Temporal Dynamics of Vaccination Decision-Making: How Trust and Risk Perception Evolved During COVID-19 in Germany

Abstract

The COVID-19 pandemic created unprecedented conditions for examining how vaccination willingness evolves during prolonged health crises. This longitudinal mixed-methods study examines temporal dynamics in COVID-19 vaccination willingness across three phases of Germany’s vaccination campaign (N = 1063 survey respondents; n = 40 interview participants). Using mixed-effects models and thematic analysis, we tested whether institutional trust and personal risk perception predict vaccination willingness and how their relative importance changes over time. Results reveal that trust in scientific institutions emerges as the strongest predictor, outperforming political trust and becoming more influential over time, while risk perceptions become less predictive with time. Qualitative analysis identified a multitude of different argumentative themes for and against COVID-19 vaccination (as well as conditional acceptance), with 30% of participants expressing both. The themes complement the quantitative analysis by demonstrating a shift from analytical, risk-focused decision-making to heuristic, trust-based processing as vaccination campaigns progress, with important implications for adaptive public health communication strategies.

1. Introduction

The COVID-19 pandemic created an unprecedented disruption in understanding how vaccination willingness evolves during a prolonged public health crisis. Unlike previous vaccination campaigns, COVID-19 presented a unique scenario where individuals formed attitudes about hypothetical vaccines before their availability and then potentially revised these views as real-world implementation unfolded across distinct rollout phases. This temporal progression offers crucial insights into a fundamental question: how does the balance between institutional trust and personal risk perception shift as a vaccination campaign progresses from hypothetical consideration to mass implementation?

A growing body of evidence documents that vaccine confidence and willingness did not uniformly increase with disease salience during COVID-19. Instead, they often declined over time. Early in the pandemic (April 2020), a multi-country European survey already showed substantial uncertainty alongside high stated acceptance (overall 73.9%; Germany with comparatively large “no” and “unsure” shares), with safety and side-effect concerns prominent among the hesitant [1]. Subsequent studies observed drops in willingness in several settings (e.g., a decline in the United States during 2020) and substantial within-country variation over time, underscoring that attitudes were dynamic rather than fixed [2,3,4]. Beyond COVID-19, recent syntheses indicate broader setbacks in routine immunization coverage and confidence since 2020, suggesting spillovers that extend past the immediate pandemic context [5]. Longitudinal panel evidence across seven European countries further shows that willingness, hesitancy, and refusal states shifted in both directions across nine waves (“comes in waves”), rather than following a simple monotonic path [6,7,8].

These temporal patterns shape the theoretical puzzle of vaccination behavior: how can shifts in vaccination willingness be explained over time? Existing research on vaccination willingness has predominantly relied on cross-sectional designs that capture attitudes at single time points, providing limited insight into how decision-making processes evolve during extended health crises [9]. Comparative analyses across eight European countries highlight heterogeneity in hesitancy and its correlates [10], while country-specific studies point to the centrality of trust and perceived risk as decision inputs [11]. In Germany, for instance, trust in medical experts and authorities, political preferences, and regional context are systematically associated with vaccination status and attitudes [12,13]. At the regional level, structural factors (demographics, education, economic context, and party preferences) account for a large share of county and state differences in uptake, yet residual variation suggests additional roles for “soft factors” such as trust in science and conspiracy beliefs [14]. Together, these findings imply that the relative weight of institutional trust versus personal risk perception may evolve as campaigns move from hypothetical vaccines to concrete choices with visible social and policy consequences. In summary, while studies consistently identify institutional trust and risk perception as key predictors [15,16,17,18], we lack understanding of how the relative importance of these factors changes as individuals move from anticipatory decision-making to concrete choices with observable social consequences.

Germany provides an ideal context for examining these temporal dynamics due to its robust scientific institutions and three-phase vaccination rollout that created policy variations throughout time. German institutions such as the Robert Koch Institute (RKI) and Standing Committee on Vaccination (STIKO) maintained scientific independence while navigating political pressures, enabling an empirical distinction between trust in scientific versus political institutions [19]. This institutional landscape allows for theoretically informed examination of how different forms of trust predict vaccination willingness over time.

This study employs a longitudinal mixed-methods design to test two primary hypotheses about vaccination willingness across Germany’s three-phase COVID-19 vaccination campaign. Building on the above literature, we leverage Germany’s phased rollout to examine whether (and how) the balance between institutional trust and personal risk perception shifts from the pre-availability stage to mass implementation. By situating individual-level trajectories within documented cross-national declines in confidence and the demonstrated impact of policy levers, our design directly addresses the need, as identified in prior longitudinal work, to explain transitions between hesitant, willing, and refusing states over time [8]. In doing so, this study clarifies mechanisms that cross-sectional snapshots cannot reveal and informs how future campaigns might sustain confidence as conditions and policies evolve.

2. Theoretical Framework

2.1. Key Determinants of Vaccination Willingness

Contemporary understanding of vaccination willingness is anchored in the World Health Organization’s 5Cs model, which identifies five psychological antecedents: confidence, complacency, constraints, calculation, and collective responsibility [20]. This framework captures the multidimensional nature of vaccination decisions while highlighting two constructs central to our study: confidence (encompassing institutional trust) and complacency (reflecting risk perception).

Confidence emerges as the strongest predictor across multiple studies, encompassing trust in vaccine effectiveness, safety, and the institutions delivering immunization [17,21]. However, institutional trust operates through distinct mechanisms depending on its source. Trust in scientific institutions derives from competence-based credibility grounded in peer review, replication, and evidence-based processes, while trust in political institutions stems from position-based authority rooted in electoral mandates and democratic procedures [12]. This fundamental difference creates distinct cognitive processing patterns, with scientific trust engaging analytical evaluation of evidence quality while political trust often involves identity-based reasoning and partisan considerations [22].

Research consistently demonstrates that scientific trust serves as a stronger and more stable predictor of vaccination willingness than political trust. Studies examining COVID-19 vaccination correlates found that institutional trust in medical experts and authorities was positively associated with vaccination status, whereas trust in political institutions showed more variable effects [12].

Risk perception, captured in the 5Cs model as complacency, reflects individuals’ assessment of disease threat and vaccination necessity. The Health Belief Model demonstrates that perceived benefits and barriers strongly predict vaccination behavior when integrated with trust-based frameworks [23].

Temporal Dynamics in Vaccination Willingness

Longitudinal research reveals systematic changes in how trust and risk perception predict vaccination willingness over time. Studies using vector autoregression network analyses demonstrate that while cross-sectional results indicate that vaccination willingness is strongly related to attitudes toward vaccines, temporal analyses show that early willingness primarily predicts subsequent vaccination-related variables rather than being determined by them [9]. This suggests fundamental shifts in decision-making processes as health crises progress.

Several mechanisms explain these temporal patterns. Information processing changes occur as individuals increasingly rely on trusted sources rather than personal risk calculations [23]. Trust serves as a cognitive heuristic to reduce the complex decision-making burden, while social learning from observing others’ vaccination experiences shifts the focus from abstract risk to concrete institutional relationships. Research identifies distinct phases: an initial assessment phase dominated by personal risk perception; a social learning phase where institutional messaging becomes more influential; and a routine acceptance phase where trust in institutions predominates [24].

