Compliance in the public versus the private realm: Economic preferences, institutional trust and COVID‐19 health behaviors

Sternberg, Henrike; Steinert, Janina Isabel; Büthe, Tim Article — Published Version Compliance in the public versus the private realm: Economic preferences, institutional trust and COVID‐19 health behaviors Health Economics Provided in Cooperation with: John Wiley & Sons Suggested Citation: Sternberg, Henrike; Steinert, Janina Isabel; Büthe, Tim (2024) : Compliance in the public versus the private realm: Economic preferences, institutional trust and COVID‐19 health behaviors, Health Economics, ISSN 1099-1050, Wiley, Hoboken, NJ, Vol. 33, Iss. 5, pp. 1055-1119, https://doi.org/10.1002/hec.4807 This Version is available at: https://hdl.handle.net/10419/293972 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/ Received: 4 April 2022 - Revised: 1 December 2023 - Accepted: 6 January 2024 DOI: 10.1002/hec.4807 RESEARCH ARTICLE Compliance in the public versus the private realm: Economic preferences, institutional trust and COVID‐19 health behaviors Henrike Sternberg 1,2,3 |Janina Isabel Steinert 1,3,4 |Tim Büthe 1,2,3,5 1 TUM School of Social Sciences and Technology, Technical University of Munich, Munich, Germany 2 TUM School of Management, Technical University of Munich, Munich, Germany 3 Munich School of Politics and Public Policy (HfP), Technical University of Munich, Munich, Germany 4 TUM School of Medicine and Health, Technical University of Munich, Munich, Germany 5 Sanford School of Public Policy, Duke University, Durham, North Carolina, USA Correspondence Henrike Sternberg, Department of Governance, School of Social Sciences and Technology, Technical University of Munich, Richard‐Wagner‐Strasse 1, Munich 80333, Germany. Email: [email protected] Funding information Horizon 2020 Framework Program, Grant/Award Number: 101016233 PERISCOPE Abstract To what extent do economic preferences and institutional trust predict compliance with physical distancing rules during the COVID‐19 pandemic? We reexamine this question by introducing the theoretical and empirical distinction between individual health behaviors in the public and in the private domain (e.g., keeping a distance from strangers vs. abstaining from private gatherings with friends). Using structural equation modeling to analyze survey data from Germany's second wave of the pandemic (N=3350), we reveal the following major differences between compliance in both domains: Social preferences, especially (positive) reciprocity, play an essential role in predicting compliance in the public domain but are barely relevant in the private domain. Conversely, individuals' degree of trust in the national government matters predominantly for increasing compliance in the private domain. The clearly strongest predictor in this domain is the perception pandemic‐related threats. Our findings encourage tailoring communication strategies to either domain‐specific circumstances or factors common across domains. Tailored communication may also help promote compliance with other health‐related regulatory policies beyond COVID‐19. KEYWORDS compliance, COVID‐19, economic preferences, health behavior, institutional trust, physical distancing JEL CLASSIFICATION D91, H12, H31, I12, I18 1 | INTRODUCTION What drives individual compliance with norms, standards, and imperfectly monitored laws and regulations? The importance of this question has long been recognized for general public policy contexts such as tax or fare avoidance as well as for health policy contexts such as vaccination mandates. More recently, the question has gained further This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial‐NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non‐commercial and no modifications or adaptations are made. © 2024 The Authors. Health Economics published by John Wiley & Sons Ltd. Health Economics. 2024;33:1055–1119. wileyonlinelibrary.com/journal/hec - 1055 significance for COVID‐19‐related physical distancing—a context in which individual behavior has a very high apparent societal relevance, but the individual and collective short‐and long‐term consequences of non‐compliance are relatively uncertain. 1 These characteristics pose a particular challenge for policy makers because they imply that a substantial amount of variance in compliance behaviors may not be exclusively driven by fully informed, rational cost‐benefit considerations. Instead, previous research suggests that compliance behavior in such contexts is shaped to a significant extent by individuals' economic preferences (i.e., social, risk and time preferences) and their degree of trust in the institutions endorsing the rules (e.g., Bargain & Aminjonov, 2020; Campos‐Mercade, Meier, Schneider, Meier, et al., 2021; Campos‐Mercade, Meier, Schneider, & Wengström, 2021; Chan, Skali, et al., 2020; Cucciniello et al., 2022; Keser & Rau, 2023; Shim et al., 2012; Sutinen & Kuperan, 1999). This paper proposes and empirically investigates a heretofore unaddressed implication emerging from the pronounced influence of these factors on compliance patterns: Compliance behavior may systematically vary between the public and the private domain, induced by the differential impact of individuals' economic preferences and institutional trust on compliance in these two domains. This is highly relevant in the context of behavioral stipulations to contain the spread of COVID‐19, which have included rules governing people's behavior in public spaces, such as requirements to wear masks and maintain physical distance from others, as well as rules governing relatively private behaviors, such as limits to the number of friends with whom to meet at home. While compliance in both the public and private domain is crucial to achieving the overarching objective of these rules, the two domains differ regarding the audiences who might observe and enforce compliance, suggesting potentially differing incentive structures. Against this background, we examined whether such a divergence in compliance behavior exists in the context of COVID‐19‐related physical distancing rules. Compliance in the public domain here comprised acting in conformity with health guidelines intended to govern behaviors that are easily observable by members of the general public, including public authorities. Examples include wearing a facemask or keeping a physical distance from people from other households in public spaces. Compliance in the private domain comprised behavior consistent with health guidelines intended to govern more private decisions about restricting social contacts and mobility altogether, which is to a large extent observed only by those who also fail to comply. Recognizing and examining a potential divergence in compliance between these two domains is important for advancing our theoretical understanding of compliance in general and moreover highly relevant for public policy in the context of COVID‐19. Given the lower COVID‐19 vaccine access and coverage in countries of the Global South and the prospect of emerging highly contagious virus variants, lockdown and physical distancing mandates remain crucial tools for containing infection rates in such scenarios. To assess the extent to which economic preferences and institutional trust might differ in their ability to predict health behaviors in the public versus in the private domain, we estimated separate structural equation models (SEMs) of self‐reported compliance with nationwide issued physical distancing rules, using original survey data from Germany's second wave of the pandemic in the winter of 2020/21 (N=3350). As for economic preferences, we examined risk aversion, patience, reciprocity, altruism and civic responsibility. As for institutional trust, we considered COVID‐19‐ related trust in the government and in scientific institutions. Our results confirm that compliance is significantly correlated with individuals' social and risk preferences and their institutional trust. This finding holds when controlling for COVID‐19 threat perception, which was revealed as the strongest predictor of compliance in both domains. More importantly, our survey data revealed that behavioral patterns differ significantly across the two compliance domains in three ways. First, respondents' level of positive reciprocity was of great importance for compliance in the public domain but barely relevant in the private domain. The same (but slightly weaker) domain‐specific differences emerged for negative reciprocity. Interestingly, correlations between reciprocity and compliance were positive in the case of positive reciprocity and negative in the case of negative reciprocity. Second, we also found domain‐specific patterns for the degree of trust in the national government and trust in scientific institutions: While trust in the government mattered only for increasing compliance in the private domain, trust in scientific institutions was an important factor in both domains, but significantly more so in the public domain. Third and more generally, our results suggest differences across domains in the relative importance of COVID‐19 threat perceptions vis‐à‐vis preferences and trust. Specifically, the dominance of COVID‐19 threat perceptions as the primary predictor of compliance was significantly more pronounced in the private domain. In contrast, individuals' social preferences were more strongly associated with compliance in the public domain. This analysis contributes to the literatures on the predictors of individual‐level outside‐the‐lab rule and norm compliance in economics, law, political science and psychology in a variety of contexts, including health behaviors 1056 - STERNBERG ET AL. during epidemics or pandemics (e.g., Algan et al., 2021; Blair et al., 2017; Böhm et al., 2016; Brodeur, Gray, et al., 2021; Galizzi et al., 2022). Furthermore, we contribute to the more specific and recently emerging literature on health behaviors in times of COVID‐19. In this literature, social preferences, risk preferences, (to a lesser extent) time preferences, institutional trust, and pandemic‐related threat perceptions have been identified both theoretically and empirically as important predictors of various types of compliance behaviors as well as mobility patterns. 2 To the best of our knowledge, this is the first paper to introduce—both in the general as well as in the more specific COVID‐19 compliance literature—the conceptual distinction between compliance in the public and the private domain, identify the implications for how compliance might be linked to individuals' economic preferences and institutional trust in distinct ways across the two domains, and systematically examine these potential differences empirically. Our findings suggest that the same individual may exhibit different degrees of compliance across these two domains. They also imply that the effectiveness of policies aimed at spurring compliance will vary across domains—or put differently: distinct policies might be needed to spur compliance in each domain. These policy implications of our findings are important for health policy well beyond behavioral stipulations during a pandemic, which is the context that allowed us to examine the issue. For instance, communication strategies aimed at improving public health through environmentally conscious consumer behavior might seek to encourage the purchase of products with a low carbon footprint at local stores, which we would classify as compliance in the public domain, because it is easily observed by fellow citizens. Conversely, communication strategies might target consumer behavior with regards to online purchases, which may be classified as compliance in the private domain. Quantifying the extent and analyzing the dynamics of a potential compliance divergence in this context would be highly relevant for designing effective environmental policies. The remainder of this paper is organized as follows. Section 2characterizes the two compliance domains and articulates expectations for (differential) impacts of individuals' economic preferences and institutional trust. Section 3 describes the survey design and the empirical strategy, that is, the SEMs. Section 4presents the main results and robustness checks. Section 5discusses the broader significance and policy implications. 2 | THEORETICAL CONSIDERATIONS 2.1 | Characterization of compliance domains The existing literature on (COVID‐19‐related) compliance does not distinguish between the public and private domain as spurring distinct logics of compliance. We now turn to this distinction. Physical distancing rules during the COVID‐19 pandemic have in numerous countries called for limiting social contacts and mobility in a variety of ways to reduce the risk of spreading the infection. Some of these rules predominantly govern behavior that inherently takes place in the public sphere, such as requirements or norms to, for example, wear a mask or keep a certain distance to persons from other households in public transport, at restaurants, in a public park, etc. Violations of these rules are easily observed (and hence enforceable), including by government authorities and by compliant fellow citizens. Other rules govern behavior that predominantly takes place in the private sphere, for example, rules asking citizens to restrict private gatherings to a maximum of two households or to only leave the house for necessary daily errands and other urgent reasons. We refer to decisions about complying or violating these rules as compliance in the private domain. Non‐compliance with such rules leaving the house to visit friends for fun instead of leaving the house only to get groceries, or attending or hosting a dinner party with friends from 10 different households is not easily observable. Moreover, it is most readily observed by individuals who have also chosen not to comply with the restrictions (the friends who themselves attend the dinner party). The two domains thus vary in terms of the observability of compliance behaviors to certain audiences. This results in differences with regard to (i) the risk of formal (i.e., state) punishment of non‐compliant behavior and (ii) the likelihood of social punishment by fellow citizens. Further, the two domains vary by (iii) the degree of social closeness of the people that seem most immediately affected by (non‐)compliance (in terms of the medical risk of getting infected with COVID‐19). Importantly, note that although compliance decisions across the two domains may in practice be correlated, they are logically orthogonal in the sense that compliance in any one realm could be practiced regardless of whether one complied with the rules for the respective other domain. STERNBERG ET AL. - 1057 As in (i), the risk of formal punishment, for example, getting fined for non‐compliance, was inherently higher in the public than in the private domain. For instance, mask‐wearing was in many places monitored through an increased police presence in subways or crowded city centers, whereas larger‐than‐allowed gatherings in the privacy of a home was subject only to the much smaller risk of reports by proactive neighbors. Thus, while the amount of fines at the time of data collection was higher for non‐compliance in the private domain, the risk of actually getting fined was higher in the public domain (see Table A30). To that end, a recent study suggests that the impact of economic preferences may be sensitive to the existence of government enforcement/punishment in the form of fines, which further strengthens the rationale for the suggested public‐private distinction (Papanastasiou et al., 2022). As in (ii), in terms of social punishment by fellow citizens, the type of audience to potentially execute such a punishment differs between both domains. While in the private realm, observable non‐compliance is subjected to disapproval by one's close peers, non‐compliance in the public realm is widely observed by the general public. One may argue that social incentives to comply in private settings may for this reason be in principle very strong (see also (iii) below). However, as highlighted above, in contrast to the public domain, compliance in the private domain is directly observed mostly by others who are also non‐compliant. Consequently, the likelihood (not necessarily the severity) of social punishment is also assumed to be lower in the private than in the public domain. As in (iii), the degree of social closeness of the people that seem most immediately affected by (non‐)compliant behavior (in terms of the risk of an infection) is higher in the private than in the public domain. Of course, a lack of compliance with the rules in either domain can cause a close family member or friend to get infected through virus transmission. However, this risk is much more salient in compliance behaviors in the private domain, where one directly decides about whether to meet with family and friends. Apart from that, this third distinctive characteristic also suggests that the personal dilemma of whether to comply or not is more substantial in the private domain: Complying means protecting one's closest friends/family but also not being able to maintain close social contact and support them. 2.2 | Logics of compliance: Distinctive decision‐making logics across domains In the following, we first present theoretically and empirically informed expectations on how economic preferences and institutional trust may affect COVID‐19 compliance overall before we then elaborate on how we expect dynamics to differ across the two compliance domains. 