Introduction

Adoption of preventive behaviors and compliance with COVID-19-related protective recommendations is particularly important for the collective management of the pandemic and for the individual reduction of the risk of having a severe form of COVID-19, especially for high-risk individuals. Older individuals represent an important part of this population since the risk of dying from COVID-19 increases with age [27]. Engaging in protective behavior can be associated with the provision of a public good: public health. It was particularly important when no vaccine was available, as was the case during the first year of the pandemic. It is therefore important to understand the individual decision to engage in protective behavior or comply with government recommendations (such as those of the World Health Organization) and its determinants to design relevant public health policies.

This paper explores three determinants of compliance with COVID-19 recommendations and other related preventive behaviors among seniors aged 50 or more in France. Compliance with preventive measures is captured by dummy variables indicating whether, since the pandemic outbreak, the individual (i) has stopped meeting family members living outside the household, (ii) has participated in gatherings with more than 5 other individuals, (iii) has worn a mask when outside, (iv) has maintained distance from others when outside, (v) has engaged in handwashing or using hand sanitizer more frequently, and (vi) has paid particular attention to covering their cough. We use these variables because they are related to recommendations by the French government, such as those on washing one’s hands regularly and reducing social contact. We particularly study how these protective behaviors or compliance with public health recommendations correlate with risk aversion, farsightedness and the perceived trustworthiness of others.

We conceptualize the decision to adopt preventive behaviors as a decision under risk that is taken when having social interactions. In the COVID-19 context, individuals also do not know for how long they will have to adopt these preventive behaviors. The likelihood of being infected increases when not adopting preventive behaviors. Being infected can have adverse consequences on the agent’s own health in the short term and potentially in the mid- to long term, and it can potentially be transmitted to others. Given the risky nature of the decision to not adopt preventive behaviors, we can expect that risk aversion influence the decision to adopt preventive behaviors. In addition, given the potential consequences and the uncertainty about the duration of the pandemic, we can expect that far-sightedness determines the decision to adopt preventive behaviors. Farsighted individuals being more likely to perceive the midterm benefits of preventive behavior. In addition, if we consider the adoption of preventive behaviors as a health behavior, the literature has found that risk aversion and farsightedness are correlated to health behaviors [22]. Finally, the virus is transmitted during different type of social interactions. Hence, individuals could be more willing to accept the risk involved in the decision to engage in social interactions, even if social interactions can increase the chance of being infected, if they perceive a higher trustworthiness of others. We therefore expect risk aversion and farsightedness positively correlate with adoption of protective behaviors, while the perceived the trustworthiness of others is negatively correlated. We also expect heterogeneous correlations across preventive behaviors. This is further explained in Theoretical framework and hypotheses.

We use individual-level panel data from the Survey of Health, Ageing and Retirement in Europe (SHARE). It is a multidisciplinary database of longitudinal microdata on health, socioeconomic status, and intergenerational transfer from individuals aged 50 or over in Europe. Particularly, we use the 8th wave, conducted just before the pandemic (October 2019 to March 2020), and one wave conducted during the pandemic (June/July 2020). All participants the 8th wave were invited to respond to the SHARE Corona survey such that it is a panel dataset. Therefore, we observe individuals before and during the pandemic. Preferences have been collected through a drop-off questionnaire in France only during the 8th wave, so we focus on the French Sample of the survey. The adoption of the preventive behaviors is observed in the wave collected by phone during the pandemic. Therefore, in our empirical, we estimate the correlation between preferences and beliefs measured before the pandemic on the adoption of preventive behaviors during the pandemic.

