz, A. D. (1986). Perceiving the community norms of alcohol

use among students: Some research implications for campus alcohol education

programming. International Journal of the Addictions, 21 (9-10), 961–976.

Pfattheicher, S., Nockur, L., Böhm, R., Sassenrath, C., & Petersen, M. B. (2020). The

emotional path to action: Empathy promotes physical distancing during the

Covid-19 pandemic.

Sasaki, S., Kurokawa, H., & Ohtake, F. (2020). Short-term responses to nudge-based

messages for preventing the spread of Covid-19 infection: Intention, behavior,

and life satisfaction (Discussion Papers in Economics and Business No. 20-11).

26
 Osaka University, Graduate School of Economics. Retrieved from:

https://EconPapers.repec.org/RePEc:osk:wpaper:2011

Sibony, A.-L. (2020). The UK Covid-19 response: A behavioural irony? European

Journal of Risk Regulation, 1–8.

Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2018). False-positive citations [PMID:

29592640]. Perspectives on Psychological Science, 13 (2), 255–259.

https://doi.org/10.1177/1745691617698146

Small, D. A., & Loewenstein, G. (2003). Helping a victim or helping the victim:

Altruism and identifiability. Journal of Risk and Uncertainty, 26 (1), 5–16.

Small, D. A., Loewenstein, G., & Slovic, P. (2007). Sympathy and callousness: The

impact of deliberative thought on donations to identifiable and statistical victims.

Organizational behavior and human decision processes, 102 (2), 143–153.

Soofi, M., Najafi, F., & Karami-Matin, B. (2020). Using insights from behavioral

economics to mitigate the spread of Covid-19. Applied Health Economics and

Health Policy, 18 (3), 345–350.

Stok, F. M., de Vet, E., de Ridder, D. T., & de Wit, J. B. (2016). The potential of peer

social norms to shape food intake in adolescents and young adults: A systematic

review of effects and moderators. Health Psychology Review, 10 (3), 326–340.

Stoker, G., & John, P. (2019). Rethinking the role of experts and expertise in behavioural

public policy. Policy and Politics, 47 (1), 209–226.

Thøgersen, J. (2006). Norms for environmentally responsible behaviour: An extended

taxonomy. Journal of Environmental Psychology, 26 (4), 247–261.

27
Vainre, M., Aaben, L., Paulus, A., Koppel, H., Tammsaar, H., Telve, K., Koppel, K.,

Beilmann, K., & Uusberg, A. (2020). Nudging towards tax compliance: A

fieldwork-informed randomised controlled trial. Journal of Behavioral Public

Administration, 3. https://doi.org/10.30636/jbpa.31. 84

Västfjäll, D., Slovic, P., Mayorga, M., & Peters, E. (2014). Compassion fade: Affect and

charity are greatest for a single child in need. PLOS ONE, 9 (6), 1–10.

https://doi.org/10.1371/journal. pone.0100115

Weinberg, J. (2020). Can political trust help to explain elite policy support and public

behaviour in times of crisis? evidence from the United Kingdom at the height of

the 2020 coronavirus pandemic. Political Studies, 0032321720980900.

West, R., Michie, S., Rubin, G. J., & Amlôt, R. (2020). Applying principles of behaviour

change to reduce Sars-Cov-2 transmission. Nature Human Behaviour, 4 (5), 451–

459. https://doi.org/10.1038/s41562-020-0887-9

Wiepking, P., & Heijnen, M. (2011). The giving standard: Conditional cooperation in the

case of charitable giving. International Journal of Nonprofit and Voluntary

Sector Marketing, 16 (1), 13– 22.

Yates, T. (2020). Why is the government relying on nudge theory to fight coronavirus?

Retrieved from: https://www.theguardian.com/commentisfree/2020 /mar/13/why-

is-the- government-relying-on-nudge-theory-to-tackle-coronavirus

28
 Appendices

Appendix A: Outcome scales

Compliance with guidelines

Items on this scale were based on Weinberg (2020) and amended to cover official and

unofficial guidelines in place in the UK at the time of the study, and to include seven

response levels rather than five. The scale was as follows:

Please indicate how often you intend to do the following activities in the next week. Please

be completely honest - your responses are anonymous.

Scale from ‘never’ (1) to ‘always’ (7) plus I don’t know (items were reverse coded where

necessary).