2.2.1 | Civic responsibility Compliance is likely more pronounced among individuals with a higher sense of civic responsibility, as the act of complying with physical distancing regulations during a pandemic resembles an act of civic responsibility (e.g., Barrios et al., 2021). Regarding differential dynamics across domains, expectations are conflicting: On the one hand, civic responsibility may be a more relevant driver of compliance in the public domain, which is the realm that social or civic duties are mainly associated with. On the other hand, civic responsibility can be viewed as an internalized, intrinsic motivation for compliance, which might thus be more important in the private domain, that is, in the absence of formal enforcement. 2.2.2 | Positive and negative reciprocity Individuals' level of positive and negative reciprocity may also affect compliance behavior: Specifically, in an environment in which compliance is generally high, a person with higher levels of positive reciprocity (i.e., a stronger willingness to return a favor) should exhibit a higher degree of compliance because compliance by others also protects this person and thus may be perceived as a favor to him or her (e.g., Nikolov et al., 2020). In contrast, negative reciprocity (the willingness to punish antisocial behavior) should in expectation not affect one's own level of compliance, because non‐compliance as an attempted punishment would, in the pandemic context, also punish those individuals who contribute to the public good (i.e., compliant individuals). However, one could argue that non‐compliant individuals might be punishable to a higher degree by one's own act of non‐compliance because compliance also yields 1058 - STERNBERG ET AL. self‐protection from the virus. This would suggest that individuals with higher levels of negative reciprocity comply relatively less with the imposed rules (e.g., Alfaro et al., 2022). In terms of differential dynamics across the two domains, a person's degree of (positive or negative) reciprocity should matter more for compliance in the public domain. Here, compliance behaviors are much more exposed to and observed by potential reciprocators than in the private domain, that is, whether a person wears a mask on the train or in the supermarket is observed by a higher number of individuals than whether a person stays at home alone and refrains from meeting with friends. 2.2.3 | Altruism We generally expect individuals with higher levels of altruism to exhibit higher compliance with COVID‐19 rules because those rules aim at reducing the spread of harmful infections among fellow citizens (e.g., Nikolov et al., 2020; Quaas et al., 2021). Pure altruism should not have any differential effects across the two compliance domains, since pure altruism refers to intrinsic values and does not include any reciprocated dynamics or incentives. As long as compliance in the public and private domain more or less equally helps to reduce the spread of the virus, higher levels of pure altruism should increase compliance regardless of whether it is relatively easily observed by others or not. However, given the personal dilemma individuals face in terms of compliance in the private domain, altruism could also have opposing effects in this domain: Altruism might not only call for protecting others from the medical consequences of the virus, but also from the social consequences, that is, social isolation. 2.2.4 | Risk preferences Risk‐averse individuals are expected to comply to a larger degree with physical distancing rules than individuals who are more risk‐accepting or risk‐seeking because higher compliance lowers the risk of getting infected, as well as the risk of getting fined for non‐compliance (e.g., Müller & Rau, 2021; Papanastasiou et al., 2022). Regarding differential effects across domains, we do not have strong expectations, given that non‐compliance in both domains can be characterized as risky behavior, only concerning differing aspects (e.g., the risk of punishment for non‐compliance vs. the risk of passing on an infection to close friends or family members). 2.2.5 | Time preferences We might expect more patient individuals (i.e., with lower discount rates) to exhibit higher levels of compliance as they are more willing to sacrifice a certain immediate reward (e.g., meeting with friends) for a later larger reward (e.g., the end of contact restrictions altogether) (e.g., Alfaro et al., 2022; Papanastasiou et al., 2022). We do not have strong expectations regarding differential effects across the two domains. 2.2.6 | Institutional trust The government and scientific institutions acted as key endorsers of the social‐distancing rules imposed during the COVID‐19 pandemic. Therefore, we expect that higher levels of COVID‐19‐related trust in governmental or scientific institutions spur compliance with physical distancing rules, which is in line with what recent empirical evidence suggests (e.g., Bargain & Aminjonov, 2020; Brodeur, Grigoryeva, & Kattan, 2021). For trust in the government, conflicting logics make the difference between the public and private realms theoretically indeterminate. On the one hand, trust in the government might have a more pronounced positive effect in the private domain, given that lower levels of monitoring and enforcement by state authorities make trust in the government as an intrinsic motivator more important. On the other hand, an understanding of the private domain as a realm in which the government has inherently no legitimate role to play might make trust in the government less relevant as a predictor. For trust in science, we do not have pronounced differential expectations, though one might argue that its relevance should be stronger in the public domain given the more technical‐scientific nature of the stipulations in this realm, for example, wearing a mask or keeping a 1.50 m distance from another. STERNBERG ET AL. - 1059 3 | MATERIAL AND METHODS 3.1 | Study setting and sampling The study was conducted as an online survey between February 3 and March 3, 2021, during the second nationwide COVID‐19 lockdown in Germany, which had begun in November 2020. The only stores fully operating at this time were those for daily necessities and medical supplies, while restaurants, retail stores and the like operated at most on a take‐away or delivery basis. Physical distancing rules for the second nationwide lockdown were put in place early and were repeatedly renewed, that is, they remained unchanged during the entire study period and we can expect the vast majority of the population to have been aware of their existence. 3 The national and state governments met to discuss potential changes in the national lockdown strategy on March 3, which marks the end of the data collection. The sample consisted of 3350 respondents recruited from a German access panel maintained by the survey company Respondi. Individuals were eligible to participate in the study if they were at least 18 years old and reported that they had spent the majority of the last 2 weeks in Germany. Quota sampling was used to obtain a representative sample of the German population with regard to (i) gender, (ii) age group, (iii) education, and (iv) state. Respondents received “mingle points” (worth between three to five Euros) for participating in the study, which they could redeem in the form of cash, vouchers, or donations. 3.2 | Survey design and outcome variables The survey was designed to collect information on the main variables of interest for this study, namely respondents' compliance with national physical distancing rules in Germany as well as their economic preferences and their level of institutional trust. We also collected information about an alternative highly relevant predictor of compliance, namely COVID‐19 threat perception, which has been shown to affect (COVID‐19) health behaviors (e.g., Jørgensen et al., 2021; Kluwe‐Schiavon et al., 2021; Papanastasiou et al., 2022; Plohl & Musil, 2021), and may be correlated with preferences and trust. The survey moreover collected information on a number of additional explanatory variables, including respondents' demographic and socioeconomic characteristics, political and ideological factors, knowledge about the efficacy of different prevention measures to reduce the spread of COVID‐19, and a scale to assess possible social desirability bias (Kemper et al., 2012). Tables A1 and A2 in Appendix A summarize all survey items employed in this paper (Table A21 reveals the exact formulation of the survey items as displayed to respondents, translated from the original German version). The full questionnaire, the pre‐analysis plan and the rationale for deviations from the latter can be retrieved from the supplementary material. 3.2.1 | Elicitation of compliance behaviors Compliance was elicited by asking respondents about their behavior in six situations governed by various physical distancing rules issued by the German national government. In each case, respondents were asked to rate on a scale from 1 to 5 the extent to which their own behavior in the past 2 weeks reflected these behaviors (ranging from never to always). Following the theoretical considerations above, three questions were intended to primarily elicit information about compliance in the public domain, asking respondents about (i) wearing a mask in public transport or when shopping, (ii) keeping the government‐stipulated distance of approximately 1.50 m in public spaces, and (iii) avoiding handshakes when greeting other people. Another three questions were primarily intended to elicit information about compliance in the private domain, asking respondents about (iv) leaving the home only when truly necessary, (v) restricting private meetings to the government‐stipulated limit of one person from one other household, and (vi) minimizing interactions with persons from outside one's own households in general. 4 Importantly, the German federal government's stipulations—and hence the requirements for compliant behavior—were equally clear and stable for the private and for the public domain during the data gathering phase. 1060 - STERNBERG ET AL. 3.2.2 | Elicitation of economic preferences and institutional trust With regards to economic preferences, we elicited respondents' level of altruism, positive and negative reciprocity, risk aversion, patience and civic responsibility. To capture institutional trust, we elicited their COVID‐19‐related trust in the national government and in the Robert‐Koch‐Institute (RKI), the latter as a proxy for trust in science. For the majority of these factors (altruism, positive and negative reciprocity, risk aversion, patience), we adapted the items and measurement procedure from the German version of the Global Preference Module (Falk et al., 2022,2018): We used both (i) attitudinal measures that ask about generally behaving in a certain way, and (ii) actual incentivized choices (such as donation decisions in the case of altruism or lottery participation in the case of risk aversion). These survey items were standardized and then used to construct one final measure for each preference, based on the weights for the survey items that emerged from the experimental validation procedure by Falk et al. (2018, p. 1653). 5 The survey items, weights, final preference measures, and general procedure are summarized in Tables A4–A6 and Figure A5 in Appendix A. 6 Civic responsibility was measured using (i) respondents' reported voter turn‐out in the last national election, (ii) their self‐reported tendency (not) to evade fares in public transport, and (iii) their self‐reported tendency (not) to litter. Responses to these three items were used to estimate a factor score of respondents' underlying level of civic responsibility, which we assumed to be the primary common factor among these indicators (for a similar approach, see, e.g., Müller and Rau (2021)). Institutional trust was elicited through questions about respondents' degree of trust in the German national government's and the RKI's ability to manage the pandemic situation. These two institutions were the primary endorsers of physical distancing rules and the main sources of official public health communications during the pandemic in Germany. Throughout the pandemic, the RKI has been the most widely known and recognized German national‐level scientific body to conduct epidemiological and medical analyses of COVID‐19 and to issue policy recommendations. It thus served as a proxy for trust in science in the German context (Betsch et al., 2021b). 3.2.3 | Elicitation of COVID‐19 threat perception COVID‐19 threat perception was captured using a battery of questions about (i) how threatening respondents perceived the COVID‐19 pandemic to be in general and (ii) how threatening they perceived it to be with regard to specific aspects of their lives, including their own health or the health of those close to them, their financial situation and their social lives. These items were used to estimate factor scores capturing respondents' underlying COVID‐19 threat perceptions to be then employed in the subsequent analysis. The majority of the items were adapted from Betsch et al. (2021a). 3.3 | Data collection and processing The survey was programmed in German using Qualtrics and piloted with 150 participants. The recruitment for the final survey was conducted by the survey company Bilendi. 7 Analyses were performed in R (version 4.1.0) and STATA17. Informed consent was obtained from all respondents before they were presented with the questionnaire, which they could interrupt or exit at any time. As part of the debriefing upon completion of the survey, participants were provided with a substantive list of resources for help and information sources about the COVID‐19 pandemic as well as mental health support services. 3.4 | Empirical strategy The empirical strategy comprised essentially two steps. First, we examined compliance across the six different physical distancing behaviors to ascertain to what extent there is empirical evidence for the existence of the conceptual distinction between compliance in the private and in the public domain. In view of this, we employed exploratory factor analyses to identify the subsets of physical distancing behaviors that reflect compliance in each domain and then derived initial estimates for compliance in the public and private domain, respectively. STERNBERG ET AL. - 1061 Second, we investigated by means of SEM techniques to what extent compliance patterns are correlated with individuals' economic preferences and institutional trust for each of the two domains of compliance. SEMs help to reduce measurement error in the underlying latent variable(s) of interest—here compliance behavior—by combining path analysis (the structural component of the SEM) with confirmatory factor analysis (the measurement component of the SEM) (Acock, 2013). In our case, the SEM simultaneously (i) fits a confirmatory factor analysis that captures compliance in the public/private domain as a latent variable and (ii) estimates effects of preferences and trust on this latent measure of compliance. The confirmatory factor analysis (i.e., the measurement component of the SEM) is defined as follows for compliance in each domain d, where d={public, private}. y0 d¼λ0 dCdþe0 dψd;ð1Þ y0 din Equation (1) denotes a vector of the subset of the six observed compliance items that reflect the respective compliance domain, using the results from the exploratory factor analyses in the first step (see Section 4.1). C d denotes the identified latent measure of domain‐specific compliance (i.e., the common factor within each item subset), and λ0 dis a vector of the regression coefficients of the model, that is, the factor pattern coefficients (loadings) of the observed items for their respective compliance domain. e0 dcorrespond to the unobserved unique factors of the six compliance items and ψ d are the coefficients relating the unique factors to the items. The variables of interest here are the factors capturing compliance in both domains, C d , which are assumed to induce observed responses to the respective subset of the six compliance items. The latter, y0 d, are therefore the dependent variables in the measurement model and constitute the reflective indicators of compliance in each domain (Acock, 2013). The structural component of the SEM regresses compliance behaviors in each domain (i.e., the latent variables of compliance) on economic preferences and institutional trust, and is defined as follows. Cd¼P0αdþγdTþX0ηdþZ0ζdþυdð2Þ P0denotes a vector of the regression coefficients of preferences and trust on the latent measure of domain‐specific compliance (C d ).