Introduction

Understanding how lockdown, a strict epidemic prevention policy, affects people’s risk attitudes is an interesting and fundamental issue. On the one hand, the lockdown policy during the COVID-19 was a strict public policy that has rarely been implemented in recent decades. The impact of this policy on human behaviour is itself an important issue. At the same time, it is also a practical issue with important reference value for policy making in the postpandemic era. On the other hand, in the field of behaviour science, risk preference is a basic determinant of individual decision-making. Whether and how lockdown policies can change people’s risk attitudes is an important issue in behaviour science.

However, since epidemics are emergent events, their occurrence time is almost unpredictable; thus, obtaining comparative data before and after the implementation of a lockdown is difficult. Specifically, obtaining microscopic comparison data via experiments is very difficult. Unlike traditional statistical data and microinvestigation data, which are periodically collected, regular experimental studies are rare. Thus, obtaining comparison data before and after lockdown in experiments is even rarer.

Fortunately, in this study, we obtained a pair of controlled experimental data before and after the lockdown policy was implemented from the bordering town of Gengma in Yunnan Province, China. In 2018, we conducted a series of behaviour experiments, including risk preference tasks, in many regions in Yunnan, and Gengma town was one of the experimental sites. In November 2020, Gengma was locked down because of the COVID-19 pandemic. Immediately after the city’s lockdown was lifted, we conducted the same risk preference experiment in Gengma. Most of the subjects also participated in the 2018 experiment. Thus, we were able to obtain rare comparison experimental data before and after the lockdown.

The most mature and normative experiment for measuring risk preference is an experimental design based on prospect theory (PT), which won the Nobel Prize in 2002. In their PT theory, Tversky and Kahneman (1992) proposed the fourfold pattern of risk attitudes, which suggests that people are risk seeking over low-probability gains and moderate-probability losses and are risk averse over moderate-probability gains and low-probability losses. Furthermore, according to PT’s fourfold risk preference theory, individuals have opposite risk attitudes in the gain domain and loss domain, that is, the reflection phenomenon. This risk attitude phenomenon occurs because people tend to overweight the probability of an event when it is low. Since fourfold risk preference theory of PT can comprehensively depict individual risk preference, in this study, we used this theory to systematically examine the subjects’ risk attitudes in the gain and loss domains under medium and low probabilities before and after the lockdown. We also assessed alterations in risk preferences at both the aggregated and individual levels. Moreover, we clarified the primary contribution of lockdown to influencing people’s risk attitudes by comparing the risk preferences of subjects who experienced lockdown during the pandemic with those of those who did not. The results of this study indicate that the lockdown significantly affected people’s risk attitudes in the gain and loss domains under medium-probability conditions, resulting in a shift from risk aversion to risk seeking in medium-probability gain domain decision-making and from risk seeking to risk aversion in medium-risk loss domain decision-making.

Literature Review

Although it is difficult to obtain pre- and post-data, some studies on the impact of major shocks on risk behaviours, mainly on the basis of post-data, have been conducted. Notably, in existing studies, no consensus on the direction of change in human risk attitudes after catastrophes has been reached.

Some studies have found that people become more risk seeking after suffering from natural disasters. For example, Hanaoka et al. (2018) used survey data to analyse the changes in people’s risk preferences before and after the Great Tohoku Earthquake in Japan and reported that people’s tolerance for risk increased after the earthquake and that the degree of change in risk attitudes was proportional to the degree of damage. This change in risk attitudes lasted up to 5 years after the disaster. However, their survey data involved only medium-risk (50% probability) options in the gain domain with a hypothetical lottery. After the Great Flood in Australia, Page et al. (2014) compared the difference in risk attitudes between those who lost their houses in the flood and those who did not. They reported that individuals who suffered real losses presented a greater level of risk seeking in the gain domain; the proportion of subjects who lost their houses and then placed bets in a small-probability-large-payoff lottery was approximately 50% greater than that of subjects who did not lose houses. Said et al. 2015 also reported in a field survey that people who suffered more losses in a disaster would make riskier choices. However, they also reported that people who have experienced more previous disasters go on to make more risk-averse decisions. After Hurricane Katrina, Eckel et al. (2009) conducted experiments on refugees who were evacuated and transferred to Houston and reported that negative traumatic emotions can predict an increase in the degree of risk seeking. Only 20% of the participants in the control group chose the risk-loving option, whereas 40% of those in the storm group chose the risk-loving choice. Cavatorta et al. (2020) used the random location of segments of the wall between the West Bank and Israel to study the effect of counterviolence initiatives on people’s risk attitudes, showing that people living close to the wall become more risk-tolerant, ambiguity averse and impatient than those unexposed to the wall, and this effect is amplified for people both exposed to and isolated (from the West Bank) by the wall. By measuring the risk preferences of Swiss fruit and grapevine producers, Finger et al. (2023) reported that the experience of individual shocks has only limited effects on farmer risk preferences. However, the experience of both frost- and pest-related damage tends to cause farmers to be more risk tolerant, and the simultaneous experience of climate- and pest-related crop damage causes farmers to be more risk tolerant in multiple domains. On the basis of census data from 742 youth business groups in five districts in the semiarid Tigray Region of Ethiopia, Stein et al. (2024) assessed the effects of idiosyncratic shock and a covariate climate (drought) shock on risk preference parameters in a rural poor and vulnerable population in a semiarid environment in sub-Saharan Africa, showing that subjects exposed to the covariate drought shock have become more willing to take risks. Thus, these studies show that experiencing a rare nature disaster or policy reform will make people more risk seeking.

