Government trust, perceptions of COVID-19 and behaviour change: cohort surveys, Singapore

Abstract Objective To evaluate how public perceptions and trust in government communications affected the adoption of protective behaviour in Singapore during the coronavirus disease 2019 (COVID-19) pandemic. Methods We launched our community-based cohort to assess public perceptions of infectious disease outbreaks in mid-2019. After the first case of COVID-19 was reported in Singapore on 23 January, we launched a series of seven COVID-19 surveys to both existing and regularly enrolled new participants every 2 weeks. As well as sociodemographic properties of the participants, we recorded changing responses to judge awareness of the situation, trust in various information sources and perceived risk. We used multivariable logistic regression models to evaluate associations with perceptions of risk and self-reported adopted frequencies of protective behaviour. Findings Our cohort of 633 participants provided 2857 unique responses during the seven COVID-19 surveys. Most agreed or strongly agreed that information from official government sources (99.1%; 528/533) and Singapore-based news agencies (97.9%; 522/533) was trustworthy. Trust in government communication was significantly associated with higher perceived threat (odds ratio, OR: 2.2; 95% confidence interval, CI: 1.6–3.0), but inversely associated with perceived risk of infection (OR: 0.6; 95% CI: 0.4–0.8) or risk of death if infected (OR: 0.6; 95% CI: 0.4–0.9). Trust in government communication was also associated with a greater likelihood of adopting protective behaviour. Conclusion Our findings show that trust is a vital commodity when managing an evolving outbreak. Our repeated surveys provided real-time feedback, allowing an improved understanding of the interplay between perceptions, trust and behaviour.


Introduction
The emergence of coronavirus disease 2019  in China in December 2019 resulted in a global and rapidly rising number of cases and deaths. 1 Many countries responded with restrictive measures of various degrees to suppress transmission. 2,3 However, because non-compliance and outright protests against more restrictive measures have been widespread, 4 understanding the factors that facilitate public adherence to such interventions is important.
Singapore, a city-state in South-East Asia and a global travel hub, reported its first imported COVID-19 case in a traveller from China on 23 January 2020. 5 Several clusters of local transmission followed, but were successfully contained without widespread use of socially disruptive measures. 6 Singapore subsequently experienced a second wave of infections from imported cases, rising local transmission and large outbreaks in migrant worker dormitories, with the number of confirmed cases exceeding 926 by 31 March 2020. 7 As cases continued to increase, the government announced the implementation of a so-called circuit-breaker to interrupt CO-VID-19 transmission on 4 April 2020. This package of measures and restrictions, combined with penalties for non-compliance, 8,9 was equivalent to a partial lockdown. Originally set to last from 7 April to 4 May 2020, the restrictions were extended until 1 June 2020; a phased return to pre-pandemic life was instigated after this date.
An emerging global issue is how to effectively communicate and ensure the adoption of public health recommendations and containment measures. 10 During the Ebola virus disease outbreak in Liberia in 2014, proper risk communication and health promotion encouraged community support and involvement, playing an important role in the adoption of key protective behaviour. 11 However, studies are lacking that demonstrate the role of effective communication and trust in the perceptions of, and protective behaviour adopted in response to, COVID-19. We need tools to rapidly assess public perceptions, not just at singular time points but across multiple time points as the pandemic evolves. We describe insights from a cohort-based study to record changes in public opinions and adopted behaviour in Singapore during the COVID-19 outbreak. We examine how perceptions of the disease and the local situation, as well as trust in the government's communications about the outbreak, affect the adoption by the general public of recommended behaviour.

Study design
Before the COVID-19 pandemic, we launched the communitybased cohort study "Strengthening our community's resilience against threats from emerging infections" (SOCRATEs) 12 to periodically assess public knowledge and perceptions of common infectious diseases (e.g. dengue and tuberculosis) and previous outbreaks. Launched on 27 June 2019, our study was designed so that the cohort could be rapidly re-surveyed in the event of a new outbreak. Consenting participants aged 16 years Public perceptions of COVID-19 and behaviour change, Singapore Vanessa W Lim et al. Fig. 1. Timeline of COVID-19 pandemic, imposed measures and seven cohort surveys to assess public perceptions and behaviour, Singapore, January-April 2020 19 Week 11 Week 12 Week 13 Week 14 Week 15 Week Participants were enrolled via a combination of door-to-door recruitment of the general population and self-referred participants, who received study information via word-of-mouth and social media posts. For door-to-door recruitment, we divided Singapore into five geographical zones and randomly selected an equal number of residential buildings within each zone using verified postal codes (available in data repository). 13 We set a maximum of four participants per household. From the launch of the study until 22 January 2020, we conducted face-toface interviews during which participants provided responses to a baseline survey and a survey about previous outbreaks of infectious diseases in Singapore. We replaced the initial outbreak survey with the first COVID-19 survey on 24 January 2020, a day after the first confirmed case of COVID-19 in Singapore (Fig. 1). From this date onwards, we conducted all surveys by messaging a link to participants' mobile phones, and replaced door-to-door recruitment with video interviews with selfreferred participants. To record changes in participants' perceptions of, and protective behaviour adopted in response to, the pandemic, we issued our outbreak survey seven different times until 29 April 2020 ( Table 1). The same basic questions (available in data repository) 13 were included in all seven surveys, with the addition of single-use questions at certain times to (i) assess awareness of current developments and (ii) gauge responses to, and support for, government initiatives (data repository). 13 Recruitment continued throughout the surveys, and those who missed any particular survey could participate in the following survey. Newly recruited participants provided responses to both a baseline survey and the current COVID-19 survey.
All surveys were available in all three key local languages (English, Mandarin and Malay).