Longitudinal studies during COVID-19 provide initial evidence for this progression. Emerging research on changes in vaccination attitudes over the first year of the pandemic indicates that complacency about disease risk, mistrust of vaccine benefit, and concerns about institutional motives may have decreased, while collective responsibility appears to have increased [23]. These preliminary patterns point toward the possibility that sustained exposure to health threats and institutional responses could lead to systematic shifts from analytical, risk-focused decision-making to heuristic, trust-based processing.

2.2. The German Context: Institutional Landscape and Study Rationale

Germany’s institutional context provides unique advantages for examining temporal dynamics in vaccination willingness. The country’s scientific institutions maintained credibility throughout the pandemic while operating within a federal structure that created variation in implementation approaches. Research conducted during the pre-approval period revealed widespread confidence in German institutions, with respondents expressing that the RKI, Paul Ehrlich Institute (PEI), and STIKO would not approve questionable vaccines [19]. This institutional trust was sharply contrasted with skepticism toward vaccine development in other countries, highlighting the importance of domestic scientific credibility [25].

Qualitative research in Germany identified four major themes in vaccination decision-making: deliberation about benefits versus risks, social and political context, emotions toward the pandemic, and trust in vaccines, science, and government-affiliated health institutions [26]. This DCET framework (Deliberation, Context, Emotion, Trust) reveals how participants navigated competing narratives shaped by emotional responses to evolving contexts, providing ideal conditions for examining how trust and risk perception interact over time.

Germany’s three-phase vaccination rollout created ideal conditions for longitudinal analysis. The progression from priority groups (December 2020) through general access (June 2021) to ongoing boosters enabled examination of how vaccination willingness predictors evolved from hypothetical consideration to mass implementation. This temporal structure, combined with Germany’s robust survey infrastructure and cultural emphasis on evidence-based decision-making, provides an optimal setting for testing theoretical predictions about changing determinants of vaccination willingness.

The mixed-methods approach is particularly well-suited to this context, as vaccination decisions involve both rational deliberation and intuitive processing, individual calculations and social influences, and stable attitudes alongside dynamic responses to changing circumstances. By combining quantitative identification of predictor patterns with qualitative understanding of decision processes, this design can capture the full complexity of vaccination willingness while providing actionable insights for public health policy.

2.3. Research Objectives and Hypotheses

Building on this theoretical framework, we examine how institutional trust and personal risk perception predict COVID-19 vaccination willingness across three phases of Germany’s vaccination campaign. Our longitudinal mixed-methods design enables testing of two primary hypotheses:

Hypothesis 1.

Institutional trust and personal risk perception will emerge as the strongest predictors of vaccination willingness, with trust in scientific institutions showing stronger effects than trust in political institutions.

Hypothesis 2.

Temporal dynamics will reveal trust in scientific institutions becoming more predictive over time while personal risk perception becomes less influential, reflecting a shift from analytical, risk-focused decision-making to heuristic, trust-based processing.

These hypotheses address a critical gap in vaccination research by examining not only which factors predict willingness, but how their relative importance evolves during prolonged health crises. The findings will contribute to both theoretical understanding of health behavior change and practical applications for designing adaptive public health communication strategies that respond to shifting psychological processes during pandemic vaccination campaigns.

3. Materials and Methods

This study employed a longitudinal, mixed-methods design combining a nationally representative panel survey with semi-structured interviews to examine COVID-19 vaccination willingness in Germany. Data collection occurred across three distinct phases corresponding to key stages of the vaccination rollout: Phase 1 (November–December 2020, pre-vaccination availability), Phase 2 (March–April 2021, early rollout with limited availability), and Phase 3 (August–September 2021, mass vaccination phase with widespread availability). This temporal design enabled examination of both vaccination willingness trajectories and the evolving influence of psychological and sociodemographic predictors. Both the panel datasets [27] as well as the interview dataset [28] are published as open access datasets and can be found alongside more information about data collection and processing in the corresponding dataset papers. For the present study, we utilized a specific subset of variables from these datasets, focusing on (1) vaccination willingness measures, (2) institutional trust scales for political and scientific actors, (3) science attitudes, (4) personal risk perception, (5) conspiracy beliefs, and (6) key sociodemographic variables (age, sex, education, political orientation, household income, immigration background, and social media usage). The complete survey questionnaire with exact item wording, response scales, and variable specifications for all measures used in this analysis is provided in Appendix A.1 

Table A1. Variable overview by study phase (Phase I–III).

Description	Phase I (November/December 2020)	Phase II (March 2021)	Phase III (August/September 2021)
	N = 1480	N = 482	N = 426
SD—Age	SD_AGE	–	–
SD—Sex	SD_SEX	–	–
SD—Immigration background	SD_IMMIGRATION_YEAR	–	–
SD—Education	SD_SCHOOL_C	–	–
SD—Political orientation	SD_POL_ORIENTATION	–	–
SD—Household income	SD_INCOME_DEC2019	–	–
SD—Twitter usage (COVID-19 info)	PHASE1_IST_SOCMEDIA_FREQ_TW	–	–
Overall Trust
	PHASE1_IST_TRUST_MERKEL; PHASE1_IST_TRUST_SPAHN; PHASE1_IST_TRUST_STATEGOV; PHASE1_IST_TRUST_HMINISTRY; PHASE1_IST_TRUST_WHO; PHASE1_IST_TRUST_RKI; PHASE1_IST_TRUST_DROSTEN; PHASE1_IST_TRUST_PRIMARY	PHASE2_IST_TRUST_MERKEL; PHASE2_IST_TRUST_SPAHN; PHASE2_IST_TRUST_STATEGOV; PHASE2_IST_TRUST_HMINISTRY; PHASE2_IST_TRUST_WHO; PHASE2_IST_TRUST_RKI; PHASE2_IST_TRUST_DROSTEN; PHASE2_IST_TRUST_PRIMARY	PHASE3_IST_TRUST_MERKEL; PHASE3_IST_TRUST_SPAHN; PHASE3_IST_TRUST_STATEGOV; PHASE3_IST_TRUST_HMINISTRY; PHASE3_IST_TRUST_WHO; PHASE3_IST_TRUST_RKI; PHASE3_IST_TRUST_DROSTEN; PHASE3_IST_TRUST_PRIMARY
Political Trust
	PHASE1_IST_TRUST_MERKEL; PHASE1_IST_TRUST_SPAHN; PHASE1_IST_TRUST_STATEGOV; PHASE1_IST_TRUST_HMINISTRY	PHASE2_IST_TRUST_MERKEL; PHASE2_IST_TRUST_SPAHN; PHASE2_IST_TRUST_STATEGOV; PHASE2_IST_TRUST_HMINISTRY	PHASE3_IST_TRUST_MERKEL; PHASE3_IST_TRUST_SPAHN; PHASE3_IST_TRUST_STATEGOV; PHASE3_IST_TRUST_HMINISTRY
Trust in Science
	PHASE1_IST_TRUST_WHO; PHASE1_IST_TRUST_RKI; PHASE1_IST_TRUST_DROSTEN	PHASE2_IST_TRUST_WHO; PHASE2_IST_TRUST_RKI; PHASE2_IST_TRUST_DROSTEN	PHASE3_IST_TRUST_WHO; PHASE3_IST_TRUST_RKI; PHASE3_IST_TRUST_DROSTEN
Science Attitudes
	PHASE1_AS_BOR_EXC; PHASE1_AS_UNIMP_IMP; PHASE1_AS_USEL_USEF; PHASE1_AS_HARM_BEN; PHASE1_AS_DIS_HON; PHASE1_AS_UNTR_TRUST	PHASE2_AS_USEL_USEF; PHASE2_AS_UNTR_TRUST; PHASE2_AS_BESCI; PHASE2_AS_WELLINF; PHASE2_AS_HARM_BEN	PHASE3_AS_USEL_USEF; PHASE3_AS_UNTR_TRUST; PHASE3_AS_BESCI; PHASE3_AS_WELLINF; PHASE3_AS_HARM_BEN
Personal Risk
	PHASE1_RA_HEALTH_PERS; PHASE1_RA_INFECT	PHASE2_RA_HEALTH_PERS; PHASE2_RA_INFECT	PHASE3_RA_HEALTH_PERS; PHASE3_RA_INFECT
Conspiracy Beliefs
	PHASE1_CT_HOAX; PHASE1_CT_5G; PHASE1_CT_BIOENGINEERED; PHASE1_CT_CHINESELAB; PHASE1_CT_NOCONSULT; PHASE1_CT_SURVEIL; PHASE1_CT_SECRETACT; PHASE1_CT_ENFORCEVACC	PHASE2_CT_HOAX; PHASE2_CT_5G; PHASE2_CT_BIOENGINEERED; PHASE2_CT_CHINESELAB; PHASE2_CT_NOCONSULT; PHASE2_CT_SURVEIL; PHASE2_CT_SECRETACT; PHASE2_CT_ENFORCEVACC	PHASE3_CT_HOAX; PHASE3_CT_5G; PHASE3_CT_BIOENGINEERED; PHASE3_CT_CHINESELAB; PHASE3_CT_NOCONSULT; PHASE3_CT_SURVEIL; PHASE3_CT_SECRETACT; PHASE3_CT_ENFORCEVACC
Vaccine Willingness	PHASE1_HM_VOLVACC	PHASE2_HM_VOLVACC	PHASE3_HM_VOLVACC