The literature has found associations between preferences and COVID-19 relate preventive behaviors. First, Müller and Rau [25] and Sheth and Wright [32], who use a sample of German and Californian students, respectively. Müller and Rau [25] find that patience increases staying at home and avoiding crowds, risk tolerance decreases avoiding crowds and panic buying at the very beginning of the COVID-19 outbreak.Footnote 1 Present bias also increases panic buying. In contrast, Sheth and Wright [32] find that risk tolerance is not associated with socialization under the stay-at-home order in California. [9] study whether regions with a higher average willingness to take risks were more likely to comply with mobility restrictions. They find that regions with a higher risk tolerance were less likely to reduce their mobility. In addition, using a survey conducted during the first lockdown in France, Guillon and Kergall [17] have found that risk aversion is positively associated with the probability of wanting the first lockdown to be extended and increases the number of trips and outside physical activities individuals. This might be explained by a higher degree of risk aversion and translate into a higher perceived benefit of engaging in protective behavior.Footnote 2 In the French context, Blayac et al. [5] finds that risk aversion in the health domain is correlated with compliance. In addition, Rafaï et al. [30] studies compliance behaviors, such as washing hands, mask wearing, respect with the lockdown, not touching face, and social distancing. They find that washing hands, mask wearing, and respect with the lockdown are correlated with risk aversion in the health domain, while risk aversion in general is correlated to social distancing, mask wearing, respecting lockdown. They also find that patience is predicting compliance. To our knowledge, there is no literature of literature on the association between the perceived trustworthiness of others and the adoption of preventive behaviors during the pandemic.Footnote 3

Our contributions can be summarized as follows. First, we contribute to the literature on the association between preferences and the adoption of preventive behaviors. Our second and main contribution is to explore the association the perceived trustworthiness of others and the adoption of preventive behaviors during the pandemic. We also focus on seniors, for whom the risk of complications or death due to the coronavirus is higher.

Our results suggest that farsightedness and risk aversion are strong predictors of individuals’ protective behavior. More farsighted individuals are more likely to not visit their family members anymore, wear a mask and keep their distance from others when outside, wash their hands more regularly. Risk aversion increases the probabilities of not meeting more than 5 other people and not meeting with family members anymore. We also find that risk aversion reduces the probability of covering one’s cough or sneeze. It is particularly interesting that cough covering is not correlated to preferences (or negatively) given that this preventive behavior is only meant to protect others from being infected. This might confirm some heterogeneity with respect to who is protected from the adoption of the considered protective behavior. Concerning the perceived trust in others, we find that a higher it reduces compliance with the recommendations about meeting with 5 or more people and family gatherings. We interpret this result as a sign that individuals with a higher trust in others perceive a lower risk of being infected by friends and family members. This result highlights the perceived trustworthiness increase the willingness to take a risk when the individual one interacts with are part of the close social network (family members and friends).

The implications of our results are that government policies and public health campaigns should consider the heterogeneity of preferences and beliefs and more carefully target individuals who underestimate the spread of the virus and the risk of infection at gatherings due to their trust in others. Providing specific information on the risks associated with COVID-19, on the health and financial risks associated with the lockdown and shortage of health healthcare supply, and on their long-term effects should also help to better target risk-seekers and more present-oriented individuals. Finally, it is important to inform about the risk of infection during social interactions, especially with close relatives that people tend to trust the most.

Theoretical framework and hypotheses

We assume that an agent decides to adopt a protective behavior or not, in a risky environment where the risk is being infected by the COVID-19. Being infected can have adverse consequences on the agent’s own health in the short term and potentially in the mid- to long term, and it can potentially be transmitted to others. Because of the potential transmission, we can consider that agents contribute to public health as a public good when they decide to adopt protective behaviors because they can reduce transmissions [11].

The protective behaviors we consider discrete behaviors capturing whether the agent has: (i) stopped meeting family members living outside the household, (ii) participated in gatherings with more than 5 other individuals, (iii) worn a mask when outside, (iv) maintained distance from others when outside, (v) engaged in handwashing or using hand sanitizer more frequently, and (vi) paid particular attention to covering their cough. The adoption of each of these protective behaviors has different consequences for the agent. Indeed, as argued by Diekmann [11], some protective behaviors have an impact on the agent’s likelihood to be infected as well others (reduction of social contacts with family members of friends, maintaining distance and hands hygiene), while others are adopted to prevent the transmission to others only (covering cough). For mask wearing, it is ambiguous whether it perceived as a way to protect itself at the time of the survey because it was argued that being infected is possible even when wearing a mask. On the contrary, mask wearing was perceived as a good way to reduce the chance of transmission to others when an individual is affected.