Data analysis
Because the start and end dates of epidemiological week numbers do not correspond to the start and end dates of surveys, we aggregated responses by epidemiological week number (where week 4 is 19-25 January 2020) based on the timestamps at submission to illustrate changes in perception and behaviour as the outbreak evolved. We calculated the proportion who selected a particular response as a percentage of the total number of responses received for that question (which varied as a result of different questions being included at different surveys, as well as a varying number of missing responses to any particular question within a particular survey), and compared these using χ 2 tests. We used multivariable logistic regression models to evaluate associations between various factors and perceptions of (i) the threat posed by COVID-19, (ii) the risk of infection and (iii) the risk of death upon contracting COVID-19, and between various factors and self-reported adopted frequencies of behaviour that may mitigate risk. We assessed and adjusted for various factors using a multilevel modelling framework, with a random intercept term to model the effects of participants' behaviour.
We considered sociodemographic properties and baseline survey responses regarding degree of trust in information from various sources to be static (level 1) variables, and other survey responses (e.g. awareness of the local situation) to be time-varying (level 2) variables. To investigate whether the stage of the epidemic influenced perceptions and self-reported behaviour, we also included variables for the number of weeks elapsed since the first case and the natural log of the numbers of new cases and cumulative deaths reported up to the day before each survey response.
Finally, we tested the hypothesis that respondents who expressed greater trust in the government's communications on COVID-19 would be more likely to adopt the recommended behaviour. As govern-ment recommendations on protective behaviour evolved during the outbreak, we stratified the effect over three key time periods: (i) before the Disease Outbreak Response System Condition 14 was upgraded from yellow to orange, when messaging largely focused on personal hygiene and socially responsible behaviour to adopt if infected; (ii) after outbreak condition orange was declared on 7 February 2020, when avoidance of largescale gatherings was recommended; 15 and (iii) after a key speech on 4 April 2020 by the Prime Minister of Singapore regarding the circuit-breaker period. In this speech, the Prime Minister not only articulated plans for a partial lockdown, but also announced a change in guidance for facemasks (to be worn in public by all persons, superseding previous guidance that facemasks were mainly for those who were unwell). 16,17 Results were expressed as odds ratios (ORs) with 95% confidence intervals (CIs).
We applied multiple imputation for several exposure variables with missing responses (ranging from 0.2%; 5/2857 to 16.6%; 474/2857) based on each participant's most recent survey response and the mean response for that question (data repository). 13 We performed all analyses using Stata software version 15.0 (StataCorp, College Station, United States of America).

Ethics
The Ethics Review Board of the National Healthcare Group, Singapore, approved this study (reference no. 2018/01203).

Results
We depict the periods spanned by the seven pandemic surveys in relation to the rising number of cases and timeline of measures implemented in Singapore, as well as epidemiological week numbers, in Fig. 1. We analysed a total of 2857 survey responses from the seven surveys, in a cohort of 633 participants from 540 households ( Table 2). We enrolled 33.2% (210/633) of the cohort participants via door-to-door recruitment before the COVID-19 outbreak, and 28.9% (183/633) and 37.9% (240/633) referred themselves to us before and after the CO-VID-19 outbreak, respectively. Our study population was evenly spread throughout the residential areas of Singapore (data repository). 13 However, participants aged 45 years or older were underrepresented according to the age structure of the Singapore general population, 18 and the number of responses to the COVID-19 surveys from the oldest age group (≥ 60 years) was lower (13.5%; 385/2857) than in the initial cohort (19.1%; 75/393). Our cohort included more women (60.8%; 385/633) than men, but ethnic distributions were comparable to those of Singapore. 18 Most of our cohort participants were employed (71.1%; 450/633), and 12.3% (78/633) were students.
Regarding sources of information, respondents could select more than one response. A total of 77.1% (452/586) of respondents selected social media as a trusted source of information on infectious disease outbreaks, followed by television programmes (62.1%; 364/586) and friends and/or colleagues (59.9%; 351/586; Table 2).
Overall, participants demonstrated a high level of knowledge of COVID-19 and its associated symptoms (data repository). 13 Awareness of the local COVID-19 situation was highly dynamic and reflected current events (Fig. 3). In the first CO-VID-19 survey, about one third answered that transmission had occurred in Singapore, but this increased to 79.8% (217/272) in week 6 after the first local cluster was reported. Similarly, following the first two fatalities on the last day of week 12, 97.4%  (Fig. 4). These proportions fluctuated with a slight dip during weeks 8-11, before increasing substantially over weeks 12-17 with trends that tracked changes in incident cases and cumulative deaths.

