The role of conspiracy beliefs for COVID-19 prevention: A meta-analysis

Kinga Bierwiaczonek1*, Jonas R. Kunst1 & Aleksander B. Gundersen1

1
Department of Psychology, University of Oslo, Postboks 1094, Blindern, 0317 Oslo,
Norway.
5
*Corresponding author. Email: k.m.bierwiaczonek@psykologi.

This is a preprint of a paper that has been submitted for publication.

10 Please check for the latest version before citing.

Abstract: While conspiracy theories about COVID-19 are proliferating, their impact on health-
related responses during the present pandemic is not yet fully understood. We meta-analyzed
correlational and longitudinal evidence from 53 studies (N = 78,625) conducted in 2020 and
15 2021, demonstrating under what conditions COVID-19 conspiracy beliefs influence prevention
responses. Conspiracy beliefs were associated with reluctance toward prevention measures both
cross-sectionally and over time. They primarily reduced high-effort and pervasive responses
(vaccination, social distancing), whereas low-effort responses (wearing masks, hygiene) seemed
unaffected. Alarmingly, conspiracy beliefs had an increasing effect on prevention responses as
20 the pandemic progressed and predicted support for alternative treatments that lack a scientific
basis (e.g., chloroquine treatment, complementary medicine). Conspiracy beliefs are a non-
negligeable and growing threat to public health.
One-Sentence Summary: Conspiracy beliefs negatively predict health responses during the
pandemic, but some beliefs are more harmful than others.
25

1
 The spread of the coronavirus has been accompanied by an unprecedented spread of

conspiracy theories. Although the study of conspiracy theories has a long-standing and

30 transdisciplinary tradition (1-3), we have only recently started to understand the impact of

conspiracy beliefs on health-related responses during the present pandemic, and uncertainty

about their role remains. Whereas some argue that conspiracy beliefs may undermine the global

efforts to contain the virus by reducing people’s adherence to prevention guidelines such as

social distancing (4), self-isolation, hygiene (5, 6), and vaccination (7, 8), others argue that the

35 role of conspiracy beliefs might be exaggerated (9).

Given the pervasive and global impact of the pandemic, understanding the conditions

under which conspiracy theories may undermine preventive health responses and sometimes

even lead people to pursue risky alternative treatment is an urgent issue. Here, we provide the

first meta-analytic test of the effect of conspiracy beliefs on people’s health-related responses

40 during the pandemic, analyzing data from 53 published and unpublished manuscripts from 2020

and 2021, including 310 effects from 93 independent samples in 23 countries (N = 78,625). In

doing so, we aim to address several questions of broad significance that have produced mixed

research results or remain unaddressed to date. Are some conspiracy beliefs more harmful than

others? Do they influence certain health-related attitudes and behaviors more than others? Do

45 conspiracy beliefs exert the same effect throughout the pandemic or are they more harmful at

later stages when people feel fatigued from prolonged restrictions? Are conspiracy beliefs

influential especially in countries with low levels of social development, or do they have a high

potential to harm precisely because their effects are relatively context independent?

Addressing the mixed results of single studies (4, 5, 8-10), our first goal was to estimate

50 the average effect of conspiracy beliefs on prevention responses across the available research.

Results from a multi-level meta-analysis (11) showed that believers in conspiracy theories
2
 overall are more reluctant toward prevention than non-believers (r = -.13, 95% CI = [-.17, -.09]).

Crucially, this negative effect was replicated in a longitudinal, cross-lagged model based on a

smaller selection of studies that employed repeated measurements (r = -.09, 95% CI = [-.11, -

55 .07], Figure 1A). People who believed in conspiracy theories at one point in the pandemic tended

to be reluctant toward COVID-19 prevention measures at later points and vice versa.

The next crucial question was whether some conspiracy beliefs are more harmful than

others. Results showed that beliefs that specifically concern COVID-19 (combined effect: r = -

.15, 95% CI = [-.19, -.11]) are more predictive of health responses than a general tendency to

60 believe in conspiracies (r = -.12, 95% CI = [-.21, -.04]; Figure 1B). Among the specific COVID-

19 conspiracy beliefs, believing that the coronavirus is a man-made bioweapon stood out: its

effect on prevention was close to zero and non-significant (r = .03, 95% CI = [-.09, .03]),

whereas the effects of other beliefs (e.g., big pharma conspiracy, political conspiracy, hoax) were

negative and significantly stronger (combined effect: r = -.20, 95% CI = [-.25, -.15]). Of all the

65 conspiracy beliefs tested, the bioweapon theory is the only one that clearly implies that COVID-

19 is extremely dangerous, explaining why those who believe in it do not show much reluctance

toward prevention (10).

3
 A) B)

0.4

Effect Size
0.0

-0.4

Hoax Mix Hidden Truth Big Pharma Political Bioweapon General

Conspiracy Conspiracy
Mentality
Type of Conspiracy Theory
C) D)

0.4 0.4
Effect Size

Effect Size

0.0 0.0

-0.4
-0.4

Vaccine Mix Social Masks Hygiene Alternative

Distancing Treatments 0 100 200
Type of Prevention Response Days since WHO proclaimed pandemic

70

Fig. 1.

Effects of Conspiracy Beliefs on Health-Related Responses Over Time (A), by Type of

Conspiracy Theory (B) and Type of Response (C), and Over the Time Since the WHO

Proclaimed the Pandemic (D).

75

Note. Mix refers to measures that combined different types of conspiracy theories (B) or

prevention responses (C) into one scale. Point estimates represent mean effect r, whereas error

bars and ribbon represent 95% confidence intervals.

4
 80
Next, we tested whether conspiracy beliefs would predict some types of prevention

responses more than others (Figure 1C). Indeed, effects were strongest for vaccination (r = -.31,

95% CI = [-.38, -.23]), followed by social distancing (r = -.10, 95% CI = [-.14, -.06]), whereas

they were non-significant for mask wearing (r = -.02, 95% CI = [-.14, .09]) and hygiene

85 measures such as frequent hand washing (r = -.02, 95% CI = [-.04, .07]). Thus, supporters of

conspiracy theories seem the most reluctant toward the COVID-19 prevention measures that are

associated with the most effort and nuisance and that are most effective in reducing the spread of

the coronavirus (12). Critically, our results further showed that the stronger one’s conspiracy

beliefs are, the more positive one tends to be toward alternative treatments that lack a scientific

90 basis (e.g., chloroquine, vitamin c, complementary medicine; r = .42, 95% CI = [.22, .59]).