However, the opposite conclusion has been reached in other studies, which have found that individuals suffering from natural disasters are more risk averse. Cassar et al. (2017) measured the risk attitudes of villagers 4–5 years after experiencing a flood in Thailand. In the experiment, they used coloured questionnaires of a ball-drawing lottery to simulate the subjects’ risk attitudes. They kept the payoff unchanged and changed only the proportion of balls; that is, they changed the probability of successfully drawing a particular ball. They reported that disaster-affected individuals have a greater degree of risk aversion, but the degree of risk aversion has nothing to do with the level of losses. Reynaud and Aubert (2020) conducted an artefactual field experiment in Vietnam to investigate whether and how experiencing a natural disaster affects individuals’ attitudes towards risks. The results revealed that households in villages that had been affected by a flood in recent years exhibit greater risk aversion than do individuals living in similar but unaffected villages. Interestingly, this result holds for the loss domain but not the gain domain. Experiments conducted by Cameron and Shah 2015 in Indonesia also revealed that people who have experienced floods or earthquakes within the past three years are more risk averse. They suggest that this change occurs because the disaster has changed people’s beliefs, making them think that the world is more dangerous than they had previously thought. Robinson et al. (2021) conducted an online experiment on insurance demand against flood risk with 1800 Dutch homeowners and reported that those who had experienced flooding in their homes in the past were more risk averse and were significantly more likely to purchase insurance. Chou CY et al. (2022); conducted a questionnaire survey of 863 respondents in Taiwan to explore the factors influencing risk perception and reported that risk perception has a significant effect on the willingness to pay for risk insurance. Previous earthquake disaster damage information may increase individuals’ risk perception, which in turn leads people to tend to avoid losses and pay higher insurance premiums to reduce earthquake risk. Beine and his collaborators (2020) also reported that after an earthquake in Albania, local residents were more risk averse. Liebenehm et al. (2023) combined individual-level panel data from 2008 to 2017 with historical rainfall data from rural Thailand and Vietnam to describe temporal changes in risk aversion. The results revealed that rainfall shocks increase individuals’ risk aversion and, in the absence of functioning credit and insurance markets, may ultimately lead to decisions that perpetuate poverty. To investigate the stability of farmers’ risk attitudes over time, Bozzola et al. (2021) used panel data from more than 36,000 Italian farms specializing in cereals from 1989-2009. They investigated changes in farmers’ risk attitudes in response to multiple shocks resulting from policy reforms over a long period and production shocks triggered by droughts and reported that the experience of shock (loss) makes farmers exhibit more obvious risk aversion behaviour. In their study, Shereen et al. confirmed that the accumulation of experiencing at least one negative event over multiple periods led to more people taking protective actions against risk (Chaudhry et al. 2020). Similarly, Malmendier et al. (2011) reported that individuals who suffered heavy losses in the stock market were less willing to make risky financial investments and were more pessimistic about future investments. Bourdeau-Brien et al. (2020) focused on the consequences of disasters on financial markets and inferred the impact of major catastrophes on the risk-taking behaviour of investors from U.S. municipal bond transactions. The findings strongly support the conjecture that natural disasters cause a statistically and economically significant increase in risk aversion at the local level.