Given that not adopting protective behaviors is a risky choice, we might expect that risk aversion affect the decision adopt them. Risk-averse individuals may benefit more from COVID-19 prevention than risk-seekers, especially because of the risk of dying (or at least having severe health consequences) and the unknown consequences of infection. In addition, farsightedness and preferences toward the future can affect the decision to adopt protective behaviors because of the potential future detrimental. Indeed, individuals with a higher time horizon and farsighted give more weight to their future utility (hence health) and might be more likely to adopt protective behaviors. In addition, farsighted individuals are more likely to consider the future consequences of infections on their health and the spread of the virus.

The adoption of protective behaviors can be perceived as an action involving social interactions with others, especially the avoiding gatherings with friends and family members or keeping its distance with others. We also know that the transmission of the virus often results from interactions with other individuals. When social interactions are involved, the perceived trustworthiness of others can play a role on individual decisions [14]. Indeed, the perceived trustworthiness the agent has in others can increase their willingness to engage in activities that involve interactions with other agents, even if there is a risk associated to that activity.Footnote 4 When individuals put a higher level of trust in others, they can perceive a lower risk of being infected by the individuals they interact with. Hence, the agent could be more willing to accept the risk involved in the decision to engage in social interactions, even if social interactions can increase the chance of being infected. We can therefore expect that individuals with a higher level of perceived others’ trustworthiness might be less likely to adopt protective behaviors. In addition, we might expect that the effect trustworthiness of others varies according to the individuals involved in the social interactions. Family members and friends are individuals with whom the agent interacts repeatedly, hence the level of trustworthiness for this group can be higher than for strangers (involved in the other type of protective behaviors). At least, the decision to accept the risk of not adopting a protective behavior should be higher with the former group of individuals.

Finally, according to Diekmann [11], the willingness to adopt a protective behavior depends on whether its purpose is to protect the agent or to protect others (potentially shaped by how much adopting the behavior can generate discomfort). Therefore, we will test if there some heterogeneity in the predictive power of our main variable of interest by protective behavior. To summarize, we test the following hypotheses.

H1

Likelihood of adopting protective behaviors increases with the degree of risk aversion.

H2

Likelihood of adopting protective behaviors increases with the degree of farsightedness.

H3a

Likelihood of adopting protective behaviors decreases with the perceived trustworthiness of others.

H3b

The correlation of trustworthiness is stronger for interactions that involve family members and friends.

H4

The correlation between our main variable of interest and the adoption of given protective behavior may vary according to whether the protective behavior’s purpose is protects the agent.

The objective of this paper is to test these hypotheses empirically with the data we present in the next section.

Data and methodology

We use panel data from the Survey of Health, Ageing and Retirement in Europe (SHARE). Particularly, we use the 8th wave of the regular SHARE survey, conducted just before the pandemic, and one wave, called SHARE Corona Survey, conducted during the pandemic. Note that the name of the survey has been changed to SHARE Corona Survey for the wave conducted during the pandemic because the questionnaire was different and aimed to understand the consequences of the pandemic on seniors’ life. Nonetheless, all the participants of the 8th wave were invited to respond to the SHARE Corona survey such that it is a panel dataset. Therefore, we observe the same individuals before and during the pandemic. In addition, we focus on the French sample because the preferences measures we use have been measured during the 8th wave in France only through a drop-off questionnaire. We describe the periods of data collection for each wave we use, as well as the source of each variable in Table A.1 in the Appendix.

Importantly, the wave conducted during the pandemic, the SHARE Corona survey, was conducted in June and July 2020 by phone. The survey provides all the information we need to measure the preventive behavior of individuals since the coronavirus outbreak. The main 8th wave was conducted from October 2019 to March 2020 with face-to-face interviews. This wave provides socioeconomic and demographic information on surveyed individuals just before the corona crisis. The drop-off questionnaire is a country-specific paper questionnaire given to participants by interviewers at the end of the face-to-face interview conducted for the 8th wave. Surveyed individuals can then complete this questionnaire directly and give it to the interviewer or send it by mail using a prepaid envelope. This last questionnaire provides information on risk and time preferences just before the coronavirus outbreak. Finally, among the 2400 French respondents of wave 8, 91% (2184) responded to the drop-off questionnaire. All the wave 8 respondents were asked to answer to the SHARE Corona Survey. 75% of them (1638) answered to this latter survey. Once we drop observations with missing values, we have a dataset with 1271 individuals observed twice: just before and during the pandemic.