γddenotes the regression coefficient of COVID‐19 threat perception (T) as an alternative predictor of compliance in each domain. X0is a vector of demographic and socioeconomic controls, namely gender, age group, state, education, employment in essential services, household size and income, and Z0is a vector of specific compliance controls, namely knowledge about COVID‐19 preventive measures and the degree of social desirability bias. υddenotes the error term. Individual subscripts are suppressed for simplicity. The measurement component and the structural component of the SEM are connected through the latent variable, that is, compliance in the public/private domain, respectively, allowing us to simultaneously estimate the above equations. We estimated the SEM separately for each compliance domain using Diagonal Weighted Least Squares on a polychoric correlation matrix while generating robust standard errors and a corrected test statistic to account for the ordinal and not normally distributed compliance items (e.g., Finney & DiStefano, 2006; Li, 2016). 4 | RESULTS Overall, 3350 respondents completed the online survey, among which 49.85% were female, 49.91% were male, and 0.24% indicated their gender to be diverse. Respondents were on average 47.83 years old. In terms of age, gender, educational attainment, and state of residence our sample was representative of the German population aged 18–74 (see Table A2 in Appendix A). All six compliance items were non‐normally distributed and skewed to the left. This means self‐reported compliance was generally high, which could be indicative of some social desirability bias in reporting but is not necessarily surprising given that data was collected in the midst of the quite intense second wave of infections in Germany (for a similar finding at that time, see Betsch et al., 2021b). Histograms and density plots of all six compliance items are presented in Figure A1 in Appendix A. Descriptive statistics for all variables employed in the core analysis are presented in Tables A1, A2 8 and A6 in Appendix A and the screeplots and factor scores of constructing the indexes for COVID‐19 threat perception and civic responsibility are shown in Tables A7, A8 and Figures A2–A4 in Appendix A. 9 1062 - STERNBERG ET AL. 4.3.2 | Additional robustness checks We conducted a number of additional analyses to assess the robustness of our results by having a closer look at the dependent variable, that is, the compliance measures, as well as at the different hypothesized predictors. For the sake of brevity, this subsection merely summarizes the different approaches and their results briefly, while Appendix Bcontains a more detailed account of the rationale behind the approaches and of their results. FIGURE 3 External relevance of compliance domains: State‐level Google mobility patterns. This figure shows scatter plots, linear (with 95% confidence intervals) regression lines, Pearson correlation coefficients, and p‐values of (i) compliance in the public and private domain, as predicted from the structural equation model (SEM) by means of Equations (1) and (2) (on the x‐axis), and (ii) phone‐tracking‐ based changes in mobility according to Google's mobility reports in the areas of Retail and Recreation, Workplace, Transit, and Residential (on the y‐axis). Higher values of the SEM‐predicted compliance measures indicate higher compliance. In the first three graphs, lower values of mobility changes (i.e., more negative values) indicate stronger reductions in mobility relative to the 2020 baseline period (correspondingly for increases in mobility for the case of Retail and Recreation, fourth graph). The unit of analysis are the 16 federal states in Germany, given that this is the lowest level of Google mobility reports available for Germany. STERNBERG ET AL. - 1069 In terms of the elicitation of compliance, we re‐estimated the SEM with an alternative measure of compliance in the private domain (i.e., altering the measurement part of the SEM), which we assume is less susceptible to social desirability bias, but still has the advantage of being available at the respondent level (as opposed to the Google mobility patterns). For this alternative measure of private compliance, we utilized three questions asking respondents about their willingness to participate in concrete social activities (for more details see Appendix B). The results of this exercise reveal estimates that are similar to the previous ones for compliance in the private domain (and that differ from the ones for public compliance in the same crucial instances), thus, strengthening the credibility of the main results. See Tables A17 and A18 and Figures A9 and A10 in Appendix A for the detailed results of the adjusted measurement model and the structural model of the SEM. In terms of the different hypothesized predictors, we conducted additional analyses to (i) account for potentially mediating effects of respondents' COVID‐19 threat perception (see Table A19 and Figure A11), and (ii) control for two more possibly important competing predictors, namely respondents' degree of generalized interpersonal trust and their residence in urban versus rural areas (see Table A22). As in (i), we argue that individual threat perceptions of the COVID‐19 pandemic may themselves be affected by economic preferences and institutional trust (Harper et al., 2020; Plohl & Musil, 2021), therefore suggesting not only direct but also indirect effects of preferences and trust through threat perception. In view of this, we re‐estimated the SEM by adding our measure of threat perception as a mediator to the structural component of the SEM. The results suggest that the differences across compliance domains predominantly stem from direct effects—thus, reinforcing the core argument and finding of this paper—while the estimated indirect effects were very similar across domains and in most cases also much smaller in magnitude. As in (ii), we find that our main results are robust to including a respondent's degree of generalized trust—as measured by means of a survey‐based item adopted from Falk et al. (2018)—or their rural as opposed to urban place of residence as additional predictors of public and private compliance. Moreover, our results do not suggest that a rural/urban setting plays a statistically significant role for compliance behaviors in either domain (if any, there is a small negative estimated effect of a rural setting on compliance). Interestingly, we find that generalized trust (the degree to which respondents believe other people to generally have good intentions) significantly decreases compliance only in the public domain, while it has no statistically significant estimated effect in the private domain (see Appendix Bfor a more detailed discussion of this additional finding). 5 | DISCUSSION AND CONCLUSION Compliance with expert behavioral recommendations and explicit mandates is crucial for a society's ability to achieve a wide range of public health objectives (and public policy objectives in general), ranging from safeguards for patient privacy or vaccination mandates to requirements for COVID‐19‐related physical distancing. It is especially crucial when the behavioral stipulations or mandates established by such norms, standards or regulations are only imperfectly enforceable, and compliance depends to a greater and substantive degree on individual choices and considerations about whether to comply. Accordingly, previous research has identified economic preferences and institutional trust as important drivers of individual‐level compliance (e.g., Algan et al., 2021; Bargain & Aminjonov, 2020; Campos‐Mercade, Meier, Schneider, Meier, et al., 2021; Campos‐Mercade, Meier, Schneider, & Wengström, 2021; Chan, Skali, et al., 2020; Cucciniello et al., 2022; Keser & Rau, 2023; Papanastasiou et al., 2022; Shim et al., 2012; Sutinen & Kuperan, 1999). In this paper, we have introduced the conceptual distinction between compliance in the public and the private domain and have explored empirically, in the context of compliance with COVID‐19‐related physical distancing rules, to what extent its correlations with economic preferences and institutional trust differ across the two domains. Understanding individual‐level compliance and recognizing potential differences between the private and public domain remains highly relevant in this context. Even though the immediate urgency of the current pandemic may seem to have passed, an increasing likelihood of novel epidemics and pandemics (e.g., Marani et al., 2021) combined with a significant degree of vaccine hesitancy “especially towards newly developed vaccines” suggests that physical distancing mandates will persist as a crucial part of governments' policy toolkit. Our fine‐grained analysis revealed systematic heterogeneity across the two identified domains, advancing our understanding of compliance and thus providing more tangible grounds for policy interventions. Specifically, while individuals' risk and time preferences appeared to be similarly relevant for compliance across domains, we found significantly different correlations across the two compliance domains in the case of (i) reciprocity (and to some extent also generally for social preferences as a whole), (ii) institutional trust and (iii) COVID‐19 threat perception. 15 1070 - STERNBERG ET AL. First, our empirical analyses suggest that relying on, or appealing to, reciprocal dynamics may only be a promising strategy for compliance in the public domain. These results are in line with our theoretical expectation laid out in Section 2: in the public realm, the reciprocation of compliance behaviors (e.g., in the form of wearing a mask in public or only entering an elevator separately) is more observable and, thus, much more salient in people's minds than in the private domain. Here, compliance occurs in the form of staying at home and isolating, but one does not directly perceive others doingthesame—atleastnottotheextentthatitisthecaseinthepublicrealm.Thisinterpretationismoreoversupported by the fact that we observed a substantively and statistically much weaker correlation between compliance in the public domain and altruism, which, in its pure form, we had hypothesized and defined without any reciprocal component. Second, the results reveal somewhat opposing patterns for trust in the national government and trust in scientific institutions (the RKI). While the former only seems to matter in the private realm, the latter plays a crucial role in both realms but more dominantly in the public domain. For trust in the RKI, we had weak expectations of a stronger relevance in the public domain as a result of the relatively more technical and specific stipulations in this domain, which may, thus, be more saliently perceived as scientifically validated regulations. We found support for this expectation in supplementary analyses, which reveal the exact same pattern for other proxies of trust in scientific institutions, namely, trust in science in general and trust in the World Health Organization (see Table A28 and Appendix B). For trust in the national government, we had indeterminate theoretical expectations. Additional supplementary analyses exploiting respondents' trust in government‐related media channels point to a possible explanation of the high relevance of trust in the national government in the private domain (see Table A29 and Appendix B). The observed dynamic could well be a result of the communication strategies employed by the national government over the course of the pandemic and chancellor Angela Merkel's public addresses urging citizens to stay at home and isolate (i.e., using a narrative along the lines of “united in separation”). This communication strategy might have worked against the perception of the private realm as a realm in which the government has no prominent role to play, especially for individuals with high levels of trust in the government. Finally, our findings suggest that policies which succeed in adequately informing individuals about the threat of (the detrimental consequences of) a COVID‐19 infection are likely to be highly effective across compliance domains but to an even greater extent in the private realm. Considering the different theoretically outlined characteristics of the two domains, this finding seems plausible. In the private domain, the perceived risk of passing on an infection to a close family member or friend through non‐compliance is much more salient than in the public domain. At the same time, the fear of losing a close family member or friend to COVID‐19 was one of the most dominant indicators of our measure for COVID‐19 threat perception. This potential for a varying effectiveness of communication strategies across the compliance domains suggests that policy makers should either tailor communications strategies to the circumstances of each domain or focus on the determinants that are common across domains. We are cautiously referring to our findings in terms of correlations rather than causal effects, given that gathering the data through a cross‐sectional survey in the midst of the pandemic did not allow for ensuring exogeneity by design. However, our results remained stable when controlling for various alternative influences and when performing a number of additional robustness checks (see Section 4). Moreover, previous literature seems to suggest that economic preferences and institutional trust are likely exogenous to the analyses conducted in this paper. 16 In addition, we also acknowledge that our paper only employs self‐reported compliance measures and utilizes data from an online survey panel. However, given the various robustness checks conducted to validate the compliance measures and the comparisons of our survey data with other representative non‐online surveys, we still believe that our research documents real behavioral mechanisms that can provide useful insights to policy makers. Finally, our paper, strictly speaking, only captures a snapshot of behaviors at the one specific point in the pandemic when the data was collected. Nevertheless, our survey was conducted in the midst of Germany's third wave, and, thus, in the midst of a phase of the pandemic, during which vaccines were not yet available and physical distancing was crucial—which is the exact phase relevant to our research question. Although we have investigated compliance only in the pandemic context of Germany, the general distinction between compliance in the public and private domain is likely also relevant for other countries that introduced a similar catalog of physical distancing rules. Moreover, the conceptual and empirical contribution of this paper may extend beyond the context of the current and possible future pandemics to other aspects of public policy more generally as well as to health policy in particular. One related example is easily monitorable versus largely unobserved compliance with different types of hygiene regulations by health personnel/professionals. Another, more general example of a contemporary and very pressing regulatory challenge with similar characteristics are policies encouraging STERNBERG ET AL. - 1071 environmentally responsible behavior. A public‐private compliance divergence in this context may for instance manifest itself in the differential behavioral predictors of environmentally responsible consumer behaviors in supermarkets versus online shopping with a home delivery option. Finally, our approach also has theoretical relevance for research and established findings on regulatory compliance by challenging the way in which compliance is conceptualized. ACKNOWLEDGMENTS The authors would like to thank the editor for their guidance and support throughout the review process, and the two anonymous reviewers for their valuable inputs. The authors moreover wish to thank the members of the International Relations Research Group at the Hochschule für Politik/TUM Department of Governance, especially Tobias Rommel and Zlatina Georgieva, as well as the organizers and the participants of the behavioral & empirical work in progress Seminar at the TUM School of Management and the 3rd DGGÖ Workshop “Health Economics and Development” for their valuable comments. We further wish to thank the participants of the TUM PhD seminar POL90002, Research Design and Empirical Methodsand the TUMResearchDesign Bachelor Seminar for fruitful discussions of the questionnaire andresearchdesign. Finally, we would like to thank lecturers and participants of the St. Gallen Global School in Empirical Research Methods (GSERM) for important discussions on Factor analyses and Structural Equation models. The authors gratefully acknowledge funding from the European Union's Horizon 2020 research and innovation program under grant agreement no. 101016233 PERISCOPE. The funding source had no involvement in the study design, collection, analysis and interpretation of data, in the writing of the report, and in the decision to submit the article for publication. Open Access funding enabled and organized by Projekt DEAL. CONFLICT OF INTEREST STATEMENT The authors declare no conflicts of interest. DATA AVAILABILITY STATEMENT Data available on request from the authors. ETHICS STATEMENT Ethics approval for this study was obtained from the committee for human subjects and research ethics review of the medical faculty at the Technical University of Munich (TUM, 20/21 S‐SR). ORCID Henrike Sternberg https://orcid.org/0000-0001-8539-6478 Janina Isabel Steinert https://orcid.org/0000-0001-7120-0075 Tim Büthe https://orcid.org/0000-0002-4724-5000 ENDNOTES 1 By uncertainty we mean long tails in the probability distribution, as in what Knight (1921) called “risk”. 