Longitudinal predictors of vaccine conspiracy beliefs: findings from a five-wave panel study in Japan

Abstract

The COVID-19 Pandemic not only posed a global health crisis but also intensified public divisions over vaccination, with conspiracy theories emerging as a significant obstacle to health communication. This study investigates the psychological and informational foundations of vaccine conspiracy beliefs using a five-wave longitudinal panel survey conducted annually in Japan between 2020 and 2024 (N=2,000 in 2020; N=600 in 2024). A factor analysis of an eight-item scale identified two distinct dimensions of conspiracy thinking: institutional conspiracy beliefs, which reflect distrust toward governments and pharmaceutical companies, and implausible conspiracy beliefs, which involve extreme claims such as microchip implantation. Latent growth modelling revealed that institutional conspiracy beliefs were associated with lower reliance on mass media, persistent and growing scepticism toward mainstream media, and younger age. These findings suggest that institutional conspiracy beliefs reflect enduring orientations of distrust rather than temporary reactions to pandemic-specific information. In contrast, implausible conspiracy beliefs were endorsed by only a small minority of respondents. This study makes three contributions. First, it demonstrates the need to distinguish institutional distrust-based conspiracies from fringe implausible narratives. Second, it provides longitudinal evidence clarifying how stable predispositions and changing attitudes jointly shape conspiracy beliefs. Third, it extends the literature by examining the Japanese context, adding cultural variation to a field dominated by Western-centric data.

Introduction

Homer’s epic The Iliad opens with a plague sent by Apollo, not only decimating the Greek army but igniting a bitter conflict between its leaders¹. In a similar vein, the COVID-19 Pandemic not only posed an unprecedented threat to global health but also exposed deep divisions in public opinion. Among the most visible signs of this division was the emergence of strong anti-vaccine sentiment, often underpinned by conspiracy theories that question the safety, efficacy, or necessity of vaccination itself². Although conspiracy beliefs are associated with broader political and ideological orientations^3–10, increasing attention has also been directed toward the role of media-related informational environments, including selective exposure and media scepticism. They are frequently associated with populist attitudes characterised by distrust toward political elites, scientific institutions, and mainstream media¹¹. Research has shown that individuals with such orientations are more likely to endorse conspiratorial narratives as alternative explanations that align with their anti-establishment perspectives. Although these studies point to multiple psychological and ideological foundations of conspiracy beliefs, the present study focuses specifically on the role of media-related informational environments, particularly media scepticism and patterns of media exposure, in shaping vaccine-related conspiracy beliefs.

Within this broader literature, research on digital media environments has highlighted the importance of informational exposure processes in the spread and persistence of conspiracy beliefs. The digital media environment also plays a crucial role in shaping the spread and entrenchment of conspiracy beliefs^12–14. Social media platforms often facilitate selective exposure to ideologically congruent content through algorithmic curation and personalization. This process reinforces preexisting beliefs and insulates users from countervailing information. Scholars refer to this phenomenon as the “echo chamber effect”^15,16. Moreover, the amplification of fringe content by recommendation algorithms and the influence of automated bot accounts can further distort perceptions of consensus and credibility online¹⁷. These structural features of digital platforms do not merely transmit misinformation but actively shape the information ecosystem in which conspiracy theories thrive. Therefore, to better understand the persistence and diffusion of vaccine-related conspiracy beliefs, researchers should examine differences in media-related informational environments, including patterns of mass and social media exposure.

Much existing research on conspiracy beliefs has focused on highly polarised Western media environments, particularly the United States, where partisan media ecosystems and ideological fragmentation are especially pronounced. By contrast, the Japanese media environment has traditionally been characterised by comparatively strong normative expectations of political neutrality in news reporting¹⁸. Previous research has also demonstrated substantial audience overlap among major Japanese media outlets and relatively limited ideology-based fragmentation in online news consumption¹⁹. Although some recent studies have suggested increasing political polarisation in specific domains^20,21, Japan nevertheless remains less characterised by highly segmented partisan media environments than many Western democracies. Despite this comparatively consensus-oriented media environment, conspiracy beliefs concerning COVID-19 still emerged and circulated during the pandemic. Examining conspiracy beliefs in Japan, therefore, provides an opportunity to investigate how conspiratorial thinking develops even in a media environment where overt partisan division and highly segmented information ecosystems are comparatively less pronounced. This context is particularly valuable for clarifying how media scepticism, subjective media literacy, and patterns of media exposure relate to conspiracy beliefs.

