key: cord-0768785-424g36uv
authors: Han, H.
title: Trust in the research community predicts intent to comply with COVID-19 prevention measures: An analysis of a large-scale international survey dataset
date: 2021-12-09
journal: nan
DOI: 10.1101/2021.12.08.21267486
sha: 81b6660554bb486742490f6760c9f2e987ded3a7
doc_id: 768785
cord_uid: 424g36uv

In the present study, I explored the relationship between people's trust in different agents related to prevention of spread of COVID-19 and their compliance with pharmaceutical and non-pharmaceutical preventive measures. The COVIDiSTRESSII Global Survey dataset, which was collected from international samples, was analysed to examine the aforementioned relationship across different countries. For data-driven exploration, network analysis and Bayesian generalized linear model (GLM) analysis were performed. The result from network analysis demonstrated that trust in the scientific research community was most central in the network of trust and compliance. In addition, the outcome from Bayesian GLM analysis indicated that the same factor, trust in the scientific research community, was most fundamental in predicting participants' intent to comply with both pharmaceutical and non-pharmaceutical preventive measures. I briefly discussed the implications of the findings, the importance of trust in the scientific research community in explaining people's compliance with measure to prevent spread of COVID-19.

pandemic has not concluded, data collected so far suggests that implementation of such 23 preventive measures have significantly contributed to prevention and mitigation of severe 24 COVID-19 outbreaks [3, 4] . 25

Given the importance of preventive measures in prevention of spread of 26 whether public is compliant with such measures would be critical in the current pandemic 27 situation [5] . Even if diverse preventive measures that have been found to be effective are 28 planned and implemented by agents, without people's compliance with the measures, successful 29 control of the pandemic could not be achieved [6] . For instance, rejection of and incompliance 30 with the r recommended and required preventive measures associated with political debates 31 resulted in the recent drastic increase in COVID-19 cases and deaths caused by the Delta variant 32 in multiple countries across the globe [5, 7] . Hence, it would be important to understand which 33 factors are involved in people's compliance as well as incompliance with preventive measures. 34

Previous research has suggested that trust in agents addressing pandemic-related matters 35 is one of the most fundamental factors predicting compliance with preventive measures [8] . For 36 instance, several researchers have examined and reported significant association between trust in 37 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 9, 2021. ; https://doi.org/10.1101/2021.12.08.21267486 doi: medRxiv preprint Running Head: TRUST IN SCIENCE AND COMPLIANCE 4 governmental agents and organizations in the domain of health care (e.g., World Health 38 Organization), and vaccination intent and compliance with non-pharmaceutical preventive 39 measures [9] [10] [11] . Furthermore, trust in science and scientific research communities, which play 40 fundamental roles in developing preventive measures and proposing guidelines based on 41 evidence, has also been considered as a central factor in predicting compliance [12, 13] . This 42 would be particularly important within the context of the current pandemic, because spread of 43 misinformation and conspiracy theories, which are closely associated with distrust in science and 44 particularly problematic in recent days, drives people's tendency to disobey mandatory 45 preventive measures and vaccination requirement [14] . 46

Although the aforementioned previous studies have examined the importance of trust in 47 compliance with preventive measures, several limitations would warrant further investigations. 48

First, the majority of the previous studies was conducted with datasets collected from single or a 49 limited number of countries. Given the current COVID-19 pandemic is a global issue [15] , it 50 would be necessary to collect data across diverse countries in examining the mechanism of 51 compliance tendency. Such relatively small-scale research based on data from a small number of 52 countries might not be sufficient to draw conclusions that can be well generalizable across the 53 globe. 54

Second, in terms of methodology, the previous studies employed conventional analysis 55 methods, which are based on frequentist perspective; such conventional methods are suitable to 56 test one specific null hypothesis and/or model, but not ideal for model exploration [16] . For 57 instance, if we are primarily interested which trust factor is central in prediction of compliance 58 tendency, the previous studies employing conventional methods might not be able to address our 59 interest in a complete manner. In fact, exploration of the best prediction model among multiple 60 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted December 9, 2021. To conduct the data-driven exploration, I plan to implement two novel analysis methods. 75