• Using the NHS for non-critical illnesses

• Working from home unless essential to do otherwise

• Following hygiene precautions like washing hands for 20 seconds

• Social distancing from others apart from those in your household

• Volunteering for the NHS

• Only going outside to exercise once a day, to collect essential food and

medicine or care for a vulnerable person

• Stockpiling food and other household goods

• Travelling to see friends and family

• Wash hands more frequently and for longer than normal

• Inform others if you develop symptoms of COVID-19 no matter how mild

• Exercising close to your house (i.e. not travelling to exercise)

• Disinfect frequently touched objects and surfaces

29
Intention to leave the house

Items were developed based on a scale by Barari et al., 2020 and amended for the UK

context. The 5-point scale was created by the authors for this study to ascertain the

frequency of actions. The scale was as follows:

Do you intend to leave the home in the next 5 days?

• Yes

• No

• I’m not sure

(If Yes/Not sure) What are the reasons that you will leave your home in the next 5 days?

Please be completely honest - your responses are anonymous.

Scale from ‘Not at all in the next 5 days’ (1) to ‘Multiple times a day in the next 5 days’

(5) plus I don’t know.

• Going to work

• Walking a pet

• Doing physical activity

• Getting food for yourself or household

• Going to the pharmacy

• Going to the hospital / to receive medical treatment

• Meeting friends or relatives

• For a change of scenery

• Boredom / just to leave the house

• Getting an adrenaline rush (from breaking the law)

• Exercising my freedoms

30
• Taking care of a dependent or someone who is vulnerable

• Other (please specify)

31
Appendix B: Regression tables

Intended compliance with the regulations

This section provides regression tables for the analysis used in the paper, as well as a range

of alternative specifications. Table 1 gives the results for our preferred outcome and

treatment specification, which is the proportion of times that a respondent answered either

‘always’/‘almost always’ or ‘never’/‘almost never’ (for reverse-coded items) to the items

in the compliance scale. Treatment assignment is coded as a factor variable with six levels,

where the reference category is those who were assigned to both the control (nudge) and

control (plus) conditions.

We present four models in the table, which are (1) model using wave 1 (immediate

post-treatment) data and treatment allocation only; (2) wave 1 data with covariates; (3)

wave 2 data with covariates; and (4) a panel specification using wave 1 and wave 2 data

(with the item-level sample mean imputed for wave 2 non-responders), clustered at the

level of the individual.

Table 2 presents the same four models, but with the treatment coded as a 3 x 2

factorial, to assess the interaction effects, while Table 3 and Table 4 repeat the above

analysis but with the outcome coded as a continuous variable from 1 to 7.

Intention to leave home

In this section we present regression tables for the second outcome: how frequently the

respondent intended to leave home. Table 5 presents models for our preferred treatment

specification (coded as a six-level factor), with the outcome coded as either a binary (set

to 1 if they intended to leave the house at all, and 0 otherwise) and as a continuous variable,

with six levels corresponding to the range of frequencies of leaving the house in the scale,

from ‘never’ (1) to ‘2-3 times a day’ (6). This outcome has been averaged across the items

32
in the scale and normalised to be between 0 and 1 for ease of graphical presentation. ‘Don’t

know’ responses were imputed with the sample mean for that item. The table presents four

models, corresponding to (1) a binary outcome and no covariates; (2) a binary outcome

with covariates; (3) a continuous outcome and no covariates; and (4) a continuous outcome

with covariates. Table 6 gives the same four models with the treatment specified as a 2 x

3 factorial.

33
 W1 (clean) W1 (covariates) W2 (covariates) W1/W2 panel (covariates)

(Intercept) 0.75∗∗∗ 0.73∗∗∗ 0.33∗∗∗ 0.75∗∗∗

(0.01) (0.01) (0.01) (0.01)

Control + Plus 0.00 0.00 −0.02 −0.01

(0.02) (0.02) (0.01) (0.01)

Norm + Control 0.02 0.02 −0.02+ −0.00

(0.02) (0.02) (0.01) (0.01)

Norm + Plus 0.01 0.01 −0.01 −0.00

(0.02) (0.02) (0.01) (0.01)

Beneficiary + Control −0.01 −0.00 −0.00 −0.00

(0.02) (0.02) (0.01) (0.01)

Beneficiary + Plus 0.03+ 0.04∗ 0.01 0.02+

(0.02) (0.02) (0.01) (0.01)

Wave 2 −0.45∗∗∗

(0.01)

Covariates No Yes Yes Yes

R2 0.00 0.04 0.01 0.65

Adj. R2 0.00 0.03 −0.00 0.65

Num. obs. 1480 1476 1246 2722

RMSE 0.19 0.18 0.13 0.16

N Clusters 1475
∗∗∗ ∗∗ ∗ +
p < 0.001; p < 0.01; p < 0.05; p < 0.1. Covariates are age, gender and ethnicity.