Since the outbreak of COVID-19, people’s lives have been greatly affected, which has changed the economic activities, behavioural choices and risk attitudes of individuals and enterprises. Several studies have evaluated changes in individual risk attitudes and behaviours after the COVID-19 pandemic, but no consistent conclusions have been reached. In a large-scale survey conducted from March to May 2020 that included 88,181 participants from 47 countries, Rachev NR et al. (2021) reported that risk aversion and the framing effect were greater than they were under typical circumstances. Furthermore, perceived stress and concerns about coronavirus were positively associated with the framing effect. Bu and colleagues (2020) showed that exposure to COVID-19 leads to increased risk aversion and that risk aversion increases with increased exposure. Using high-frequency data from the S&P 500 Index, Nisani et al. (2022) estimated investors’ risk and ambiguity aversion and reported that in the prepandemic period, investors exhibited risk aversion as well as an ambiguity-seeking attitude, whereas during the pandemic, they demonstrated risk- and ambiguity-neutral behaviour. Cicerale et al. (2022) ran three partial replications of the Tversky and Kahneman (1992) paper, focusing on a set of eight prospects, after a terror attack (Paris, November 2015, 134 subjects) and during the COVID-19 pandemic, both during the first lockdown in Italy (spring 2020, 176 subjects) and after the first reopening (140 subjects). They noted a significant increase in risk aversion, both in the gain and in the loss domains, that consistently emerged in the three replications. Yun Wang et al. (2020) adopted a multiple-price-list elicitation method with real money incentives to precisely measure individuals’ risk attitudes at different stake levels and the extent to which they were affected by personal and social shocks following the COVID-19 outbreak in China. Subjects who had previously experienced negative personal shocks are more risk averse at medium and large stakes but more risk loving at very small stakes. The results indicate that the impact of COVID-19 on individual risk attitudes is not as influential as expected unless an individual’s personal life is affected directly. Ikeda et al. (2020) reported that owing to the pandemic, subjects became less sensitive to an increase in losses and felt less displeasure owing to losses, especially large losses. Moreover, they became more pessimistic towards losses occurring with tiny probabilities and more optimistic towards losses with larger probabilities. Shachat et al. (2020) reported that in the early stages of the COVID-19 pandemic, risk tolerance decreased in the loss domain, in addition to increased prosocial behaviours and cooperative behaviours and increased risk tolerance in the gain domain. Tienhua Wu (2022) employed the policy Delphi method and survey data to examine long-term care facility (LTCF) managers’ perceptions and attitudes during the COVID-19 pandemic and explore the effects of sociodemographic characteristics on healthcare decisions. Their findings showed that participants exhibited risk aversion for small losses but became risk neutral when devastating damage was considered. LTCF managers exhibited perception bias that led to over- and underestimation of the occurrence of infection risk. At the same time, some studies have failed to find any changes in risk appetite before and after the COVID-19 pandemic. Angrisani and colleagues (2020) used a laboratory task to test subjects’ risk appetite before and after the COVID-19 outbreak and reported no change in risk appetite. Similarly, Drichoutis, Nayga, and Lohmann noted a similar stability in risk appetite (Drichoutis and Nayga 2020, Lohmann et al. 2020). Hao Luo et al. (2023) employed a DID methodology to investigate the impact of the COVID-19 pandemic and associated lockdowns on risk attitudes among rural populations in Thailand. Their study indicated no substantial overall shift in risk-taking propensity among rural households post-lockdown. However, individuals working outside the agricultural sector have a statistically significant reduction in their willingness to take risks after experiencing a lockdown. Xavier Gassmann et al. (2022) estimated risk and ambiguity avoidance before, during, and after the COVID-19 lockdown in France; they conducted an online questionnaire of students at Burgundy Business School using incentive-compatible tasks, which revealed that patience, risk aversion, and ambiguity aversion fell during lockdown and then gradually returned to their initial levels 4 months later. Moreover, studies have tried to discuss the relationship between the behavioural impacts of COVID-19 and public policy. Using Jeju Island, South Korea, as a case study, Yuqian Lin et al. (2024) collected a nearly two-year smart card dataset collected from the beginning of 2019 until nine months into the pandemic, focusing on changes in frequent transit users and how these changes were affected by key government policies. Sun et al. (2024) investigated the vaccination situation and reasons for 14,001 local residents in the 2021–2022 flu season in Shanghai. Their results revealed that the influenza vaccination rate in the 2021-2022 season was higher than that in the 2018–2019 season in the same area, and this trend was found among populations of different age groups. This suggests that the COVID-19 pandemic has resulted in an increase in public awareness regarding the prevention and control of infectious diseases and changes in people’s health behaviours.

There are several limitations regarding the existing studies that have focused on the impact of natural disasters on risk preference. First, the unpredictable features of disasters make it difficult to obtain pre-disaster data. Previous studies usually use subjects who are not influenced or are less affected by disasters as the baseline sample, and research that directly compares risk preferences before and after lockdown is rare. Second, although the fourfold risk attitude theory of PT has already been the main theory used to analyse risk preference, few studies are based on the typical PT framework, not to mention following the standard experimental procedure. Most previous studies measured only risk attitude changes in the gain domain, and some were based on hypothetical lotteries. Research in the loss domain is still lacking. Third, the literature investigating the impact of great shocks on risk attitudes has focused on geographic disasters such as earthquakes, floods, and storms or financial crises. There is literature on the impact of COVID-19 on people’s risk attitudes, but it focuses mainly on changes in risk attitudes before and after infection with COVID-19 or the impact on society as a whole during the COVID-19 pandemic. There is limited research on whether or how being affected by the lockdown policy could influence people’s risk attitudes.