First, network analysis will be performed to explore how trust in different agents and compliance 76 with different types of preventive measures are associated with each other. In this exploration, I 77 intend to examine which factor is positioned in the most central and influential position in the 78 network [19] . Second, I will explore the best model predicting compliance with different types of 79 preventive measures with Bayesian model exploration [20] . Through this process, all possible 80 candidate regression models in terms of all possible combinations of trust in different agents will 81 be tested, and the most probable model given data will be identified. Finally, based on results 82 from the aforementioned processes employing data-driven methods, I will examine which trust 83 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted December 9, 2021. The employed items were developed by the COVIDiSTRESS Global Survey Consortium 104 members. They were translated and back translated by the consortium members from different 105 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. staying at least the recommended distance (Compliance 5); staying at home unless going out for 127 essential reasons (Compliance 6); self-isolating if you suspected that you had been in contact 128 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted December 9, 2021. participants' education level: PhD or doctorate; university degree (e.g., MA, MSc, BA, BSc); 138 some university or equivalent (still ongoing, or completed a module or more, but did not 139 graduate); up to 12 years of school; up to 9 years of school; up to 6 years of school; none. 140

To examine the overall association between responses to the seven trust and eight 143 compliance items, I conducted network analysis with bootnet R package. The main purpose of 144 network analysis is to demonstrate associations between nodes, trust and compliance in the case 145 of the present study. A connection between two specific node is defined as an edge, which has a 146 weight representing the strength of the association [22] . An edge weight is quantified in term of 147 partial correlation between two nodes by bootnet. As an illustrative example, in the case of the 148 edge between Trust 1 and Compliance 1, the edge weight can be understood in terms of 149 correlation between Trust 1 and Compliance 1 after controlling for correlation with all other 150 items in the same network (i.e., Trust 2 … Compliance 8). In a network plot, which visualizes 151 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 9, 2021. ; https://doi.org/10.1101/2021.12.08.21267486 doi: medRxiv preprint the result of network analysis, an edge between two nodes is presented in the format of a line 152 with a specific thickness, which represents its edge weight, the strength of the association. 153

While exploring a partial correlation network, bootnet employs one technique, graphical 154 LASSO (GLASSO), to identify a regularized partial correlation network through penalizing 155 spurious edge weights [19] . Implementation of GLASSO is required to minimize false positives 156 that may exist in a network of interest. For instance, we can imagine that there is no true non-157 zero partial correlation between two specific nodes. In the reality, possibly due to noise and/or 158 measurement error, even after controlling for association with other nodes, the edge weight 159 between the two nodes could not exactly become zero, although that is a false positive Closeness is defined in terms of the inverse of summed distances from one specific node to the 170 other nodes in the same network. Finally, betweenness is estimated in terms of how many times 171 one specific node is in the shortest path between two other nodes in the whole network. In the 172 present study, I examined which node reported the highest strength, closeness, and betweenness 173 values. 174 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 9, 2021. Bayesian GLM analysis for each compliance dependent variable to identify which trust 197 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. In terms of BFM0 and BFMF, all best models identified by Bayesian GLM analysis, except 256 for the best model predicting vaccination intent (Compliance 1), were supported by very strong 257 evidence compared with both the null and full models. In the case of Compliance 1, the full 258 model including all seven trust predictors was identified as the best model. 259

