We used the health belief model (HBM) adapted for use in COVID-19 research by Hsing et al. (35) and further expanded it by taking into consideration potential modifying aspects such as demographics, and time-dependent aspects such as corona fatigue. Figure 1 shows the key components of the HBM model used in this project.

The presented model (Figure 1) considers behavioral aspects and attitudes. These aspects also take the respondent's knowledge into consideration. The KAP-survey concept (knowledge, attitude, practice) was therefore used in the development of the questionnaire (2). To create an item pool for the COVI-Ad questionnaire, a literature review was carried out. Questionnaires that were based either on the health belief model or single aspects of it, and that had already been used during the COVID-19 pandemic and other pandemics or epidemics, were screened. New items were formulated for aspects that were not covered in these questionnaires. To make it easier to respond to the items during telephone interviews, the number of response categories was kept to a minimum. The resulting questionnaire was discussed within an expert group meeting (psychologist, medical doctors). After minor changes, eight telephone interviews were carried out in advance to assess the comprehensibility and feasibility of the questionnaire. The final questionnaire consisted of 68 items with a closed- and two items with an open-response format. A translated version of the German questionnaire can be found in the Supplementary material. Since the aim of the questionnaire is to map the relevant aspects of the adapted health belief model, the items were analyzed separately for each aspect. Explorative factor analysis (VARIMAX rotation) was carried out separately for all aspects apart from sociodemographic variables and single-item aspects. Internal consistency (Cronbach's alpha) was calculated for each resulting factor.

The survey was conducted by two professional call centers in two Austrian states from April 20 to June 9. Overall, 500 volunteers that were representative of the population of Carinthia in terms of age, gender, and educational status and were ≥16 years, and 503 from the population of Vorarlberg, participated in the interview study. To achieve this sample size 3,690 persons in Carinthia (response rate 13.6%) and 3,526 in Vorarlberg (response rate 14.3%) were contacted. Participation was voluntary and participants received no incentives.

Baseline characteristics are presented as mean ± SD or median (IQR), as appropriate. Categorical variables are provided as absolute and relative numbers. In a first step univariate linear regression analysis was performed, whereby adherence served as the outcome. Predictors were the factors and the single-item aspects described above, along with sociodemographic variables (age, gender, employment status, living with children, living status, education). Dummy coding was used for categorical variables with more than two categories. To enhance comparability, all factors and single-item aspects apart from age were transformed to range from 0 to 1. To ensure the resulting betas were comparable, the age variable was therefore divided by 100. Univariate significant predictors were checked for multicollinearity (variance inflation factor < 2.5). Remaining variables were included in a multivariate regression analysis (backwards selection). Exploratory data analysis was used to assess the influence of the predictors on the single measures by using logistic regression analysis. For this analysis, univariate significant predictors were also checked for multicollinearity (variance inflation factor < 2.5). The remaining variables were subjected to multivariate logistic regression analysis (backwards selection). SPSS 26 was used for data analysis (36), a value of p < 0.05 was considered significant.

The median age of participants was 50 (38–64) years and 52% of respondents were female. Female respondents were older (female: median 54 years IQR: 41–66; male: 44, 35–62). About 1/3 had a university or high school diploma (EL4 and EL5), while 40% were employed and 30% had retired (Table 2).

Overall, respondents' median adherence was 0.75 (IQR: 0.5–1.0). Social norms (median: 1, IQR: 0.67–1.00) and trust in institutions (median: 0.64, IQR: 0.5–0.83) were also rated highly. Respondents rated a COVID-19 infection as more dangerous than an influenza infection (median: 1.0, 0.5–1.0). We also measured self-efficacy (median: 0.5, IQR: 0.5–1.0), personal health risk (median: 0.50, IQR: 0.25–0.75), perceived barriers due to health-promoting measure (median: 0.50, IQR: 0.25–0.75), perceived incentives to engaging in health-promoting measures (median: 0.50, IQR: 0.25–0.75), economic risk stemming from the measures to combat the coronavirus (median: 0.50 IQR: 0.2–0.7), and perceived susceptibility (median: 0.5, IQR: 0.0–1.0). Low ratings were observed for information fatigue (median: 0.4, IQR: 0.2–0.7) and behavioral fatigue (median: 0.25, IQR: 0.00–0.38).