2 An extensive, but not complete list of conducted works in this regard include Barrios et al. (2021), Campos‐Mercade, Meier, Schneider, and Wengström (2021), Durante et al. (2021), Müller and Rau (2021), Bartscher et al. (2021), Borgonovi and Andrieu (2020), Nikolov et al. (2020), Quaas et al. (2021), Sheth and Wright (2020), and van Hulsen et al. (2020) for various types of social preferences; Papanastasiou et al. (2022), Andersson et al. (2021), Müller and Rau (2021), Schunk and Wagner (2021), Alfaro et al. (2022), Chan, Skali, et al. (2020), Nikolov et al. (2020), Pullano et al. (2020), Xie et al. (2020), and Xu and Cheng (2021) for risk preferences; Fang et al. (2022), Papanastasiou et al. (2022), Andersson et al. (2021), Müller and Rau (2021), Schunk and Wagner (2021), Alfaro et al. (2022), and Nikolov et al. (2020) for time preferences; Brodeur, Grigoryeva, and Kattan (2021), Farzanegan and Hofmann (2022), Fazio et al. (2021), Granados Samayoa et al. (2021), Kazemian et al. (2021), Koetke et al. (2021), Plohl and Musil (2021), Bargain and Aminjonov (2020), Chan, Brumpton, et al. (2020), and Goldstein and Wiedemann (2022) for institutional trust; and Papanastasiou et al. (2022), Algan et al. (2021), Jørgensen et al. (2021), Kluwe‐Schiavon et al. (2021), Plohl and Musil (2021), Harper et al. (2020), Vai et al. (2020), and Van Bavel et al. (2020) for COVID‐19‐related threat perceptions. 3 In addition to the nationwide rules of interest for this paper, there were some minor differences across the German states in the specific rules and recommendations regarding for example, school/nursery restrictions, contact restrictions for young children and disabled individuals, or the specifics of quarantining after returning from a trip outside Germany (Press and Information Office of the Federal Government, 2021a,2021b). 4 The behaviors and the corresponding questions were adapted from Betsch et al. (2021a) and slightly adjusted. See Table A3 in Appendix A for the exact wording. As part of our robustness checks, we consider an alternative way of distinguishing between compliance in the public and the private realm; see Section 4.3. 1072 - STERNBERG ET AL. 5 The experimental validation procedure enabled Falk et al. (2018) to analyze which linear combination of the different survey items performed best in predicting the corresponding behavior in an experimental setting in the lab. We used these same identified weights to form our preference measures. Note that Falk et al. (2018) conducted the validation procedure with a German sample and thus, in the same country context as this study. 6 For positive reciprocity, we were only able to collect one of the two survey items intended to form the final measure for positive reciprocity. We therefore proceeded with this single item and further assessed the results for robustness when using only a single survey item for all the other preferences as well (selecting the item that had been assigned the highest weight in the experimental validation procedure by Falk et al. (2018)). Our core findings remained robust, see Table A20 and Figure A12 in Appendix A. 7 Prior to the pilot launch in the field, the survey was moreover piloted and discussed in two research design seminars at the authors' university. 8 In order to increase the credibility of and validate the main variables of interest for our empirical analysis (i.e., compliance, economic preferences and threat perception), we compare descriptive statistics of our survey data with other representative surveys that collected data on presumably comparable items, specifically data from the COVID‐19 Snapshot Monitoring, the World Value Survey (WVS) and the Global Preference Survey. We generally find a high similarity between our survey data and the other datasets (see Tables A24, A25 and Figure A13), while the similarities are highest for our measures of compliance and threat perception and slightly less so for our measures of economic preferences. Specifically, as intuitively to be expected, the similarity in descriptive statistics is slightly lower, the more a measure deviates from those employed by the Global Preference Survey. 9 We examine, in an additional analysis, to what extent state‐level averaged COVID‐19 threat perception is correlated with state‐level COVID‐19 case incidence rates. For this analysis, we employ data about the COVID‐19 incidence date by region (number of registered COVID‐19 infections in a state within the past 7 days/100,000 inhabitants) from the COVID‐19 dataset by the Federal Statistical Office of Germany (“Statistisches Bundesamt”). We find a positive (statistically insignificant) correlation between COVID‐19 case incidence during the time of the data collection and average threat perception at the state level (Pearson correlation coefficient: þ0.2928). We, moreover, find a negative (statistically insignificant) correlation between state‐level threat perception and the average case incidence in a state throughout the infection waves since the start of the pandemic and therefore also prior to the start of our data collection (Pearson correlation coefficient: −0.3847). See Table A23 in Appendix A as well as Appendix Bfor a detailed summary and interpretation of results. 10 Degrees of freedom for SEMs are differently calculated (see e.g., Rigdon, 1994). 11 See Table A16 in Appendix A for the Pearson's correlation matrix of the core explanatory variables, which suggests that we were not facing a case of highly correlated explanatory variables. With the exception of trust in government/trust in the RKI (corr. coeff.: 0.71), all other pairwise correlations were of a low/moderate degree (corr. coeff.:0.01–0.38). 12 We conduct a number of additional analyses with potential proxies for trust in the RKI (trust in science, trust in WHO) and potential channels for trust in the government (trust in state‐level government, trust in established media channels) to better understand the differential effects observed for the institutional trust variables. The results reveal that the estimated effects for trust in the RKI are indeed robust, that is, we observe the same pattern for the utilized proxies: a high, statistically significant relevance in both domains, but more so in the public domain. For trust in the government, the additional analyses seem to suggest that the large, statistically significant estimated effects only in the private domain may be a result of the communication strategy of particularly the national government, whose narrative focused predominantly on the general message to stay at home and isolate (i.e., closely related to our definition of private compliance). See Tables A28 and A29 for details. 13 Our findings for respondents' age align well with those for COVID‐19 threat perceptions, which previous studies have found to be stronger among older citizens: Both factors were a stronger predictor of compliance in the private domain. However, the fact that both variables remain statistically significant when added simultaneously suggests that the age effect captures a dynamic that is somewhat distinct from respondents' pandemic‐related threat perceptions (e.g., older people being more rule‐compliant in general, especially in private settings, while younger people largely comply only when substantially monitored). 14 Mobility changes in going to the grocery store/pharmacy are conceptually unrelated to the compliance items in our survey. Mobility changes in parks would be hard to interpret given very different weather conditions throughout weeks of the year, which predominantly determine outside activities in Germany during these months. 15 Our results are particularly interesting in light of a recent contribution by Papanastasiou et al. (2022): The authors find that economic preferences, specifically, risk and time preferences, become less relevant as predictors of compliance behaviors if respondents are presented with the hypothetical prospect of being fined for non‐compliance. While they conducted their data collection at a point in time when fines had not yet been introduced, fines had already come into effect when we conducted our study. The comparison with Papanastasiou et al. (2022) thus seems to permit the following two additional interpretations of our results: First, the fact that we find economic preferences in general to be statistically relevant despite the existence of fees suggests that the estimated effects are rather a lower bound for their relevance in the absence of fees. Second, the finding by Papanastasiou et al. (2022), who examined risk and time preferences in particular, may also serve as a partial explanation for why social preferences as opposed to risk and time preferences are more important for compliance behaviors in our study. This applies especially for compliance in the public realm, which we had not only assigned a larger likelihood of formal (state‐) punishment through fines, but also of social punishment to which especially the impacts of social preferences seem to be sensitive. STERNBERG ET AL. - 1073 16 Betsch et al. (2021b), Drichoutis and Nayga (2021), Shachat et al. (2021), Angrisani et al. (2020), Bu et al. (2020), Ikeda et al. (2023), Lotti and Pethiyagoda (2021), van de Groep et al. (2020), Habibpour et al. (2018), Chuang and Schechter (2015), Meier and Sprenger (2015), Carlsson et al. (2014), Volk et al. (2012), and Andersen et al. (2008), see Appendix Bfor a structured and more detailed overview of these articles' relevance for our argument. 17 Data was retrieved from https://www.corona‐daten‐deutschland.de/dataset/infektionen_bundeslaender. REFERENCES Acock, A. C. (2013). Discovering structural equation modeling using Stata. Stata Press Books. Alfaro, L., Faia, E., Lamersdorf, N., & Saidi, F. (2022). Health externalities and policy: The role of social preferences. Management Science, 68(9), 6751–6761. https://doi.org/10.1287/mnsc.2022.4461 Algan, Y., Cohen, D., Davoine, E., Foucault, M., & Stantcheva, S. (2021). Trust in scientists in times of pandemic: Panel evidence from 12 countries. Proceedings of the National Academy of Sciences,118(40), e2108576118. https://doi.org/10.1073/pnas.2108576118 Andersen, S., Harrison, G. W., Lau, M. I., & Elisabet Rutström, E. (2008). Lost in state space: Are preferences stable? International Economic Review,49(3), 1091–1112. https://doi.org/10.1111/j.1468‐2354.2008.00507.x Andersson, O., Campos‐Mercade, P., Meier, A. N., & Wengström, E. (2021). Anticipation of COVID‐19 vaccines reduces willingness to socially distance. Journal of Health Economics,80, 102530. https://doi.org/10.1016/j.jhealeco.2021.102530 Angrisani, M., Cipriani, M., Guarino, A., Kendall, R., & Ortiz de Zarate, J. (2020). Risk preferences at the time of COVID‐19: An experiment with professional traders and students (Federal Reserve Bank of New York Staff Report 927). https://ssrn.com/abstract=3609586 Bargain, O., & Aminjonov, U. (2020). Trust and compliance to public health policies in times of COVID‐19. Journal of Public Economics,192, 104316. https://doi.org/10.1016/j.jpubeco.2020.104316 Barrios, J. M., Benmelech, E., Hochberg, Y. V., Sapienza, P., & Zingales, L. (2021). Civic capital and social distancing during the COVID‐19 pandemic. Journal of Public Economics,193, 104310. https://doi.org/10.1016/j.jpubeco.2020.104310 Bartscher, A. K., Seitz, S., Siegloch, S., Slotwinski, M., & Wehrhöfer, N. (2021). Social capital and the spread of COVID‐19: Insights from European countries. Journal of Health Economics,80, 102531. https://doi.org/10.1016/j.jhealeco.2021.102531 Betsch, C., Wieler, L., Bosnjak, M., Ramharter, M., Stollorz, V., Omer, S., Korn, L., Sprengholz, P., Felgendreff, L., Eitze, S., & Schmid, P. (2021a). Germany COVID‐19 Snapshot MOnitoring (COSMO Germany): Monitoring knowledge, risk perceptions, preventive behaviours, and public trust in the current coronavirus outbreak in Germany – Wave 27. Retrieved from https://projekte.uni‐erfurt.de/ cosmo2020/web/summary/27/ Betsch, C., Wieler, L., Bosnjak, M., Ramharter, M., Stollorz, V., Omer, S., Korn, L., Sprengholz, P., Felgendreff, L., Eitze, S., & Schmid, P. (2021b). Germany COVID‐19 Snapshot MOnitoring (COSMO Germany): Monitoring knowledge, risk perceptions, preventive behaviours, and public trust in the current coronavirus outbreak in Germany – Wave 38 summary. Retrieved from https://projekte.uni‐erfurt. de/cosmo2020/web/summary/38/ Blair, R. A., Morse, B. S., & Tsai, L. L. (2017). Public health and public trust: Survey evidence from the Ebola Virus Disease epidemic in Liberia. Social Science & Medicine,172, 89–97. https://doi.org/10.1016/j.socscimed.2016.11.016 Böhm, R., Betsch, C., & Korn, L. (2016). Selfish‐rational non‐vaccination: Experimental evidence from an interactive vaccination game. Journal of Economic Behavior & Organization,131, 183–195. https://doi.org/10.1016/j.jebo.2015.11.008 Borgonovi, F., & Andrieu, E. (2020). Bowling together by bowling alone: Social capital and COVID‐19. Social Science & Medicine,265, 113501. https://doi.org/10.1016/j.socscimed.2020.113501 Brodeur, A., Gray, D., Islam, A., & Bhuiyan, S. (2021). A literature review of the economics of COVID‐19. Journal of Economic Surveys,35(4), 1007–1044. https://doi.org/10.1111/joes.12423 Brodeur, A., Grigoryeva, I., & Kattan, L. (2021). Stay‐at‐home orders, social distancing, and trust. Journal of Population Economics,34(4), 1–34. https://doi.org/10.1007/s00148‐021‐00848‐z Bu, D., Hanspal, T., Liao, Y., & Liu, Y. (2020). Risk taking during a global crisis: Evidence from Wuhan. Covid Economics,5, 106–146. Campos‐Mercade, P., Meier, A. N., Schneider, F. H., Meier, S., Pope, D., & Wengström, E. (2021). Monetary incentives increase COVID‐19 vaccinations. Science,374(6569), 879–882. https://doi.org/10.1126/science.abm0475 Campos‐Mercade, P., Meier, A. N., Schneider, F. H., & Wengström, E. (2021). Prosociality predicts health behaviors during the COVID‐19 pandemic. Journal of Public Economics,195, 104367. https://doi.org/10.1016/j.jpubeco.2021.104367 Carlsson, F., Johansson‐Stenman, O., & Nam, P. K. (2014). Social preferences are stable over long periods of time. Journal of Public Economics,117, 104–114. https://doi.org/10.1016/j.jpubeco.2014.05.009 Chan, H. F., Brumpton, M., Macintyre, A., Arapoc, J., Savage, D. A., Skali, A., Stadelmann, D., & Torgler, B. (2020). How confidence in health care systems affects mobility and compliance during the COVID‐19 pandemic. PLoS One,15(10), e0240644. https://doi.org/10.1371/ journal.pone.0240644 Chan, H. F., Skali, A., Savage, D. A., Stadelmann, D., & Torgler, B. (2020). Risk attitudes and human mobility during the COVID‐19 pandemic. Scientific Reports,10(1), 1–13. https://doi.org/10.1038/s41598‐020‐76763‐2 Chuang, Y., & Schechter, L. (2015). Stability of experimental and survey measures of risk, time, and social preferences: A review and some new results. Journal of Development Economics,117, 151–170. https://doi.org/10.1016/j.jdeveco.2015.07.008 Cucciniello, M., Pin, P., Imre, B., Porumbescu, G. A., & Melegaro, A. (2022). Altruism and vaccination intentions: Evidence from behavioral experiments. Social Science & Medicine,292, 114195. https://doi.org/10.1016/j.socscimed.2021.114195 1074 - STERNBERG ET AL. Destatis Statistisches Bundesamt. (2021). Bevölkerungsstand: Amtliche Einwohnerzahl Deutschlands 2022. [Online]. Retrieved January, 29, 2021 from https://www.destatis.de/DE/Themen/Gesellschaft‐Umwelt/Bevoelkerung/Bevoelkerungsstand/_inhalt.html Drichoutis, A. C., & Nayga, R. M. (2021). On the stability of risk and time preferences amid the COVID‐19 pandemic. Experimental Economics,25(3), 1–36. https://doi.org/10.1007/s10683‐021‐09727‐6 Durante, R., Guiso, L., & Gulino, G. (2021). Asocial