The TPB (15) posits that intentions are the most proximate and robust predictor of actual behavior. They represent an individual’s motivation and preparedness to engage in a particular action. The TPB also proposes that intentions are formed by three determinants: attitude, subjective norms, and perceived behavioral control (PBC). Attitude is the value that an individual ascribes to a given behavior. An individual’s intention to perform a given behavior is strengthened by a more favorable attitude toward that behavior. This is because individuals tend to intend to engage in behaviors perceived as favorable. The subjective norms are the normative expectations of the social group to which the individual belongs. The greater social pressure to engage in a behavior is associated with enhanced intentions. When individuals believe that significant others approve of a given behavior, they are more likely to intend to perform it. The PBC is the individual’s belief in their capacity to control the performance of the behavior in question. PBC is comprised of two key elements: self-efficacy and controllability. Self-efficacy can be defined as an individual’s perception of their capacity to achieve a given behavior successfully. Controllability pertains to an individual’s judgment regarding the availability of resources and opportunities to achieve a given behavior. Higher perceived behavioral control is associated with increased intention. When individuals believe they have the ability and resources to perform a behavior, they are more likely to form intentions. In turn, the TPB posits that an individual’s beliefs about the probable consequences of the behavior, particularly expected benefits and costs, shape the attitude. Finally, the TPB suggests that the relative importance of these relationships can vary depending on the specific behavior and context.

Under the TPB umbrella, this study conceptualizes vaccination attitude as the value of support or hesitancy related to vaccination among the general public. Subjective norms are the perceived social pressure from significant others to get vaccinated. Some empirical studies have supported the relationships between attitude and subjective norms on intention in the context of vaccination in particular (13, 26) and preventive behaviors in general (handwashing and limitation of social contacts) (27). Following the TPB rationale, it is expected that:

As previously stated, TPB also considers that beliefs influence attitude formation. In this study, perceived benefits are defined as the belief that vaccination will reduce the threat of disease. In contrast, perceived costs are defined as the belief that vaccination will entail monetary costs or require exertion of effort (19). These constructs are salient aspects in the health prevention literature (28). Even more empirical studies have supported that the perception of benefits and costs influences attitudes toward vaccination (19, 29). Following the tenets of TPB, it is anticipated that:

It should be noted that this study does not consider PBC, as some empirical studies have found a marginal additional impact on vaccination intention. Possibly unlike other behaviors that require ongoing effort or complex skills, vaccination is typically a simple, one-time action that is widely accessible in many contexts. Most individuals may perceive minimal difficulty in obtaining vaccination, resulting in low variability in PBC across the population (13).

While the TPB offers a robust foundation for understanding vaccination behavior, integrating specific constructs from health literature could facilitate a more comprehensive and nuanced understanding of the vaccination phenomenon. The model incorporates risk perceptions and response efficacy, as outlined in Protection Motivation Theory (PMT) (17, 30).

PMT posits that individuals will be self-protective when they perceive a health threat. The Protection Motivation Theory (PMT) posits that two appraisal processes—threat appraisal and coping appraisal—drive the decision to engage in self-protective behaviors.

Threat appraisal depends on the knowledge about the threat that the individual has. Therefore, it is formed by three dimensions of risks. First, perceived likelihood refers to the probability of being harmed by a hazard if no preventive action is taken, as illustrated by the question, “What is the likelihood that you will get the flu this year if you do not get a flu shot?.” Second, perceived susceptibility focuses on individual constitutional vulnerability to an illness, exemplified by questions like “Are you more likely to get the flu than others?” Finally, perceived severity refers to the seriousness of contracting an illness, considering medical and social consequences (e.g., death or impairment in daily activities), such as asking, “If I do not get vaccinated, would my health be seriously endangered? “(31, 32).

The coping appraisal refers to the ability of the individual to cope and avert health threats successfully through self-efficacy, response efficacy, and response cost. This study excludes self-efficacy and costs because the former is part of PCB, and the latter is similar to the TPB construct costs. Perceived response efficacy refers to the belief about how effective the result of preventive behavior, such as getting the vaccine, is in reducing harm.