However, this effect was weaker when it referred to self-reported behavior (r = .23, 95% CI =

[.06, .39]), than when it referred to attitudes toward such treatments (r = .52, 95% CI = [.25,

.71]). Hence, believers of conspiracy theories seem more likely to have favorable attitudes

toward alternative treatments than to actually practice them.

95 Many governments reported a “pandemic fatigue” of their populations as the pandemic

progressed (13), and one possible reason for why people adhered less to health measures at later

stages of the pandemic may be that conspiracy theories gained influence. In support of this, the

time of data collection (coded as the number of days after the WHO proclaimed a world

pandemic) significantly predicted the strength of the observed effects (b = -.23, 95% CI = [-.43, -

100 .01], Figure 1D). The later into the pandemic, the stronger became the observed negative

association between conspiracy beliefs and prevention responses. As the pandemic progressed,

societies tended to polarize into those who accepted the official discourse about the pandemic

and those who rejected it (14). This polarization may have led to more congruency in both

5
 groups: more compliance with prevention measures among those accepting the official discourse,

105 and more reluctance among those believing in conspiracy theories (15, 16).

A final question of broad significance was whether conspiracy beliefs exert an influence

on prevention responses primarily in countries with low social development or whether effects

are relatively context independent. To answer this question, we tested the influence of seven

major indicators of social development: inequality (the Gini index), the quality of democracy,

110 freedom of press, access to education at a primary and higher level, citizens’ trust in institutions,

as well as the stringency of the government response to the pandemic. None of these indicators

moderated the effects. While it is possible that such macro-level factors influence a population’s

average endorsement of conspiracy theories, the degree to which these beliefs influence

prevention seems relatively independent of a country’s social development. This once more

115 emphasizes the threat of conspiracy beliefs to public health. Once disseminated among a

population, the theories may negatively influence prevention irrespective of factors that

otherwise can make societies resilient to misinformation.

In sum, our meta-analysis shows that conspiracy beliefs predict people’s reluctance

toward COVID-19 prevention measures, both cross-sectionally and over time. Even though

120 effects tend to be small, at a large scale, such a reluctance poses a non-negligeable threat to

public health and undermines global efforts to contain the virus. Crucially, the negative effects of

conspiracy beliefs are the strongest for vaccination and social distancing – the measures that in

combination are the most effective in reducing the spread of the virus (12), and that rely heavily

on the compliance of all. In addition to reducing adherence to prevention measures, conspiracy

125 beliefs may harm people by making them pursue alternative treatments that are ineffective at

best. Alarmingly, the negative influence of conspiracy beliefs seems to get stronger as the

pandemic progresses, suggesting that conspiracy theories are a growing threat to public health.