This study investigates the psychological and informational foundations of belief in vaccine conspiracy theories. We aim to clarify whether these beliefs primarily stem from selective media consumption during the pandemic^22,23 or reflect pre-existing scepticism towards information provided by mainstream media^24,25. Specifically, we ask: (1) To what extent is belief in vaccine conspiracy theories shaped by selective exposure to media content? (2) To what extent are psychological attitudes such as media scepticism, considering both their initial levels and their changes over time, associated with a greater proneness to such beliefs?

To answer these questions, we analyse five waves of panel survey data collected annually from April 2020 to April 2024 in Japan. This rich longitudinal dataset enables us to examine how psychological attitudes and informational environments, such as media scepticism and media use, evolved over the course of the pandemic and how these trajectories related to subsequent vaccine conspiracy beliefs. By observing the same individuals over time, we can identify temporal fluctuations and stable attitudinal patterns that may be obscured in cross-sectional analyses. We apply multivariate statistical techniques to examine the associations between media exposure, media scepticism, and individual predispositions, leveraging the panel structure to enhance our findings’ robustness and temporal depth. We employ latent growth modelling (LGM) as our primary multivariate technique, which enables us to separate initial levels (intercepts) from temporal changes (slopes) in key predictors, including mass media use, social media use, media scepticism, and subjective media literacy. By modelling both stable predispositions and within-individual changes, this approach provides a more precise account of how informational and psychological environments influence proneness to conspiracy beliefs.

This study illuminates the mechanisms behind the rise of vaccine conspiracy beliefs, contributing to a more nuanced understanding of public opinion fragmentation in the digital age. The findings offer practical implications for public health communication strategies and the design of interventions to curb misinformation and foster societal resilience^26,27.

Results

Respondents were recruited using the website “Yahoo! Crowdsourcing” in Japan. We conducted a five-wave longitudinal panel survey in April of each year: on 3 in 2020, 13–16 in 2021, 13–17 in 2022, 11–17 in 2023, and 14–18 in 2024. Participant characteristics for each year were as follows. 2020 (N=2,000, male=61.70%, mean age=46.48, s.d.=11.10), 2021 (N=1156, male=66.00%, mean age=48.65, s.d.=10.64), 2022 (N=781, male=68.89%, mean age=50.49, s.d.=10.12), 2023 (N=664, male=70.93%, mean age=52.07, s.d.=10.05), and 2024 (N=600, male=71.83%, mean age=53.14, s.d.=10.04). In the first year of the survey, 2,000 participants were recruited. In each subsequent year, only those who had participated in the preceding year were invited to take part again. To examine potential attrition bias, we conducted a logistic regression analysis predicting attrition before Wave 5 using data from all respondents in Wave 1 (see Supplementary Information, Table S3, S4 and S5). Although younger participants and women were more likely to drop out, attrition was not substantially associated with the main predictors examined in the present study.

Outcome

Using the eight-item Anti-vaccine Conspiracy Belief Scale²⁸, we measured conspiracy beliefs in the fifth wave of the survey and conducted a factor analysis with maximum likelihood estimation and Promax rotation. On the basis of eigenvalues and interpretability, we identified a two-factor model as the most appropriate solution. The analysis extracted two factors: institutional conspiracy, which reflects beliefs that governments and pharmaceutical companies systematically conceal unfavourable vaccine-related information, and implausible conspiracy, which reflects beliefs that vaccines contain mechanisms for monitoring people (Table 1). We calculated factor scores for each factor and used them as outcome variables in subsequent analyses (Fig. 1). The correlation between the two factors was moderate (r =.543), suggesting that they were related but not redundant. Both factors also demonstrated high internal consistency, with Cronbach’s alpha coefficients of.90 for institutional conspiracy beliefs and.88 for implausible conspiracy beliefs.