In general, the PMT suggests that individuals, when faced with a threat to their health, engage in preventive behaviors if they believe that this will reduce their risk or that not performing these behaviors may cause them harm (33).

Based on conceptual and empirical results (19, 34), this study posits that PMT constructs could be conceptualized as beliefs that contribute to shaping the attitude toward vaccination and, therefore, have an indirect effect on intention. Both PMT constructs and TPB beliefs are fundamentally cognitive in nature. They represent mental schemas that individuals use to understand and anticipate the consequences of their actions. This shared cognitive basis reinforces the rationale for incorporating PMT constructs as beliefs within the TPB framework. Also, within TPB, beliefs contribute to attitudes by providing the evaluative information necessary to form an overall judgment about a behavior. Similarly, the PMT constructs provide evaluative information about health threats and protective behaviors, which can directly shape attitudes (35).

Some studies have shown that beliefs about risks may affect attitude, especially in preventive behaviors (26, 36, 37).

Following the TPB tenets and the previous discussion and findings, it is expected that:

PMT, in general, and specifically the work of Witte (18), proposes that message components are the critical elements of a fear appeal communication designed to influence cognitive appraisals. These components can be categorized into two main groups: threat components and efficacy components. The first one comprises three types of messages: Communication on the probability of occurrence: Messages that convey information about the likelihood of the threat occurring if no preventive action is taken. Communication on noxiousness: Messages that represent the severity or harm of the threat’s consequences. Communication on the susceptibility of vulnerable groups: Messages that convey information about the vulnerability of specific groups or individuals to the threat. The efficacy component comprises Communication on the response efficacy, which consists of messages that convey information about the effectiveness of the recommended preventive action in averting or reducing the threat.

The components of the fear appeal message function as deliberate stimuli designed to trigger specific cognitive processes. These components demonstrate a direct correspondence with cognitive appraisals: threat elements (probability of occurrence and noxiousness) align with threat appraisal dimensions (likelihood and severity), while efficacy information maps to coping appraisal (response efficacy). This relationship is characterized by proportionality, where the intensity of message components directly influences the strength of corresponding appraisals. Thus, more vivid and powerful message elements evoke stronger cognitive responses, establishing a predictable link between external communication and internal cognitive processes. This structured alignment and proportional influence form the foundation for understanding how fear appeals shape recipients’ perceptions and subsequent behaviors (18, 38, 39).

Therefore,

However, the relationship between “Communication on the susceptibility of vulnerable groups” and “Perceived susceptibility” is complex and contingent upon individual conditions, particularly one’s membership in a vulnerable group. The comparative measure of perceived susceptibility would influence this relationship. The message of susceptibility content compares the risk for vulnerable groups to the general population. At the same time, individuals are asked to assess their perceived susceptibility relative to the general population (or age group). For vulnerable group members, the impact of such communication on perceived susceptibility is likely direct and positive. The emphasis on higher risk for their group aligns with their circumstances, potentially elevating their perception of susceptibility. Conversely, for non-vulnerable individuals, the effect may be negative or neutral. By highlighting the increased risk for vulnerable groups, the message implicitly suggests lower risk for others, potentially decreasing perceived susceptibility among those who do not identify as vulnerable (18, 40).

Therefore,

The moderating role of information seeking in the relationship between threat/response messages and cognitive appraisals could be elicited by examining how individuals process persuasive communications under different levels of information seeking.

The Elaboration Likelihood Model (ELM) (41, 42) is a comprehensive framework for understanding persuasion processes. The ELM has become one of the most influential social psychology and communication theories, providing insights into how people process persuasive messages and form or change their attitudes (43).

This study highlights four core tenets of ELM. First, at the ELM’s core is an elaboration continuum, which represents the degree of thought or cognitive effort an individual dedicates to processing a persuasive message. This continuum ranges from low to high elaboration, with different processes of persuasion operating at various points along this spectrum. Second, the ELM posits two primary persuasion routes: the central and peripheral. The central route, associated with high elaboration, involves scrutiny and evaluation of message arguments. In contrast, the peripheral route, linked to low elaboration, relies on simple cues or heuristics rather than effortful processing of message content. Third, where individuals fall on this elaboration continuum is largely determined by their motivation and ability to process the message. Factors such as personal relevance, need for cognition, and distractions can influence one’s motivation and ability to elaborate on a message. Fourth, the ELM proposes that the route through which an attitude is formed or changed has significant implications for the strength of that attitude. Attitudes formed through the central route (high elaboration) tend to be stronger, more enduring, more resistant to counter-persuasion, and more predictive of behavior than those formed through the peripheral route (low elaboration). This distinction in attitude strength highlights the importance of understanding whether attitude change occurs and the process through which it happens.