6
 1. J. M. Bale, Political paranoia v. political realism: on distinguishing between bogus
130 conspiracy theories and genuine conspiratorial politics. Patterns of Prejudice 41, 45-60
(2007).
2. D. Pipes, Conspiracy: How the paranoid style flourishes and where it comes from.
(Simon & Schusters, New York, 1997).
3. J. W. van Prooijen, K. M. Douglas, Belief in conspiracy theories: Basic principles of an
135 emerging research domain. European Journal of Social Psychology 48, 897-908 (2018).
4. K. Bierwiaczonek, J. R. Kunst, O. Pich, Belief in COVID-19 Conspiracy Theories
Reduces Social Distancing over Time. Applied Psychology Health and Well Being 12,
1270-1285 (2020).
5. R. Imhoff, P. Lamberty, A Bioweapon or a Hoax? The Link Between Distinct Conspiracy
140 Beliefs About the Coronavirus Disease (COVID-19) Outbreak and Pandemic Behavior.
Social Psychological & Personality Science 11, 1110-1118 (2020).
6. L. Pummerer et al., Conspiracy Theories and Their Societal Effects During the COVID-
19 Pandemic. Social Psychological & Personality Science, (2021).
7. P. Bertin, K. Nera, S. Delouvee, Conspiracy Beliefs, Rejection of Vaccination, and
145 Support for hydroxychloroquine: A Conceptual Replication-Extension in the COVID-19
Pandemic Context. Frontiers in Psychology 11, 565128-565128 (2020).
8. D. Freeman et al., COVID-19 vaccine hesitancy in the UK: the Oxford coronavirus
explanations, attitudes, and narratives survey (Oceans) II. Psychological Medicine, 1-15
(2020).
150 9. R. M. Sutton, K. M. Douglas, Agreeing to disagree: reports of the popularity of Covid-19
conspiracy theories are greatly exaggerated. Psychological Medicine, 1-3 (2020).
10. H.-W. Chan et al., Not-so-straightforward links between believing in COVID-19-related
conspiracy theories and engaging in disease-preventive behaviours. Humanities and
Social Sciences Communications 8, 1-10 (2021).
155 11. W. Van den Noortgate, J. A. López-López, F. Marín-Martínez, J. Sánchez-Meca, Meta-
analysis of multiple outcomes: a multilevel approach. Behavioral Research Methods 47,
1274-1294 (2015).
12. B. Huang et al., Integrated vaccination and physical distancing interventions to prevent
future COVID-19 waves in Chinese cities. Nature Human Behavior 5, 695-705 (2021).
160 13. World Health Organization, Pandemic fatigue. Reinvigorating the public to prevent
COVID-19. (2020).
14. J. J. Van Bavel et al., Using social and behavioural science to support COVID-19
pandemic response. Nature Human Behaviour 4, 460-471 (2020).
15. W. Bruine de Bruin, H.-W. Saw, D. P. Goldman, Political polarization in US residents'
165 COVID-19 risk perceptions, policy preferences, and protective behaviors. Journal of risk
and uncertainty 61, 177-194 (2020).
16. J. Kerr, C. Panagopoulos, S. van der Linden, Political polarization on COVID-19
pandemic response in the United States. Personality and Individual Differences 179,
110892 (2021).
170 17. A. Jovancevic, N. Milicevic, Optimism-pessimism, conspiracy theories and general trust
as factors contributing to COVID-19 related behavior – A cross-cultural study.
Personality and Individual Differences 167, (2020).
18. D. Moher, Liberati, A., Tetzlaff, J., Altman, D. G., & The Prisma Group. , Preferred
reporting items for systematic reviews and meta-analyses: The PRISMA statement. .
175 PLoS Medicine 6, e1000097 (2009).
7
 19. M. O. Rieger, Y. He-Ulbricht, German and Chinese dataset on attitudes regarding
COVID-19 policies, perception of the crisis, and belief in conspiracy theories. Data in
Brief 33, (2020).
20. M. Borenstein, L. V. Hedges, J. P. T. Higgins, H. R. Rothstein. (John Wiley & Sons,
180 Chichester, West Sussex, U.K.,Hoboken, 2009).
21. R. A. Peterson, S. P. Brown, On the Use of Beta Coefficients in Meta-Analysis. Journal
of Applied Psychology 90, 175-181 (2005).
22. Chair of Comparative Politics and German Government University of Würzburg, Project
Democracy Matrix. (2021).
185 23. Reporters without Borders, 2021 World Press Freedom Index. 2021.
24. The World Bank, Gini index (World Bank estimate). (2021).
25. Social Progress Imperative, Social Progress Index. (2020).
26. T. Hale et al., A global panel database of pandemic policies (Oxford COVID-19
Government Response Tracker). Nature Human Behavior 5, 529-538 (2021).
190 27. OECD, OECD Data: Trust in government. 2021.
28. L. V. Hedges, I. Olkin, Statistical methods for meta-analysis. (Academic Press, Orlando,
1985).
29. W. Viechtbauer, Conducting Meta-Analyses in R with the metafor Package. Journal of
Statistical Software 36, 1-48 (2010).
195 30. J. E. Pustejovsky, M. A. Rodgers, Testing for funnel plot asymmetry of standardized
mean differences. Research Synthesis Methods 10, 57-71 (2019).
31. U. Simonsohn, J. P. Simmons, L. D. Nelson, Better p-curves: Making p-curve analysis
more robust to errors, fraud, and ambitious p-hacking, A reply to Ulrich and Miller
(2015). Journal of Experimental Psychology: General, 144, 1146-1152 (2015).
200 32. T. D. Stanley, Limitations of PET-PEESE and Other Meta-Analysis Methods. Social
Psychological & Personality Science 8, 581-591 (2017).
33. M. W. L. Cheung, Meta-analysis: A structural equation modeling approach. (Wiley,
Chichester, UK, ed. 1, 2015).
34. Z. Fisher, E. Tipton, robumeta: An R-package for robust variance estimation in meta-
205 analysis. 2015.
35. L. V. Hedges, E. Tipton, M. C. Johnson, Robust variance estimation in meta-regression
with dependent effect size estimates. Research Synthesis Methods 1, 39-65 (2010).
36. J. E. Pustejovsky, E. Tipton, Small-Sample Methods for Cluster-Robust Variance
Estimation and Hypothesis Testing in Fixed Effects Models. Journal of Business &
210 Economic Statistics 36, 672-683 (2018).
37. J. P. Selig, K. J. Preacher, Mediation Models for Longitudinal Data in Developmental
Research. Research in Human Development 6, 144-164 (2009).
38. T. D. Stanley, H. Doucouliagos, Meta-regression approximations to reduce publication
selection bias. Res. Syn. Meth 5, 60-78 (2014).
215 39. D. Allington, B. Duffy, S. Wessely, N. Dhavan, J. Rubin, Health-protective behaviour,
social media usage and conspiracy belief during the COVID-19 public health emergency.
Psychological Medicine 51, 1-7 (2020).
40. D. Allington, S. McAndrew, V. Moxham-Hall, B. Duffy, Coronavirus conspiracy
suspicions, general vaccine attitudes, trust and coronavirus information source as
220 predictors of vaccine hesitancy among UK residents during the COVID-19 pandemic.
Psychological Medicine, 1-12 (2021).

8
 41. S. Alper, F. Bayrak, O. Yilmaz, Psychological correlates of COVID-19 conspiracy
beliefs and preventive measures: Evidence from Turkey. Current Psychology, 1-10
(2020).
225 42. I. P. Banai, Banai, B., Mikloušić, I., Beliefs in COVID-19 conspiracy theories,
compliance with the preventive measures, and trust in government medical officials.
Current Psychology, (2021).
43. P. Bertin, K. Nera, S. Delouvee, Conspiracy Beliefs, Rejection of Vaccination, and
Support for hydroxychloroquine: A Conceptual Replication-Extension in the COVID-19
230 Pandemic Context. Frontiers in Psychology 11, 565128-565128 (2020).
44. M. Biddlestone, R. Green, K. M. Douglas, Cultural orientation, power, belief in
conspiracy theories, and intentions to reduce the spread of COVID‐19. British Journal of
Social Psychology 59, 663-673 (2020).
45. A. B. Boot, A. Eerland, J. Jongerling, P. P. J. L. Verkoeijen, R. A. Zwaan, Gathering,
235 processing, and interpreting information about COVID-19. Scientific Reports 11, 6569-
6569 (2021).
46. M. Bruder, L. Kunert, The conspiracy hoax? Testing key hypotheses about the correlates
of generic beliefs in conspiracy theories during the COVID-19 pandemic. International
Journal of Psychology, (2021).
240 47. V. Cavojova, J. Srol, E. Ballova Mikuskova, How scientific reasoning correlates with
health-related beliefs and behaviors during the COVID-19 pandemic? Journal of Health
Psychology, 1359105320962266-1359105320962266 (2020).
48. N. Corbu, E. Negrea-Busuioc, G. Udrea, L. Radu, Romanians' willingness to comply
with restrictive measures during the COVID-19 pandemic: evidence from an online
245 survey. Journal of Applied Communication Research 49, 369-386 (2021).
49. V. A. Earnshaw et al., COVID-19 conspiracy beliefs, health behaviors, and policy
support. Translational Behavioral Medicine 10, 850-856 (2020).
50. T. El-Elimat, M. M. AbuAlSamen, B. A. Almomani, N. A. Al-Sawalha, F. Q. Alali,
Acceptance and attitudes toward COVID-19 vaccines: A cross-sectional study from
250 Jordan. PLoS One 16, e0250555-e0250555 (2021).
51. J. Farias, R. Pilati, COVID-19 as an undesirable political issue: Conspiracy beliefs and
intolerance of uncertainty predict adhesion to prevention measures. Current Psychology,
1-11 (2021).
52. K. N. Fountoulakis et al., Self-reported changes in anxiety, depression and suicidality
255 during the COVID-19 lockdown in Greece. Journal of Affective Disorders 279, 624-629
(2021).
53. T. K. Hartman et al., Different Conspiracy Theories Have Different Psychological and
Social Determinants: Comparison of Three Theories About the Origins of the COVID-19
Virus in a Representative Sample of the UK Population. Frontiers in Political Science 3,
260 (2021).
54. M. J. Hornsey et al., To what extent are conspiracy theorists concerned for self versus
others? A COVID-19 test case. European Journal of Social Psychology, (2021).
55. S. Hughes, L. Machan, It's a conspiracy: Covid-19 conspiracies link to psychopathy,
Machiavellianism and collective narcissism. Personality and Individual Differences 171,
265 110559-110559 (2021).
56. T. Karic, J. Mededovic, Covid-19 conspiracy beliefs and containment-related behaviour:
The role of political trust. Personality and Individual Differences 175, 110697-110697
(2021).