Outcomes

Our outcome variables, protective behavior against COVID-19 or compliance with the recommendations, are derived from the COVID-19 questionnaire.

Concerning recommendations with respect to social distancing, we use two dummy variables indicating whether individuals no longer engage in the following different activities since the outbreak: (i) meeting with more than 5 people from outside their household and (ii) visiting other family members. To construct these two variables, we proceed as follows. Individuals are asked whether they have ever left their home since the COVID-19 outbreak. If they respond yes, then they are asked, “Since the outbreak of Corona, how often have you done the following activities, as compared to before the outbreak? Not anymore, less often, about the same, or more often?” Then, individuals could respond for each of the two activities we consider. We construct a variable for each of these two activities that is equal to one if the individual responded that they had never left their home since the beginning of the outbreak or do not do this given activity anymore, and 0 otherwise. We should note that even though not meeting with more than 5 people from outside one’s household was a clear recommendation of the government, whatever the period, not visiting other family members was recommended only during the first lockdown (17th of March–11th of May).

Other recommendations are made for individuals when they go outside to prevent the spread of the virus: (i) maintaining distance from others, (ii) washing hands more regularly, (iii) covering one’s cough and (iv) wearing a mask. For maintaining distance, because almost all the sample (95%) reported doing this always or often, we focus on the most extreme case: always maintaining distance. With respect to washing hands, there were two questions on whether respondents washed their hands more regularly and used hands sanitizer more regularly. Because sanitizer is used for hand washing and is recommended, at least in France, as a substitute for soap when outside, we consider a dummy variable equal to 1 if the individual reports more regular sanitizer use or hand washing. We also construct a dummy variable for whether they pay particular attention to covering their cough or sneeze. Concerning wearing masks and maintaining distance from others, respondents are asked how often they wear a mask or keep their distance from others when they are outside, with the following response options: always, often, sometimes and never. For masks, we construct a dummy variable equal to 1 when the reported frequency is always or often and 0 otherwise. Note that wearing a mask was recommended only after the first lockdown when surgical masks became available for the general population.

To summarize, we have seven outcomes that capture individuals’ preventive behavior since the beginning of the COVID-19 outbreak: (i) does not meet family members living outside the household anymore, (ii) does not participate in gatherings with more than 5 other individuals, (iii) always keeps distance from others when outside, (iv) always or often wears a mask when outside, (v) washes hands or use hands sanitizer more frequently, and (vi) pays a particular attention to covering coughs. We argue that the three first variables measure behaviors related to social distancing, although the three last variables measure protective behaviors that relate to hygiene.

Variables of interest

As explained in the theoretical background, we study the correlation between the adoption of protective behaviors and risk aversion, farsightedness and the perceived trustworthiness of others. These variables were measured before the corona crisis. With respect to risk aversion and farsightedness, which are derived from a questionnaire specific to France (also called the drop-off questionnaire) from the 8th wave of the SHARE survey conducted before the corona crisis. The continuous variables are derived from self-reported measures, scored on a scale from 0 to 10, that are not specific to a given domain such that they capture broad risk tolerance and patience.

Risk aversion: on a scale from 0 to 10, do you generally consider yourself to be a cautious person, limiting risks as much as possible, or, conversely, do you consider yourself to be someone who likes to take risks, likes adventure, and seeks novelty and challenges?.

Farsightedness: on a scale from 0 to 10, do you consider yourself more as someone who lives from day to day and takes life as it comes, without thinking too much about tomorrow, or, conversely, as someone who thinks about the future and is farsighted?

These variables are therefore not specific to health or economic planning/risks, as is the case in some surveys, and they are not incentivized. Nonetheless, Dohmen et al. [10] have shown that these broad measures are highly correlated with different specific dimensions and can explain different human behaviors, including health behaviors. These measures might also be better at predicting health behaviors than specific financial risk and time preferences. The exact wordings of the questions are as follows:

With respect to the perceived trustworthiness of others, it is derived from the regular questionnaire of SHARE.Footnote 5 It is measured by a continuous variable derived from a self-reported score from 0 to 10. Such question is validated in Fehr [14] who suggest self-reported others’ trustworthiness should be asked on a continuous scale. In addition, Sapienza et al. [28] has shown that such captures well beliefs about others’ trustworthiness in a trust game.