Dashes (–) indicate that the variable was not asked/recorded in that phase. Codes follow dataset naming.

and

Table A2. Survey items: descriptions, item phrasings, and response scales.

Description	Item Name (Phase I)	Item Phrasing	Scale
Socio demographic
Age	SD_AGE	Age	Open [Continuous scale]
Sex	SD_SEX	What sex were you assigned at birth? (biological sex, e.g., on birth certificate)	1 = Female; 2 = Male; 3 = Diverse; 4 = No entry
Immigration background	SD_IMMIGRATION	Did you or your parents move to Germany after 1955?	1 = Yes; 2 = No; -98 = Other; −97 = Prefer not to say; −96 = Not applicable/No opinion; −95 = Unsure; −94 = None of the above
Education	SD_SCHOOL_C	What’s your highest school leaving qualification?	1 = No school leaving qualification; 2 = Still in school; 3 = Haupt-/Volksschulabschluss; 4 = Realschulabschluss; 5 = Fachhochschulreife; 6 = Abitur
Political orientation	SD_POL_ORIENTATION	Left–Right self-placement.	−3 (Left) to 3 (Right); special codes −98 to −94
Household income	SD_INCOME_DEC2019	Monthly household income (December 2019).	1 = <€500; 2 = €500–€1000; 3 = €1000–€1500; 4 = €1500–€2000; 5 = €2000–€2500; 6 = €2500–€3000; 7 = €3000–€3500; 8 = €3500–€4000; 9 = €4000–€4500; 10 = €4500–€5000; 11 = €5000–€5500; 12 = €5500–€6000; 13 = €6000–€10,000; 14 = €10,000–€18,000; 15 = >€18,000; special codes −98 to −94
Twitter usage	PHASE1_IST_SOCMEDIA_FREQ_TW	How frequently do you use Twitter for COVID-19 information?	1 = Never to 7 = Always; special codes −98 to −94
Trust indices		Please indicate to what extent you distrust or trust the following sources for reliable information about the Coronavirus (COVID-19) situation.	−2 = Completely distrust to 2 = Completely trust; −96 = Not applicable/No option
Angela Merkel	PHASE1_IST_TRUST_MERKEL	Angela Merkel (Chancellor)
Jens Spahn	PHASE1_IST_TRUST_SPAHN	Jens Spahn (Health Minister)
State government	PHASE1_IST_TRUST_STATEGOV	State government
Health Ministry	PHASE1_IST_TRUST_HMINISTRY	German Public Health Ministry
WHO	PHASE1_IST_TRUST_WHO	World Health Organization
RKI	PHASE1_IST_TRUST_RKI	Robert Koch Institute
Christian Drosten	PHASE1_IST_TRUST_DROSTEN	Christian Drosten (virologist)
Primary news source	PHASE1_IST_TRUST_PRIMARY	Primary news source [pre-selected]
Science attitudes		For each pair of words below, please select the point between them that you think best describes SCIENCE.	7-point scale: −3 to 3; special codes −98 to −94
Excitement	PHASE1_AS_BOR_EXC	Boring–Exciting
Importance	PHASE1_AS_UNIMP_IMP	Unimportant–Important
Usefulness	PHASE1_AS_USEL_USEF	Useless–Useful
Harmfulness	PHASE1_AS_HARM_BEN	Harmful–Beneficial
Honesty	PHASE1_AS_DIS_HON	Dishonest–Honest
Trustworthiness	PHASE1_AS_UNTR_TRUST	Untrustworthy–Trustworthy
Personal risk		Thinking about the Coronavirus (COVID-19) situation, please indicate your level of agreement with the following statements.	1 = Strongly Disagree to 7 = Strongly Agree; special codes −98 to −94
Concern about own health	PHASE1_RA_HEALTH_PERS	I am concerned about my own health.
Concern about infecting others	PHASE1_RA_INFECT	I am concerned that I could infect others with Coronavirus (COVID-19).
Conspiracy beliefs		Which of the following statements about the Coronavirus (COVID-19) do you agree with?	1 = Strongly Disagree to 7 = Strongly Agree; special codes −98 to −94
Hoax	PHASE1_CT_HOAX	The Coronavirus (COVID-19) is a hoax.
5G	PHASE1_CT_5G	The new 5G network is making us more susceptible to the virus.
Bioengineered	PHASE1_CT_BIOENGINEERED	The coronavirus was bioengineered in a military lab.
Chinese lab	PHASE1_CT_CHINESELAB	The Coronavirus (COVID-19) originated in a Chinese lab.
Non-transparent Decisions	PHASE1_CT_NOCONSULT	Many important decisions about the Coronavirus (COVID-19) situation are made without the public ever being informed.
Surveillance	PHASE1_CT_SURVEIL	The Coronavirus (COVID-19) situation has provided an excuse for government agencies to closely monitor all citizens.
Secret activities	PHASE1_CT_SECRETACT	The Coronavirus (COVID-19) situation has happened because of secret activities outside of Germany.
Mandatory vaccination	PHASE1_CT_ENFORCEVACC	The coronavirus is part of a global effort to enforce mandatory vaccination.
Vaccine willingness	PHASE1_HM_VOLVACC	How would you feel about taking the following step on a voluntary basis? Coronavirus (COVID-19) vaccination.	1 = Definitely not; 2 = Probably not; 3 = Maybe; 4 = Probably; 5 = Definitely; −97/−96

Item stems/scales reproduced verbatim or lightly formatted for better visualization within the table.