9
 57. J. Kowalski, M. Marchlewska, Z. Molenda, P. Górska, Ł. Gawęda, Adherence to safety
270 and self-isolation guidelines, conspiracy and paranoia-like beliefs during COVID-19
pandemic in Poland - associations and moderators. Psychiatry Research 294, 113540-
113540 (2020).
58. C. A. Latkin, L. Dayton, M. Moran, J. C. Strickland, K. Collins, Behavioral and
psychosocial factors associated with COVID-19 skepticism in the United States. Current
275 Psychology, 1-9 (2021).
59. L. B. Lazarević et al., What drives us to be (ir)responsible for our health during the
COVID-19 pandemic? The role of personality, thinking styles, and conspiracy mentality.
Personality and Individual Differences 176, 110771-110771 (2021).
60. G. Marinthe, G. Brown, S. Delouvée, D. Jolley, Looking out for myself: Exploring the
280 relationship between conspiracy mentality, perceived personal risk, and COVID‐19
prevention measures. British Journal of Health Psychology 25, 957-980 (2020).
61. Ş. B. Nazlı, F. Yığman, M. Sevindik, D. Deniz Özturan, Psychological factors affecting
COVID-19 vaccine hesitancy. Irish Journal of Medical Science, (2021).
62. T. Oleksy, A. Wnuk, D. Maison, A. Łyś, Content matters. Different predictors and social
285 consequences of general and government-related conspiracy theories on COVID-19.
Personality and Individual Differences 168, 110289-110289 (2021).
63. T. Oleksy, A. Wnuk, M. Gambin, A. Łyś, Dynamic relationships between different types
of conspiracy theories about COVID-19 and protective behaviour: A four-wave panel
study in Poland. Social Science & Medicine 280, 114028-114028 (2021).
290 64. M. Pivetti, G. Melotti, M. Bonomo, E. Hakoköngäs, Conspiracy Beliefs and Acceptance
of COVID-Vaccine: An Exploratory Study in Italy. Social Sciences (Basel) 10, 108
(2021).
65. N. Plohl, B. Musil, Modeling compliance with COVID-19 prevention guidelines: the
critical role of trust in science. Psychology, Health & Medicine 26, 1-12 (2021).
295 66. G. Prati, Intention to receive a vaccine against SARS-CoV-2 in Italy and its association
with trust, worry and beliefs about the origin of the virus. Health Education Research 35,
505-511 (2020).
67. E. C. Prichard, S. D. Christman, Authoritarianism, Conspiracy Beliefs, Gender and
COVID-19: Links Between Individual Differences and Concern About COVID-19, Mask
300 Wearing Behaviors, and the Tendency to Blame China for the Virus. Frontiers in
Psychology 11, 597671-597671 (2020).
68. K. Resnicow et al., Novel Predictors of COVID-19 Protective Behaviors Among US
Adults: Cross-sectional Survey. Journal of Medical Internet Research 23, e23488 (2021).
69. M. O. Rieger, Y. He-Ulbricht, German and Chinese dataset on attitudes regarding
305 COVID-19 policies, perception of the crisis, and belief in conspiracy theories. Data in
Brief 33, (2020).
70. D. Romer, K. H. Jamieson, Patterns of Media Use, Strength of Belief in COVID-19
Conspiracy Theories, and the Prevention of COVID-19 From March to July 2020 in the
United States: Survey Study. Journal of Medical Internet Research 23, e25215-e25215
310 (2021).
71. T. Rothmund, Farkhari, F., Azevedo, F., Ziemer, C. T. , Scientific trust, risk assessment,
and conspiracy beliefs about COVID-19-four patterns of consensus and disagreement
between scientific experts and the German public [unpublished preprint]. (2020).
72. J. B. Ruiz, R. A. Bell, Predictors of intention to vaccinate against COVID-19: Results of
315 a nationwide survey. Vaccine 39, 1080-1086 (2021).