capital: Civic culture and social distancing during COVID‐19. Journal of Public Economics,194, 104342. https://doi.org/10.1016/j.jpubeco.2020.104342 Falk, A., Becker, A., Dohmen, T., Enke, B., Huffman, D., & Sunde, U. (2018). Global evidence on economic preferences. Quarterly Journal of Economics,133(4), 1645–1692. https://doi.org/10.1093/qje/qjy013 Falk, A., Becker, A., Dohmen, T. J., Huffman, D., & Sunde, U. (2022). The preference survey Module: A validated instrument for measuring risk, time, and social preferences. Management Science,69(4), 1935–1950. https://doi.org/10.1287/mnsc.2022.4455 Fang, X., Freyer, T., Ho, C. Y., Chen, Z., & Goette, L. (2022). Prosociality predicts individual behavior and collective outcomes in the COVID‐ 19 pandemic. Social Science & Medicine,308, 115192. https://doi.org/10.1016/j.socscimed.2022.115192 Farzanegan, M. R., & Hofmann, H. P. (2022). A matter of trust? Political trust and the COVID‐19 pandemic. International Journal of Sociology,52(6), 476–499. https://doi.org/10.1080/00207659.2022.2086729 Fazio, R. H., Ruisch, B. C., Moore, C. A., Granados Samayoa, J. A., Boggs, S. T., & Ladanyi, J. T. (2021). Who is (not) complying with the US social distancing directive and why? Testing a general framework of compliance with virtual measures of social distancing. PLoS One, 16(2), e0247520. https://doi.org/10.1371/journal.pone.0247520 Finney, S. J., & DiStefano, C. (2006). Non‐normal and categorical data in structural equation modeling. In G. R. Hancock & R. O. Mueller (Eds.), Structural equation modeling: A second course (Chap. 9, pp. 269–314). Information Age Publishing. Galizzi, M. M., W. Lau, K., Miraldo, M., & Hauck, K. (2022). Bandwagoning, free‐riding and heterogeneity in influenza vaccine decisions: An online experiment. Health Economics,31(4), 614–646. https://doi.org/10.1002/hec.4467 Goldstein, D. A., & Wiedemann, J. (2022). Who do you trust? The consequences of political and social trust for public responsiveness to COVID‐19 orders. The Consequences of Political and Social Trust for Public Responsiveness to COVID‐19 Orders. Perspectives on Politics, 20(2), 412–438. https://doi.org10.1017/S1537592721000049 Google LLC. (2021). Google COVID‐19 community mobility reports [online]. Retrieved July 9, 2021, from https://www.google.com/covid19/ mobility/ Granados Samayoa, J. A., Ruisch, B. C., Moore, C. A., Boggs, S. T., Ladanyi, J. T., & Fazio, R. H. (2021). When does knowing better mean doing better? Trust in President Trump and in scientists moderates the relation between COVID‐19 knowledge and social distancing. Journal of Elections, Public Opinion, and Parties,31(sup1), 218–231. https://doi.org/10.1080/17457289.2021.1924744 Habibpour, M. M., Peiffer, M., Pepermans, R., & Jegers, M. (2018). How giving affects giving: A long‐term analysis of donations. Applied Economics,50(21), 2402–2413. https://doi.org/10.1080/00036846.2017.1397853 Harper, C. A., Satchell, L. P., Fido, D., & Latzman, R. D. (2020). Functional fear predicts public health compliance in the COVID‐19 pandemic. International Journal of Mental Health and Addiction,19(5), 1–14. https://doi.org/10.1007/s11469‐020‐00281‐5 Ikeda, S., Yamamura, E., & Tsutsui, Y. (2023). COVID‐19 Enhanced Diminishing Sensitivity in Prospect‐Theory Risk Preferences: A Panel Analysis. Review of Behavioral Economics,10(4), 287–313. Jørgensen, F., Bor, A., & Petersen, M. B. (2021). Compliance without fear: Individual‐level protective behaviour during the first wave of the COVID‐19 pandemic. British Journal of Health Psychology,26(2), 679–696. https://doi.org/10.1111/bjhp.12519 Kazemian, S., Fuller, S., & Algara, C. (2021). The role of race and scientific trust on support for COVID‐19 social distancing measures in the United States. PLoS One,16(7), e0254127. https://doi.org/10.1371/journal.pone.0254127 Kemper, C. J., Beierlein, C., Bensch, D., Kovaleva, A., & Rammstedt, B. (2012). Eine Kurzskala zur Erfassung des Gamma‐Faktors sozial erwünschten Antwortverhaltens: Die Kurzskala Soziale Erwünschtheit‐Gamma (KSE‐G) (GESIS‐Leibniz‐Institut für Sozialwissenschaften Working Papers, 2012/25). GESIS ‐ Leibniz‐Institut für Sozialwissenschaften. Keser, C., & Rau, H. A. (2023). Determinants of people’s motivations to approach COVID‐19 vaccination centers. Scientific Reports,13(1), 5282. https://doi.org/10.1038/s41598‐023‐30244‐4 Klopp, E. (2022). A tutorial on testing the equality of standardized regression coefficients in structural equation models using Wald tests with lavaan. The Quantitative Methods for Psychology,16(4), 315–333. Kluwe‐Schiavon, B., Viola, T. W., Bandinelli, L. P., Castro, S. C. C., Kristensen, C. H., Costa da Costa, J., & Grassi‐Oliveira, R. (2021). A behavioral economic risk aversion experiment in the context of the COVID‐19 pandemic. PLoS One,16(1), e0245261. https://doi.org/10. 1371/journal.pone.0245261 Knight, F. H. (1921). Risk, uncertainty and profit (Vol. 31). Houghton Mifflin. Koetke, J., Schumann, K., & Porter, T. (2021). Trust in science increases conservative support for social distancing. Group Processes & Intergroup Relations,24(4), 680–697. https://doi.org/10.1177/1368430220985918 Li, C.‐H. (2016). Confirmatory factor analysis with ordinal data: Comparing robust maximum likelihood and diagonally weighted least squares. Behavior Research Methods,48(3), 936–949. https://doi.org/10.3758/s13428‐015‐0619‐7 Lotti, L., & Pethiyagoda, S. (2021). Generosity during COIVID‐19: Investigating socioeconomic shocks and game framing. Humanities and Social Sciences Communications,9(1). https://doi.org/10.1057/s41599‐022‐01200‐w Marani, M., Katul, G. G., Pan, W. K., & Parolari, A. J. (2021). Intensity and frequency of extreme novel epidemics. Proceedings of the National Academy of Sciences,118(35), e2105482118. https://doi.org/10.1073/pnas.2105482118 STERNBERG ET AL. - 1075 Meier, S., & Sprenger, C. D. (2015). Temporal stability of time preferences. The Review of Economics and Statistics,97(2), 273–286. https://doi. org/10.1162/rest_a_00433 Merkel, A. (Televised speech). (2020). Fernsehansprache von Angela Merkel (Pressemitteilung 100). [Online]. Retrieved May, 10, 2023, from https://www.bundesregierung.de/breg‐de/aktuelles/fernsehansprache‐von‐bundeskanzlerin‐angela‐merkel‐1732134 Müller, S., & Rau, H. A. (2021). Economic preferences and compliance in the social stress test of the COVID‐19 crisis. Journal of Public Economics,194, 104322. https://doi.org/10.1016/j.jpubeco.2020.104322 Nikolov, P., Pape, A., Tonguc, O., & Williams, C. (2020). Predictors of social distancing and mask‐wearing behavior: Panel survey in seven US states. Human Capital and Economic Opportunity Working Group. Papanastasiou, A., Ruffle, B. J., & Zheng, A. (2022). Compliance with social distancing: Theory and empirical evidence from Ontario during COVID‐19. Canadian Journal of Economics/Revue canadienne d’économique,55(S1), 705–734. https://doi.org/10.1111/caje.12565 Plohl, N., & Musil, B. (2021). Modeling compliance with COVID‐19 prevention guidelines: The critical role of trust in science. Psychology Health & Medicine,26(1), 1–12. https://doi.org/10.1080/13548506.2020.1772988 Press and Information Office of the Federal Government. (2021a). Coronavirus in Deutschland [online]. Retrieved September 19, 2021, from https://www.bundesregierung.de/breg‐de/themen/coronavirus/corona‐diese‐regeln‐und‐einschraenkung‐gelten‐1734724 Press and Information Office of the Federal Government. (2021b). Regeln in den Bundesländern [online]. Retrieved September 19, 2021, from https://www.bundesregierung.de/breg‐de/themen/coronavirus/corona‐bundeslaender‐1745198 Pullano, G., Valdano, E., Scarpa, N., Rubrichi, S., & Colizza, V. (2020). Evaluating the effect of demographic factors, socioeconomic factors, and risk aversion on mobility during the COVID‐19 epidemic in France under lockdown: A population‐based study. The Lancet Digital Health,2(12), e638–e649. https://doi.org/10.1016/s2589‐7500(20)30243‐0 Quaas, M., Meya, J., Schenk, H., Bos, B., Drupp, M. A., & Requate, T. (2021). The social cost of contacts: Theory and evidence for the COVID‐ 19 pandemic in Germany. PLoS One,16(3), e0248288. https://doi.org/10.1371/journal.pone.0248288 Rigdon, E. E. (1994). Calculating degrees of freedom for a structural equation model. Structural Equation Modeling: A Multidisciplinary Journal,1(3), 274–278. https://doi.org/10.1080/10705519409539979 Schunk, D., & Wagner, V. (2021). What determines the willingness to sanction violations of newly introduced social norms: Personality traits or economic preferences? Evidence from the COVID‐19 crisis. Journal of Behavioral and Experimental Economics,93, 101716. https:// doi.org/10.1016/j.socec.2021.101716 Shachat, J., Walker, M., & Wei, L. (2021). How the onset of the COVID‐19 pandemic impacted pro‐social behaviour and individual preferences: Experimental evidence from China. Journal of Economic Behavior & Organization,190, 480–494. https://doi.org/10.1016/j.jebo. 2021.08.001 Sheth, K., & Wright, G. C. (2020). The usual suspects: Do risk tolerance, altruism, and health predict the response to COVID‐19? Review of Economics of the Household,18(4), 1041–1052. https://doi.org/10.1007/s11150‐020‐09515‐w Shim, E., Chapman, G. B., Townsend, J. P., & Galvani, A. P. (2012). The influence of altruism on influenza vaccination decisions. Journal of The Royal Society Interface,9(74), 2234–2243. https://doi.org/10.1098/rsif.2012.0115 Sutinen, J. G., & Kuperan, K. (1999). A socio‐economic theory of regulatory compliance. International Journal of Social Economics,26(1–2), 174–193. https://doi.org/10.1108/03068299910229569 Vai, B., Cazzetta, S., Ghiglino, D., Parenti, L., Saibene, G., Toti, M., Verga, C., Wykowska, A., & Benedetti, F. (2020). Risk perception and media in shaping protective behaviors: Insights from the early phase of COVID‐19 Italian outbreak. Frontiers in Psychology,11.https:// doi.org/10.3389/fpsyg.2020.563426 Van Bavel, J. J., Baicker, K., Boggio, P. S., Capraro, V., Cichocka, A., Cikara, M., Crockett, M. J., Crum, A. J., Douglas, K. M., Druckman, J. N., Drury, J., Dube, O., Ellemers, N., Finkel, E. J., Fowler, J. H., Gelfand, M., Han, S., Haslam, S. A., Jetten, J., … Willer, R. (2020). Using social and behavioural science to support COVID‐19 pandemic response. Nature Human Behaviour,4(5), 460–471. https://doi.org/10. 1038/s41562‐020‐0884‐z van de Groep, S., Zanolie, K., Green, K. H., Sweijen, S. W., & Crone, E. A. (2020). A daily diary study on adolescents’ mood, empathy, and prosocial behavior during the COVID‐19 pandemic. PLoS One,15(10), e0240349. https://doi.org/10.1371/journal.pone.0240349 van Hulsen, M., Rohde, K. I., & van Exel, J. (2020). Inter‐temporal and social preferences predict compliance in a social dilemma: An application in the context of COVID‐19 (Tinbergen Institute Discussion Paper 2020‐047/I). Tinbergen Institute. http://hdl.handle.net/ 10419/229667 Volk, S., Thöni, C., & Ruigrok, W. (2012). Temporal stability and psychological foundations of cooperation preferences. Journal of Economic Behavior & Organization,81(2), 664–676. https://doi.org/10.1016/j.jebo.2011.10.006 Xie, K., Liang, B., Dulebenets, M. A., & Mei, Y. (2020). The impact of risk perception on social distancing during the COVID‐19 pandemic in China. International Journal of Environmental Research and Public Health,17(17), 6256. https://doi.org/10.3390/ijerph17176256 Xu, P., & Cheng, J. (2021). Individual differences in social distancing and mask‐wearing in the pandemic of COVID‐19: The role of need for cognition, self‐control and risk attitude. Personality and Individual Differences,175, 110706. https://doi.org/10.1016/j.paid.2021.110706 How to cite this article: Sternberg, H., Steinert, J. I., & Büthe, T. (2024). Compliance in the public versus the private realm: Economic preferences, institutional trust and COVID‐19 health behaviors. Health Economics, 33(5), 1055–1119. https://doi.org/10.1002/hec.4807 1076 - STERNBERG ET AL. APPENDIX A TABLE A1 Overview and descriptive statistics of survey items. Variable Values/description Mean (SD) Min Max N Female 0: Male (50.03%) 0.499 (0.50) 0 1 3342 1: Female (49.97%) Age group 1: 18–29 (18.99%) 3.12 (1.44) 1 5 3350 2: 30–39 (17.85%) 3: 40–49 (18.06%) 4: 50–59 (22.09%) 5: 60þ(23.01%) Education 1: Low (33.50%) 2.02 (0.83) 1 3 3349 2: Medium (31.11%) 3: High (35.38%) Employed in essential services 0: No (78.81%) 0.21 (0.41) 0 1 3350 1: Yes (21.19%) Household income (monthly) 1: Less than 900€ (9.30%) 4.54 (2.06) 1 10 3333 2: 900–1299€ (10.68%) 3: 1300–1699€ (10.56%) 4: 1700–2299€ (16.17%) 5: 2300–3199€ (20.70%) 6: 3200–3999€ (15.39%) 7: 4000–4999€ (10.23%) 8: 5000–5999€ (3.96%) 9: 6000–9999€ (2.61%) 10: More than 10,000€ (0.39%) Household size 1: Just me (28.90%) 2.06 (0.85) 1 4 3350 2: 2–3 people (40.15%) 3: 3–4 people (26.81%) 4: More than 4 people (4.15%) Compliance items Keeping distance Scale from 1 (applies never) to 5 (applies always in last 2 weeks) 4.44 (0.87) 1 5 3346 Wearing masks Scale from 1 (applies never) to 5 (applies always in last 2 weeks) 4.78 (0.71) 1 5 3347 Avoiding handshakes Scale from 1 (applies never) to 5 (applies always in last 2 weeks) 4.70 (0.79) 1 5 3347 Only leaving home when necessary Scale from 1 (applies never) to 5 (applies always in last 2 weeks) 4.21 (1.11) 1 5 3346 Avoiding other households Scale from 1 (applies never) to 5 (applies always in last 2 weeks) 4.15 (1.11) 1 5 3345 (Continues) STERNBERG ET AL. - 1077 TABLE A1 (Continued) Variable Values/description Mean (SD) Min Max N Restricting private meetings Scale from 1 (applies never) to 5 (applies always in last 2 weeks) 4.14 (1.16) 1 5 3347 Alternative compliance items Willingness to join friends’ sledding trip next weekend (rev. coded) Scale from 1 (not at all willing) to 7 (very willing) 2.33 (1.91) 1 7 2742 Willingness to join friends’ surprise birthday caroling (rev. coded) Scale from 1 (not at all willing) to 7 (very willing) 3.32 (2.18) 1 7 3002 Willingness to join friends’ dinner/movie night (rev. coded) Scale from 1 (not at all willing) to 7 (very willing) 2.40 (1.96) 1 7 2944 Economic preference items Altruism Incentivized donation decision 0–10 (donation from 10€ windfall endowment) 3.88 (3.46) 0 10 3320 Willingness to give to good causes Scale from 1 (not at all willing) to 11 (very willing) 7.82 (2.82) 1 11 3349 Pos. reciprocity Willingness to return a favor Scale from 1 (not at all willing) to 11 (very willing) 9.44 (1.98) 1 11 3349 Neg. reciprocity Willingness to take revenge Scale from 1 (not at all willing) to 11 (very willing) 3.80 (2.78) 1 11 3348 Willingness to punish unfair behavior toward self Scale from 1 (not at all willing) to 11 (very willing) 4.88 (2.87) 1 11 3349 Willingness to punish unfair behavior toward others Scale from 1 (not at all willing) to 11 (very willing) 5.28 (2.83) 1 11 3349 Risk aversion Willingness to take risks (rev. coded) Scale from 1 (not at all willing) to 11 (very willing) 6.45 (2.63) 1 11 3350 Incentivized lottery choice 0: Lottery with 5% chance of 10 Euro payment (54.45%) 0.46 (0.50) 0 1 3339 1: Guaranteed payment of 0.50 Euro (45.55%) Patience Intertemporal choice sequence using staircase method 1–32 (see Figure A5 for details) 17.20 (12.11) 1 32 3111 Willingness to wait Scale from 1 (not at all willing) to 11 (very willing) 7.53 (2.48) 1 11 3349 Civic responsibility Voter turnout 0: Did not vote in the last national election 2017 (14.94%) 0.85 (0.36) 0 1 3200 1: Voted in the last national election 2017 (85.06%) Frequency to evade fares (rev. coded) Scale from 1 (never) to 6 (always) 1.71 (0.98) 1 6 3350 Frequency of littering (rev. coded) Scale from 1 (applies not at all) to 4 (applies always) 1.78 (0.93) 1 4 3350 Institutional trust items Trust in RKI Scale from 1 (very low trust) to 7 (very high trust in this institution) 4.91 (1.91) 1 7 3350 Trust in government Scale from 1 (very low trust) to 7 (very high trust in this institution) 4.10 (1.90) 1 7 3349 1078 - STERNBERG ET AL. TABLE A12 Structural equation model (SEM) measurement component (confirmatory factor analysis). Latent variable Compliance in the public domain Compliance in the private domain Wearing a mask in public transport/when shopping 0.892*** (0.035) Keeping a 1.5 m distance in public spaces (whenever possible) 0.833*** (0.027) Avoiding handshake greetings 0.886 (const.) Leaving the home only when absolutely necessary 0.775*** (0.028) Generally avoiding other households 0.895*** (0.038) Restricting private meetings to 1 person from another household 0.743 (const.) Observations =2922 Note: The table shows the results purely of the measurement component of the SEM for each compliance domain, as in Equation (1) (but resulting from estimating both equations simultaneously). Coefficient estimates are standardized and estimated using diagonal weighted least squares and a polychoric correlation matrix. See Table 2for the fit statistics of the full SEM. *p<0.05, **p<0.01, ***p<0.001. TABLE A11 Compliance: Summated rating scale reliability analysis. Item