The exact wording provided below—

Perceived trustworthiness: now I would like to ask a question about how you view other people. Generally speaking, would you say that most people can be trusted or that you can't be too careful in dealing with people? Please tell me on a scale from 0 to 10, where 0 means you can't be too careful and 10 means that most people can be trusted.

Control variables

We control for several characteristics that might be correlated with our key variables and the outcomes. We control for age (using spline functions to take into account potential nonlinearity), education, and gender. We also control for dummy variables indicating whether the individual’s household has difficulties to make ends meet. This indicator provides a subjective measure of living conditions that has been shown to be robust to cultural norms and to be correlated to both income and health-related behaviors [2, 13]. This measure may also better reflect the current living conditions of the individuals during the crisis than the income they reported before the pandemic.Footnote 6 Additionally, we control for determinants of the demand for healthcare and the risk of developing severe COVID-19: whether the individual’s body mass index (BMI) is higher than or equal to 30, whether the individual has at least one chronic condition, whether the person has at least one limitation in their activities of daily living (ADL limitations) and whether they visited at least one specialist during the year before the coronavirus outbreak.Footnote 7 One should note that we have tested several health measures and specifications on the current variables we use (e.g., the number of chronic conditions), and our conclusions remained unchanged for our main variable of interest.

As noted by Fehr [14], the self-reported measure of trustworthiness is potentially correlated correlate with risk aversion and “preferences toward taking social risks”. We already control for risk aversion in the analysis given that it is one of the main variables of interest. We additionally control for variables that might correlate with social preferences and the adoption of protective behaviors: political opinions [16, 21] and religiosity [3, 16, 18, 26]. Political opinions come from a self-reported variable scored from 0 (left) to 10 (right) that is discretized to capture extreme political orientation. Indeed, those with extreme opinions may be less likely to comply with some public health measures, such as vaccination [20]. [33] also finds that individuals who feel marginalized and do not feel represented by the government have a lower propensity to be vaccinated. This indicates that comparing extreme opinions might be of interest when studying protective behavior since conspiracy thinking might correlate with individuals’ level of compliance. We therefore construct a variable taking three different values: 0 if the individual responded 0, 1 if the individual responded with a value from 1 to 9, and 2 if the individual responded 10.Footnote 8 Finally, for religiosity, individuals were asked at what frequency they currently pray. We construct a binary variable indicating whether the individuals pray or not. These variable captures whether individuals are religious enough to pray. Descriptive statistics can be found in Table 1.

Table 1 Descriptive statistics
Full size table

Methodology

Because we explore that our outcomes, preventive behaviors, are dummy variables, we will explore the correlation with our variable of interests using probit models. The models are estimated with and without controls (results without controls are presented in the appendix).

Results

We present the results for our variables of interest in different graphs, and full tables are provided in the Appendix. Concerning risk aversion (Fig. 1), it increases the probability of not participating in meeting with more than 5 other persons and no longer visiting family members anymore. These results highlight the fact that risk preferences are related to fear of the coronavirus. Indeed, these two outcomes relate to (usually inside) social gatherings that represent greater risk due to a higher transmission rate. Risk-averse individuals might therefore greatly fear transmitting the virus to their friends and family members. These results are in line with Müller and Rau [25], who find that risk aversion tends to have a positive or null correlation with compliance. In contrast, it contradicts Sheth and Wright [32], who find that risk aversion does not correlate with socialization among students in California. Our results are also in contradiction with Guillon and Kergall [17]. One should note that, with respect to the latter, the population of interest (they do not focus on old age population) and the dimensions of behaviors with respect to COVID that are used are different.

Fig. 1
figure 1

Marginal effect of risk aversion on the different outcomes. N = 1271. Marginal effects estimated from a probit model. These results are obtained when controlling for age, education, patience, trust in others, the use of specialist care wave 8, the existence of a chronic condition, gender, political opinions, BMI, religiosity, and economic difficulties. ***p < 0.01, **p < 0.05, *p < 0.1

Full size image

Regarding the magnitude, we find that, on average, increasing the risk aversion score by one unit increases the probability of not meeting with more than 5 people and not visiting family members by 1.5 and 1.3 percentage points, respectively. This implies substantial effects since, for each activity, the average difference in the probability of complying between the most risk-averse and least risk-averse is approximately 15 and 13 percentage points, respectively.