In this study, for the first time, we systematically evaluated the change in the fourfold risk attitudes of people before and after the implementation of the strict lockdown policy during the COVID-19 pandemic. The research was carried out in a city bordering Myanmar, Gengma, in Yunnan Province, China. Gengma County is affiliated with Lincang city, Yunnan Province. It is located on the southwestern border of China and is connected to the mountains and rivers of Myanmar for 47.35 kilometres. It is the most convenient land passage from Lincang and Kunming to Yangon, Myanmar. Although the overall threat of COVID-19 is quite low in China because of its strict prevention policies, Gengma residents have been under great pressure from the COVID-19 pandemic since its outbreak in 2019 because of its special geometrical location.

We conducted two rounds of experiments with real payoffs to evaluate the fourfold risk attitude change of Gengma residents. The first round of the study was performed in 2018, when we conducted a series of behaviour experiments, including risk preference experiments, just before the outbreak of the COVID-19 pandemic. The second round of the experiment was performed during the pandemic in 2020, after the lockdown of Gengma County caused by the discovery of asymptomatic infected persons from Myanmar. This provided us with a good opportunity to measure and compare people’s risk attitudes before and after the lockdown policy during COVID-19 and to see how this freshly experienced emergency and shock changed the fourfold risk attitude.

Considering the limitations of the previous studies, on the basis of the above experiments, we aimed to answer the following questions:

  1. (1)

    In a controlled experimental setting with pre-test and post-test data, what effects does lockdown have on individuals’ risk behaviour?

  2. (2)

    On the basis of the fourfold framework of prospect theory, how does lockdown affect individuals’ risk behaviours in the gain‒loss dimension and at different probability levels?

  3. (3)

    What are the potential policy implications of this change in risk attitudes related to the lockdown?

Methods

Ethics Statement

The questionnaire and methodology for this study was approved by Medical Ethics Committee of Kunming Medical University (Approval number: KMMU2014004, Date: 12.2014). The research has been conducted in accordance with the relevant guidelines (including Declaration of Helsinki, etc.) applied by Medical Ethics Committee of Kunming Medical University when human participants are involved. The scope of approval covered research locations, procedures, methods, and data monitoring plans, as well as participant-related protocols, including recruitment methods, sample size, consent procedures, compensation, and disclosure of potential risks and benefits.

Subjects

Meng-ding town experiment: The first round of the experiment was carried out with 65 residents of Gengma County, Meng-ding town of Yunnan Province, in September 2018. The demographics of these subjects are summarized as follows: mean age, 35 years (range: 19–55 years); 38 males and 27 females. The second round of the experiment was carried out in November 2020, immediately after the 14-day quarantine was lifted from Gengma County. Fifty-five residents from the same county as those in the first-round experiment were recruited. The demographics of these subjects are summarized as follows: mean age, 35 years (range: 20–56 years); 35 males; and 20 females. All the subjects had a junior high school education level or more. The subjects were mainly civil servants, community workers and ordinary residents. Their monthly income ranged from 3000-6000 RMB.

To further confirm whether the observed changes in risk attitudes were driven primarily by the lockdown policies themselves or by the broader psychological and emotional impacts of the pandemic, we obtained identical experimental data from Meng-sa, a nearby town that experienced the pandemic but did not impose a lockdown during the study period.

Meng-sa town experiment: The first round of the experiment was carried out with 59 residents of Gengma County, Meng-sa town of Yunnan Province, in September 2018. The demographics of these subjects are summarized as follows: mean age, 36 years (range: 20–59 years); 33 males and 26 females. The second round of the experiment was carried out in January 2021. Fifty-five residents from the same county as those in the first-round experiment were recruited. The demographics of these subjects are summarized as follows: mean age, 35 years (range: 20–55 years); 33 males and 18 females. All the subjects had a junior high school education level or more. The subjects were mainly civil servants, community workers and ordinary residents. Their monthly income ranged from 3000 to 6000 RMB.

All of the subjects had similar occupations, comparable monthly incomes, and the same education level as the subjects in the first-round experiment. Specifically, in the Mend-ding town experiment, 39 of the 55 subjects in the second-round experiment also participated in the first-round experiment. The investigators adhered to the practice in experimental economics of applying monetary incentives to motivate decision-making without using deception. Additionally, we compared the demographic profiles of our participants in two experimental studies with the demographic characteristics of China’s urban employed population on the basis of data from the Seventh National Population Census and the 2021 China Statistical Yearbook and compared these findings (see Table 1 for details). Overall, natural attributes such as age and gender, as well as social attributes such as income, education, and employment status, essentially reflect the characteristics of the employed population residing in urban areas of China. This group constitutes the mainstream Chinese population, being the primary creators of social wealth and key participants in social life, as well as the principal formulators and bearers of various economic and social policies.