Table 1: Analysis of intended compliance with guidelines, with treatments coded separately

34
 W1 (clean) W1 (covariates) W2 (covariates) W1/W2 panel (covariates)

(Intercept) 0.75∗∗∗ 0.73∗∗∗ 0.33∗∗∗ 0.75∗∗∗

(0.01) (0.01) (0.01) (0.01)

Nudge: Norm 0.02 0.02 −0.02+ −0.00

(0.02) (0.02) (0.01) (0.01)

Nudge: Beneficiary −0.01 −0.00 −0.00 −0.00

(0.02) (0.02) (0.01) (0.01)

Plus 0.00 0.00 −0.02 −0.01

(0.02) (0.02) (0.01) (0.01)

Norm x Plus −0.01 −0.01 0.02 0.01

(0.02) (0.02) (0.02) (0.02)

Beneficiary x Plus 0.04 0.04 0.02 0.03∗

(0.02) (0.02) (0.02) (0.02)

Wave 2 −0.45∗∗∗

(0.01)

Covariates No Yes Yes Yes

R2 0.00 0.04 0.01 0.65

Adj. R2 0.00 0.03 −0.00 0.65

Num. obs. 1480 1476 1246 2722

RMSE 0.19 0.18 0.13 0.16

N Clusters 1475
∗∗∗ ∗∗ ∗ +
p < 0.001; p < 0.01; p < 0.05; p < 0.1. Covariates are age, gender and ethnicity.

Table 2: Analysis of intended compliance with guidelines, with factorial treatments

35
 W1 (clean) W1 (covariates) W2 (covariates) W1/W2 panel (covariates)

(Intercept) 6.17∗∗∗ 6.08∗∗∗ 5.14∗∗∗ 6.14∗∗∗

(0.04) (0.05) (0.03) (0.04)

Control + Plus 0.00 −0.00 −0.03 −0.02

(0.06) (0.06) (0.03) (0.04)

Norm + Control 0.04 0.04 −0.04 −0.00

(0.06) (0.06) (0.03) (0.04)

Norm + Plus 0.02 0.01 −0.07+ −0.03

(0.07) (0.06) (0.04) (0.04)

Beneficiary + Control −0.00 0.00 −0.04 −0.02

(0.06) (0.06) (0.04) (0.04)

Beneficiary + Plus 0.09 0.09 −0.03 0.04

(0.06) (0.06) (0.04) (0.04)

Covariates No Yes Yes Yes

R2 0.00 0.03 0.01 0.47

Adj. R2 −0.00 0.02 0.01 0.47

Num. obs. 1480 1476 1246 2722

RMSE 0.70 0.69 0.36 0.56

N Clusters 1475

∗∗∗ ∗∗ ∗ +
p < 0.001; p < 0.01; p < 0.05; p < 0.1. Covariates are age, gender and ethnicity.

Table 3: Analysis of intended compliance with guidelines, with treatments coded separately, and continuous
outcome

36
 W1 (clean) W1 (covariates) W2 (covariates) W1/W2 panel (covariates)

(Intercept) 6.17∗∗∗ 6.08∗∗∗ 5.14∗∗∗ 6.14∗∗∗

(0.04) (0.05) (0.03) (0.04)

Nudge: Norm 0.04 0.04 −0.04 −0.00

(0.06) (0.06) (0.03) (0.04)

Nudge: Beneficiary −0.00 0.00 −0.04 −0.02

(0.06) (0.06) (0.04) (0.04)

Plus 0.00 −0.00 −0.03 −0.02

(0.06) (0.06) (0.03) (0.04)

Norm x Plus −0.02 −0.03 0.01 −0.01

(0.09) (0.09) (0.05) (0.06)

Beneficiary x Plus 0.09 0.09 0.04 0.07

(0.09) (0.09) (0.05) (0.06)

Covariates No Yes Yes Yes

R2 0.00 0.03 0.01 0.47

Adj. R2 −0.00 0.02 0.01 0.47

Num. obs. 1480 1476 1246 2722

RMSE 0.70 0.69 0.36 0.56

N Clusters 1475

∗∗∗ ∗∗ ∗ +
p < 0.001; p < 0.01; p < 0.05; p < 0.1. Covariates are age, gender and ethnicity.