Trust and behaviour
Most respondents (93.5%; 202/216 to 98.9%; 185/187) agreed or strongly agreed that they could trust the government to communicate the facts about COVID-19 (data repository). 13 Regarding pandemicrelated changes in behaviour, the likelihood of being socially responsible when infected was consistently high from weeks 4 to 13 (data repository). 13 5). An increasing proportion reported changing their plans, and 90.5% (57/63) of those who had planned to travel out of Singapore had changed their plans by week 13 (data repository). 13 Regarding views on governmentimplemented measures, 80.6% (382/474) to 95.5% (512/536) agreed in surveys 2-7 that various government-implemented measures were needed (data repository). 13 We observed the lowest support for penalties for not complying with social distancing (80.6%; 382/474), and the need for 5 days of sick leave for acute respiratory infections (82.4%; 291/353). Sizeable proportions felt that some measures were either implemented too late or that more should be done.

Risk-associated factors
We evaluated factors associated with the three perception variables of respondents by dichotomizing these responses appropriately ( Fig. 6; data repository). 13 Participants who enrolled by selfreferral were more likely to perceive CO-VID-19 as a threat compared with those who were recruited during door-to-door interviews. Women were less likely than men to perceive COVID-19 as a threat. Older age and higher educational levels were inversely associated with perceived risk of COVID-19 infection and risk of death if infected, respectively. Multiple indicators of epidemic stage were positively associated with perceptions of threat and risk in univariable analyses (data repository), 13 but in multivariable analyses the only significant associations were for the log of new cases with increased perceived threat and risk of infection. However, being aware that COVID-19 deaths had occurred was positively associated with perceived threat, risk of infection and risk of death if infected. We observed significant associations between trust in information from family and/or relatives with an increased perceived risk of infection and risk of death if infected. Conversely, trust in information from friends and/or colleagues was significantly associated with a decreased perceived risk of death.  All three perception variables were significantly correlated (data repository) 13 and also significantly associated (P < 0.001) with key protective behaviour (data repository) 13 Given their potential role on causal pathways, these were omitted in multivariable analyses on the frequency of hand washing, avoiding crowded places and wearing facemasks (Fig. 6).
In univariable analysis, significant positive associations between wearing facemasks and both time since the first case and being aware that COVID-19 deaths had occurred (data repository) 13 became significant inverse associations after adjusting for the log of numbers of new cases and cumulative deaths reported. Our data show that wearing facemasks was positively associated with both the log of number of new cases (OR: 1.8; 95% CI: 1.2-2.5) and the log of number of cumulative deaths reported (OR: 2.9; 95% CI: 2.4-3.5); avoiding crowded places was also positively associated with the log of number of cumulative deaths reported (OR: 1.6; 95% CI: 1.4-1.8). Trust in information from family and/or relatives was positively associated with avoiding crowded places and wearing facemasks. Trust in government communication on COVID-19 was positively associated with hand washing both before and after the declaration of outbreak condition orange; we only observed significant positive associations with avoiding crowded places after the declaration of outbreak condition orange and the circuit-breaker broadcast. Regarding the wearing of facemasks, an initially nonsignificant inverse association (OR < 1.0) became a significant positive association (OR: 2.1; 95% CI: 1.2-3.9) after the circuit-breaker broadcast (data repository). 13 Deeper questioning about facemasks in surveys 3 and 7 revealed that those with greater trust in government communication were more likely to agree with both the earlier recommendation that facemasks were mainly for those who were sick, and the revised recommendation for facemasks to be worn in public at all times (data repository), 13 with the association being statistically significant (OR: 2.9; 95% CI: 1.8-4.8; data repository). 13 We also observed significant positive associations between trust in government communication on COVID-19 and other socially responsible behaviour, such as the covering of mouths when coughing and avoiding social gatherings.