Table 1 Analysis of the Representativeness of Experimental Samples.
Full size table

Experimental design

We performed two rounds of experiments. The first round of the experiment was conducted in September 2018, before the outbreak of the COVID-19 pandemic. The second round of the experiment was conducted in November 2020 after the outbreak of COVID-19. Notably, on the evening of November 9, 2020, Gengma County reported one case of imported asymptomatic infection from Myanmar and two cases of asymptomatic infection in Myanmar nationals in the area. It was immediately announced that from 24:00 on November 9 to 24:00 on November 23, residents in the dam districts of Meng-ding Town, Qing-shui-he and Ban-xing districts would be quarantined at home for 14 days. The second round of the experiment was conducted after the quarantine of our experimental subjects was lifted.

The experiment was conducted using a questionnaire comprising two parts. The first part utilized a questionnaire format with four sets of simple choice tasks to measure participants’ four-quadrant risk preferences, following the methodologies of Cohen et al. (1987), Tversky and Kahneman (1992), Holt and Laury (2002), Zhong et al. (2009), and Chew et al. (2021). The second part consisted of a questionnaire for collecting basic demographic information. The first part of the lottery choice questionnaire included four tasks, each of which contained 10 pairs of choices with a lottery and a certain outcome. The lottery did not change in the 10 pairs of choices, but from the first to last pairs of choices, the specific outcome gradually increased in the pure gain experiments and gradually decreased in the pure loss experiments. The average duration of the experiment was approximately half an hour per participant. Detailed information on the four tasks can be found in Appendix Tables 1–4.

The experiment was conducted in a one-on-one manner. For each subject, the experimenter individually filled out the questionnaire with him or her. All the experimental instruments were explained to the subjects orally. The following is the content of the instrument of the moderate-probability pure-gain lottery.

In this scenario, there are 100 cards: 50 black cards and 50 red cards. Please randomly select one of them and guess whether it will be a black card or a red card before the draw.

Option A: If the card you choose is red, you will receive 100 RMB; if it is black, you will receive 0 RMB. In other words, you have a 50% probability of getting 100 RMB and a 50% probability of getting 0 RMB.

Option B: Ten types of amounts are listed (in ascending order), corresponding to the amount you will definitely obtain if you choose this item.

Decision: For the following 10 rows, please mark your choice with a check mark (√) in the last column of each row.

To ensure that the participants accurately understood the probabilities presented in the experiment, we employed a card-drawing format to illustrate the experimental tasks. Notably, for the majority of tasks, instructions for drawing cards were provided to facilitate the participants’ understanding of the probabilities involved in each task, whereas the actual tasks were conducted in a questionnaire format. Only during a single instance when a real payment was made did the card drawing occur as an actual procedure. Through this method, the participants were able to fully grasp that altering the probability was accomplished by modifying the number of black and red cards within the deck.

Each participant received a participation fee of 20 RMB. Furthermore, upon completion of the experiment, the experimenter randomly selected one of the 40 choices from the 4 experiments to determine an additional payoff on the basis of the participants’ decisions. For example, for a particular participant, the experimenter might draw the first choice pair from Task 1 for payment (Option A: 1% chance of receiving 800 RMB, 99% chance of receiving 0 RMB; Option B: a certain gain of 2 RMB). If the participant had previously chosen Option B, she would receive 2 RMB. If the participant had chosen Option A, a card draw would be conducted as a lottery (as described in the introduction), and the payoff would be made on the basis of the actual outcome. All incentives were disbursed in cash immediately after the participants completed the entire experiment. In our 2018 Meng-ding experiment, a total of 65 individuals participated, with an average earning of 31.3 RMB per person (including the 20 RMB participation fee). In our 2020 Meng-ding experiment, 55 participants were included, with an average earning of 35.8 RMB per person (including the 20 RMB participation fee).

The risk attitudes were determined following previous studies (Tversky and Kahneman 1992; Zhong et al. 2009). In all of our treatments, the majority of the subjects chose option A at the beginning and then crossed over to option B without ever returning to option A. There were a few cases of all-A or all-B choices. Additionally, a very small number of participants (only 2 in total) switched from B back to A after transitioning from A to B. These questionnaires were considered invalid and were not included in the statistics. The total number of risky choices is an indicator of the degree of risk seeking. In the 4th pair of tasks, the EV (expected value) of the lottery (which is calculated according to the Von Neumann–Morgenstern utility theorem with a risk-neutral assumption) equals a certain outcome.

Thus, at the aggregate level, a risk-neutral cohort should choose option A in the first 3 pairs of choices and switch to option B at the 5th pair of choices, as indicated by the dashed line in Fig. 1. The risk-neutral cohort should show 50% of choosing A, as well as B, in the 4th pair of choices because the two options are equal to each other.