Table 4: Analysis of intended compliance with guidelines, with factorial treatments and continuous outcome

37
 Binary (clean) Binary (covariates) Continuous (clean) Continuous (covariates)

(Intercept) 0.68∗∗∗ 0.73∗∗∗ 0.17∗∗∗ 0.19∗∗∗

(0.03) (0.03) (0.01) (0.01)

Control + Plus −0.03 −0.04 −0.01 −0.01

(0.04) (0.04) (0.01) (0.01)

Norm + Control −0.03 −0.04 −0.01 −0.01

(0.04) (0.04) (0.01) (0.01)

Norm + Plus −0.03 −0.03 −0.02 −0.02

(0.04) (0.04) (0.01) (0.01)

Beneficiary + Control −0.08+ −0.08+ −0.02 −0.02

(0.04) (0.04) (0.01) (0.01)

Beneficiary + Plus −0.08+ −0.09∗ −0.02∗ −0.03∗

(0.04) (0.04) (0.01) (0.01)

Covariates No Yes No Yes

R2 0.00 0.02 0.00 0.02

Adj. R2 0.00 0.01 0.00 0.02

Num. obs. 1480 1476 1480 1476

RMSE 0.48 0.48 0.13 0.13

∗∗∗ ∗∗ ∗ +
p < 0.001; p < 0.01; p < 0.05; p < 0.1. Covariates are age, gender and ethnicity.

Table 5: Analysis of intention to leave the house, with treatments coded separately

38
 Binary (clean) Binary (covariates) Continuous (clean) Continuous (covariates)

(Intercept) 0.68∗∗∗ 0.73∗∗∗ 0.17∗∗∗ 0.19∗∗∗

(0.03) (0.03) (0.01) (0.01)

Nudge: Norm −0.03 −0.04 −0.01 −0.01

(0.04) (0.04) (0.01) (0.01)

Nudge: Beneficiary −0.08+ −0.08+ −0.02 −0.02

(0.04) (0.04) (0.01) (0.01)

Plus −0.03 −0.04 −0.01 −0.01

(0.04) (0.04) (0.01) (0.01)

Norm x Plus 0.03 0.04 0.00 0.00

(0.06) (0.06) (0.02) (0.02)

Beneficiary x Plus 0.03 0.03 0.00 0.00

(0.06) (0.06) (0.02) (0.02)

Covariates No Yes No Yes

R2 0.00 0.02 0.00 0.02

Adj. R2 0.00 0.01 0.00 0.02

Num. obs. 1480 1476 1480 1476

RMSE 0.48 0.48 0.13 0.13

∗∗∗ ∗∗ ∗ +
p < 0.001; p < 0.01; p < 0.05; p < 0.1. Covariates are age, gender and ethnicity.

Table 6: Analysis of intention to leave the house, with factorial treatments

39
Appendix C: Pre-registration text submitted to OSF:

https://osf.io/mzyj9/?view_only=f85f707c010e4002bb3f0d27625d2cb7

C.1 Hypotheses
H1: Participants who viewed either the identifiable victim or classic social norm

message would subsequently report a greater intention to comply with public health

guidelines than those participants in the control group.

H2: Participants who are asked to complete a self-persuasion task will subsequently report

a greater intention to comply with public health guidelines than those who do not complete

a self-persuasion task.

We will also examine, through an interaction, whether some social norm nudges are more

susceptible to reflective thinking (through self-persuasion) than others. We do not have a strong

prior opinion as to whether there will be a difference and, if so, what that difference will be.

C.2 Design plan

C.2.1 Study type

Experiment - A researcher randomly assigns treatments to study subjects, this includes field or

lab experiments. This is also known as an intervention experiment and includes randomized

controlled trials.

C.2.2 Blinding
For studies that involve human subjects, they will not know the treatment group to which they

have been assigned.

C.2.3 Is there any additional blinding in this study?

No response

C.2.4 Study design

The project is a randomised survey experiment involving two rounds of randomisation and is a

between subjects design.