.

4. Item Overview

4.1. Quantitative Component: Longitudinal Panel Survey

4.1.1. Sampling and Recruitment

The survey targeted the German population aged 16 and over as part of the Viral Communication project (https://www.viralcomm.info/, accessed on 6 September 2025). Recruitment involved mailing postcard invitations to 30,000 randomly selected households using the German postal service database, with addresses stratified by federal state population distribution to ensure geographical representativeness.

From 1480 initial survey entries, 417 respondents were excluded for incomplete demographic data required for census weighting, yielding a final sample of N = 1063 respondents across all three phases. The weighted sample comprised 53% female participants with a mean age of 48.9 years (SD = 18.6). Phase-specific sample sizes varied due to panel attrition and the natural reduction in unvaccinated respondents over time: Phase 1 (N = 1106), Phase 2 (N = 416), and Phase 3 (N = 40 unvaccinated respondents).

Survey weights were applied to align the sample with German census data (Zensus 2011) across multiple dimensions: age, sex, nationality, migration background, federal state, and education level. This weighting procedure corrected for sampling biases inherent in postal recruitment and ensured that findings could be generalized to the broader German population.

4.1.2. Measures

The following section details the key variables extracted from the larger Viral Communication dataset for this specific analysis. All measures were administered across multiple study phases as indicated, with complete question wording and response options provided in Appendix A.1 Table A1 and Table A2.

Vaccination Willingness. The primary outcome variable assessed willingness to receive a COVID-19 vaccination using the question “How would you feel about taking the following steps on a voluntary basis? Coronavirus (COVID-19) vaccination”, with responses on a 5-point ordinal scale: 1 = definitely not, 2 = probably not, 3 = maybe, 4 = probably yes, and 5 = definitely yes. In Phase 1, this represented hypothetical willingness; in subsequent phases, it applied only to unvaccinated respondents.

Institutional Trust. Trust was measured using separate scales for scientific and political institutions. Trust in scientific actors included confidence ratings for the World Health Organization (WHO), Robert Koch Institute (RKI), and virologist Christian Drosten using 5-point scales from −2 (completely distrust) to 2 (completely trust). Trust in political institutions measured confidence in Chancellor Angela Merkel, Health Minister Jens Spahn, state governments, and the German Public Health Ministry using identical scales. An overall trust index combined both dimensions for parsimony in primary analyses. All trust scales demonstrated high internal consistency across study phases (Cronbach’s $\alpha$ = 0.84–0.94).

Science Attitudes. Attitudes toward science were assessed using a validated scale measuring general confidence in scientific methods, research processes, and scientific institutions. Items used semantic differential scales (7-point, −3 to 3) asking respondents to rate science on dimensions such as unimportant–important, useless–useful, harmful–beneficial, and dishonest–honest. The scale showed acceptable reliability ($\alpha$ = 0.73–0.80).

Personal Risk Perception. COVID-19 risk perception was measured through items assessing perceived personal vulnerability to infection, severity of potential illness, and concern about health consequences. Key items included “I am concerned about my own health” and “I am concerned that I could infect others with Coronavirus (COVID-19)” rated on 7-point Likert scales from 1 (strongly disagree) to 7 (strongly agree). Despite moderate reliability ($\alpha$ = 0.56–0.62), this scale captured meaningful individual differences in risk assessment.

Conspiracy Beliefs. Conspiracy-related thinking was assessed using items measuring belief in alternative explanations for COVID-19 origins, distrust of official narratives, and endorsement of conspiracy theories related to the pandemic. Items included beliefs that COVID-19 is a hoax, that 5G networks increase susceptibility, that the virus was bioengineered in a military lab, originated in a Chinese lab, and represents a global effort to enforce mandatory vaccination, and that important decisions are made without public consultation (all rated on 7-point Likert scales from 1 = strongly disagree to 7 = strongly agree). The scale demonstrated good reliability ($\alpha$ = 0.80–0.86).

Sociodemographic Variables. Standard demographic measures included age (continuous), gender (binary: male/female), education level (ordinal), household income (ordinal), political orientation (7-point left–right scale), migration background (binary), and social media usage patterns.

4.1.3. Statistical Analysis

The analysis strategy employed multiple complementary approaches to address the ordinal nature of the dependent variable and the longitudinal design. Given the ordered categorical nature of vaccination willingness, both linear and ordinal mixed-effects models were estimated to provide comprehensive insight into predictor effects.

Mixed-Effects Modeling. Linear mixed-effects models were fitted using the lme4 package in R, incorporating random intercepts for individuals to account for within-person clustering across time points. Ordinal mixed-effects models were estimated using the ordinal package to respect the categorical nature of the willingness scale. All models included fixed effects for study phase, psychological predictors, and sociodemographic covariates.

Model Specifications. Multiple model specifications were tested to ensure robustness. Primary analyses focused on weighted models to correct for sampling biases, with unweighted results provided in the Appendices for comparison. Models examined both overall trust indices and disaggregated scientific versus political trust to test theoretical predictions about differential effects. Interaction models tested whether predictor effects varied significantly across study phases.

Model Selection and Diagnostics. Model fit was assessed using likelihood ratio tests, AIC comparisons, and residual diagnostics. Random effects intraclass correlation coefficients (ICCs) quantified the proportion of variance attributable to between-person differences. Diagnostic plots examined residual patterns, outliers, and assumption violations.

Phase-Specific Analysis. Primary analyses focused on Phases 1 and 2 due to the small sample of unvaccinated respondents in Phase 3 (N = 40), which was insufficient for reliable mixed-effects modeling. Phase 3 data were treated qualitatively to understand the characteristics of persistent vaccination hesitancy, with quantitative summaries relegated to the Appendix C.

4.2. Qualitative Component: Semi-Structured Interviews

4.2.1. Participant Selection and Sampling

From the 936 survey respondents who provided complete vaccination willingness data, 278 expressed willingness to participate in follow-up interviews. Participants were selected using a two-stage purposive sampling approach designed to ensure both demographic representativeness and theoretical diversity in vaccination attitudes.

Primary sampling criteria balanced representation across sociodemographic variables: age group (16–29, 30–44, 45–59, and 60+ years), gender, and socioeconomic status (based on median income split). Secondary criteria considered attitudinal factors, including institutional trust levels, migration background, vaccination willingness, and attitudes toward protective measures. This approach ensured adequate representation of vaccination-hesitant perspectives, which comprised 22.5% of the interview sample compared to 15% in the broader survey.

The final interview sample comprised 40 participants across the three study phases: December 2020 (n = 40), April 2021 (n = 38), and September 2021 (n = 38). The sample included 55% women, with a balanced age distribution (30% aged 16–29, 25% aged 30–44, 22.5% aged 45–59, and 22.5% aged 60+) and 50% high-socioeconomic-status participants.