In a first step, the following variables were significant univariate predictors of adherence to health-promoting behaviors: age, gender, employment status (retirement, employed, short-time work, student), university degree (EL 5), living with children, two perceived severity items (comparison to influenza, personal health risk), self-efficacy, perceived barriers due to health-promoting measure, trust in institutions, social norms, information fatigue, and behavior (Figure 3, Supplementary Table 1).

In a second step, multivariate regression analysis indicated that six independent predictors explained 29% of the variance in adherence [Radjusted2 = 0.285, F₍₁₎ = 59.85, p < 0.001]. Higher age (β: 0.43 95%CI: 0.33–0.54; p < 0.001), social norms (β: 0.33 95%CI: 0.27–0.40; p < 0.001), perceived personal health risk (β: 0.12 95%CI: 0.05–0.18; p < 0.001), self-efficacy (β: 0.06 95%CI: 0.02–0.10; p = 0.002), female gender (β: 0.05 95%CI: 0.01–0.08; p = 0.002), and decreased behavioral fatigue (β: −0.11 95%CI: −0.18 to −0.03; p = 0.045) were associated with increased health-promoting behavior (Figure 4).

Multivariate regression analysis of individual measures indicated that two to seven independent predictors explained 9–27% of variance. Twelve different predictors were included in the final six models. No predictor was included in all final models. The predictors that were most often included were behavioral fatigue (four times) and age (three times) (Table 3, Supplementary Figure 2; univariate results: Supplementary Figure 1).

The survey revealed that higher age and female gender were independent predictors of adherent behavior. Even though previously published studies were inconclusive, a large percentage of studies support our results (23, 24, 37–39). For example, one Canadian study of over 2,000 persons between 18 and 100 years old showed that age and male gender were associated with lower adherence to different COVID-19 protective measures such as working remotely from home, social distancing, and maintaining a physical distance of 2 m from others (23). Another study that used cluster analysis to compare adopters and non-adopters of COVID-19 measures in 5,893 persons between 18 and 94 years old confirmed that older and female persons had lower odds of being in the non-adapter cluster (37). No influence of age or gender was found in a survey of elderly persons (aged over 60 years), which may reflect homogeneity across these variables within the study group (22). Wolfe's paper, which focused on age differences in COVID-19 risk-taking, also revealed that risk perception for the self and others partially mediated the effect of age differences on taking risks (38). One reason why the younger population seems to be less adherent to protection measures may be that they are less vulnerable to the consequences of an infection with SARS-Covid-2. It is well-known that the likelihood of complications, hospitalization, and death is dependent on age, and this has been extensively communicated in the media and by public institutions. Another reason may be that people of younger age are still actively involved in the workforce and may frequently feel that the risk of financial loss offsets concerns about becoming infected.

Results on the effect of gender differences are contradictory. One study of 21,649 persons from eight OECD countries (Australia, Austria, France, Germany, Italy, New Zealand, the UK, and the US) underpins our findings that women have been more adherent to pandemic rules in all countries and take the pandemic more seriously (40). In addition, the paper by Abd Elhameed Ali et al., which also presented results from over 700 people, shows female gender to be positively related to better knowledge about COVID-19 measures and greater adherence to containment measures (26). In contrast, an online survey of 893 Brazilians by Carvalho and Machado that was primarily concerned with the correlation of adherence to pandemic rules and psychopathy traits showed no gender differences (25). In summary, it can be seen that the influence of gender and age found in our study is found in many but not all studies.

In our study we could also show that social norms are strongly associated with increased health-promoting behavior. Even though the results found in the literature appear to be inconsistent (20, 28), this paper supports recent findings indicating that social norms have a significant impact on adherence to COVID-19 measures in the general population (31, 41). In the Corona pandemic, social norms have played an important role in reducing individual transmission risk, as well as the transmission rate in the population as a whole. As studies suggest that social norms and social identities influence behavioral changes, it is important to mention the potential impact of influencing social norms and attitudes to specific COVID-19 measures, especially in vulnerable groups. According to Neville et al. (42), public health messages aimed at changing behaviors should focus on specific groups and present the desired behavior, without including any reference to unwanted behaviors. These messages should be presented by people that are perceived as “one of us.” The intended behavioral change should be framed in an identity-affirming manner, and group members should be seen to change their behavior without losing their influence and without polarization within the group. As social norms are formed by all members of a group, each individual has an influence. According to a review by Tankard and Paluck (43), understanding norms requires information on individual behavior, the group as a whole, and institutional signals. Each of these may be influenced by COVID-19 measures and information strategies that focus on providing consistent information that takes into account group identities and aim to enhance people's self-efficacy. Even if an influence of social norms on behavior cannot be found in all studies, as in our study, social norms may play an important role in adherence.