40
 The steps are as follows (see also flow diagram):

1) All participants will complete initial questions on their awareness of COVID-19, level of

concern and adherence to guidelines.

2) They will then be randomly allocated to one of 3 social norm conditions:

Control: Mock up of a news article with information on COVID-19 guidelines.

Classic social norm (T1): Mock up of a news article with true figures about adherence to

elements of the public health guidelines.

Identifiable victim (T2): Mock up of a news article with a story about someone who is

vulnerable to COVID-19.

3) Participants will then be randomly allocated to complete a self-persuasion task or will see
no task.

4) All participants will complete the final questionnaires on intentions to adhere to guidelines.

See “Testing the effectiveness of social norm messages in affecting adherence to public

health guidelines”, 2020 for the flow diagram of the study.

C.2.5 Randomization
Qualtrics survey software will randomly allocate participants to conditions (keeping groups

roughly even in size) using simple randomisation techniques. This will occur twice: once to

allocate to a social norm mock news article and once to allocate to a self-persuasion task or not.

C.2.6 Sampling Plan

C.2.7 Existing Data

Registration prior to creation of data

41
C.2.8 Explanation of existing data
N/A

C.2.9 Data collection procedures

Participants will be recruited through Prolific. Prolific is a dedicated online experiment

platform with 70,000 participants internationally, including UK representative samples and is

considered one of the more robust platforms for this type of research (Palan & Schitter, 2018;

Peer et al., 2017). Participants will be paid £0.75 for a 7 minute study which is deemed ‘fair’

by Prolific. Participants must be based in the UK (given public health guidelines differ between

countries the focus here is on the UK) and be aged over 18 years old (due to restrictions on

Prolific Academic for recruiting participants).

C.2.10 Sample size

Approx. 1,500 participants is the target sample size.

C.2.11 Sample size rationale

Our goal was to establish reasonable power given time and financial constraints.

C.2.12 Stopping rule

Prolific will remove the advert for the study when the desired number of participants are

recruited.

C.3 Variables

C.3.1 Manipulated variables

The initial manipulation is in social norms message, people will see either:

1) A control (information on health guidelines)

2) A ’classic’ social norm (statistics on adherence to health guidelines)

3) A identifiable beneficiary message (a quote from a vulnerable person about how they are

coping with COVID-19)

The second manipulation is in self-persuasion (Nudge Plus):

42
1) Self-persuasion written task

2) No self-persuasion written task

C.3.2 Measured variables

Initial measures:

• Understanding of COVID-19 (several true or false questions)

• Level of concern for self and the country (measured separately on a scale of 1-4, 1 being

‘not at all concerned’ and 4 being ’very concerned’ there is a 5th ‘I don’t know’ option).

• Level of concern questions (multiple questions on a scale from 1-7, 1 being ‘strongly

disagree’ and 7 being ‘strongly agree’).

• Length of time expecting to follow social distancing and public health guidelines (options

are ‘another week’ another 2 weeks’ ‘another month’ ‘a few more month’ ‘more than 6

months more’ and ‘I don’t know’.

• Adherence to public health guidelines (multiple questions on a scale from 1-7, 1 being

‘never’ and 7 being ‘always’ with an additional option to select ‘I don’t know).

Endline measures:

• Do you need to leave the home in the next 5 days? (Yes, No, I’m not sure).

• What are the reasons that you will leave your home in the next 5 days (check all options
that apply)

• Please indicate how often you intend to do the following activities next week (multiple

questions on a scale of 1-7, 1 being ‘never’ and 7 being ‘always’ with an additional ‘I

don’t know’ option

43
 Demographics (e.g. age, gender)

C.3.3 Indices
We will generate mean scores for the measures which include multiple items including

adherence, intention to adhere, knowledge of COVID-19 and level of concern. In places this

will require the reverse coding of some items.

C.4 Analysis Plan

C.4.1 Statistical models

We will use regression analysis.

We will regress the intention to adhere to guidelines on a vector of baseline adherence to

guidelines for each of the social norms (control, classic social norm, identifiable victim) and

each of the nudge plus interventions (self-persuasion, no self-persuasion). We will also run an

interaction using the same models but adding the interaction between social norm and nudge

plus condition.

C.4.2 Transformations
A dummy code for intention to go out in the next 5 days (with not going out as the reference)

and gender (with male as the reference) may be created.