4.2.2. Data Collection Procedures

Interviews were conducted in German via telephone or Zoom by two trained interviewers: a female psychologist with extensive qualitative research experience (primary interviewer) and a male psychology undergraduate. Regular debriefing sessions ensured consistency and quality control across interviewers.

Semi-structured interviews followed a standardized guide with open-ended questions, allowing flexibility to pursue emerging themes while maintaining consistency across participants. The central vaccination question was personalized based on participants’ prior survey responses: “In your survey response, you mentioned that you’d [X] get a voluntary coronavirus vaccination. Could you explain why you’re feeling that way?”, where [X] represented each participant’s specific response level.

Interviews averaged 41 min and 56 s (range: 22–88 min). All sessions were audio-recorded with explicit informed consent and transcribed using an intelligent verbatim approach. Initial automatic transcription (f4x software, Version 2020) was followed by manual correction and verification using MAXQDA 2020. Transcripts removed pauses and filler words while anonymizing all identifying information.

4.2.3. Qualitative Analysis

Coding employed a multi-step inductive approach that prioritized participants’ perspectives through in vivo coding processes. Initial open coding was conducted by the principal investigator with support from three additional trained coders.

Quality control involved strategic team meetings and independent verification at three checkpoints (after 10, 21, and 40 interviews). At each checkpoint, an independent second coder applied the developing code system to previously coded interviews, with discrepancies discussed until consensus was reached (inter-rater $\kappa >0.70$). This iterative process ensured inter-rater reliability and definitional clarity of emerging codes.

Coding subsequently organized open codes into broader thematic categories while maintaining close connection to participant voices. The final thematic structure emerged through constant comparative analysis, identifying patterns within and across vaccination attitude groups and temporal phases.

4.3. Ethical Considerations and Data Management

The study protocol was approved by the institutional ethics committee. All participants provided informed consent for survey participation and, separately, for interview participation and audio recording. Data were stored securely with restricted access, and all interview transcripts were anonymized to protect participant confidentiality.

4.4. Integration of Quantitative and Qualitative Components

The mixed-methods design enabled triangulation of findings across data sources. Quantitative models identified significant predictors and temporal trends, while qualitative analysis illuminated the underlying reasoning processes and decision-making dynamics. Integration occurred through comparison of thematic findings with statistical results, examination of attitude stability and change across methods, and exploration of mechanisms underlying quantitative associations.

This complementary approach provided both population-level insights about vaccination willingness trajectories and in-depth understanding of individual decision-making processes, enhancing the comprehensiveness and validity of the study conclusions.

5. Results

5.1. Vaccination Willingness Trends Across Study Phases

Vaccination willingness varied substantially across the three phases of data collection, revealing the dynamic nature of vaccination attitudes (

Table 1. Descriptive statistics for vaccination willingness and uptake across study phases.

Phase	Willingness Mean (SD)	Vaccination Rate
Phase 1 (Pre-rollout)	3.55 (1.44)	—
Phase 2 (Early rollout)	4.36 (1.19)	11.6%
Phase 3 (Mass rollout)	2.38 (1.56)	89.5%

Note. Willingness measured on 5-point scale (1 = definitely not, 5 = definitely yes). Phase 3 willingness reflects only unvaccinated respondents.

and Figure 1). Before vaccine availability (Phase 1), respondents expressed moderate willingness (M = 3.55, SD = 1.44, N = 1106). Willingness increased significantly during early rollout (Phase 2; M = 4.36, SD = 1.19, N = 416), when 11.6% had already been vaccinated. By mass rollout (Phase 3), vaccine uptake reached 89.5%, but willingness among the remaining unvaccinated individuals was markedly low (M = 2.38, SD = 1.56, N = 40), indicating that hesitant individuals constituted the majority of those still unvaccinated.

5.2. Mixed-Effects Models: Predictors of Vaccination Willingness

We estimated both linear and ordinal mixed-effects models to account for the longitudinal design and examine predictor stability over time. Linear models provide easily interpretable effect sizes and allow straightforward comparison of predictor magnitudes, while ordinal models better respect the categorical nature of the willingness scale and account for potential threshold effects between response categories. The analysis focused on data from Phases 1 and 2, as Phase 3 had insufficient unvaccinated respondents (N = 40) for reliable modeling. Primary results focus on weighted models to correct for sampling biases, with unweighted results provided in the Appendix A.2 and Appendix A.3.

5.2.1. Core Psychological Predictors

Trust in institutions emerged as the strongest predictor across all model specifications. In the weighted ordinal model with overall trust (

Table 2. Ordinal mixed-effects model results: predictors of vaccination willingness (weighted analysis with overall trust).

Predictor	Coefficient $\mathit{\beta }$ (SE)	p-Value
Phase 2 vs. Phase 1	3.15 (0.30)	<0.001 ***
Trust (Overall Index)	1.45 (0.21)	<0.001 ***
Gender (Female)	1.35 (0.32)	<0.001 ***
Personal Risk Perception	0.63 (0.17)	<0.001 ***
Conspiracy Beliefs	$-0.44$ (0.18)	0.016 *
Science Attitudes	0.43 (0.16)	0.006 **
Income	0.16 (0.05)	0.002 **
Political Orientation	$-0.13$ (0.14)	0.367
Age	0.06 (0.01)	<0.001 ***

Note. *** p < 0.001, ** p < 0.01, * p < 0.05. Rows ordered by ∣β∣ (largest to smallest). Education (schoolingcategories), migration background, and Twitter use were non-significant (p > 0.10) and are reported in the Appendix A.3 together with threshold and random-effect estimates.

), institutional trust had the largest effect ($\beta =1.45$, p $< 0.001$), followed by gender ($\beta =1.35$, p = 0.001) and personal risk perception ($\beta =0.64$, p $< 0.001$). When trust was disaggregated into scientific and political components, trust in science consistently outperformed trust in political institutions, with scientific trust maintaining significance across weighted specifications ($\beta =1.27$, p $< 0.001$ in weighted models) while political trust showed more variable effects.

Additionally, we find that individuals with fewer conspiracy beliefs are more likely to get vaccinated against COVID-19 ($\beta =-0.44$, p < 0.05).

5.2.2. Sociodemographic Influences

Age and gender showed consistent effects across all specifications. Each additional year of age was associated with increased willingness ($\beta =0.06$, p $< 0.001$), and women showed substantially higher willingness than men ($\beta =1.35$, p $< 0.001$). Income had a modest positive effect ($\beta =0.16$, $p = 0.002$). Contrary to expectations, political orientation, education level, migration background, and Twitter usage showed weak or inconsistent associations with vaccination willingness across model specifications.

5.2.3. Temporal Dynamics and Phase Effects

A substantial main effect of study phase emerged across all models, with willingness increasing markedly from Phase 1 to Phase 2 ($\beta =3.15$, p $< 0.001$ in weighted ordinal models). This large effect indicates that vaccination willingness was highly responsive to changing contexts, including vaccine availability, emerging safety data, and evolving public discourse.