10
 73. M. Sallam et al., High Rates of COVID-19 Vaccine Hesitancy and Its Association with
Conspiracy Beliefs: A Study in Jordan and Kuwait among Other Arab Countries.
Vaccines 9, 42 (2021).
74. M. Sallam et al., Low COVID-19 Vaccine Acceptance Is Correlated with Conspiracy
320 Beliefs among University Students in Jordan. Int J Environ Res Public Health 18, 2407
(2021).
75. A. Soveri, L. C. Karlsson, J. Antfolk, M. Lindfelt, S. Lewandowsky, Unwillingness to
engage in behaviors that protect against COVID-19: the role of conspiracy beliefs, trust,
and endorsement of complementary and alternative medicine. BMC Public Health 21,
325 684-684 (2021).
76. J. Srol, E. Ballova Mikuskova, V. Cavojova, When we are worried, what are we
thinking? Anxiety, lack of control, and conspiracy beliefs amidst the COVID‐19
pandemic. Applied Cognitive Psychology 35, 720-729 (2021).
77. A. C. Sternisko, A. Cislak, A. Van Bavel, J. J., Collective narcissism predicts the belief
330 and dissemination of conspiracy theories during the COVID-19 pandemic. (2021).
78. V. Swami, Barron, D. , Analytic thinking, rejection of coronavirus (COVID-19)
conspiracy theories, and compliance with mandated social-distancing: Direct and indirect
relationships in a nationally representative sample of adults in the United Kingdom
[unpublished preprint]. (2020).
335 79. P. Teovanović et al., Irrational beliefs differentially predict adherence to guidelines and
pseudoscientific practices during the COVID‐19 pandemic. Applied Cognitive
Psychology 35, 486-496 (2021).
80. V. van Mulukom, Low Levels of Trust affect Guideline Adherence and Conspiracy
Belief during the COVID-19 Pandemic [unpublished preprint]. (2020).
340 81. G. B. S. Wirawan, P. N. T. Y. Mahardani, M. R. K. Cahyani, N. L. P. S. P. Laksmi, P. P.
Januraga, Conspiracy beliefs and trust as determinants of COVID-19 vaccine acceptance
in Bali, Indonesia: Cross-sectional study. Personality and Individual Differences 180,
110995-110995 (2021).
82. L. Pummerer, Sassenberg, K. , Conspiracy theories and social norms [unpublished data].
345 (2020).
83. K. Winter, L. Pummerer, M. J. Hornsey, K. Sassenberg, Pro-vaccination subjective
norms moderate the relationship between conspiracy mentality and vaccination
intentions. British Journal of Health Psychology, (2021).
84. L. Liekefett, Can Conspiracy Beliefs Be Beneficial? [Unpublished manuscript]. (2021).
350 85. P. Hanel, Unpublished Dataset. (2021).
86. M. Chayinska et al., Coronavirus conspiracy beliefs and distrust of science predict risky
public health behaviours through optimistically biased risk perceptions in Ukraine,
Turkey, and Germany. Group Processes & Intergroup Relations, 136843022097827
(2021).
355
Note. References 4-8, 10, 17, and 39 – 86 are reports included in the meta-analyses (see Data S1,
S2).

11
 Acknowledgments: We would like to thank all authors of primary studies that shared their data
with us.
360 Funding: This research was not funded.
Author contributions:
Conceptualization: KB, JRK, ABG
Methodology: KB
Data coding: KB, JRK, ABG
365 Data analysis: KB
Visualization: JRK
Writing – original draft: KB, JRK,
Writing – review & editing: KB, JRK, ABG
Competing interests: Authors declare that they have no competing interests.
370 Data and materials availability: All data are available in the main text supplementary
materials.

Supplementary Materials
Materials and Methods
Supplementary Text
375 Figs. S1 to S2
Tables S1 to S2
Data S1 to S2

12
 1

Supplementary Materials for

5
The role of conspiracy beliefs for COVID-19 prevention: A meta-analysis

Kinga Bierwiaczonek, Jonas R. Kunst & Aleksander B. Gundersen

10 Correspondence to: k.m.bierwiaczonek@psykologi.uio.no

This PDF file includes:

15 Materials and Methods

Supplementary Text
Fig. S1 to S2
Tables S1 to S3

20 Other Supplementary Materials for this manuscript include the following:

Data S1 to S2
Data Used in Cross-Sectional Meta-Analyses
Data Used in the Longitudinal Meta-Analysis
25
 2

Materials and Methods

30 Inclusion Criteria

We included published and unpublished quantitative studies that reported (a) at least one

measure of conspiracy beliefs related to the coronavirus (e.g., “Coronavirus was created in a

lab”)(17) or a general conspiracy mindset (e.g., conspiracy mentality questionnaire, 5), (b) at

least one measure of COVID-19-related health behaviors, intentions and/or attitudes, either

35 generally recommended (i.e., social distancing, self-isolation, mask wearing, hygiene,

vaccination; we further refer to those as prevention) or non-recommended (e.g., using

chloroquine, vitamin C, ginger tea, essential oils as treatments for COVID-19; we further refer to

those as alternative treatments); and (c) a statistic reflecting the association between these

variables (e.g., correlation coefficient, odds ratio, regression coefficient). We defined conspiracy

40 beliefs as beliefs in secretive, intentional, and malevolent behavior of one or more actors (1), and

we excluded beliefs that did not involve these three elements (e.g., the belief that SARS-CoV-2

first appeared at the Wuhan meat market). Because we focused on health-related outcomes, we

included only those outcome measures that assessed attitudes, intentions or behavior toward

prevention or treatment of COVID-19 and we excluded other types of pandemic-related

45 behaviors (e.g., stocking up on toilet paper, hoarding food).

Search Procedures

The database search followed the PRISMA guidelines (18) and was completed on May

23, 2021. We searched for relevant papers in the following databases: PubMed, Ovid (including

Global Health, Health and Psychosocial Instruments, International Political Science Abstract,

50 Ovid MEDLINE, PsycInfo), Google Scholar and PsyArxiv. We used a broad search string (title

and abstract) to gather as many studies as possible at the first screening stage: (covid-19 OR
 3

corona OR sars-cov-2) AND conspiracy. Whenever necessary, the search string was adapted to

the format required by the respective database. Additionally, we sent out a call for unpublished

studies via two relevant scientific associations: Society for Personality and Social Psychology

55 and International Society of Political Psychology. Figure S1 summarizes the results of each

phase of our search. Abstract and full text screening was performed by the first author (PhD in

psychology). Additionally, 10% of the identified abstracts and full texts were screened by a

second coder to ensure high reliability of the screening process. The interrater agreement was

substantial (k = .76) for abstract inclusion and perfect (k = 1) for full text inclusion. All abstracts

60 for which the two coders disagreed were further evaluated in the full text screening phase.