C.4.3 Inference criteria

We will use the standard p<0.05 criteria for determining if the results are significantly different

from those expected if the null hypothesis were correct. We will also use R-squared and effect

sizes.

C.4.4 Data exclusion

We will include a manipulation check to ensure that people have read the written information.

This will consist of 4 questions and participants will be removed if they get 2 or more incorrect.

Outliers will be included in the analysis.

44
C.4.5 Missing data

If a subject does not complete a questionnaire their data will be removed from the analysis.

C.4.6 Exploratory analysis

We may look at differences in demographic traits (particularly age and gender) and how these

affect intentions to adhere.

Level of concern and understanding for the COVID-19 virus may also affect intentions to

adhere (and current adherence).

45
Appendix D: Covid-19 context

Date (2020) The UK lockdown (2020) COVID-19 study New COVID-

19cases (and
deaths)

Feb The Government advises on a range of

social distancing restrictions as well as self-
isolation and quarantine for those with
symptoms.

10-Feb Strengthened powers are issued to

quarantine people against their will if
necessary.

01-Feb Matt Hancock warns that over 70s would be

asked to ‘self-isolate’ within weeks.

16-Mar Boris Johnson advises against ‘nonessential’

travel and contact with others – this is just a
suggestion at this point.

20-Mar Boris Johnson orders all pubs, cafes,

restaurants, bars and gyms to close and the
Chancellor announces the taxpayer will
meet 80% of the wages of employees.

23-Mar Boris Johnson announces a nationwide

lockdown granting strengthened powers to
enforce social distancing (many of which
come into effect on 26 May). People are
only allowed to leave the house for set
reasons and 1.5 million people are told to
‘shield’ themselves. This is not currently a
legal requirement but is widely adhered to
from the get go.

Continued on next page

46
 Date (2020) The UK lockdown (2020) COVID-19 study New COVID-19cases
(and deaths)

The Coronavirus Act 2020 was passed

25 March (26 March granting powers related to the 23 March
police’s powers come announcement.
into effect)

05-Apr Queen makes a rare broadcast about

COVID-19.

09-Apr 4,852 (1,116)

Pilot goes live and is
closed

Easter weekend
10 April – 13 April

Apr-11 3,577 (843)

Pilot goes live and is
closed

15-Apr There is growing speculation about an exit

plan and Keir Starmer calls for the
Government to publish one.

16-Apr Dominic Raab announces lockdown will be

in place for “at least the next three weeks”.

22-Apr In a Commons statement Matt Hancock tells

MPs that “we are at the peak” although says
that social distancing cannot yet be relaxed.

23-Apr Wave 1 goes live 5,143 (682)

24-Apr Wave 1 is closed 4,973 (1,010)

Continued on next page

47
Date (2020) The UK lockdown (2020) COVID-19 study
New COVID-19cases
(and deaths)

27-Apr Boris Johnson makes his first public

statement since his return to work following
hospitalisation with COVID-19. He says that
the UK is in a “moment of maximum risk”
but suggests that “we are not beginning to
turn the tide”.

07-May Wave 2 goes live 3,767 (579)

09-May Wave 2 is closed 2,150 (275)

10-May After speculation over previous days, Boris

Johnson announces his three-step plan to get
out of lockdown (the first of which take
effect on 13 May).

10-May The UK updates its message from “stay at

home, protect the NHS, save lives” to “stay
alert, control the virus, save lives”.

13-May People can meet with one person outside the

home, providing they remain in a public
outdoor space and social distance.

22-May The Daily Mirror and The Guardian publish

articles relating to Dominic Cummings’
(Senior Advisor to Prime Minister, Boris
Johnson) alleged breach of the
Government’s lockdown rules back in
March by traveling from London to Durham.

Continued on next page

48
Date (2020) The UK lockdown (2020) COVID-19 study
New COVID-19

cases (and deaths)

24-May The Observer and Sunday Mirror allege that

Dominic Cummings made a second trip
Durham in April.

25-May After mounting pressure, Dominic

Cummings gives a press conference about
his alleged breach of lockdown rules.

25-May Boris Johnson outlines plans to open

outdoor markets and car showrooms from 1
June and other non-effectual shops from 15
June.

28-May Boris Johnson announces that people can

meet in groups of 6 outdoors (from 1
June).

30-May Those shielding can now leave the house

(from 1 June).

Schools and nursery start phased opening.

01-Jun
Table 7: The UK COVID-19 context at the time of the research

49