In contrast, risk aversion decreases the probability of covering one’s cough or sneeze. This negative effect can be explained by different mechanisms. This result is difficult to understand, and more research needs to be done to better understand this result. Nonetheless, one explanation can be that risk aversion does not matter for a preventive measure that protects others only.

Concerning farsightedness, it positively correlates with most of the preventive measures (Fig. 2). This confirms the findings of Müller and Rau [25]. Patience significantly increases not visiting family members anymore, paying particular attention to covering coughs and sneezes, always wearing a mask when outside, always keeping distance from others when outside, and washing hands more frequently than before the corona crisis. This shows that patience is a major determinant of compliance with respect to recommendations and might reflect that our general measure of patience captures preferences or patience related to health decisions, especially in the uncertain context of the coronavirus, for which long-term effects are unknown, just as its duration. One should also note that because the survey was conducted in June and July, when wearing masks when outside was not necessarily promoted by the French government, our result could indicate that patience increases persistence in the diligent practice of protective behaviors.

Fig. 2
figure 2

Marginal effects of farsightedness on the different outcomes. N = 1271. Marginal effects estimated from a probit model. These results are obtained when controlling for age, education, risk aversion, trust in others, the use of specialist care in wave 8, the existence of a chronic condition, gender, political opinions, BMI, religiosity, and economic difficulties. ***p < 0.01, **p < 0.05, *p < 0.1

Full size image

The effects are also substantial, since, for example, the average difference between the more patient and least patient individuals in the probability of always wearing a mask is approximately 19 percentage points. This means that if all individuals were patient, the mean probability of wearing masks would be approximately 77% (note that it is 70% in the data), and it would be 58% if all individuals were impatient. The difference is approximately 12 and 10 for not visiting family members anymore and always keeping its distance when outside, respectively.

The marginal effects of trust in others are displayed in Fig. 3. The correlation is significantly different from zero and negative for the probabilities of no longer visiting family members and no longer participating in gatherings with more than 5 other people. This suggests that individuals with a high level of trust in others are less likely to comply with the recommendations to have fewer (and smaller) social gatherings. This result could be explained by the fact that individuals with a higher level of trust expect their friends/family members to be less likely to transmit the coronavirus.

Fig. 3
figure 3

Marginal effects of trust in others on the different outcomes. N = 1271. Marginal effects estimated from a probit model. These results are obtained when controlling for age, education, risk aversion, patience, trust in others, the use of specialist care in wave 8, the existence of a chronic condition, gender, political opinions, BMI, religiosity, and economic difficulties. ***p < 0.01, **p < 0.05, *p < 0.1

Full size image

Regarding the magnitude, it is again substantial since the average predicted probability of not visiting family members (not participating in gatherings with more than 5 other people) is 0.59 (0.67) for those with the lowest level of trust and 0.39 (0.56) for those with the highest.Footnote 9

Discussion

Using an original survey conducted among French participants in the European SHARE Survey, this article explores how the adoption of preventive behaviors and compliance with COVID-19-related protective recommendations correlate with several preferences and beliefs measured before the corona crisis, namely, risk aversion, patience, trust in others, political opinion, and religiosity. We find evidence that patience, risk aversion, and perceived trustworthiness have substantial correlations with protective behaviors. We also find that the identity of the predictive power of our main variable of interest vary with respect to the individuals involved.

Our results suggest that farsightedness and risk aversion are strong predictors of individuals’ protective behavior. Risk aversion increases the probabilities of not meeting more than 5 other people and not meeting with family members anymore, but decreases the probability of covering one’s cough or sneeze. For the other preventive behaviors, the association is positive but not significantly different from zero. This is a partial validation of hypothesis H1. As suggested in assumption H4, some heterogeneity could be expected. We particularly find that risk aversion does not correlate with preventive behaviors that aim to protect others only (covering one’s cough and sneeze). In addition, the identity of the individuals involved in the social interaction seems to matter. Indeed, a correlation is found only when the interaction involves explicit individuals who are close to the agent (family members or friends). This result might capture the fact that risk-averse individuals do not only take into account their own risk of being infected, but the risk to transmit to their friends and family members. When the individuals are not explicitly identified or unknown to the agent, risk aversion does seem to matter. Another potential explanation is that gatherings represent a higher risk of transmission than for exposure situations captured by the other protective behaviors, hence a higher risk of being infected.