Coding

To ensure high coding reliability, data from 52 included reports were extracted by two

independent coders: the first author and either the second or the third author. The interrater

agreement was almost perfect (k = .92). Any discrepancies were carefully checked by the first

65 author and discussed by the research team to reach agreement. One study (19) was excluded

from double coding and carefully coded by the first author because it required calculating

correlations from primary data with a highly complex structure.

From each sample, the coders extracted the following information: study design, sample

size, percentage of males, average age, percentage of participants without higher education,

70 country of data collection, sampling strategy (e.g., via social media, via participant recruitment

platforms such as Amazon MTurk, representative sampling), date of data collection; name,

number of items and reliability of the scale(s) used to measure conspiracy beliefs; the type of

conspiracy beliefs measured (e.g., general conspiracy mindset vs. scales comprising several

COVID-19-beliefs vs. measures focusing on one specific conspiracy theory only); the content of
 4

75 conspiracy beliefs, (e.g., coronavirus is a hoax, coronavirus a bioweapon); name, number of

items and reliability of the scale(s) used to measure health behaviors; the type of health

behaviors measured (e.g., prevention vs. alternative treatment; mask wearing, vaccination,

hygiene, social distancing, general adherence to/support for health guidelines, vitamin C use,

chloroquine use).

80 Next, all available correlations of interest were extracted. Whenever possible, we used

correlations reported in the primary studies. Whenever correlations were not reported but public

data were available (e.g., via the Open Science Framework), we calculated correlations ourselves

(keffects = 114). Whenever correlations were not reported and data were not available, we utilized

the available statistical information. Here, we used Borenstein and colleagues’ (20) formulas to

85 estimate correlation from odds ratios (keffects = 17) and from means and standard deviations (keffects

= 1), and Peterson and Brown’s (21) formula to approximate r values from standardized

regression coefficients (keffects = 15). We also contacted four authors by email, requesting

additional information, and three authors provided it.

We coded information from external sources to be used in moderation analyses. These

90 moderators were infection rates at the time of data collection, as well as seven variables

reflecting a country’s social development: democracy quality index (22), freedom of press index

(23), the Gini coefficient (24), access to education and access to knowledge (25), government

response stringency index (26), and trust in institutions (27).

Additionally, we coded effect sizes from eight identified longitudinal studies into a

95 separate database. These included correlations between conspiracy beliefs at each wave and

health behaviors at each wave, as well as autocorrelations: correlations between T1 conspiracy

beliefs and T1 health behaviors; T1 conspiracy beliefs and T2 health behaviors; T2 conspiracy
 5

beliefs and T1 health behaviors; T2 conspiracy beliefs and T2 health behaviors. Since the

number of waves in longitudinal studies ranged from two to five, with most studies reporting two

100 waves, for those studies that reported more than two waves (k = 3), we averaged effects within

and between each wave, except for the correlations within T1. For example, in a study with three

waves, correlations between T1 and T2 and correlations between T2 and T3 were averaged;

correlations within T2 and within T3 were averaged; and correlations within T1 were used as

extracted. This procedure is conceptually similar to the standard practice of constraining paths to

105 equality in a cross-lagged panel model, where all path coefficients between and within the

different waves are averaged except for the correlations within T1 to obtain a conservative

estimate of the effect over time (4).

Analytic Procedures

In all analyses, correlations were transformed to z before model estimation, and back to r

110 after model estimation (28). First, we meta-analyzed cross-sectional data using the multilevel

approach. Specifically, we specified a four-level meta-analytical model in the metafor package

for R (29) to account for the hierarchical structure of effects (11). The first level corresponded to

the sampling variance of the effects, the second level to the variation between effect sizes (effect

level), the third to the variation between participant samples (sample level), and the fourth level

115 to the variation between countries (country level).

We started by specifying the intercept-only model (Model 1 in Table S2), that is, a model

without any exploratory variables, to assess the overall strength of the correlation between

conspiracy beliefs and prevention. Then, we fitted three meta-regression models including

categorical moderators (dummy-coded) and continuous moderators (rescaled to a 0-1 scale to

120 avoid inflated variance estimates). Model 2 was a meta-regression model testing for effect-level
 6

moderators (general conspiracy beliefs vs. beliefs specific to COVID-19; types of health-level

responses: vaccination, social distancing, hygiene, mask wearing, a mix of different responses;

outcome measures referring to self-reported behavior, intentions, attitudes) and sample-level

moderators (date of data collection, infection rate at data collection). Model 3 was a meta-

125 regression model fitted on a subset of effects referring to COVID-19 conspiracy beliefs and

excluding general conspiracy mentality. This model tested for the effects of the different kinds of

conspiracy beliefs (bioweapon conspiracy, hoax, big pharma conspiracy, political conspiracy).

Model 4 was a meta-regression testing for the moderating role of seven country level moderators

(Gini coefficient, democracy quality, freedom of press, access to education, access to knowledge,

130 government response stringency index, trust in institutions).

To assess publication bias, we used PET-PEESE (30) and cross-validated it with the p-

curve analysis (31). These results must be taken with extreme caution as there is currently no

method of publication bias assessment that would be reliable with high heterogeneity of effects,

and the recommended cut-off point for PET-PEESE is I2 = 80% (32). This cut-off point was

135 largely exceeded in our study pool.

The longitudinal studies were analyzed using meta-analytical structural equation

modeling (MASEM) (33). Because only one longitudinal study included alternative treatment as

outcome, we could not meta-analyze this outcome longitudinally and we focused on prevention

outcomes. First, we used the robumeta v. 2.0 package for R (34) to pool the correlations of

140 interest with robust variance estimation (35): T1 conspiracy belief with T1 outcome, T1

conspiracy belief with T2 outcome, T2 conspiracy belief with T1 outcome, T2 conspiracy belief

with T2 outcome. Robust variance estimation is similar to multilevel meta-analysis in that it

accounts for the hierarchical structure of effects, but it is more suited for small study pools (36)
 7

such as our subset of longitudinal studies (ksamples = 8). Then, we used the metaSEM v. 1.2.4

145 package for R (33) to specify a meta-analytic cross-lagged panel model with two waves (37),

including all correlations between conspiracy beliefs at T1 and T2, and outcomes at T1 and T2.