Fig. 1: Proportion of risk choices in each decision in 2018 and 2020 experiment in Meng-ding.
figure 1

A Data averages for moderate probability pure gain lotteries (MG); B Data averages for low probability pure gain lotteries (LG); C Data averages for moderate probability pure loss lotteries (ML); D Data averages for low probability pure gain lotteries (LL). Orange colored solid line represents choices in 2018 experiment. Blue solid line represents choices in 2020 experiment. Dashed line represents risk-neutral predictions. X axis represents the number of the pair of prospects in each task, and Y axis represents the percentage of the A choice at each pair of prospects. N = 65 for 2018 experiment, N = 55 for 2020 experiment. ** represents P < 0.01. ns represents not significant.

Full size image

At the individual level, the risk attitude is measured by comparing the EV and the CE of the lotteries for each subject. The risk attitude depends on the switch point of the subject in each experiment. For example, in the MG experiment, if the subject chooses A in the first 5 pairs of choices and then switches to B at the 6th pair of choices, then her CE for this lottery is calculated by dividing the sum of the certain outcomes of the 5th and 6th pairs of choices. This is because the subject’s choice of A in the 5th pair of choices indicates that she thinks the certainty equivalent (CE) of the lottery is greater than the sure outcome of B, which is 55. However, her choice of B in the 6th pair of choices indicates that they believe that the CE of the lottery is less than the sure outcome of B at that point, which is 60. Thus, from this subject’s choices, it can be inferred that they consider the CE of the lottery to be greater than 55 and less than 60. In this experiment, we take the average of these two values as the CE of the lottery. By comparing this CE with the expected value (EV) of the lottery, which is 50 RMB, we can determine that the subject perceives the value of the lottery to be higher than its EV; thus, this individual is classified as a risk-seeking person. The calculated CEs are listed in Appendix Tables 1–4. According to our experimental design, if the subject switches his or her choice from A to B (the relationship between the switch point and individual risk attitudes is summarized in Appendix Table 2):

Before the 4th trial (included), the individual is risk averse,

After the 5th trial (included), the individual is risk seeking

Statistical tools

Nonparametric Kruskal‒Wallis tests were used to assess significant differences (Figs. 1, 2). Pearson correlation analysis was applied to investigate the influence of subject demographic characteristics.

Fig. 2: Proportion of risk choices in each decision in 2018 and 2021 experiment in Meng-Sa.
figure 2

A Data averages for moderate probability pure gain lotteries (MG); B Data averages for low probability pure gain lotteries (LG); C Data averages for moderate probability pure loss lotteries (ML); D Data averages for low probability pure gain lotteries (LL). Orange colored solid line represents choices in 2018 experiment. Blue solid line represents choices in 2020 experiment. Dashed line represents risk-neutral predictions. X axis represents the number of the pair of prospects in each task, and Y axis represents the percentage of the A choice at each pair of prospects. N = 59 for 2018 experiment, N = 55 for 2021 experiment. ns represents not significant.

Full size image

Results

In the present study, we used choices in lotteries with real monetary rewards to elicit risk attitudes before and after the lockdown of the COVID-19 pandemic in a border city of China.

Changes in aggregated risk attitudes after the lockdown of COVID-19

We conducted two rounds of experiments to evaluate the changes in people’s risk attitudes before and after the strict lockdown during the COVID-19 pandemic. The first-round experiment, conducted in September 2018 before the outbreak of the COVID-19 pandemic, is one of a series of economic behaviour experiments that we have conducted in several regions of Yunnan Province. By the end of 2020, when the pandemic in most regions of China was effectively controlled, a new wave of outbreak had appeared in Gengma, Yunnan Province. As a border city with Myanmar, Gengma has always been at high risk of epidemics. When we launched the second round of experiments in November 2020, the subjects, who were local residents of Gengma, had just completed a 14-day home quarantine and endured frequent nucleic acid tests due to the emergence of two cases of imported cases of COVID-19 in Myanmar nationals who had travelled to the local area.

Both rounds of experiments had the same setting, which included four sets of simple choice tasks in the form of questionnaires: moderate-probability pure gain lottery (MG), low-probability pure gain lottery (LG), moderate-probability pure loss lottery (ML), and low-probability pure loss lottery (LL). Each set of tasks contained 10 pairs of choices with a lottery (option A) and a certain outcome (option B) (detailed in Appendix Tables 1–4). In the pure gain experiments, the expected gain of option A was greater than that of option B in the first three choices, equal to that of option B in the fourth choice, and greater than that of option B in the last six choices. Thus, as predicted by expected utility theory (EUT), a risk-neutral person should choose option A in the first three choices and option B in the 5th–10th choices, as shown in Fig. 1A, B by the dashed line. We then calculated the percentage of option A, the risk option, in each pair of choices in the two sets of pure gain experiments in both the 2018 and 2020 experiments, which is shown as the solid line (Fig. 1A, B).