According to ELM, in scenarios of high information seeking, which represents a motivated state for systematic information processing (44), individuals are interested in processing messages deeply, engaging in the central route of persuasion. They pay close attention to the message components that directly correspond to cognitive appraisals. This deep processing leads to stronger cognitive appraisals because the individuals thoughtfully consider the merits of the information presented. Consequently, the impact of messages on cognitive appraisals is amplified.

Conversely, in scenarios of low information seeking, individuals are less motivated to process information deeply and are more likely to engage in peripheral route processing. They focus on superficial cues rather than the substantive content of the threat messages. As a result, the direct influence of message components on cognitive appraisals is attenuated because the individuals do not elaborate on the critical elements that align with threat appraisal dimensions (41, 45).

Therefore, information-seeking moderates the relationship between communications and cognitive appraisals by determining the route of message processing. Through central processing, high information seekers experience a stronger linkage between messages and cognitive appraisals due to deliberate and thoughtful evaluation of the message content. Low information seekers, processing messages peripherally, exhibit a weaker connection as they rely on external cues rather than engaging with the core message components. Thus,

Also, tenets of ELM can support the moderating role of information processing ease in the relationship between messages and their corresponding cognitive appraisals.

In high information processing ease scenarios, individuals find public health messages easy to understand, requiring minimal cognitive effort to comprehend the basic message content. This initial message comprehensibility enhances individuals’ ability to process information—a key determinant of ELM. In this situation, individuals are more able to process messages deeply, engaging in the central route of persuasion (42). In this path, individuals can scrutinize the message content thoroughly. They pay close attention to the substantive elements of the messages, which directly correspond to cognitive appraisals. Individuals can elaborate on the information because the messages are clear and easily digestible, leading to stronger cognitive appraisals. Consequently, the positive effects of communications on appraisals are amplified.

Conversely, in low information processing ease scenarios, individuals find the public health messages difficult to comprehend due to complexity, technical jargon, or poor presentation. This hinders their ability to process information centrally, as the increased cognitive effort reduces their capacity to engage deeply with the message content. As a result, individuals are more likely to rely on peripheral cues—such as aesthetic elements—rather than the substantive message components. The direct influence of threat and efficacy messages on cognitive appraisals is diminished because individuals cannot effectively map the message elements to their corresponding cognitive appraisals (41, 45).

Therefore, information processing ease moderates the relationship between messages and cognitive appraisals by influencing the route of message processing. High information processing ease facilitates central processing, enhancing the alignment between message components and cognitive appraisals. Low information processing ease leads to reliance on peripheral processing, weakening the direct effect of message content on cognitive appraisals. Thus,

Research suggests two distinct mechanisms influence the formation of risk perceptions from COVID-19 messages: understanding the message content (learning and comprehension of facts about threat or response efficacy) and the acceptance of this content (46, 47). The first mechanism—understanding the message—operates independently of trust in government, as empirical evidence demonstrates no significant differences in factual information acquisition between high and low-credibility sources. In other words, individuals comprehend the content of risk messages similarly, regardless of their trust level in the government as an information source.

In the high-trust scenario, when individuals trust the government (analogous to high source credibility), the acceptance mechanism operates without interference. Evidence shows the net change in agreement with advocated positions for high-credibility sources, indicating that trust enables a direct pathway between message comprehension and acceptance (46, 48). When people trust their government, understanding COVID-19 messages flows smoothly into the formation of perceptions, unimpeded by doubts about the source’s credibility. This clean cognitive pathway allows the message’s content about threat severity and likelihood to shape perceptions directly.