Please note that the pooled correlations used for this model were based on different numbers of

studies due to missing data, which is common in multivariate meta-analyses. The two-stage

approach to meta-analytical structural equations models takes such differences into account,

150 preventing them from biasing the results (see (37) for technical details).

To estimate the effects on alternative treatments, we again used robust variance

estimation with small-sample adjustment (35, 36) instead of the multilevel approach due to the

low number of studies including alternative treatment outcomes (ksamples = 8). As in the multi-

level analyses, we additionally fitted a robust variance estimation meta-regression model with

155 the type of conspiracy beliefs (general vs. specific to COVID-19) as the moderator. Because the

dataset with alternative treatments only consisted of 15 individual effects, this was the only

moderation effect that we were able to test while maintaining the recommended threshold of df

>= 4 (36).

Sample Descriptives

160 The final pool of studies comprised of 53 papers, seven of which were unpublished at the

time of the literature search. The full dataset with all coded variables is available as Data S1

(cross-sectional dataset) and Data S2 (longitudinal dataset). The included papers reported 310

relevant effect sizes from 78,625 participants in 93 independent samples covering 23 different

countries, with four papers reporting samples from mixed countries (recoded for analysis as a

165 separate cluster, hence kcountries = 24). The average sample size was 836.45 (SD = 892.62), and

the most represented countries were the United States (ksamples = 12), Germany (ksamples = 10) and
 8

the United Kingdom (ksamples = 10). The earliest beginning of data collection was January 19th,

2020, and the latest was December 19th, 2020. The majority of studies were cross-sectional

(ksamples = 77). Nine studies were longitudinal, but one of them only reported cross-sectional

170 results. Four studies were experimental, but the experimental manipulation was only successful

in one study. All studies measured attitudes, intentions and/or behaviors related to recommended

prevention guidelines, and eight studies additionally reported attitudes, intentions and/or

behaviors related to alternative treatments. On average, 39.7% of the participants across the

primary studies were male and 28.53% of the participants had no higher education (but note a

175 large proportion of missing data for the latter variable). The average participant age was 34.96

(SD = 8.87).

Technical description of results

Prevention, cross-sectional results. In the multilevel meta-analysis of effects of

conspiracy beliefs on COVID-19 prevention, we found a significant weak and negative effect,

180 suggesting that believers in conspiracy beliefs are overall slightly more reluctant toward

prevention than non-believers (Table S1, Model 1).

The meta-regression moderated by effect-level and sample-level factors (Table S1,

Model 2) revealed that the negative effects of conspiracy beliefs on prevention were significantly

weaker when these beliefs were measured with scales assessing a general conspiracy mentality

185 than when they were measured with scales referring to single or multiple (referred to as ‘mixed’)

COVID-19 theories. The negative effects of conspiracy theories were significantly stronger for

vaccination than for the remaining health responses, and significantly weaker for hygiene than

for the remaining health responses. The date of data collection significantly predicted the
 9

strength of the effect. The later into the pandemic the data were collected, the stronger negative

190 effects were found.

In the meta-regression model fitted to the subset of effects referring only to COVID-19

conspiracy beliefs and excluding measures of general conspiracy mentality (Table S1, Model 3),

the effect was significantly weaker for the belief that COVID-19 is a man-made bioweapon than

for the remaining beliefs. Finally, in the meta-regression model with country-level moderators

195 (Table S1, Model 4), we did not find any significant effects.

In all analyses, effects were highly heterogenous, with the total percentage of true

variance (I2) exceeding 90%. The effects varied the most between samples (all I2 > 50%) and

between individual effects; much less variance was found at the country level, and sampling

variance accounted for a non-significant percentage of the overall variance of effects in all cases.

200 Prevention, longitudinal results. In the longitudinal meta-analysis conducted on a subset

of nine longitudinal studies from eight manuscripts (Figure 1A in the main manuscript, see also

Table S2 for detailed results), the effects of conspiracy beliefs on COVID-19 prevention over

time were negative and significant yet weak in size. The relationship was reciprocal. People who

believed in conspiracy theories at one point in the pandemic tended to be reluctant toward

205 prevention measures at a later point, and people who were reluctant toward prevention at an

earlier point showed a slight tendency to hold more conspiracy beliefs at a later point. This

model was fully saturated (df = 0) and had perfect fit with the data, hence we do not report fit

indices.

Alternative treatment, cross-sectional results. The robust variance estimation meta-

210 analysis showed that conspiracy beliefs had a medium-sized positive effect on alternative

COVID-19 treatments (Table S4). That is, the stronger participants’ conspiracy beliefs were, the
 10

more positive they tended to be toward alternative treatments for COVID-19. However, this

effect was highly heterogenous and meta-regressions showed that it was weaker when it referred

to self-reported behavior of engaging in alternative treatments than when it referred to attitudes

215 toward such treatments.

Publication Bias

A p-value of PET intercept smaller than .10 indicates that PEESE results, rather than PET

results, should be interpreted (38). In our PET-PEESE analyses, PET intercepts were significant

at p < .01 both for prevention and alternative treatment, we therefore report and interpret PEESE

220 results here. The unstandardized estimate of bias was non-significant for both prevention (b =

5.63, 95% CI = [-1.32, 12.57]) and alternative treatment (b = -96.23, 95% CI = [-206.80, 14.32]),

indicating that our data did not bear a noteworthy amount of publication bias in neither of these

meta-analyses. However, the bias-corrected estimate of the overall meta-analytical effect of

conspiracy beliefs was slightly larger than in the main analysis, both for prevention (r = -.15,

225 95% CI = [-.20, -.10], compared to r = -.13 in the main analysis, see Table S1) and for treatment

(r = .60, 95% CI = [.33, .78], compared to r = .42 in the main analysis, see Table S4). These

results suggest that the publication bias, although negligible, was in the direction of making the

meta-analytical effect slightly weaker that the true effect.