Table 1.

Factor loadings of the Anti-vaccine Conspiracy Belief Scale (fifth wave): Maximum likelihood estimation with Promax rotation identified two factors: institutional conspiracy and implausible conspiracy.

Variable	Factor1	Factor2
Pharmaceutical companies, scientists, and academics work together to cover up the dangers of vaccines.	1.013	−0.091
Vaccines are harmful, and this fact is covered up.	0.961	−0.105
Vaccines are not tampered with. *	0.861	−0.045
The government is trying to cover up the link between vaccines and autism.	0.715	0.123
Immunizations allow governments to track and control people.	0.351	0.340
Tiny devices are placed in vaccines to track people.	−0.042	0.970
Tiny devices are implanted in vaccines for use in mind-control experiments.	−0.077	0.992
The vaccine allows the government to monitor the elderly through the implantation of tiny tracking devices.	−0.033	0.938

* Item reverse-scored.

Fig. 1.

Distribution of conspiracy belief factors across the sample: Panel (A) Institutional conspiracy belief scores. Panel (B) Implausible conspiracy belief scores. Most participants rejected implausible conspiracy beliefs, whereas institutional conspiracy beliefs showed greater variance across individuals.

These two factors may reflect qualitatively distinct domains of conspiratorial thinking rather than merely different degrees of the same underlying belief system. The first factor, institutional conspiracy, reflects distrust toward established authorities and institutions. It comprises items suggesting that real-world entities, such as governments and pharmaceutical companies, systematically conceal vaccine-related information (e.g., “The government is trying to cover up the link between vaccines and autism”). In contrast, the second factor, implausible conspiracy, involves highly fringe narratives widely regarded as scientifically implausible, such as claims of microchip-based tracking or mind-control experiments. The differentiation in factor loadings, in which items load strongly on one factor while showing minimal loadings on the other, is consistent with interpreting these factors as distinct forms of conspiracy-oriented beliefs. Furthermore, the distribution patterns shown in Fig. 1 reinforce this distinction. Institutional conspiracy beliefs show substantial variation across the sample, whereas implausible conspiracy beliefs exhibit a pronounced floor effect, endorsed by only a very small minority of respondents. One item, “Immunisations allow governments to track and control people”, showed moderate loadings on both factors. This pattern is theoretically understandable because the item combines elements of institutional distrust, specifically concerns about governmental control, with more implausible tracking-related narratives.

Repeatedly measured predictors

To measure media exposure, we categorised media use into two types: mass media and social media. The frequency of mass media use was calculated as the simple sum of self-reported usage frequencies for four media types: newspapers, television, news websites operated by traditional media outlets, and Yahoo! News. The frequency of social media use was calculated as the sum of usage frequencies for six platforms: Facebook, X(Twitter), LINE, YouTube, other news aggregation sites, and blogs. We designed the media-use measures to capture general patterns of informational exposure within mass media and social media environments rather than platform-specific effects. For this reason, the study combined use frequencies across multiple media sources into additive indices. Although the included platforms differ in informational content and interaction structure, the present study focused on the overall level of exposure to broader media environments across the longitudinal observation period. Similarly, the study operationalised subjective media literacy and media scepticism as additive indices reflecting generalised orientations toward information evaluation and media credibility. Because the present study employed five-wave longitudinal data, additive scores helped maintain comparability across waves and avoided instability associated with wave-specific weighting or factor-loading differences. Table 2 summarises the predictors used for the analysis.

Table 2.

Predictor variables included in the analysis.