Cronbach's α α If item is deleted Scale: Compliance in the public domain 0.797 Wearing a mask in crowded places 0.765 Keeping a 1.5 m distance in public spaces (whenever possible) 0.712 Avoiding handshakes 0.693 Scale: Compliance in the private domain 0.780 Leaving the home only when absolutely necessary 0.709 Generally avoiding other households 0.642 Restricting private meetings to 1 person from another household 0.753 Correlation between scales: 0.643 Observations: 3340 Note: The table shows the results of a reliability analysis (i.e., overall Cronbach's αand Cronbach's αwhen dropping a certain item) of two summated rating scales using the two item triplets as indicated in the table and identified in the exploratory factor analysis in Table 1. TABLE A13 Decision‐making across compliance domains—extended structural equation model (SEM) results (also showing coefficients for control variables). Outcome Compliance in the public domain Compliance in the private domain COVID‐19 threat perception 0.297*** 0.365*** (0.026) (0.017) Altruism 0.037 0.012 (0.031) (0.020) (Continues) STERNBERG ET AL. - 1085 TABLE A13 (Continued) Outcome Compliance in the public domain Compliance in the private domain Civic responsibility 0.127*** 0.108*** (0.024) (0.016) Pos. reciprocity 0.180*** 0.030 (0.024) (0.017) Neg. reciprocity −0.149*** −0.086*** (0.029) (0.019) Risk aversion 0.065** 0.055** (0.034) (0.022) Patience −0.001 0.028 (0.031) (0.020) Trust in RKI 0.204*** 0.068** (0.032) (0.021) Trust in government 0.012 0.138*** (0.033) (0.021) Female 0.117*** 0.039* (0.025) (0.016) Age group 0.057* 0.135*** (0.027) (0.018) High education −0.002 0.019 (0.030) (0.020) Medium education 0.027 0.016 (0.028) 0.018 Employed essential services 0.020 −0.021 (0.024) (0.016) Household income 0.003 −0.034 (0.030) (0.019) Household size −0.044 −0.013 (0.028) (0.018) Knowledge preventive measures 0.117*** 0.087*** (0.025) (0.016) Social desirability index 0.092*** 0.102*** (0.024) (0.016) R 2 0.492 0.411 Observations 2918 2918 Note: Displayed are standardized coefficient estimates and standard errors in parentheses for the results of the SEMs estimated by means of Equations (1) and (2) (shown are only the structural component results, see Table A12 for the results of the measurement component). The SEM was estimated separately for each compliance domain, using Diagonal Weighted Least Squares, a polychoric correlation matrix, robust standard errors and a corrected test statistic. All estimations moreover control for the respondents' state of residence. Table A14 repeats the same analyses with the final‐model sample of N=2922 throughout. See Tables A1, A15 and A21 and Figure A8 for the source survey items and factor scores of the social desirability bias index. *p<0.05, **p<0.01, ***p<0.001. 1086 - STERNBERG ET AL. TABLE A14 Decision‐making across compliance domains—structural equation model (SEM) results (same observations in all submodels). Outcome Compliance in the public domain Compliance in the private domain (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) COVID‐19 threat perception 0.464*** 0.321*** 0.304*** 0.314*** 0.297*** 0.502*** 0.387*** 0.372*** 0.381*** 0.365*** (0.022) (0.025) (0.026) (0.025) (0.026) (0.016) (0.017) (0.017) (0.017) (0.017) Altruism 0.164*** 0.060** 0.044 0.053* 0.037 0.127*** 0.030 0.017 0.024 0.012 (0.028) (0.030) (0.031) (0.030) (0.031) (0.019) (0.019) (0.020) (0.019) (0.020) Civic responsibility 0.196*** 0.172*** 0.147*** 0.151*** 0.127*** 0.202*** 0.169*** 0.127*** 0.148*** 0.108*** (0.023) (0.023) (0.024) (0.023) (0.024) (0.016) (0.016) (0.016) (0.016) (0.016) Pos. reciprocity 0.248*** 0.218*** 0.205*** 0.193*** 0.180*** 0.115*** 0.066** 0.056** 0.039* 0.030 (0.023) (0.023) (0.024) (0.024) (0.024) (0.018) (0.017) (0.017) (0.017) (0.017) Neg. reciprocity −0.193*** −0.184*** −0.170*** −0.165*** −0.149*** −0.125*** −0.109*** −0.103*** −0.092*** −0.086*** (0.027) (0.028) (0.029) (0.028) (0.029) (0.019) (0.019) (0.019) (0.019) (0.019) Risk aversion 0.169*** 0.085*** 0.066** 0.085*** 0.065** 0.168*** 0.080*** 0.055** 0.081*** 0.055** (0.031) (0.033) (0.034) (0.033) (0.034) (0.022) (0.021) (0.022) (0.021) (0.022) Patience 0.141*** −0.014 0.004 −0.019 −0.001 0.115*** 0.010 0.032 0.006 0.028 (0.027) (0.029) (0.031) (0.029) (0.031) (0.019) (0.019) (0.020) (0.019) (0.020) Trust in RKI 0.385*** 0.223*** 0.216*** 0.210*** 0.204*** 0.239*** 0.087** 0.075** 0.079** 0.068** (0.030) (0.031) (0.032) (0.031) (0.032) (0.021) (0.021) (0.021) (0.021) (0.021) Trust in government 0.042 0.014 0.017 0.008 0.012 0.186*** 0.137*** 0.145*** 0.129*** 0.138*** (0.031) (0.031) (0.033) (0.031) (0.033) (0.022) (0.020) (0.021) (0.020) (0.021) Demogr. and socioecon. controls No No No No No Yes No Yes No No No No No Yes No Yes Compliance‐specific controls No No No No No No Yes Yes No No No No No No Yes Yes R 2 0.215 0.222 0.043 0.173 0.448 0.477 0.464 0.492 0.252 0.116 0.037 0.155 0.366 0.396 0.382 0.411 Observations 2918 2918 2918 2918 2918 2918 2918 2918 2918 2918 2918 2918 2918 2918 2918 2918 Note: Displayed are standardized coefficient estimates and standard errors in parentheses for the results of the SEMs estimated by means of Equations (1) and (2) (shown are only the structural component results, see Table A12 for the results of the measurement component), replicating Table 2but using the same observations for all estimations. The SEM was estimated separately for each compliance domain, using Diagonal Weighted Least Squares, a polychoric correlation matrix, robust standard errors and a corrected test statistic. Demographic controls contain respondents' gender, age group and state. Socioeconomic controls include education, employment in essential services, household size and income. Compliance‐specific controls contain knowledge about COVID‐19 preventive measures and the degree of social desirability bias (factor scores as in Figure A8 and Table A15). *p<0.05, **p<0.01, ***p<0.001. STERNBERG ET AL. - 1087 TABLE A15 Social desirability: Factor loadings. Factor 1 Social desirability items (positive) Unconditional objectivity in disputes 0.658 Unconditional kindness under stress 0.772 Unconditional attention in conversations 0.564 Eigenvalue 1.347 Percent shared variance accounted for 44.90 Multiple R 2 of scores with factors 0.730 Observations: 3339 Note: The table shows standardized factor pattern coefficients from a 1‐factor solution of the survey items aimed at capturing social desirability bias (exaggerating positive characteristics). Estimated using iterated principal axis factoring, no rotation and polychoric correlations. TABLE A16 Correlation matrix of core explanatory variables. CV19‐threat Pos. Reci. Neg. Reci. Altruism Civ. Resp. Risk aversion Patience Trust (RKI) Trust (Govt.) CV19‐threat 1.00 Pos. Reci. 0.16*** 1.00 Neg. Reci. 0.02 −0.03 1.00 Altruism 0.18*** 0.24*** −0.03 1.00 Civ. Resp. 0.08*** 0.13*** −0.19*** 0.12*** 1.00 Risk aversion 0.08*** −0.05** −0.21*** −0.12*** 0.12*** 1.00 Patience 0.06** 0.14*** 0.02 0.22*** 0.05** −0.11*** 1.00 Trust (RKI) 0.38*** 0.14*** −0.04* 0.23*** 0.10*** −0.00 0.20*** 1.00 Trust (Govt.) 0.33*** 0.06*** −0.03 0.21*** 0.06** −0.02 0.20*** 0.71*** 1.00 Source: Shown are Pearson correlations between the main explanatory variables of the SEM, that is, the final preference measures as well as the COVID‐19 threat perception measure. *p<0.05, **p<0.01, ***p<0.001. TABLE A17 Alternative compliance measure: Structural equation model (SEM) measurement component (confirmatory factor analysis). Latent variable Alternative compliance measure Sledding trip (reverse coded) 0.896 (const.) Surprise birthday caroling (reverse coded) 0.830*** (0.018) Dinner or movie night (reverse coded) 0.932*** (0.024) Observations =2141 Note: The table purely shows the results of the measurement component of the SEM, as in Equation (1) (but resulting from estimating both equations simultaneously). However, instead of the initial compliance items, we used three alternative survey items as reflective indicators, as displayed in the table (see Section 4.3 and Tables A1 and A21 for detailed descriptions). Coefficient estimates are standardized and estimated using diagonal weighted least squares and a polychoric correlation matrix. *p<0.05, **p<0.01, ***p<0.001. 1088 - STERNBERG ET AL. TABLE A18 Robustness: Decision‐making across compliance domains—structural equation model (SEM) results with alternative compliance measure. Outcome: Compliance (alternative measure) COVID‐19 threat perception 0.319*** (0.025) Altruism 0.025 (0.030) Civic responsibility 0.173*** (0.024) Pos. reciprocity −0.053* (0.026) Neg. reciprocity −0.164*** (0.028) Risk aversion 0.099*** (0.031) Patience 0.010 (0.030) Trust in RKI 0.156*** (0.033) Trust in government 0.079** (0.034) Female −0.034 (0.023) Age group 0.207*** (0.026) High education 0.013 (0.029) Medium education 0.009 (0.027) Employed essential services −0.040* (0.022) Household income 0.014 (0.029) Household size 0.008 (0.026) Knowledge preventive measures 0.073*** (0.023) Social desirability index −0.014 (0.023) (Continues) STERNBERG ET AL. - 1089 TABLE A18 (Continued) Outcome: Compliance (alternative measure) R 2 0.469 Fit statistics (full SEM) Robust χ 2 (66) =126.907; p<0.001 Robust RMSEA =0.021 Robust TLI =0.999 Robust CFI =0.987 SRMR =0.007 R 2 =0.469 Observations =2141 Note: Displayed are standardized coefficient estimates and standard errors in parentheses for the results of the SEM estimated by means of Equations (1) and (2) (shown are only the structural component results, see Table A17 for the results of the measurement component) and fit statistics for the full SEM. However, instead of the initial compliance items, we used three alternative survey items as reflective indicators (see Section 4.3 and Tables A1 and A21 for detailed descriptions). The SEM was estimated using Diagonal Weighted Least Squares, a polychoric correlation matrix, robust standard errors and a corrected test statistic. All estimations moreover control for the respondents' state of residence. See Tables A1, A15 and A21 and Figure A8 for the source survey items and factor scores of the social desirability bias index. Abbreviations: CFI, comparative fit index; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residual; TLI, Tucker‐Lewis index. *p<0.05, **p<0.01, ***p<0.001. TABLE A19 Robustness: Decision‐making across compliance domains—structural equation model (SEM) results with mediation. Model Compliance in the public domain Compliance in the private domain Direct effect Indirect effect Total effect Direct effect Indirect effect Total effect COVID‐19 threat perception 0.286*** ‐ 0.286*** 0.359*** ‐ 0.359*** (0.021) (0.021) (0.017) (0.017) Altruism 0.044 0.019*** 0.063** 0.013 0.024*** 0.037 (0.031) (0.007) (0.031) (0.024) (0.008) (0.026) Civic responsibility 0.121*** −0.004 0.105*** 0.110*** −0.005 0.100*** (0.024) (0.006) (0.024) (0.019) (0.007) (0.021) Pos. reciprocity 0.172*** 0.021*** 0.193*** 0.027 0.026*** 0.053* (0.024) (0.006) (0.025) (0.021) (0.007) (0.022) Neg. reciprocity −0.142*** 0.020*** −0.122*** −0.085*** 0.025*** −0.060** (0.028) (0.007) (0.029) (0.023) (0.008) (0.025) Risk aversion 0.069** 0.028*** 0.097*** 0.056** 0.035*** 0.092*** (0.033) (0.008) (0.034) (0.026) (0.009) (0.028) Patience 0.003 −0.003 0.000 0.029 −0.004 0.025 (0.030) (0.007) (0.031) (0.024) (0.008) (0.026) Trust in RKI 0.201*** 0.068*** 0.270*** 0.069** 0.086*** 0.156*** (0.031) (0.009) (0.032) (0.026) (0.009) (0.027) Trust in government 0.031 0.038*** 0.069* 0.144*** 0.047*** 0.192*** (0.032) (0.008) (0.033) (0.026) (0.009) (0.027) 1090 - STERNBERG ET AL. TABLE A19 (Continued) Model Compliance in the public domain Compliance in the private domain Direct effect Indirect effect Total effect Direct effect Indirect effect Total effect Female 0.116*** 0.031*** 0.147*** 0.039* 0.039*** 0.078*** (0.024) (0.006) (0.025) (0.020) (0.007) (0.021) Age group 0.057* 0.020*** 0.078** 0.138*** 0.026*** 0.163*** (0.027) (0.006) (0.027) (0.021) (0.007) (0.023) High education −0.003 −0.011 −0.014 0.020 −0.013 0.007 (0.030) (0.007) (0.030) (0.024) (0.008) (0.026) Medium education 0.023 −0.002 0.021 0.016 −0.003 0.013 (0.027) (0.006) (0.028) (0.022) (0.007) (0.024) Employed essential services 0.019 −0.016** 0.003 −0.020 −0.020** −0.040* (0.023) (0.006) (0.024) (0.020) (0.007) (0.021) Household income 0.009 −0.005 0.004 −0.033 −0.007 −0.039 (0.029) (0.007) (0.030) (0.024) (0.008) (0.025) Household size −0.046 0.015** −0.030 −0.015 0.019 0.004 (0.027) (0.006) (0.028) (0.022) (0.008) (0.024) Knowledge measures 0.130*** ‐ 0.130*** 0.102*** ‐ 0.102*** (0.025) ‐ (0.025) (0.021) ‐ (0.021) Social desirability index 0.099*** 0.099*** 0.110*** ‐ 0.110*** (0.024) ‐ (0.024) (0.021) ‐ (0.021) Fit statistics (full SEM) Public Private Robust χ 2 (68) =167.704; p<0.001 Robust χ 2 (68) =116.578; p<0.001 Robust RMSEA =0.022 Robust RMSEA =0.016 Robust TLI =0.997 Robust TLI =0.999 Robust CFI =0.965 Robust CFI =0.988 SRMR =0.025 SRMR =0.013 R 2 =0.487 R 2 =0.412 Observations =2918 Observations =2918 Note: Displayed are standardized coefficient estimates and standard errors in parentheses of estimating the SEMs by means of Equations (1) and (2), and additionally allowing for a mediating effect of COVID‐19 threat perceptions. Shown are the resulting estimated direct, indirect and indirect effects (i.e., the results of the structural component) as well as fit statistics of the full SEM. The SEM was estimated separately for each compliance domain, using Diagonal Weighted Least Squares, a polychoric correlation matrix, robust standard errors and a corrected test statistic. All estimations moreover control for the respondents' state of residence. See Tables A1, A15 and A21 and Figure A8 for the source survey items and factor scores of the social desirability bias index. *p<0.05, **p<0.01, ***p<0.001. STERNBERG ET AL. - 1091 TABLE A20 Robustness: Decision‐making across compliance domains—structural equation model (SEM) results using single‐item preference measures. Outcome Compliance in the public domain Compliance in the private domain COVID‐19 threat perception 0.290*** 0.361*** (0.026) (0.018) Altruism 0.026 0.006 (0.025) (0.016) Civic responsibility 0.122*** 0.104*** (0.024) (0.016) Pos. reciprocity 0.179*** 0.032 (0.024) (0.017) Neg. reciprocity −0.155*** −0.088*** (0.024) (0.016) Risk aversion 0.094*** 0.081*** (0.026) (0.017) Patience −0.033 −0.010 (0.025) (0.016) Trust in RKI 0.200*** 0.066** (0.032) (0.021) Trust in government 0.016 0.144*** (0.033) (0.021) Female 0.110*** 0.032 (0.025) (0.016) Age group 0.050* 0.128*** (0.027) (0.018) High education 0.003 0.028 (0.030) (0.020) Medium education 0.023 0.017 (0.028) (0.018) Employed essential services 0.020 −0.019 (0.024) (0.016) Household income 0.016 −0.024 (0.029) (0.019) Household size −0.048 −0.014 (0.028) (0.018) Knowledge preventive measures 0.114*** 0.084*** (0.025) (0.016) Social desirability index 0.086*** 0.098*** (0.024) (0.016) 1092 - STERNBERG ET AL. TABLE A20 (Continued) Outcome Compliance in the public domain Compliance in the private domain R 2 0.500 0.414 Observations 2918 2918 Fit statistics (full SEM) Public Private Robust χ 2 (66) =160.35; p<0.001 Robust χ 2 (66) =114.982; p<0.001 Robust RMSEA =0.022 Robust RMSEA =0.016 Robust TLI =0.998 Robust TLI =0.999 Robust CFI =0.956 Robust CFI =0.983 SRMR =0.019 SRMR =0.009 Note: Displayed are standardized coefficient estimates and standard errors in parentheses for the results of the SEMs estimated by means of Equations (1) and (2) (shown are only the structural component results) and fit statistics for each SEM. However, instead of the initially constructed preference measures, we used only a single survey item for each type of preference to assess the robustness of the singular survey item of positive reciprocity. For each preference type, we used the survey item with the highest weight in the experimental validation procedure developed by Falk et al. (2018). See Table A4 for these items, the respective items are marked in italic font. The SEM was estimated separately for each compliance domain, using Diagonal Weighted Least Squares, a polychoric correlation matrix, robust standard errors and a corrected test statistic. All estimations moreover control for the respondents state of residence. See Tables A1, A15 and A21 and Figure A8 for the source survey items and factor scores of the social desirability bias index. Abbreviations: CFI, comparative fit index; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residual; TLI, Tucker‐Lewis index. *p<0.05, **p<0.01, ***p<0.001. TABLE A21 Questionnaire items (translated from original German version). Question (Variable name as in Table A1) Response options Compliance Below you find a list of behavioral patterns. How well does each reflect your behavior in the past 2 weeks? All information you provide in the questionnaire will be treated anonymously and cannot be linked to your identity. I have kept a distance of at least 1.50 m to