The p-curve analyses (Fig. S2) supported the evidential value of our data for both

230 prevention and alternative treatment. In both cases, both half- and full p-curve tests indicated

right-skewness with p <.001, thus fulfilling the two conditions of evidential value. That is, the

conclusions of PET-PEESE analyses and p-curve analyses converged in suggesting minimal

publication bias in our data.

 11

235

Records identified from: Records removed before

Identification

Ovid databases* (k = 229) screening:

240 PubMed (k = 154) Duplicate records removed
PsyArxiv (k = 24) from database search (k =
Google Scholar (k = 139) 296)
Call for studies (k = 10)

245

Records screened (abstract and Records excluded

title; k = 260) (k = 169)

250

Reports sought for retrieval Reports not retrieved

(k = 103) (k = 1)
Screening

255
Reports excluded:
Reports assessed for eligibility
• Not quantitative (k = 2)
(full text; k = 102)
• Statistical information
insufficient (k = 5)
• Missing eligible prevention
260 and/or treatment measures
(k = 26)
• Missing eligible conspiracy
belief measures (k = 5)
• Sample and measures
overlap with another report
Included

Independent samples included in

265 (k = 11)
review (k = 93)
Reports of included studies
(k = 53)

Fig. S1.
PRISMA Diagram Summarizing Literature Search

270 Note: * Ovid databases searched: Global Health, Health and Psychosocial Instruments,
International Political Science Abstract, Ovid MEDLINE, PsycInfo.
 1

Fig. S2.

(B) Prevention

(A) Alternative Treatment

Fig. S2.
Results of P-curve Analyses for Data on Prevention (A) and Alternative Treatment (B)
 1

Table S1.

Results of Cross-Sectional Multilevel Meta-Analyses of the Association between Conspiracy Beliefs and Prevention

% % % %
ksampl kcountri Total variance variance variance variance
Model Predictor kES es es r 95% CI p(r) Q(r) df(Q) I2 Level 1 Level 2 Level 3 Level 4

1 Intercept only 295 93 24 -.131 [-.171, -.091] <.0001 9322.32 294 96.85 -0.01 39.37 56.18 4.46

2 Intercept 265 83 22 -.121 [-.212, -.029] .011 3860.98 254 93.42 -0.02 40.29 55.75 3.98
General conspiracy
mentalitya .124 [.066, .180] <.0001
Specific COVID-19
beliefsa .018 [-.034, .070] .489
Outcome: vaccinationb -.107 [-.183, -.029] .008
Outcome: mask wearingb .062 [-.006, .130] .076
Outcome: social
distancingb .024 [-.026, .075] .340
Outcome: hygieneb .132 [.066, .197] .000
Outcome: intentionsc -.011 [-.068, .046] .711
Outcome: self-reported
behaviorc .060 [.000, .120] .051
Infection rated -.065 [-.280, .155] .562
Date of data collectiond -.229 [-.430, -.007] .044

3 Intercept 214 76 24 -.195 [-.243, -.146] <.0001 5283.43 235 95.55 -0.01 46.56 53.45 0.00
Conspiracy: hoaxe -.058 [-.148, .033] .211
Conspiracy: man-made bioweapone .137 [.072, .201] <.0001
Conspiracy: political
interestse .030 [-.043, .104] .414
Conspiracy: hidden truthe .056 [-.032, .143] .215
Conspiracy: big pharmae .013 [-.084, .110] .786
 2

4 Intercept 207 59 15 .503 [-.685, .960] .434 5892.29 199 96.62 -0.01 31.32 55.07 13.62
d
Gini Index .071 [-.676, .747] .875
d
Quality of democracy -.559 [-.946, .484] .284
d
Freedom of press -.617 [-.973, .611] .322
d
Access to education .455 [-.291, .857] .222
d
Access to knowledge -.196 [-.792, .591] .656
Government Response Stringency
Indexd -.007 [-.296, .283] .962
d
Trust in government -.238 [-.642, .269] .357
Note. Model 1 is the main effects model with intercept only; Model 2 is a meta-regression model with effect-level and sample-level moderators;
Model 3 is a meta-regression model fitted to a subset of studies including specific COVID-19 conspiracy beliefs and excluding general conspiracy
beliefs; Model 4 is a meta-regression model with country-level moderators. Moderators a, b, c, e were dummy coded; a baseline is measures
consisting of mixed/multiple COVID-19 conspiracy beliefs; b baseline is measures consisting of mixed/other outcomes; c baseline is attitudes; e
baseline is measures consisting of mixed/other conspiracy theories; d predictors were rescaled to a 0-1 scale to avoid inflated variance estimates.
Effects significant at p < .05 are presented in bold.
 3

Table S2.
Results of the Meta-analytical Cross-Lagged Panel Model of the Effects of Conspiracy Beliefs on Prevention Over Time

Path ksamples keffects ES SE 95% CI p

Lagged paths
Conspiracy T1-> Prevention T2 9 25 -.091 .009 [-.109, -.074] <.0001
Prevention T1 -> Conspiracy T2 7 27 -.073 .008 [-.088, -.057] <.0001
Autoregressions
Conspiracy T1 -> Conspiracy T2 6 22 .691 .006 [.680, .702] <.0001
Prevention T1 -> Prevention T2 8 31 .571 .007 [.556, .585] <.0001
Correlations
Conspiracy T1 and Prevention T1 8 22 -.122 .011 [-.143, -.101] <.0001
Conspiracy T2 and Prevention T2 8 28 -.066 .006 [-.079, .054] <.0001
 4

Table S3.

Results of Cross-Sectional Meta-Analyses of the Association between Conspiracy Beliefs and Alternative Treatment

Model Predictor kES ksamples r df(r) 95% CI p(r) I2

Main effects
Intercept only 15 8 .421 7 [.223, .586] .002 98.56

Meta-regresssion
Intercept 15 8 .519 4 [.251, .713] .007 98.27
Outcome: self-reported behaviora -.326 4.3 [-.584, -.007] .047
Note. Predictors a was dummy coded: 1 – self reported behavior, 0 – attitudes.
 5

Data S1. (separate file)

Data Used in Cross-Sectional Meta-Analyses

Data S2. (separate file)

Data Used in the Longitudinal Meta-Analysis