The results show that the participants’ risk preference for moderate probability pure gain prospects dramatically changed after the lockdown in the MG task. In the 2018 MG task (Fig. 1A, orange line), the participants’ preference for the 6th-10th options was consistent with the risk-neutral estimate. However, in the 1st-5th choices, that is, when the expected value (EV) of the lottery is greater than the certain outcome, a certain proportion of the subjects still choose the sure outcome option B, which means that the subjects show an overall risk aversive attitude. This finding is consistent with the fourfold risk preference hypothesis of PT theory. However, in the 2020 MG task (Fig. 1A, blue line), the subjects’ choice curve is significantly greater than the risk-neutral expectation in the 5th-9th choices; that is, when the value of a certain gain in choice B is already greater than the EV of choice A, many subjects still choose risky choice A. This result indicates that the lockdown experience increases people’s risk-seeking tendency for the moderate-probability pure gain lottery. In the LG task, the participants’ risk attitudes were different from those in the MG task. In both the 2018 (Fig. 1B, Orange Line) and the 2020 (Fig. 1B, Blue Line) LG tasks, the right half of the selection curve is significantly higher than the risk-neutral estimation curve; that is, when the expected value (EV) of the lottery is lower than the sure outcome, many subjects still choose A, the lottery. This means that, in the LG task, the subjects as a whole show the characteristics of risk-seeking, and this risk attitude was not changed by lockdown.

In the pure loss experiments, the expected loss of the lottery in the 1st–3rd choices is smaller than the certain loss. Thus, risk-neutral subjects should all choose option A in the first 3 choices, and with the gradual decrease in the sure loss, risk-neutral subjects should switch to option B in the 5th-10th pairs of choices. The risk-neutral choice is represented as a dashed line in Fig. 1C, D. We found that in the 2018 experiment (Fig. 1C, 1D, orange line), the risk attitudes of our subjects in the pure loss domain are also in line with the PT prediction, that is, risk seeking in the medium-probability pure loss experiment and risk aversion in the small-probability pure loss experiment. However, the choice curve of the 2020-ML task (Fig. 1C, blue line) is much lower than that of the 2018-ML task, which means that the proportions of risky choices in the first 1st-5th choices were significantly lower than those in 2018. This result indicates that the participants in the 2020 experiment had a greater level of risk aversion in the ML task. Moreover, similar to the pure gain task, the subjects’ risk preference for the LL task was not altered (Fig. 1D).

The above results show that in our first round of experiments (2018), the aggregated risk preferences of our subjects in the four quadrants were consistent with the expectations of PT; that is, in the gain domain, the participants showed risk-seeking for low probability tasks, and for moderate probability tasks, they showed risk aversion; in the loss domain, they showed risk aversion for low probability tasks and risk-seeking for moderate probability tasks. The PT explanation for this phenomenon is that when faced with a risky choice that has a small probability and a large payoff, either gain or loss, people tend to overweight the probability, which makes them risk seeking for gain and risk averse for loss. Moreover, this result is also in line with PT’s reflection hypothesis; that is, the risk preference in the loss domain is a mirror image of the risk preference in the gain domain.

However, when we compared the 2018 and 2020 experiments, we found that the strict lockdown policy during the COVID-19 pandemic greatly affected people’s risk preferences. First, in the medium-probability experiments, either in the gain domain or in the loss domain, the risk attitudes of the subjects underwent a substantial change: they became more risk seeking for potential gains and more risk averse for potential losses. To further confirm this observation, we conducted a nonparametric statistical analysis on overall risk attitudes in 2018 and 2020 via nonparametric tests (Supplemental Tables 1–4). The analysis results indicate that there was a significant change in risk preference exhibited by participants in the moderate risk-gain and moderate risk-loss experimental groups, whereas no significant change in risk preference was observed in the low risk-gain and low risk-loss experimental groups. Notably, their risk attitudes towards moderate probability lotteries became similar to their attitudes towards low probability lotteries, both in the gain and loss domains. Second, the directions of the changes in participants’ risk attitudes in the gain domain and loss domain were opposite. This finding supports the prediction of PT’s reflection hypothesis that when the risk attitude changes dynamically due to external stimulation, its change in the loss domain mirrors that in the gain domain.

In summary, our results showed that people’s risk attitudes before COVID-19 were consistent with the PT hypothesis, but after they experienced a 14-day lockdown under the long-term stress of COVID-19, their risk attitudes for moderate-probability choice dramatically changed to become similar to their low-probability risk attitudes. Furthermore, people’s risk attitudes in the gain domain, either for low probability or moderate probability, and their risk-attitude changes mirrored those in the loss domain.