5.2.4. Interaction Effects and Temporal Stability

Because the Phase 3 unvaccinated sample was small ($N=40$), interaction models were estimated on Phases 1–2 only. In the weighted linear mixed-effects model, predictor effects changed between phases: trust in science became more predictive ($\beta =0.31$, $p < 0.001$), personal risk became less predictive ($\beta =-0.26$, $p < 0.001$), and trust in political institutions showed a negative Phase 2 interaction ($\beta =-0.22$, $p=0.030$). The Phase 2 interaction for conspiracy beliefs was small and marginal ($\beta =0.13$, $p=0.054$), and science attitudes did not interact with phase. A likelihood ratio test favored the interaction model over the main-effects model (${\chi }^2\left(5\right)=39.23$, $p=2.14\times {10}^{-7}$; $\Delta \mathrm{AIC}=-29.2$).

Figure 2 visualizes these effects using phase-specific simple slopes ($\beta$) from the same weighted mixed-effects model. Error bars are model-based standard errors computed with the Kenward–Roger small-sample correction for linear mixed models [29,30]. The full model is presented in Appendix A.4 

Table A5. Interaction effects: phase × predictor interactions.

Interaction Term	Linear Weighted	Ordinal Weighted
Phase 2 × Trust Science	0.31 ***	0.21
Phase 2 × Trust Political	$-0.22$ *	$-0.22$
Phase 2 × Science Attitudes	$-0.01$	$-0.05$
Phase 2 × Personal Risk	$-0.26$ ***	$-0.04$
Phase 2 × Conspiracy Beliefs	0.13 †	$-0.27$
Model comparison
${\chi }^2\left(5\right)$	39.23 ***	—
$\Delta$AIC	$-29.20$	—

Coefficients are standardized (predictors z-scored), rounded to two decimals. Interactions compare Phase 2 to Phase 1. *** p < 0.001, * p < 0.05, ^† p < 0.10.

.

The previous phase × predictor interactions ask whether each predictor’s slope changed over time (e.g., science trust increased; personal risk decreased). The phase-stratified “relative importance” ranks predictors at a given time point. It is descriptive and does not test change. Thus a predictor can weaken over time (negative interaction) yet still rank highly within a phase and vice versa. In Phase 1, personal risk and conspiracy beliefs were clearly the dominant predictors (largest $\left|\beta \right|$), with trust in politics smaller and trust in science/science attitudes negligible. In Phase 2, trust in science moved to the top rank, conspiracy beliefs remained strongly negative, while personal risk dropped in influence; trust in politics stayed modest. Full estimates are reported in Appendix A.5 

Table A6. Phase-stratified relative importance (weighted linear models; standardized coefficients).

Predictor	Phase 1			Phase 2
	$\mathit{\beta }$ (SE)	p	Rank	$\mathit{\beta }$ (SE)	p	Rank
Personal Risk	0.38 (0.05) ***	<0.001	1	0.18 (0.06) **	0.001	4
Conspiracy Beliefs	$-0.35$ (0.05) ***	<0.001	2	$-0.32$ (0.07) ***	<0.001	2
Trust in Politics	0.15 (0.07) *	0.029	3	0.16${\left(0.08\right)}^{ †}$	0.056	5
Trust in Science	0.11 (0.07)	0.144	4	0.34 (0.07) ***	<0.001	1
Science Attitudes	$-0.04$ (0.05)	0.378	5	$-0.23$ (0.05) ***	<0.001	3
Model spec: phase-specific weighted OLS with controls (age, gender, political orientation, and income).
Predictors are z-scored within phase; ranks are by absolute $\left|\beta \right|$ within phase.

*** $p<0.001$, ** $p<0.01$, * $p<0.05$, ^† $p<0.10$.

.

5.3. Model Performance and Diagnostics

Linear mixed-effects models explained substantial variance in vaccination willingness, with random effects ICC values indicating meaningful between-person differences that persisted over time. Diagnostic analyses revealed acceptable residual patterns with no evidence of major assumption violations, though some outliers were identified. Detailed diagnostic information is provided in Appendix B.

Post Hoc Analysis: Political Orientation and Conspiracy Beliefs

Given documented political polarization in COVID-19 vaccine attitudes, we examined temporal changes in political effects and the structure of conspiracy beliefs. Political orientation and conspiracy beliefs were significantly correlated across both phases (Phase 1: r = 0.31, p $< 0.001$; Phase 2: r = 0.24, p $< 0.001$), confirming that conspiracy beliefs follow political rather than random patterns.

A three-way interaction model (political orientation × conspiracy beliefs × phase) revealed significant temporal changes. The political orientation × phase interaction ($\beta =-0.10$, p = 0.033) demonstrated increasing political polarization, with conservative individuals showing relatively decreased willingness in Phase 2. The conspiracy beliefs × phase interaction ($\beta =0.11$, p = 0.032) indicated that conspiracy beliefs became stronger predictors over time. The three-way interaction was non-significant (p = 0.951), suggesting straightforward rather than complex relationships (

Table 3. Political orientation and conspiracy beliefs: temporal interaction effects.

Interaction Term	Coefficient ($\mathit{\beta }$)	p-Value
Pol. Orientation × Phase 2	−0.10	0.033
Consp. Beliefs × Phase 2	0.11	0.032
Pol. × Consp. × Phase 2	0.003	0.951

Note. “Pol.” = Political, “Consp.” = Conspiracy. Results from weighted linear mixed-effects model. Negative political coefficient indicates increased conservative hesitancy over time.

).

These findings provide direct evidence of increasing political polarization in vaccination attitudes and confirm that conspiracy beliefs are politically structured rather than randomly distributed.

5.4. Qualitative Analysis: Understanding Decision-Making Processes

To complement the quantitative analysis and understand the reasoning behind vaccination decisions, in-depth interviews were conducted with 40 participants across three phases (December 2020, April 2021, and September 2021). The qualitative sample provided an enhanced representation of hesitant and resistant attitudes (27.5% contra, 22.5% maybe) compared to the general population survey, enabling deeper exploration of vaccination decision-making processes.

5.4.1. Vaccination Reasoning Model

Thematic analysis revealed distinct motivational themes that participants used when reasoning about COVID-19 vaccination, consistent with the quantitative trust and risk perception findings (see Figure 3). They can broadly be grouped in three broad categories of reasoning.

Pro-vaccination reasoning centered on three themes: (1) high institutional trust in research bodies, monitoring systems, and vaccination programs generally; (2) collective focus, emphasizing pandemic control, herd immunity, protecting others, and social obligation; and (3) personal risk assessment, involving fear of infection, risk–benefit calculations favoring vaccination, and self-identification as vulnerable.

Anti-vaccination reasoning similarly organized around three themes: (1) safety concerns about rapid development, insufficient testing, side effects, long-term consequences, and mRNA technology; (2) low personal risk perception, including not considering oneself in a risk group and risk–benefit calculations opposing vaccination; and (3) general vaccine skepticism, favoring natural immunity, questioning effectiveness due to mutations, and expressing general distrust of vaccines.

Conditional reasoning characterized participants who wanted to “wait and see,” expressed mixed feelings, preferred specific vaccines, sought more information, or leaned toward vaccination while harboring doubts.