More farsighted individuals are more likely to not visit their family members anymore, wear a mask and keep their distance from others when outside and wash their hands more regularly. Although it is significant at the 10 percent level only, it is positively correlated to covering one’s cough and keeping its distance when outside. Again, this is a partial validation of the assumption H2. One potential explanation to the irrelevance of farsightedness to explain the reduction of the probability to not meet 5+ people can be explained by two potential mechanisms compensating each other. First, more farsighted might perceive better the potential detrimental effects of being infected in the midterm, which could make them more likely to reduce their social interactions. Second, they can also be more likely to perceive the future detrimental effects on their mental health to reduce gatherings. Hence, it is possible the effect of farsightedness on the participation in social gatherings is the result of the two opposite effects that compensate each other.

Regarding the perceived trustworthiness of others, we find a negative association with the adoption of preventive behaviors related to gatherings and visiting family members. This is in line with assumption H3b. We interpret this result as a sign that individuals with a higher trust in others perceive a lower risk of being infected by friends and family members and that individuals are more willing to accept the risk of not adopting a protective behavior when interacting with them. This result highlights the fact that perceived trustworthiness increases the willingness to take a risk when the individual one interacts with are part of the close social network (family members and friends).

Overall, our results imply that preferences and the perceived trustworthiness of others are predictors of individuals’ behavior in the pandemic context. They also suggest that they predict better preventive behaviors that are meant individuals themselves rather than the others. Finally, we can also observe that the identity of the individuals who are involved in the interaction associated to the preventive behavior matters a lot.

Our work suffers from several limitations that need to be discussed. First, the focus on France might appear less relevant than investigation of other countries. It should be emphasized that Fetzer et al. [15] conducted a survey in 175 countries (between March 20 and April 5, 2020) and revealed that France is an average country in terms of staying at home and more regular handwashing. France was also close to the mean in terms of not attending gatherings and keeping distance to distance from others. One should also note that France was close to the average in terms of trust in the government. Finally, France seems to have been slightly less severely affected by the virus than its neighbors in terms of cases and deaths, except for Germany and Switzerland (see Table F.1). One could therefore speculate that the French population might have perceived a lower risk than the residents in some of these countries.

Another limitation of this study is the use of preferences measured before the pandemic while the pandemic may have impacted individuals’ preferences. Although we still find significant correlations, this could be a limitation for assessing the contemporaneous correlation between protective behaviors and preferences. Indeed, the pandemic is an exogenous shock, between the time preferences are collected and protective behaviors are observed, that might have affected individuals’ preferences. It is difficult if such change in preferences occurred in our sample because preferences are observed only before the pandemic and, if so, how this would bias the results. Indeed, the literature on the effect of the pandemic on preferences is still inconclusive given that some papers find that the COVID-19 crisis increases risk aversion [8] for men only (Lohmann et al. [23]), decreases risk aversion [19], has no effect on risk preferences [1, 12], or no consistent effect on risk present in men (Lohnmann et al. [23], or no effect on time preferences [12], increases trust (Ben-David and Sade [4]), or no consistent effect on trust [31]. Finally, even if preferences would change, at it is not clear if this should bias the results. Indeed, we study the correlation between protective behaviors and the risk aversion or patience linearly. In such a case, how individuals are ranked between each other are not changed, i.e., the distribution is simply shifted, and it is not clear that this would create a bias.

Despite these limitations, our results suggest that government policies and public health campaigns should consider the heterogeneity of preferences and beliefs and more carefully target individuals who underestimate the spread of the virus and the risk of infection at gatherings due to their trust in others. Providing specific information on the risks associated with COVID-19, on the health and financial risks associated with the lockdown and shortage of health healthcare supply, and on their long-term effects should also help to better target risk-seekers and more present-oriented individuals. Finally, it is important to inform about the risk of infection during social interactions, especially with close relatives that people tend to trust the most.