In the present study, first, I conducted network analysis to understand association 261 between participants' trust in seven different agents addressing the COVID-19 pandemic and 262 their intent to comply with eight different types of preventive measures. Second, Bayesian GLM 263 analysis was performed to explore the best model predicting each compliance intent variable 264 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 9, 2021. The findings suggest that in predicting people's intent to comply with both 276 pharmaceutical (e.g., vaccination) and non-pharmaceutical measures to prevent spread of 277 COVID-19 (e.g., hand washing, mask use, social distancing, self-isolation), trust in scientific 278 research and the community of scientists play the most fundamental role compared with trust in 279 other agents, e.g., government, healthcare system, health organization [12, 13] . Given that such 280 measures were primarily tested and suggested by scientific studies with empirical evidence, even 281 if their implementation and enforcement are tasks to be done by other agents, trust in science is 282 expected to make the greatest, fundamental influence on people's compliance [33] . Hence, if 283 people do not have robust trust in scientific research regarding COVID-19, then they are unlikely 284 to abide by preventive measures implemented by governments and health-related organizations 285

[13]. 286 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 9, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 9, 2021. ; https://doi.org/10.1101/2021.12.08.21267486 doi: medRxiv preprint many of the current social issues related to incompliance with COVID-19 preventive measures 310 have been emerged from and reinforced by misinformation shared through diverse forms of 311 media, improvement of science communication would be required to address the issues [40] . 312

We may consider several strengths of the present study and how it could make significant 313 contributions to literature. First, a large-scale international survey dataset, the COVIDiSTRESSII 314 Global Survey dataset, was analysed instead of a relatively small-size dataset collected from a 315 limited number of countries. Because the COVID-19 pandemic is a global issue, findings from 316 the current study will be able to provide researchers and policy makers across the globe with 317 useful insights about how to promote people's compliance with preventive measures based on 318 generalizable evidence from a cross-national investigation. Second, I explored the overall 319 association between compliance and trust in different agents instead of testing specific 320 hypotheses. With novel quantitative methods, network analysis and Bayesian GLM analysis, I 321 was able to demonstrate that trust in scientific research is most influential and fundamental in 322 predicting compliance. 323

However, several limitations in the present study may warrant further investigations. 324

First, while collecting data regarding compliance, the project consortium employed the self-325 report method for the feasibility of the global survey project. Hence, whether the reported 326 compliance intent predicts compliance behaviour in the reality could be questionable. Second, 327 only one item per trust in each specific agent or compliance with each specific preventive 328 measure was employed in the survey. Because the consortium was not able to use multiple items 329 per construct due to the feasibility issue, the psychometrical aspects of the trust and compliance 330 items could not be tested in a complete manner. Thus, future studies shall employ more direct 331 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 9, 2021. ; https://doi.org/10.1101/2021.12.08.21267486 doi: medRxiv preprint measures for compliance and conduct psychometrics tests by employing multiple items per 332 construct to address the limitations in the current study. 333

The data and codes that support the findings of this study are openly available in the 335 Open Science Framework project page (https://doi.org/10.17605/OSF.IO/Y4KGH). 336

This research received no specific grant from any funding agency, commercial or not-for-338 profit sectors. 339 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 9, 2021. Trust . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted December 9, 2021. ; https://doi.org/10.1101/2021.12.08.21267486 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted December 9, 2021. ; https://doi.org/10.1101/2021.12.08.21267486 doi: medRxiv preprint

Modeling compliance with COVID-19 prevention guidelines: the 378 critical role of trust in science

In science we (should) trust: Expectations and compliance across nine 381 countries during the COVID-19 pandemic

Science skepticism in times of COVID-384 19

COVIDiSTRESS Global Survey dataset on psychological and 387 behavioural consequences of the COVID-19 outbreak

Improved model exploration for the relationship between moral 389 foundations and moral judgment development using Bayesian Model Averaging

Compliance 1-Compliance 8. β (SE): Standardized regression coefficient (standard error). ES (ROPE %): Effect 467 size in Cohen's D (% of 95% HDI within the defined region of practical equivalence)

References 342

Tregoning JS, et al. Progress CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted December 9, 2021. ; https://doi.org/10.1101/2021.12.08.21267486 doi: medRxiv preprint