Variables	Description
Frequency of mass media use	Simple addition of frequencies of using four types of media (e.g., newspapers, TV, news sites operated by newspapers and TV stations, Yahoo! News).
Frequency of social media use	Simple addition of frequencies of using six types of social media (e.g., Facebook, X(Twitter), LINE, YouTube, other news sites, summary sites and blogs).
Media scepticism	Simple addition of three measures assessing scepticism toward the validity of information in mass media.
Subjective media literacy	Simple addition of five measures about the subjective belief of being able to correctly judge the accuracy of information on mass media.

Predicting outcomes using latent growth modelling

For the observed values of each time-varying predictor in the five-wave panel survey, we estimated both the intercept and the linear slope using LGM²⁹. We carried out the analysis within the framework of LGM with distal outcomes^30–32, which employs latent factors (the intercept factor and the slope factor) estimated from the growth model to predict outcomes.

The advantage of this model is that it distinguishes between the effects of initial levels of media usage frequency, literacy, and media scepticism at the start of the pandemic and the effects of their subsequent growth rates. Compared with panel designs that rely on only two measurement points, a five-wave design allows us to estimate growth rates more precisely and reduce the influence of measurement error, thereby producing more robust results. As prior research has noted³², a key issue with this method is that results can vary depending on the chosen time slice. To obtain time-coding-invariant results that remain interpretable regardless of arbitrary coding decisions, we used an incremental validity approach (see the supplementary material for details). We conducted estimation with the robust maximum likelihood (MLR) method. Full details of the model specification, time coding, and fit indices are provided in the Supplementary Information. We compared linear and quadratic growth specifications for all repeatedly measured predictor variables (see Supplementary Information). Because the quadratic models frequently produced estimation and identification problems, we retained the linear growth specifications as a more stable modelling approach.

To examine the respective effects of media usage behaviour and the cognitive-attitudinal constructs of subjective literacy and media scepticism on conspiracy theory beliefs, we estimated three models: Model 1 included only media usage variables, Model 2 included only subjective literacy and media scepticism, and Model 3 incorporated both sets of predictors simultaneously. Age and gender were included as time-invariant covariates in all models. Only the regression results predicting the distal outcome from latent growth factors for each model are presented in Tables 3 and 4.

Table 3.

LGM predicting institutional conspiracy beliefs (2020-2024): Regression results from LGM analyses. Initial levels and growth rates of media usage, subjective media literacy, and media scepticism were used as predictors, alongside age and gender. Institutional conspiracy beliefs were primarily associated with low mass media use and heightened media scepticism over time.

	Model 1			Model 2			Model 3
	Estimate		SE	Estimate		SE	Estimate		SE
Mass media intercept	−0.91	**	(0.27)				−0.39		(0.37)
Mass media slope	−6.46		(4.89)				−4.19		(6.75)
Social media intercept	0.41	+	(0.22)				0.22		(0.30)
Social media slope	0.62		(2.60)				−1.45		(3.06)
Media literacy intercept				−0.19		(0.36)	−0.20		(0.38)
Media literacy slope				−3.69		(7.23)	−0.86		(9.26)
Media scepticism intercept				2.12	**	(0.33)	2.03	**	(0.34)
Media scepticism slope				20.23	*	(8.64)	19.75	*	(8.85)
Age	−0.53	*	(0.22)	−0.77	**	(0.19)	−0.62	**	(0.20)
Gender	0.01		(0.09)	0.00		(0.08)	−0.01		(0.08)
	0.07			0.35			0.36

.

Table 4.

LGM predicting implausible conspiracy beliefs (2020-2024): Regression results from LGM analyses for the implausible conspiracy factor. Younger age predicted higher scores, but overall endorsement was rare. Estimates should be interpreted with caution given the limited explanatory power of this outcome.

	Model 1			Model 2			Model 3
	Estimate		SE	Estimate		SE	Estimate		SE
Mass media intercept	−0.53	+	(0.30)				−0.41		(0.33)
Mass media slope	−3.16		(5.52)				−3.31		(7.03)
Social media intercept	0.20		(0.24)				0.27		(0.28)
Social media slope	−4.17		(2.75)				−5.21	+	(3.10)
Media literacy intercept				0.09		(0.28)	−0.01		(0.31)
Media literacy slope				−1.88		(5.72)	2.61		(7.84)
Media scepticism intercept				0.05		(0.26)	0.01		(0.29)
Media scepticism slope				8.64		(5.67)	8.81		(6.00)
Age	−0.64	**	(0.21)	−0.85	**	(0.20)	−0.68	**	(0.21)
Gender	−0.11		(0.09)	−0.10		(0.09)	−0.11		(0.09)
	0.06			0.06			0.11

.