5.4.2. Joint Analysis: Linking Quantitative Trends to Qualitative Themes

Integration of quantitative and qualitative findings reveals how statistical trends manifest in individual reasoning processes (

Table 4. Joint display: quantitative trends and qualitative themes across study phases.

Phase	Quantitative Pattern	Dominant Qualitative Codes
Phase 1	High risk perception effects	Rapid development concerns (37.5%)
	Moderate trust effects	Trust in research institutions (27.5%)
Phase 2	Increasing trust effects	Fear of infection (39.5%)
	Declining risk effects	Favorable risk–benefit calculation (36.8%)
		Reduced safety concerns (18.4%)
Phase 3		Social/practical challenges
		Family conflicts over status
		Adaptation to access rules

Note. Percentages reflect proportion of interview participants expressing each theme.

). The declining influence of risk perception in quantitative models corresponds to qualitative shifts from abstract safety fears in Phase 1 (37.5% citing rapid development concerns) to more concrete, trust-based reasoning in Phase 2 (39.5% citing infection fear and 36.8% favorable risk–benefit assessments). Similarly, the increasing predictive power of scientific trust aligns with participants’ growing emphasis on institutional credibility as real-world vaccination experiences accumulated.

The longitudinal interview data revealed significant shifts in vaccination discourse across the three phases, paralleling these quantitative trends. Phase 1 was characterized by dominant concerns about rapid development and insufficient testing, with substantial ambivalence about hypothetical vaccines and trust in research institutions emerging as a key differentiator (27.5% expressing high confidence in scientific oversight). Phase 2 saw fear of personal infection becoming the most prominent theme, with safety concerns about rapid development declining substantially as real-world vaccination experience reduced abstract safety fears. Phase 3 was dominated by social and practical challenges, with participants describing family conflicts over vaccination status and adapting to constantly changing access rules, reflecting the shift toward trust-based rather than risk-based decision-making processes.

5.4.3. Dynamic and Dialogical Decision-Making

A crucial finding was that approximately 30% of participants expressed both pro- and anti-vaccination arguments during interviews, revealing that vaccination willingness often involved literal internal dialogues where individuals actively weighed competing considerations within themselves. Rather than simply holding ambivalent attitudes, participants engaged in ongoing internal negotiations, voicing different arguments and counterarguments as they worked through their decision-making process.

This dialogical quality provides a crucial mechanism for understanding the temporal changes observed in the quantitative models. Because vaccination attitudes operated as dynamic internal negotiations rather than fixed positions, participants could readily incorporate new information and experiences into their ongoing decision-making processes. As external contexts evolved—from hypothetical vaccines to real-world implementation and from abstract safety concerns to observable outcomes in their communities—different arguments gained or lost salience within individuals’ internal dialogues. This explains how trust in scientific institutions became increasingly influential over time while personal risk assessments became less central: participants’ internal conversations shifted to emphasize different considerations as new evidence and social experiences accumulated. The dialogical nature of decision-making thus enabled the systematic temporal shifts we observed, with individuals updating the relative weight they gave to different arguments as the vaccination landscape evolved around them.

In addition, participants frequently referenced the same information sources to support opposing conclusions. For example, Robert Koch Institute (RKI) statistics were cited both to justify vaccination necessity and to question official narratives. As one participant noted, “I just kept myself very well informed and share information with my work colleagues”, while another stated “Yes, trying to keep a cool head in all the reporting.” This pattern illuminates how information processing styles, rather than information exposure per se, influenced vaccination decisions.

5.5. Summary of Key Findings

The longitudinal analysis revealed that COVID-19 vaccination willingness in Germany was primarily determined by institutional trust and personal risk perception, embedded within a rapidly evolving temporal context. The dramatic increase in willingness between study phases underscores the dynamic nature of vaccination attitudes. Trust in scientific institutions proved more predictive than political trust, while demographic factors showed modest but consistent effects. The analysis challenges narratives emphasizing political polarization or social media influence, instead highlighting the central importance of institutional credibility and risk communication in vaccination decision-making. The detection of significant interaction effects reveals that the relative importance of different factors shifted as the vaccination campaign progressed, with scientific trust becoming increasingly central while personal risk perceptions became less decisive.

6. Discussion

This longitudinal mixed-methods study provides novel insights into how vaccination willingness evolves during prolonged health crises by examining temporal dynamics across Germany’s three-phase COVID-19 vaccination campaign. Our findings support both primary hypotheses and reveal important theoretical implications for understanding health behavior change during pandemics. In particular, we observe strong and distinctive effects of trust in scientific actors that intensify over time, while the role of personal risk wanes. This pattern aligns with multi-country evidence documenting dynamic and sometimes declining willingness and confidence during the pandemic [1,2,3,4,5,6,7,8]. Our study adds an explanation by showing how the relative importance of trust versus risk shifts across rollout phases.

Support for Hypothesis 1 was demonstrated through the consistent emergence of institutional trust and personal risk perception as the strongest predictors of vaccination willingness, with trust in scientific institutions significantly outperforming trust in political institutions across all model specifications. The distinction between scientific and political trust proved crucial, with scientific trust maintaining robust predictive power while political trust showed variable effects. This finding extends theoretical understanding by demonstrating that competence-based credibility (scientific trust) operates through different psychological mechanisms than position-based authority (political trust), with the former showing greater predictive validity for health behaviors over time. This pattern is consistent with European comparative work showing marked heterogeneity in hesitancy and its determinants across countries [10] and with German studies in which trust in medical experts and authorities is on average strongly linked to vaccination status and acceptance [12,13]. Meta-analytic syntheses conducted during the pandemic further indicate that acceptance levels and correlates vary substantially across settings, underscoring the central role of confidence as a psychological antecedent [6,7]. At the structural level, subnational analyses in Germany highlight that educational and political contexts explain large shares of spatial variation in uptake, with remaining differences plausibly related to “soft factors” such as trust in science and conspiracy beliefs [14].

Support for Hypothesis 2 emerged through the significant interaction effects revealing that trust in scientific institutions became more predictive over time while personal risk perception became less influential. This temporal shift from analytical, risk-focused decision-making to heuristic, trust-based processing aligns with dual-process theories of cognition and extends them to the domain of health behavior during extended crises. The findings suggest that as individuals experience sustained uncertainty and information overload, they increasingly rely on trust-based cognitive shortcuts rather than detailed risk–benefit calculations. Longitudinal panel evidence across seven European countries likewise documents frequent transitions among willing/hesitant/unwilling states (“comes in waves”), with trajectories shaped by trust and perceived risk [8]; early U.S. panels similarly recorded sizable declines in willingness from April to October 2020 before later recoveries, underscoring temporal sensitivity in determinants [3,4].

The qualitative analysis provided crucial mechanistic insights into these temporal dynamics. The vaccination reasoning model (pro-vaccination, anti-vaccination, and conditional reasoning) revealed how participants organized decision-making around institutional trust, collective considerations, and risk perceptions. Importantly, 30% of participants expressed dialogical reasoning with both pro- and anti-vaccination arguments, indicating that vaccination attitudes represent dynamic negotiations rather than fixed positions. This finding challenges static models of health behavior and highlights the importance of longitudinal designs for understanding attitude change processes. The shift we observe, from an early emphasis on abstract safety concerns to later reliance on institutional credibility and social/practical considerations, mirrors early European survey evidence showing uncertainty despite relatively high stated acceptance [1] and complements German studies that identify trust in experts versus alternative information sources, autonomy concerns, and political preferences as salient correlates [12,13].