Another factor we found to have an impact on adherence is perceived personal health risk. This means the more dangerous the virus is considered to be, the more willing a person is to take protective action. These findings confirm further cross-sectional studies such as Lang et al. (37) showing that people who were unconcerned that an infection with the virus might lead to severe symptoms had higher odds of being non-adopters of non-pharmaceutical preventive interventions. Furthermore, in a sample of over 6,600 persons in the US, Bruine de Bruin and Bennett also found people that considered high risk to be associated with an infection with the coronavirus to be more likely to adopt protective behaviors (44). It is interesting, however, that the perceived threat of a SARS-COVID-19 infection having serious repercussions seems to have declined over the course of the pandemic. Results from a longitudinal survey from three rounds of interviews in Spain conducted between July and November 2020 revealed that the perceived threat of becoming seriously ill if infected with COVID-19 infected decreased over time, although the probability of becoming infected remained stable (7). In addition, another study involving 30,000 interviews conducted in 39 rounds in Hong Kong also analyzed temporal changes in the perceived severity of the disease and found it to be positively associated with the incidence of infected people (33).

Besides the perceived health risk, there are also known differences in the health risk due to COVID-19 between groups. In the case of diabetics, for example, it is possible to determine the individual risk with the help of models (45). Furthermore, modern technologies can be useful in early diagnosis and accurate classification of COVID-19 patients (46) and combat COVID-19 (47, 48).

In our study, lower behavioral fatigue was associated with greater health-promoting behavior. These results have been confirmed in further studies showing that behavioral fatigue associated with the Corona pandemic impacted people's adherence to measures to reduce transmission risk in the population (5, 32, 33).

Martinez-Garcia et al. analyzed data from a survey of 20,054 persons that was conducted in Spain from April to September 2020 and showed that adherence to containment measures declined over time (49). While they found that the psychological impact was the most important predictor of adherence to containment measures in the beginning, the economic impact played a greater role at the end of the period under review. The authors recommended the use of psychological and economic support programs to enhance adherence in the population. Reicher and Drury also concluded that lower adherence may be related to the availability of financial resources in the population and not only to psychology. Measures to counteract behavioral fatigue should therefore consider the specific needs of communities (50).

Liao et al. showed that psychological fatigue is also associated with public confidence in government, and psychological distress. Thus, fatigue is not only a predictor of adherence but also has an effect on other health-related aspects and may be influenced by official measures and strategies (33).

Based on qualitative data from our survey (not shown), we would also suggest that behavioral fatigue is influenced by changes in behavior. People may, for example, develop strategies to reduce their individual transmission risk (e.g., daily testing rather than wearing FFP2 masks), which may explain why some people do not follow all governmental measures. Behavioral fatigue may therefore be lower and adherence higher than shown in the results, as it is generally based on a measurement of adherence to concrete measures.

Comparable to other studies [e.g., (51)] self-efficacy was also found to be a predictor of adherence to COVID-19 measures. Even though the other predictors in the model have a larger influence on adherence, self-efficacy nonetheless plays an important role in dampening the Corona pandemic, as self-efficacy enhances adherence and reduces an individual's risk of infection. Moreover, COVID-19-related self-efficacy is also reported to be positively correlated with mental health, preventive behavior, and knowledge about COVID-19 (52). Additionally, enhancing self-efficacy not only influences an individual's transmission risk, but may also reduce the rate of new infections in the population as a whole. This is because self-efficacy appears to strengthen social norms and lead to more preventive behavior. On the other hand, Alemany-Arrebola et al. found that self-efficacy was sometimes negatively affected by aspects related to COVID-19, such as perceived stress associated with the pandemic, confinement, and critical events (illness and death of a relative/friend due to COVID-19) (53). These aspects increase individuals' anxiety levels and reduce their self-reported perceptions of (academic) self-efficacy. In summary, self-efficacy is an important aspect of adherence that was also found in other studies.