other people in public, whenever possible. (Keeping distance) 1: Applied never; 2: Applied rarely; 3: Applied sometimes; 4: Applied often; 5: Applied always I have worn a mask in public transport or when shopping. (Wearing masks) 1: Applied never; 2: Applied rarely; 3: Applied sometimes; 4: Applied often; 5: Applied always I have only left the house when absolutely necessary (such as groceries, medical reasons or sports). (Only leaving home when necessary) 1: Applied never; 2: Applied rarely; 3: Applied sometimes; 4: Applied often; 5: Applied always I have deliberately avoided physical contact with people from other households. (Avoiding other households) 1: Applied never; 2: Applied rarely; 3: Applied sometimes; 4: Applied often; 5: Applied always I have restricted private gatherings to only one other person from another household. (Restricting private meetings) 1: Applied never; 2: Applied rarely; 3: Applied sometimes; 4: Applied often; 5: Applied always I have avoided shaking other people's hands when greeting them. (Avoiding handshakes) 1: Applied never; 2: Applied rarely; 3: Applied sometimes; 4: Applied often; 5: Applied always Compliance (alternative measures) When answering the next questions, please remember that all the information you provide in the questionnaire will be treated anonymously by us and cannot be linked to your identity. How would you act in the following situations? If it snows heavily next weekend and some friends of yours organize a sledding trip (there is still room for you on their Scale from 1 (Definitely not) to 7 (Certainly); 99 (I am generally not very enthusiastic about this type of activity.) (Continues) STERNBERG ET AL. - 1093 TABLE A21 (Continued) Question (Variable name as in Table A1) Response options sleds), would you join them? (Willingness to join sledding trip with friends next weekend) If a friend invited you to have dinner with three other friends next weekend and then watch soccer or a movie, would you accept the invitation? (Willingness to join surprise birthday caroling for a friend) Scale from 1 (Definitely not) to 7 (Certainly); 99 (I am generally not very enthusiastic about this type of activity.) Suppose one or more of your best friends has a birthday next week. A group of friends is planning to get together in the morning to serenade the birthday girl or boy at the door or under the window. Will you join them? (Willingness to join a dinner/movie night at friends' house) Scale from 1 (Definitely not) to 7 (Certainly); 99 (I am generally not very enthusiastic about this type of activity.) Economic preferences We now ask for your willingness to act in a certain way in different areas. Please indicate your answer on a scale from 0 to 10, where 0 means you are “completely unwilling to do so” and a 10 means you are “very willing to do so”. How willing are you to give up something that is beneficial for you today in order to benefit more from that in the future? (Willingness to wait) Scale from 0 (completely unwilling to do so) to 10 (very willing to do so) How willing are you to punish someone who treats you unfairly, even if there may be costs for you? (Willingness to punish unfair behavior toward self) Scale from 0 (completely unwilling to do so) to 10 (very willing to do so) How willing are you to punish someone who treats others unfairly, even if there may be costs for you? (Willingness to punish unfair behavior toward others) Scale from 0 (completely unwilling to do so) to 10 (very willing to do so) How willing are you to give to good causes without expecting anything in return? (Willingness to give to good causes) Scale from 0 (completely unwilling to do so) to 10 (very willing to do so) How well do the following statements describe you as a person? Please indicate your answer on a scale from 0 to 10. A 0 means “does not describe me at all” and a 10 means “describes me perfectly”. When someone does me a favor, I am willing to return it. (Willingness to return a favor) Scale from 0 (does not describe me at all) to 10 (describes me perfectly) If I am treated very unjustly, I will take revenge at the very first occasion, even if there is a cost to do so. (Willingness to take revenge) Scale from 0 (does not describe me at all) to 10 (describes me perfectly) In general, how willing or unwilling you are to take risks? Please use a scale from 0 to 10, where 0 means you are “completely unwilling to take risks” and a 10 means you are “very willing to take risks”. (Willingness to take risks) Scale from 0 (completely unwilling to take risks) to 10 (very willing to take risks) By answering the following questions, you have the chance to win a bonus payment of 10 euros. After completing the survey, 5% of the participants (i.e., 1 out of 20) will be randomly selected to receive this bonus payment of 10 euros. You have the option to donate any portion of the bonus payment to a charitable organization, while you would receive the remaining amount from Respondi. If you decide to make a donation, we will do so after the survey is 1094 - STERNBERG ET AL. TABLE A26 Robustness check Google mobility patterns: Retail and recreation and transit. Outcome: Mobility changes in Retail and recreation areas Transit areas (1) (2) (3) (4) (5) (6) (7) (8) Private compliance −16.058** −14.030* −10.410 −31.044** −24.835 −21.600 (4.772) (6.348) (5.331) (9.540) (12.523) (12.965) Public compliance −6.807* −1.797 −2.769 −14.370* −5.502 −7.957 (3.118) (3.571) (3.095) (5.986) (7.044) (7.527) Population density (inhabitants/km) −0.003 −0.004 (0.001) (0.003) Stadtstaat 3.193 8.649 (3.817) (9.283) Constant −46.774*** −45.191*** −44.924*** −44.375*** −25.619** −20.430 −19.958 −16.525 (3.165) (5.545) (4.908) (4.45349) (6.328) (10.644) (9.681) (10.832) Adj. R 2 0.408 0.201 0.374 0.584 0.584 0.241 0.373 0.366 Observations 16 16 16 16 16 16 16 16 Note: Shown are the results of regressing (i) state‐level phone‐tracking‐based changes in mobility according to Google's mobility reports in the areas of Retail and Recreation (columns 2–5) and Transit (columns 6–9) and on (ii) state‐level averages of our measures of compliance in the public and private domain, as predicted from the SEM by means of Equations (1) and (2) (controlling for a state's population density and whether the state is a city state, i.e., Hamburg, Berlin and Bremen). Mobility data retrieved from Google LLC (2021). Changes in mobility refer to the 2020 baseline period before the start of the pandemic. *p<0.05, **p<0.01, ***p<0.001. TABLE A27 Robustness check Google mobility patterns: Workplace and residential. Outcome: Mobility changes in Workplace areas Residential areas (1) (2) (3) (4) (5) (6) (7) (8) Private compliance −14.731* −15.859 −9.420** 5.756** 7.613** 6.109*** (5.726) (7.675) (1.723) (2.167) (1.374) Public compliance −4.664 0.999 0.083 1.073 −1.645 −1.134 (3.705) (4.318) (1.570) (1.267) (1.219) (0.798) Population density (inhabitants/km) −0.004*** 0.001** (0.001) (0.000) Stadtstaat 2.711 −1.721 (1.936) (0.984) Constant −18.931* −20.261** −19.959** −20.490*** 8.366*** 10.204*** 10.059*** 9.631*** (3.798) (6.588) (5.934) (2.259) (1.143) (2.253) (1.675) (1.148) Adj. R 2 0.273 0.038 0.220 0.909 0.404 −0.019 0.437 0.787 Observations 16 16 16 16 16 16 16 16 Note: Shown are the results of regressing (i) state‐level phone‐tracking‐based changes in mobility according to Google's mobility reports in the areas of Workplace (columns 2–5) and Residential (columns 6–9) and on (ii) state‐level averages of our measures of compliance in the public and private domain, as predicted from the SEM by means of Equations (1) and (2) (controlling for a state's population density and whether the state is a city state, i.e., Hamburg, Berlin and Bremen). Mobility data retrieved from Google LLC (2021). Changes in mobility refer to the 2020 baseline period before the start of the pandemic. *p<0.05, **p<0.01, ***p<0.001. STERNBERG ET AL. - 1101 TABLE A29 Extension institutional trust: Understanding effects of trust in government. Outcome Public compliance Private compliance (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Trust in government 0.012 −0.007 0.030 0.138*** 0.145*** 0.118*** (0.033) (0.021) (0.019) (0.021) (0.013) (0.012) Trust in state govt. 0.031 0.034 0.070** −0.010 (0.017) (0.022) (0.011) (0.013) Trust in estab. media −0.017 −0.026 0.081*** 0.045* (0.017) (0.019) (0.010) (0.011) Trust in RKI 0.204*** 0.190*** 0.194*** 0.219*** 0.207*** 0.068** 0.112*** 0.124*** 0.070** 0.062* (0.032) (0.017) (0.018) (0.015) (0.017) (0.021) (0.011) (0.010) (0.012) (0.011) Other main behavioral predictors Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Demogr. and socioecon. controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Compliance‐specific controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes R 2 0.492 0.490 0.491 0.492 0.494 0.411 0.406 0.413 0.407 0.413 Observations 2918 2917 2917 2918 2918 2918 2917 2917 2918 2918 Note: Displayed are standardized coefficient estimates and standard errors in parentheses for the institutional trust variables of the SEM, as estimated by means of Equations (1) and (2) (shown are only the structural component results). Columns 1 and 6 therefore merely repeat the results from Table 2. The SEM was estimated separately for each compliance domain, using Diagonal Weighted Least Squares, a polychoric correlation matrix, robust standard errors and a corrected test statistic. Other main behavioral predictors contain COVID‐19 threat perception, pos. and neg. reciprocity, altruism, civic responsibility, risk aversion and patience. Demographic controls contain respondents' gender, age group and state. Socioeconomic controls include education, employment in essential services, household size and income. Compliance‐specific controls contain knowledge about COVID‐19 preventive measures and the degree of social desirability bias (factor scores as in Figure A8 and Table A15). *p<0.05, **p<0.01, ***p<0.001. TABLE A28 Extension institutional trust: Proxies for trust in Robert Koch Institute (RKI). Outcome Public compliance Private compliance (1) (2) (3) (4) (5) (6) Trust in RKI 0.204*** 0.068** (0.032) (0.021) Trust in science 0.182*** 0.053* (0.018) 0.012 Trust in WHO 0.165*** 0.055* (0.018) (0.011) Trust in government 0.012 0.050 0.044 0.138*** 0.153*** 0.147*** (0.033) (0.016) (0.017) (0.021) (0.010) (0.011) Other main behavioral predictors Yes Yes Yes Yes Yes Yes Demogr. and socioecon. controls Yes Yes Yes Yes Yes Yes Compliance‐specific controls Yes Yes Yes Yes Yes Yes R 2 0.492 0.493 0.490 0.411 0.411 0.410 Observations 2918 2918 2918 2918 2918 2918 Note: Displayed are standardized coefficient estimates and standard errors in parentheses for the institutional trust variables of the SEM, as estimated by means of Equations (1) and (2) (shown are only the structural component results). Columns 1 and 4 therefore merely repeat the results from Table 2. The SEM was estimated separately for each compliance domain, using Diagonal Weighted Least Squares, a polychoric correlation matrix, robust standard errors and a corrected test statistic. Other main behavioral predictors contain COVID19 threat perception, pos. and neg. reciprocity, altruism, civic responsibility, risk aversion and patience. Demographic controls contain respondents' gender, age group and state. Socioeconomic controls include education, employment in essential services, household size and income. Compliance‐specific controls contain knowledge about COVID‐19 preventive measures and the degree of social desirability bias (factor scores as in Figure A8 and Table A15). *p<0.05, **p<0.01, ***p<0.001. 1102 - STERNBERG ET AL. TABLE A30 Fines for non‐compliance across German states. State Behavior Date of regulations Source Not wearing a face‐mask Not keeping distance of 1.50 m Meeting with more than allowed number of people Hosting an illegal gathering Baden‐ Württemberg 50–250 Euro n.a. n.a. 100–1000 Euro September 15, 2021 https://www.bussgeldkatalog.org/corona‐ baden‐wuerttemberg/ Bayern 250 Euro 150 Euro 150 Euro 5000 Euro November 20, 2020 https://www.bussgeldkatalog.org/corona‐ bayern/ Berlin 50–500 Euro 100–500 Euro 25–500 Euro n.a. November 20, 2020 https://www.bussgeldkatalog.org/corona‐ berlin/ Brandenburg 50–250 Euro n.a. 50–250 Euro 1000–5000 Euro August 25, 2020 https://www.bussgeldkatalog.org/corona‐ brandenburg/ Bremen n.a. n.a. 50–250 Euro 250–2500 Euro November 23, 2020 https://www.bussgeldkatalog.org/corona‐ bremen/ Hamburg 150 Euro 150 Euro 150–500 Euro 1000 Euro November 23, 2020 https://www.bussgeldkatalog.org/corona‐ hamburg/ Hessen 200 Euro n.a. 200 Euro 500–1000 Euro April 20, 2020 https://www.bussgeldkatalog.org/corona‐ hessen/ Mecklenburg‐ Vorpommern 50–150 Euro n.a. 50–500 Euro n.a. November 23, 2020 https://www.bussgeldkatalog.org/corona‐ mecklenburg‐vorpommern/ Niedersachsen 100–150 Euro 100–400 Euro 150–400 Euro 300–3000 Euro November 23, 2020 https://www.bussgeldkatalog.org/corona‐ niedersachsen/ Nordrhein‐ Westfalen 50–150 Euro n.a. 250 Euro 1000–5000 Euro September 6, 2020 https://www.bussgeldkatalog.org/ corona‐nrw/ Rheinland‐Pfalz 50 Euro 50 Euro 100 Euro 500 Euro November 23, 2020 https://www.bussgeldkatalog.org/corona‐ rheinland‐pfalz/ (Continues) STERNBERG ET AL. - 1103 TABLE A30 (Continued) State Behavior Date of regulations Source Not wearing a face‐mask Not keeping distance of 1.50 m Meeting with more than allowed number of people Hosting an illegal gathering Saarland 50–100 Euro n.a. 200 Euro 1000–4000 Euro November 23, 2020 https://www.bussgeldkatalog.org/corona‐ saarland/ Sachsen 60 Euro 150 Euro 150 Euro 5000 Euro November 23, 2020 https://www.bussgeldkatalog.org/corona‐ sachsen/ Sachsen‐Anhalt 50–75 Euro n.a. 50 Euro 1000 Euro September 14, 2021 https://www.bussgeldkatalog.org/corona‐ sachsen‐anhalt/ Schleswig‐ Holstein 150 Euro 150 Euro 150 Euro 1000–2000 Euro November 23, 2020 https://www.bussgeldkatalog.org/corona‐ schleswig‐holstein/ Thüringen 60 Euro 100 Euro n.a. 1000–3000 Euro November 20, 2020 https://www.bussgeldkatalog.org/corona‐ thueringen/ Note: The table shows fines for non‐compliance with COVID‐19 regulations across federal states in Germany. Note that the time at which regulations came into place differ across states, for example, in Baden‐Württemberg and Sachsen‐Anhalt, we were only able to obtain information about the regulations at the later stages in the pandemic, such that they did not anymore include fines for most rules that were relevant during our survey data collection. “n.a.” indicates that we were not able to obtain information about the amount of the specific fine in this state, given the dynamically changing conditions during the pandemic and difficulties to find information about regulations that are no longer in place. Coherently, an “n.a.” entry does not mean that there was no fine for the respective behavior. 1104 - STERNBERG ET AL. FIGURE A2 COVID‐19 threat perception: screeplot factor analysis. This figure shows a screeplot of an exploratory factor analysis of the 8 survey items aimed at capturing COVID‐19 threat perception (see Tables A1, A6 and A21 for the items and procedure), with Eigenvalues of factors on the y‐axis and the identified factors on the x‐axis. The factor analysis was performed using iterated principal axis factoring, no rotation, and a mixed correlation matrix (continuous and polychoric). FIGURE A1 Distribution of self‐reported compliance items. This figure shows the scaled density distribution of each of the six survey items aimed to quantify compliance behavior. The bold line indicates the mean and the non‐bold line the median in each case. Participants were asked the following question: “Below you find a list of behavioral patterns. How well does each reflect your behavior in the past 2 weeks?” Response options were 1 =“Applied never”; 2 =“Applied rarely”; 3 =“Applied sometimes”; 4 =“Applied often”; 5 =“Applied always”. STERNBERG ET AL. - 1105 FIGURE A4 Civic responsibility: screeplot factor analysis. This figure shows a screeplot of an exploratory factor analysis of the 3 survey items aimed at capturing civic responsibility (see Tables A1, A6 and A21 for the items and procedure), with Eigenvalues of factors on the y‐axis and the identified factors on the x‐axis. The factor analysis was performed using iterated principal axis factoring, no rotation, and a polychoric correlation matrix. FIGURE A3 Distribution of COVID‐19 threat perception factor scores. This figure shows a scaled density and histogram of the factor scores capturing COVID‐19 threat perception, as in Table A7. 