In the low-trust scenario, when individuals have low trust in the government (analogous to low source credibility), the acceptance mechanism is disrupted by interference. While these individuals still comprehend the messages, low trust creates a cognitive barrier between understanding and acceptance. Evidence shows no change in agreement with advocated positions for low-credibility sources (46, 48). This interference means that even though people understand the message content, their low trust in the government prevents this understanding from fully translating into perception formation. This phenomenon manifests as understanding without believing, highlighting how low trust disrupts the acceptance process while leaving comprehension intact. Thus, the relationship between messages and perception formation becomes weaker not due to failed understanding but because low trust interferes with the acceptance of the understood content. Therefore,

Considering the above, Figure 1 summarizes the research model showing the contributions of each of the source theories.

The instrument was developed by using diverse tools from different authors. Sources of information were evaluated through Probability of occurrence, Response efficacy depictions, Magnitude of noxiousness and Susceptibility, which were evaluated by the tools reported by Demirtaş-Madran (38). Benefits and barriers were evaluated through Perceived benefits and Perceived barriers, for which the present study used the tool designed by Cheng and Ng (49). Risk beliefs was analyzed through different constructs. First, Perceived severity, which was evaluated through the tool developed by Yueng et al. and Zhang, et al. (50, 51). The Perceived likelihood was analyzed by using the instrument designed by Weinstein et al. (52). Response efficacy was evaluated by the tool reported by Witte, Cameron (53). Perceived susceptibility was evaluated by the tool reported by Trifiletti, Shamloo (32). About the TPB constructs, Attitude was evaluated through the tool designed by Park and Oh (54), Subjective norms were evaluated by through the instrument reported by Li, Liu (55) and Intention was evaluated by the tool developed by Park and Trifiletti (32, 54).

For improving content validity, the scales were adapted to the COVID-19 context. Annex presents the scales’ constituent measurement items, response formats, and sources. The researchers released a pilot test before obtaining the full-scale sample, collecting 50 preliminary observations. The analysis of internal consistency, reliability, and validity shows satisfactory indicators. Hence, minor changes were necessary for the complete sample collection.

Following literature recommendations (56), several procedures were implemented to mitigate the potential effects of common method bias (CMB). These procedures included separating the measures of the predictor from the independent variables to avoid mental connections. Additionally, various scale types (Likert, semantic differential, rating) were employed alongside diverse anchor labels (e.g., agreement, frequency, likely options) to prevent respondents from falling into a response pattern. Furthermore, the questions are presented in a manner that intersperses items from different variables throughout the questionnaire to reduce the effects of proximity.

Data was collected using Amazon’s Mechanical Turk (MTurk) platform, targeting the adult American population. MTurk is particularly suitable for studies requiring large samples without specialized expertise (57). Recent research has shown that implementing certain practices can maintain the reliability of MTurk-based surveys (58, 59). Specifically, the review of Mortensen and Hughes (60) of health research found MTurk to be as efficient and reliable as traditional methods for data collection.

The researchers followed recent recommendations (57, 58, 61) to ensure data quality by implementing several measures. MTurk settings were adjusted to include only US residents with a high task completion rate (at least 98%) and a minimum of 500 completed tasks. SurveyMonkey settings prevented multiple responses from the same IP address. The questionnaire introduction emphasized the importance of attentive responses and the scientific nature of the study. Participants were informed about anonymous analysis to reduce social desirability bias and that the responses would be scrutinized, warning of rejecting invalid responses. Three attention check questions and a captcha verification were incorporated to deter bots. Compensation was set following US minimum wage laws for standard respondents.

The researchers created the survey using SurveyMonkey and made it accessible to MTurk workers. Data collection occurred between November 5th and December 10th 2023. After completing the survey, the researchers extracted the data from SurveyMonkey and compensated participants through MTurk.

Following the ethical principles put forth in the Belmont Report, the participants gave their written informed consent before answering the questionnaire. In the informed consent, they were informed about the project, the absence of risks in answering and the privacy with which their data would be handled. The project was approved by the institutional review board of the Pontificia Universidad Católica del Perú (PUCP) (010-2023-CEI-CCSSHHyAA/PUCP).

PLS is an appropriate method for examining complex relationships, handling non-normal data, and it is less demanding regarding sample size, particularly when considering multigroup analysis (61). The software used was SmartPLS 4.1.0.6.