6.1. Practical Implications for Public Health Communication

The temporal dynamics identified in this study have important implications for designing adaptive public health communication strategies. Our findings suggest that vaccination communication should be tailored to different phases of vaccination campaigns, with early messaging emphasizing personal risk and disease severity while later messaging reinforces institutional credibility and collective responsibility. This phase-sensitive approach is consistent with evidence that intentions and confidence evolved over time across countries [8,10] and with German and European work showing that both psychological antecedents (e.g., confidence and perceived benefits) and contextual factors (e.g., political alignment and regional information environments) underpin acceptance [6,7,12,13,14].

During the initial assessment phase (Phase 1), when risk perception strongly predicts willingness, communication strategies should focus on clear, transparent information about disease threats and vaccination benefits. The high prevalence of concerns about rapid development (37.5% of interview participants) suggests that addressing safety concerns through detailed explanation of approval processes and safety monitoring systems may be particularly effective early in campaigns. Such concerns were widespread in early European surveys [1] and are a recurrent theme in acceptance research, where perceived benefits and barriers (including safety) are central (see also [6,7]).

As campaigns progress to implementation phases (Phase 2), the increasing importance of scientific trust suggests that communication should emphasize institutional credibility, scientific oversight, and real-world safety data. Crucially, our findings underscore the importance of maintaining clear separation between scientific and political actors. Where scientific figures became politicized (exemplified by Dr. Fauci in the United States) vaccination campaigns suffered. Germans appeared to distinguish more clearly between scientific institutions (such as the Robert Koch Institute and Paul Ehrlich Institute) and political actors, with the former maintaining stronger predictive power for vaccination willingness. This separation protected scientific credibility from political polarization.

The shift from abstract safety fears to concrete trust-based reasoning indicates that highlighting institutional competence becomes more influential than detailed risk–benefit presentations. Communication strategies should therefore emphasize scientific credentials, peer review processes, and evidence-based decision-making while avoiding partisan framings that might undermine perceived independence. In terms of concrete instruments, experimental studies in Germany could already show that preserving freedom of choice between vaccines reduces reactance and raises stated uptake [31] and that incentives and access via familiar providers (e.g., local doctors) can spur vaccination, especially among the undecided [32]. Later in the campaign, willingness to receive annual boosters in the D-A-CH region was high but patterned by approval of mitigation measures and political engagement, suggesting that messaging should also speak to civic norms and policy acceptance in late phases [33].

During mass vaccination phases (Phase 3), when social dynamics become central, community-based approaches addressing practical barriers and family conflicts may be more effective than individual-focused messaging. In this regard, the subnational patterning of uptake in Germany suggests targeting by regional context (education, political preferences, and economic conditions), while addressing residual trust-related gaps [14].

6.2. Methodological Contributions

This study demonstrates the value of mixed-methods longitudinal designs for understanding complex health behaviors. The joint analysis linking quantitative trends to qualitative themes (Table 4) provides a model for integrating statistical patterns with individual reasoning processes. This approach reveals not only which factors predict behavior but how and why their influence changes over time. Our results complement longitudinal European panel evidence showing intention “waves” [8] and answer calls for designs that move beyond cross-sectional snapshots to explain within-person change.

The temporal focus addresses a critical gap in vaccination research, which has predominantly relied on cross-sectional designs. Our findings demonstrate that predictor effects are not static but evolve systematically as health crises progress. This has important implications for intervention timing and design, suggesting that strategies effective early in pandemics may be less effective later and vice versa. The broader literature’s mixed patterns in acceptance across countries and time reinforce the need for phase-tailored interventions [6,7,10].

6.3. Limitations and External Validity

Several limitations should be considered when interpreting these findings. First, the observational design precludes causal inferences about the relationships between trust, risk perception, and vaccination willingness. While our temporal design strengthens inferences about the direction of effects, we cannot definitively establish that changes in trust cause changes in willingness versus alternative explanations such as common cause variables or reverse causation. Where causal claims are needed, experimental and quasi-experimental studies point to promising levers (choice, incentives, and proximal access) [31,32].

Second, Germany’s unique institutional landscape may limit generalizability to other contexts. The high baseline trust in German scientific institutions and strong cultural emphasis on evidence-based decision-making may amplify the effects observed in this study. Whether similar temporal dynamics would emerge in lower-trust or more polarized settings remains an empirical question requiring cross-national research. So far, cross-country evidence documents substantial heterogeneity in both baseline hesitancy and its drivers [10], and meta-analyses show wide between-study variance in acceptance [6,7].

Third, the interview sample, while purposively diverse, was limited to German speakers and may have excluded important perspectives from immigrant communities or digitally excluded populations. The telephone and Zoom interview formats may have systematically excluded individuals without reliable Internet or phone access, potentially limiting representation of socioeconomically disadvantaged groups.

Fourth, panel attrition over the three phases may have introduced selection biases, with the most engaged or vaccination-willing participants more likely to continue participation. The dramatic reduction in unvaccinated respondents by Phase 3 (N = 40) meant that late-stage analyses necessarily focused on a highly selected sample of persistent vaccine-hesitant individuals.

6.4. Future Directions and Policy Recommendations

These findings suggest several important directions for future research and policy development. First, cross-national replication is needed to establish whether the temporal dynamics observed in Germany generalize to other institutional and cultural contexts. Comparative studies examining countries with different institutional trust levels and vaccination rollout strategies could illuminate the boundary conditions for these effects.

Second, intervention studies testing adaptive communication strategies based on these temporal dynamics could provide causal evidence for the practical implications suggested by our findings. Randomized trials comparing risk-focused versus trust-focused messaging at different campaign stages could directly test the hypothesis that optimal communication strategies change as vaccination campaigns progress.

Third, longer-term follow-up is needed to understand how the trust-based decision-making patterns established during COVID-19 vaccination affect subsequent health behaviors. Whether the increased reliance on scientific trust generalizes to other health decisions or remains specific to pandemic contexts has important implications for public health preparedness.

From a policy perspective, our findings suggest that public health agencies should develop adaptive communication frameworks that adjust messaging strategies based on campaign phase and real-time monitoring of public attitudes. Investment in institutional credibility and transparent communication processes may pay dividends not only for current campaigns but for future health emergencies by establishing trust relationships that facilitate rapid public response. Practically, this implies (i) early-phase safety transparency addressing development speed concerns [1]; (ii) mid-phase emphasis on the competence and independence of scientific institutions [12,13]; and (iii) late-phase support for pragmatic access and autonomy-preserving choice, potentially complemented by non-coercive incentives [31,32,33].

Finally, the dialogical nature of vaccination decision-making observed in our qualitative analysis suggests that interventions acknowledging and working with attitude ambivalence may be more effective than approaches that assume fixed positions. Communication strategies that validate uncertainty while providing clear guidance for navigating complex decisions may be particularly effective for reaching ambivalent populations.