1106 - STERNBERG ET AL. FIGURE A5 Tree for the staircase time task (extracted from Falk et al. (2022), p. 66). This figure shows the intertemporal choice sequence of the staircase method in detail, as developed by Falk et al. (2022), and employed in this paper as part of the preference measure for patience. The numbers indicate the payment in 12 months, “A” denotes the respondent's choice of “100 euros today,” while “B” denotes the respondent's choice of “x euros in 12 months”. All respondents start with the left‐most decision, the remaining intertemporal choice sequence is revealed by going from left to right through the tree. STERNBERG ET AL. - 1107 FIGURE A6 Compliance: screeplot exploratory unrotated factor analysis. This figure shows a screeplot of an exploratory factor analysis of the 6 survey items aimed at capturing compliance behavior (see Tables A1, A3 and A21 for the items), with Eigenvalues of factors on the y‐axis and the identified factors on the x‐axis. The factor analysis was performed using iterated principal axis factoring, no rotation, and a polychoric correlation matrix. 1108 - STERNBERG ET AL. FIGURE A7 Compliance: summated rating scale monotone homogeneity assumption. This figure illustrates an assessment of the monotone homogeneity assumption of the two summated rating scales for compliance in the public (first row of graphs) and private (second row of graphs) domain, as in Table A11. Plotted are rest scores on the x‐axis (i.e., the public/private scale score if dropping one of the items) and the value of the respective dropped item on the y‐axis. The monotone homogeneity assumption thereby requires expected responses to each of the three items of compliance in the public (private) domain to be increasing/decreasing as the true dimension of compliance in the public (private) domain also increases/decreases (i.e., the rest scores). This increase/decrease has to be of a monotone nature, which here seems to be the case (all lines are monotonously increasing). The red line shows a linear fit, the blue line a smoothed fit. [Colour figure can be viewed at wileyonlinelibrary.com] STERNBERG ET AL. - 1109 FIGURE A8 Social desirability: screeplot factor analysis. This figure shows a screeplot of an exploratory factor analysis of the 3 survey items aimed at capturing social desirability bias (exaggerating positive characteristics) (see Tables A1, A6 and A21 for the items and procedure), with Eigenvalues of factors on the y‐axis and the identified factors on the x‐axis. The factor analysis was performed using iterated principal axis factoring, no rotation, and a polychoric correlation matrix. FIGURE A9 Alternative compliance measure: screeplot factor analysis. This figure shows a screeplot of an exploratory factor analysis of 3 alternative survey items aimed at capturing compliance behavior (see Tables A1 and A21 for the items), with Eigenvalues of factors on the y‐axis and the identified factors on the x‐axis. The factor analysis was performed using iterated principal axis factoring, no rotation, and a polychoric correlation matrix. 1110 - STERNBERG ET AL. higher due to the higher density and observability of the behavior of others, this line of argument may be even more relevant for the private domain given the contrast between crowded living situations in the city and the much more spaced, rural housing. On the other hand, one could argue that residents in smaller towns know each other (and especially their neighbors) substantially better and are less anonymous than those living in a larger city. In that case, our expectations would suggest the opposite, that is, (private) compliance being higher in the rural context. Weexaminethepotential importance of respondents' residenceinruralversus urban settings as an additionalpredictor for compliance in the public and private domain by using a survey‐based measure of the size of the community a respondent lives in. Specifically, as part of our survey, we asked respondents how many inhabitants live in their community/municipality, with response options being “below 5000,” “5001–20,000,” “20,001–100,000,” “100,001–500,000,” “above 500,000.” Based on this question, we constructed an indicator variable that takes on the value of 1 if a respondent reported living in a municipality with less than or equal to 20,000 inhabitants and that takes on the value of 0 if (s)he reported living in a municipality with more than 20,000 inhabitants. We chose this definition of the variable because, in Germany, municipalities with less than 20,000 inhabitants are counted as “Kleinstädte,” that is, small towns/villages. Table A22 (Columns 3 and 6) presents the results of including this measure of rural/urban residence setting as an additional predictor in the full model of compliance in the public and private domain, respectively. We find a negative, but no statistically significant association between a respondent living in a rural as opposed to a more urban setting and their degree of compliance in the public/private domain. However, in terms of magnitude, the effect size of the coefficient is much larger for compliance in the private than in the public domain (though this difference is also not statistically significant according to a Wald test testing for the equality of coefficients). Thus, the direction and pattern of this finding is generally in line with the first theoretical expectation outlined above, even though they lack statistical significance. Finally and importantly, the main results in terms of our previously examined core predictors (economic preferences, institutional trust, threat perception) remain stable when including respondents' urban/rural residence setting as an additional predictor. Correlation between regional CV‐19 threat perception and case incidence rate We examine the relation between our measure of respondents' COVID‐19 threat perception and the incidence rate of COVID‐19 cases, using data from the Federal Statistical Office of Germany (“Statistisches Bundesamt”) about the number of registered COVID‐19 cases in the past 7 days per 100 people. 17 To do so, we employ federal state level averages of our threat perception measure and the state level averaged COVID‐19 case incidence (i) across the time period of the data collection (February 2021) and (ii) across all previous infection waves (i.e., from the beginning of the pandemic until the start of the data collection). We chose to examine the case incidence level during both of these time periods because it seems likely that, if individuals' threat perceptions are indeed related to the actual number of cases, both, the trajectory and intensity of previous infection waves as well as the severity of the infection wave in the midst of our data collection, play a role (in potentially opposing ways). In addition to reporting Pearson correlation coefficients, we also show the results of regressing state level threat perception on state level case incidence rates while controlling for possible confounders at the state level (population density, share of the German population living in this state, the state being a Stadtstaat, i.e., one of the three German cities that are also a federal state). The results of this exercise are shown in Table A23 in Appendix A. We find that neither the average COVID‐19 case incidence during the data collection nor the case incidence during the prior infection waves are significantly related to respondents' COVID‐19 threat perception at the state level. However, the findings do reveal a pattern with respect to the direction of the correlations: While there is a small, positive correlation between respondents' threat perception and the case incidence rate during our data collection period, its correlation with the case incidence rate during prior infection waves is slightly negative. We, first, interpret the former result as providing some form of validation for our threat perception measure. Second, in combination, these findings interestingly also suggest that individuals' threat perception may be, on the one hand, highly dependent on the immediate pandemic situation, and, on the other hand, particularly sensitive to retrospective adjustments. Specifically, we observe that respondents who experienced higher case numbers in the first waves of the pandemic actually reported a slightly lower fear of COVID‐19 during our survey data collection, while those that experienced higher case incidence levels during the data collection also revealed a slightly higher perceived threat. To that end, these findings moreover confirm the conceptualization of our measure of threat perception as a not necessarily rationally or objectively informed concept, but rather as a subjective account of how threatening respondents perceived the COVID‐19 pandemic to be in general and how threatening they perceived it to be with regard to specific aspects of their lives (health, financial, social; see also the factor scores we used to construct the threat perception measure). This STERNBERG ET AL. - 1117 conceptualization of threat perception may also contribute to explaining why it appears to play such a crucial role for individuals' compliance behaviors during the pandemic, as identified in the main analysis. Finally, we remain, for the following reasons, cautious not to overstate or over‐interpret the outlined correlations between our measure of threat perception and COVID‐19 case incidence rates—neither in terms of the absence of statistical significance nor in terms of the mere directions of the correlations. First, the analysis was conducted at the state‐level and is, thus, based on N=16. It is therefore statistically underpowered. Second, during the pandemic there was a lot of variation within federal states themselves and it is likely that living in a large city within a certain state as opposed to living in a rural area within a certain state plays a crucial role for the correlation between case numbers and threat perception. A more thorough analysis would therefore be to go even lower than state‐level. Unfortunately, we did not ask respondents about the specific district they live in and, thus, could not conduct such an analysis. Third, and relatedly, patterns at individual level and the aggregate level can differ. Additional analyses for the institutional trust variables To better understand the differential observed dynamics for trust in the RKI and trust in the government across the two domains, we conducted a number of supplementary analyses that exploit additional information collected in our survey. In terms of trust in the RKI, we had, in our theory section, argued that we would, if any, expect a larger impact on compliance in the public domain given the more technical stipulations in this realm, for example, wearing a mask or keeping a 1.50 m distance from another as opposed to the general recommendation of “staying at home” in the private domain. Recall that we had chosen trust in the RKI as a measure of trust in scientific institutions in the German case, because in Germany the RKI was the dominant scientific institution that was heavily involved in policy consultation and communication with regards to the COVID‐19 pandemic. In line with our theoretical expectations, our core results suggest that trust in the RKI, indeed, plays a stronger role for public compliance than for private compliance, though its estimated effects are statistically significant in both realms. Thus, to assess whether the above explanation is plausible, we draw on two alternative proxies for trust in scientific institutions, namely (i) trust in science (in general) and (ii) trust in the WHO and examine whether the observed pattern holds. Table A28 in Appendix A reports the results of this exercise. Estimating the SEM with trust in science and trust in the WHO, respectively, instead of trust in the RKI as a predictor of public and private compliance (all other variables remaining as in the main specification) reveals the same pattern, which we also observed for trust in the RKI: Both proxies reveal statistically significant correlations with compliance in the two domains, but much larger estimated effects in the public domain. The magnitudes and levels of statistical significance are strikingly similar to those of trust in the RKI (trust in science: β pub =0.182, p‐value <0.001; β priv =0.053, p‐value <0.05; trust in the WHO: β pub =0.165, p‐value <0.001; β priv =0.055, p‐value <0.05). Moreover, including trust in science and trust in the WHO did not change the pattern identified for trust in the government (or any of the other core predictors) substantially. In terms of trust in the national government, we had, in our theory section hypothesized conflicting logics: On the one hand, the absence of formal monitoring by state authorities in the private realm might make trust as an intrinsic motivator more important. On the other hand, the private realm may be understood as a realm in which the government has inherently no legitimate role to play. Our results revealed that trust in the government is only significantly correlated with compliance in the private domain, not with compliance in the public domain. To examine this finding further and understand how it relates to our initial expectations, we utilize two additional institutional trust variables, which aim at quantifying trust in institutions or actors that are related to the German government, but in different ways: First, we repeat the main SEM specification, but replace trust in the government with trust in the state‐level government. We use this variable since they allow us to investigate whether the observed pattern is a general pattern for government‐related institutions or whether it is a specific pattern of citizens' trust in the national government in particular. The results of this exercise are reported in Table A29 (Columns 2, 3, 7 and 8) in Appendix A. For trust in the state‐level government, we initially find a similar pattern as for trust in the (national) government, that is, there is only a significant correlation with compliance in the private, but not in the public domain. However, the estimated effects are smaller and also slightly less statistically meaningful (β pub =0.031, p‐value =0.267; β priv =0.070, p‐ value <0.01). Moreover, when adding both, trust in the state‐level government and trust in the national government, simultaneously, the statistical significance of the estimated effects for trust in the state‐level government disappears, while those for trust in the national government remains (see Columns 3 and 8). These findings suggest that the pattern observed in our main analysis may be specific to and driven by trust in the national government (“Bundesregierung”). Given these findings, we propose an alternative explanation for the observed dynamics, which we aim to investigate by repeating the above exercise but instead including trust in Germany's established media channels as a predictor in 1118 - STERNBERG ET AL. the model. Specifically, we argue that the high relevance of trust in the national government for citizens' compliance in the private realm could be a result of the specific policy communication and messaging campaigns issued during the pandemic in Germany. For instance, the narratives of the regular public addresses by Chancellor Angela Merkel to the German population were much more focused on the general aspect of staying at home (i.e., along the lines of being united in isolation), which is closely related to our definition of private compliance, and less so on the more technical rules of mask‐wearing or avoiding handshakes (i.e., our definition of public compliance) (see e.g., Angela Merkel, televised speech, 2020). Thus, it may be that, as a result of such communication strategies throughout the entire pandemic, the view of the private realm as a realm in which the government has no prominent role to play was somewhat loosened, especially for individuals with a high level of trust in the national government. We try to assess the plausibility of this alternative interpretation by drawing on another variable, namely trust in established media channels, which are widely broadcasted and supported the mentioned public addresses by Angela Merkel. Thus, if the observed patterns are a result of policy communication during the pandemic, we should observe a very similar pattern for citizens' trust in the said government‐related media channels—which is what we indeed observe: The estimated coefficients of citizens' trust in established media channels show the same pattern as those for trust in the national government, that is, they are statistically significant for compliance the private realm, but not for compliance in the public realm (β pub = −0.017, p‐value =0.499; β priv =0.081, p‐value <0.001). Moreover, while the coefficients are smaller in magnitude, they are, unlike the coefficients of trust in the state‐level government, robust to a simultaneous estimation together with trust in the national government (see Columns 4, 5, 9 and 10 of Table A29). We therefore interpret these findings as support for the alternative explanation outlined above. Finally, it is noteworthy that including the mentioned additional variables did not change the patterns identified for trust in the RKI (or any of the other core predictors) substantially. STERNBERG ET AL. - 1119