The sample size was determined using the inverse square root method proposed by Kock and Hadaya (62). This method is particularly appropriate for the analysis as it accounts for the unique characteristics of PLS-SEM modeling. The calculated sample size requirement is 275 participants, using this approach with a minimum path coefficient of 0.15, a significance level of 0.05, and a statistical power of 80%.

Moderating effects were evaluated following the Multigroup Analysis (MGA) technique (63). The MGA method is developed specifically for the PLS-SEM context to test moderating effects by examining whether model relationships differ significantly across predefined groups. Conceptually, this aligns with moderation theory, where a third variable affects the relationship between two variables. Rather than using interaction terms, MGA directly tests whether path coefficients vary between groups formed based on the moderator variable values. The statistical significance of path differences between groups provides evidence of the moderating effect (64).

In the case of H12, the groups were divided according to the dichotomous answers of the participants regarding their health condition. For the rest of the hypotheses, following previous methodological studies, MGA focuses on extreme groups to detect differentiated effects (65). Therefore, the sample was divided into three segments based on the moderator variable: Upper, median, and lower groups, where the intermediate group corresponds to values in the median of the variable (66). The multigroup analysis was performed with the upper and lower groups.

The researchers’ choice to use extreme groups rather than a simple median split is grounded in established methodological literature. DeCoster, Iselin (65) explain that focusing “on the extreme ends of the distribution, researchers increase the differences found within their samples, which enhances the observed effects.” This approach enhances statistical power by increasing their ability to detect meaningful differences in path coefficients. It creates clearer group differentiation by excluding cases in the “gray area” near the median. This methodological strategy maximizes their ability to detect moderating effects in the PLS-SEM context.

Table 1 provides a comprehensive sample description encompassing essential demographic, socioeconomic, and health-related characteristics. These descriptors collectively represent the participants’ backgrounds and experiences pertinent to the study.

The reliability, convergent validity, and discriminant validity were analyzed considering the cut points recommended by the literature (61). Table 2 shows that Cronbach’s alpha values were greater than 0.7 or close to that threshold, and composite reliability indicators were superior to 0.7, showing that scales had internal consistency. Also, AVE values were greater than 0.5, and the ranges of the factor loadings were over 0.7, which showed convergent validity. Table 3 shows that HTMT values were lower than 0.85 or close to that cutoff, indicating adequate discriminant validity.

Harman’s single-factor test assessed potential common method bias (CMB). Table 3 shows the analysis, which revealed that the single unrotated factor extracted accounted for 39.5% of the total variance, below the commonly accepted threshold of 50%. This result suggests that CMB is unlikely to be a significant concern in this study.

Considering the satisfactory measurement model results, the following section examines direct effects in the entire sample to evaluate hypotheses H1 through H11. Subsequently, a multigroup analysis assesses the moderation effects proposed in hypotheses H12–H15.

Before assessing the structural relationship, Variance Inflation Factors (VIFs) were calculated to evaluate potential multicollinearity issues. The analysis showed that all VIF values were below the cutoff of 5. The highest observed VIF was 2.38. These results indicate that multicollinearity was not a significant concern in the analysis.

Table 4 shows empirical support for six of the 11 hypotheses tested. Although the hypotheses linking the communication components and their respective appraisals are rejected at the level of the complete sample, as will be shown subsequently, these hypotheses are supported when analyzed in the different sample subgroups proposed in this study.

Previous to multigroup analysis, measurement invariance across each pair of groups should be established through the Measurement Invariance of Composite Models (MICOM) procedure (67). Compositional invariance is established if the original correlation is greater than or equal to the 5% quantile of the empirical distribution of the permutation correlations for each construct. In the analysis, most of the quantiles are greater than 90% across all group comparisons and constructs, and the minimum quantile observed was 10%, above the generally accepted threshold. Therefore, measurement invariance is established.

According to the analysis procedure, an MGA was conducted for the four pairs of groups. Table 5 shows empirical support for hypotheses H12, H13, and H15. The results support partially H14. As mentioned, the hypotheses linking the communication components and their respective appraisals are supported in the upper groups of information-seeking and trust in government.