Overall, younger age was associated with higher scores on institutional conspiracy beliefs. When considering only media usage, individuals who did not typically use mass media scored higher on this factor. When considering only subjective literacy and media scepticism, individuals who were already highly sceptical and those whose scepticism increased over the five-year survey period scored higher. Subjective media literacy and scepticism demonstrated greater explanatory power; once these factors were controlled for, the effect of media usage disappeared. Similarly, younger individuals tended to score higher on implausible conspiracy beliefs. However, because only a small minority of individuals endorsed implausible conspiracy beliefs, the results should be interpreted with caution, given their limited explanatory power.

Discussion

Our longitudinal study investigated the determinants of belief in vaccine-related conspiracy theories during the COVID-19 Pandemic, drawing on five annual panel surveys conducted from 2020 to 2024. The analyses revealed that vaccine-related conspiracy beliefs consist of two distinct factors: one concerning institutional distrust (the institutional conspiracy beliefs factor) and another reflecting more extreme and implausible claims (the implausible conspiracy beliefs factor). Importantly, these two forms of conspiracy thinking are shaped by different psychological attitudes and patterns of media use.

Although the two-factor distinction emerged empirically rather than from a preregistered theoretical framework, it aligns with broader literature suggesting that conspiracy beliefs are psychologically heterogeneous rather than fully reducible to a single continuum. Previous research has argued that conspiracy beliefs arise from multiple motives³³. In our data, institutional conspiracy beliefs primarily reflected distrust toward existing authorities and institutions, whereas implausible conspiracy beliefs centred on highly fringe narratives involving microchips and mind-control devices. The latter may reflect stronger tendencies toward illusory pattern perception or meaning-seeking under conditions of uncertainty, consistent with previous findings linking lack of control to conspiracy perception and pattern detection³⁴. Although our study did not directly examine these underlying mechanisms, the substantial differences in content, prevalence, and distributional characteristics between the two factors suggest that they represent qualitatively distinct forms of conspiracy-oriented thinking rather than merely stronger or weaker manifestations of the same underlying worldview.

Institutional conspiracy beliefs were strongly associated with a persistent scepticism toward mass media, a lack of engagement with traditional media, and an increase in media scepticism throughout the pandemic. Younger individuals also tended to score higher on this factor. These findings suggest that institutional distrust does not merely reflect responses to pandemic-related information but may instead reflect broader, relatively stable orientations toward social and informational institutions. This aligns with prior research linking conspiracy beliefs to anti-establishment attitudes and populist worldviews^4,5,35. Thus, institutional conspiracy beliefs appear to be rooted in stable predispositions that shaped people’s responses to health communication during the pandemic.

In contrast, implausible conspiracy beliefs, such as claims that vaccines contain microchips or mind-control devices, were endorsed only by a very small minority of respondents. Although younger individuals tended to report higher scores on this factor, the models had limited explanatory power. Given the low prevalence of such beliefs in our sample, statistical analyses should be interpreted with caution. While some results hinted at associations with changing social media use, our data do not allow us to determine clear causal mechanisms. Disengagement from mainstream or widely used platforms may possibly reflect exposure to fringe content through alternative online or offline networks, but this remains speculative and requires further investigation.

Importantly, these findings underscore the need for caution when interpreting social media content analyses or computational studies that rely heavily on high-visibility posts. Our results raise the possibility that extreme and highly implausible conspiracy beliefs, although endorsed by only a small minority, may nevertheless attain disproportionate visibility in online environments. This pattern may have implications for computational analyses of online discourse, particularly those based on platforms such as X(Twitter) that amplify highly visible content^36,37. Future research should examine whether highly visible fringe content accurately reflects broader patterns of public endorsement.