Discussion
Our online survey approach allowed the public's knowledge of, and support for, the recommendations and actions of health authorities to be rapidly assessed, and allowed us to examine the factors influencing behaviour adopted in response. We have demonstrated how the perceptions and behaviour of individuals are strongly influenced by the combination of local outbreak conditions and trust in the authorities' communication on  Overall, respondents demonstrated high levels of knowledge of the current COVID-19 outbreak, which improved further as the pandemic progressed. We have therefore shown that real-time feedback via repeated surveys can help to identify aspects needing clarification or more emphasis in public messaging campaigns. We found that most respondents were up to date with local developments (e.g. the surge in proportions of respondents being aware that deaths had been reported). Variables reflecting counts of cases and deaths and awareness of local deaths were important predictors of perceived threat and risk, as well as the likelihood of adopting key protective behaviour such as avoiding crowded places and wearing facemasks. This result emphasizes the role of timely and accurate detection of infections, as well as transparent reporting of local cases and deaths, in ensuring compliance with public health recommendations. Since the first COVID-19 case in Singapore, health authorities have been using social messaging platforms (e.g. WhatsApp, Facebook) to communicate with the public about the outbreak on a daily basis and to provide advice on how to reduce the risk of infection. 19 Provision of daily outbreak-related messages from the government to the public could explain why those with greater trust in government communication recognized the threat from COVID- 19, and yet perceived their risk of infection and death to be lower, whereas greater trust in information from family and/or relatives appeared to lead to increased perceptions of risk.
Our other noteworthy finding was how trust in government communication on COVID-19 influenced the risk-avoiding behaviour of the public. While trust in advice from family and/ or relatives influenced behaviour, greater trust in government communication had temporally nuanced associations corresponding to when this behaviour was recommended by health authorities. In particular, earlier messaging had largely convinced the public that facemasks were only required for those who were sick (data repository). 13 Subsequently, increased local transmission, and emerging evidence on the role of pre-symptomatic infection 20 and how facemasks can reduce the transmission of the virus, 21 led to a decision to recommend universal facemask use in public places. This messaging was largely successful in that it correlated with a substantial rise in facemask use from week 14, even before penalties for non-compliance were introduced in the latter half of week 15. Changes to public information may be needed in an evolving pandemic where new evidence emerges and risk assessments change. We note that, unlike advice from family and/or relatives, government messaging must be rationally calibrated. For instance, when

More frequent Increased
Less frequent Decreased CI: confidence interval. Note: Multivariable analysis adjusted for recruitment method, age, sex, race, level of education, occupational status, other medical conditions, indicators of epidemic stage in Singapore, awareness of virus and local situation, trust in information from different sources and trust in government to communicate facts during various outbreak phases. Analysis was based on 2517 responses from 572 participants in all cases except for the risk of dying, which was based on 2124 responses from 565 participants.
Public perceptions of COVID-19 and behaviour change, Singapore Vanessa W Lim et al.
there is evidence that transmission is increasing, health authorities may need to proactively escalate measures even if the public perceives the risk of infection to be low. Health authorities may also need to de-escalate more disruptive measures that are no longer warranted, even when public fears persist. Maintaining public trust in health authorities, in both their response and communications, is therefore crucial during an outbreak so that protective behaviour that is appropriate to the situation is adopted. 22 Given the complexities of communicating the need for outbreak interventions, a cohort-based approach such as ours has several advantages. Unlike post-outbreak studies, 23,24 we were able to influence real-time decision-making and facilitate improvements in communication strategies, allowing authorities to influence public acceptance. While crosssectional studies during the COVID-19 outbreak have yielded insights on public adoption of preventive measures, 25,26 a cohort-based study can track changes in the perceptions and opinions of individu-als and also disentangle the underlying effects of participant characteristics and awareness of outbreak developments.
However, the representativeness of our cohort was an important limitation. Reliance on self-referrals and online survey responses, as opposed to more traditional methods such as door-todoor recruitment, may have introduced biases, with evidence that those enrolled through self-referrals were more likely to view COVID-19 as a threat. Moreover, the migrant workers most affected in Singapore's outbreak were not included in our study; such initiatives must also sample neglected but vulnerable populations. Finally, social desirability bias may partially explain some associations between recommended behaviour and trust in government response.
Métodos A mediados de 2019, se inició el estudio de cohorte de la población local para evaluar la percepción que tiene el público sobre los brotes epidémicos de las enfermedades infecciosas. Tras la notificación del primer caso de la COVID-19 en Singapur el 23 de enero, se iniciaron Public perceptions of COVID-19 and behaviour change, Singapore Vanessa W Lim et al.