Taken together, these findings highlight the need to treat institutional and implausible conspiracy beliefs as qualitatively distinct phenomena. Institutional conspiracy beliefs were widespread and systematically related to media scepticism and distrust toward authorities, whereas implausible conspiracy beliefs remained rare and should therefore be interpreted more cautiously. For future research and public health communication, this distinction may have important implications. Addressing institutional conspiracy beliefs may require rebuilding trust in mainstream media and political institutions, whereas responding to highly implausible narratives may require more targeted approaches directed at relatively small, potentially isolated groups.

Several limitations should be acknowledged. First, the media-use variables consisted of broad additive indices that combined heterogeneous platforms. Although this approach allowed us to capture general patterns of informational exposure across five longitudinal waves, it may obscure important differences in the informational content, exposure dynamics, and social functions of specific platforms. Future research should therefore examine platform-specific effects in greater detail. Second, although the present study focused on media scepticism and media exposure as predictors of conspiracy beliefs, conspiracy-related attitudes likely emerge from broader social and psychological processes. Previous research has shown that trust in science, perceived ingroup norms, and national solidarity also shape conspiracy beliefs during public health crises³⁸. The present findings therefore should not be interpreted as suggesting that media scepticism alone explains vaccine-related conspiracy beliefs. Future research would benefit from integrating media-related variables with broader trust, identity, and social-norm frameworks to better understand the multifaceted origins of conspiracy beliefs. Third, because the sample was recruited through an online crowdsourcing platform, the findings may not fully generalise to populations with different media-use patterns or levels of digital engagement. Finally, the observed age effects should be interpreted with caution. In the present panel design, age may reflect a combination of cohort-related differences in media environments and selective retention processes. Consequently, the observed age effects should not be interpreted as reflecting simple age-related differences in conspiracy beliefs.

Finally, our study contributes in three ways. First, it identifies two distinct types of vaccine-related conspiracy beliefs shaped by different psychological and informational factors. Second, by employing a five-wave longitudinal design, we were able to distinguish relatively stable individual predispositions from pandemic-specific changes, thereby providing stronger temporal evidence than cross-sectional studies. Third, by focusing on the Japanese context, this study broadens a field often dominated by Western-centric data and adds valuable cultural variation. In particular, the present findings suggest that conspiracy beliefs can emerge even in comparatively consensus-oriented media environments characterised by stronger normative expectations of media neutrality and lower levels of overt partisan fragmentation than those typically observed in Western contexts. Overall, our findings suggest that although extreme and implausible conspiracy beliefs attract disproportionate visibility in digital environments, the more consequential and widespread challenge for public health communication may stem from persistent scepticism towards mainstream media information and related informational environments.

Author contributions

H.Y. initiated and performed the project; H.Y., T.S. Y.O. and R.U. designed the survey questionnaire. T.S. analysed the data; H.Y. and T.S. wrote the paper; H.Y. and T.S. contributed equally to this work; All authors reviewed the manuscript.

Funding

This work was supported by JSPS KAKENHI Grant Numbers 21KK0027 (H.Y. and T.S.) and 23K25160 (H.Y.).

Data availability

A part of data of this study are stored in an OSF data package titled “Data of Longitudinal predictors of vaccine conspiracy beliefs: Findings from a five-wave panel study in Japan,” which can be accessed at the below link. Other data is available from the corresponding author on reasonable request. https://doi.org/10.17605/OSF.IO/9EAYJ

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

The present series of experiments was approved by the Ethics Committee of Rissho University (Ethics approval number: 02-1, 03-2, 03-7, 05-2, and 06-2) and conducted in accordance with the requirements of the Declaration of Helsinki. All participants were informed about the purpose of the study as well as the ways the data would be used. They agreed that the data would be used only for scientific research, that all data would be anonymised, and that they had the right to stop responding at any time. Informed consent was obtained from all participants.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-57049-5.