To further clarify the impact of the lockdown itself on human risk preferences and distinguish between the effects of COVID-19 and the lockdown, we analysed the experimental results from Meng-sa town in Gengma County during the same period. Meng-sa Town, like Meng-ding Town, is under the administration of Gengma County in Lincang city and has the same demographic composition and economic structure. We also conducted the same risk preference experiments in Meng-sa in 2018 and 2021. Unlike Meng-ding, however, the subjects in Meng-sa did not experience a lockdown during the period of the experiment or between the two experiments. This means that the main influencing factor on the risk attitudes of Meng-sa subjects was the experience of a pandemic rather than that of a lockdown. Through a comparison of the experimental results between Meng-ding and Meng-sa, we can isolate the impact of the lockdown. The experimental results from Meng-sa are shown in Fig. 2, where the orange line represents the proportion of Choice A in 2018, and the blue line represents the proportion of Choice A in 2021. The aggregated risk attitude curves from the two rounds of experiments almost overlap (Fig. 2). We also conducted a nonparametric statistical analysis on the overall risk attitudes of Meng-sa participants in 2018 and 2021, and the results revealed no significant differences in risk preferences across the four domains between the two rounds of experiments (Supplemental Tables 5–8). These results indicate that the risk preferences of Meng-sa subjects, who did not experience the lockdown, did not significantly change from before to after the pandemic. This finding also supports our findings that the change in risk attitudes of Meng-ding subjects was influenced by the lockdown rather than the pandemic.

Changes in individual risk attitudes after the COVID-19 lockdown

The above analysis reveals that, at the aggregate level, experiencing lockdown during the COVID-19 pandemic changed people’s risky behaviours with moderate probability, making them more risk seeking in pure gain tasks and more risk averse in pure loss tasks. We conducted a further analysis to determine whether the alteration of their individual-level behaviour was consistent with the aggregate-level results. We counted the proportion of subjects who showed risk aversion in the four quadrants. The risk preference is determined on the basis of the participant’s switching point from choice A to B. If the switching point is earlier than the 4th choice (inclusive), this subject is defined as risk averse. If the switching point is later than the 5th choice (inclusive), then this subject is defined as risk seeking (detailed in Appendix Tables 1–4). The results show that in the 2018 experiment, 76.9% of the subjects showed a risk-averse attitude in the MG task, which dropped to 41.8% in 2020 (Fig. 3A). In the 2018 ML task, 40% of the subjects were risk averse. However, in the 2020 experiment, the proportion of risk-averse subjects increased to 63.6% (Fig. 3B). These results showed that at the individual level, before COVID-19, most of the subjects were risk averse in the MG task, and they turned risk seeking after lockdown. In contrast, in the ML task, risk-chasing subjects accounted for the majority before the COVID-19 lockdown, whereas most subjects subsequently experienced risk aversion. In LG and LL, there was no significant difference in the proportion of risk-averse subjects between 2018 and 2020. The proportion of risk-averse subjects in the LG task was 41.5% in 2018 and 38.2% in 2020, and the proportion of risk-averse subjects in the LL task was 75.4% in 2018 and 67.2% in the 2020 task (Fig. 3C, D). This finding indicates that the risk preference of our subjects in low-probability tasks is more stable. Moreover, the proportion of risk-averse subjects in MG tasks in 2020 decreased to 41.8%, which is close to the proportion demonstrating risk aversion in low-probability tasks (41.5% in 2018 and 38.2% in 2020). Additionally, in 2020, the proportion of risk-averse subjects in ML tasks increased to 63%, which is also numerically close to the proportion in low-probability tasks (75.4% in 2018 and 67.2% in 2020). Thus, at the individual level, the alterations in the risk attitudes of the participants before and after COVID-19 are consistent with the aggregate-level results.

Fig. 3: Individual level analysis of risk preference.
figure 3

A Proportion of risk averse subjects in moderate probability pure gain lotteries (MG) in 2018 (orange) and 2020 (blue) experiments. B Proportion of risk averse subjects in low probability pure gain lotteries (LG) in 2018 and 2020 experiments. C Proportion of risk averse subjects in moderate probability pure loss lotteries (ML) in 2018 and 2020 experiments. D Proportion of risk averse subjects in low probability pure loss lotteries (LL) in 2018 and 2020 experiments. N = 65 for 2018 experiment, and N = 55 for 2020 experiment.

Full size image

Next, we wanted to better understand how many subjects at the individual level changed their risk attitudes in the two experiments and the direction of their changes in risk preference. Our first experiment included 65 subjects, while our second experiment included 55 subjects. The subjects in these two experiments were not completely identical, and only 39 subjects participated in both experiments. Thus, we wanted to conduct an individual analysis of the changes in the risk attitudes of these 39 subjects. First, we analysed the overall change in the risk attitudes of these 39 subjects in the two experiments. We used nonparametric statistics to analyse the experimental data of the 39 subjects, and the overall change in their risk attitudes was consistent with the results of the full sample (Supplemental Fig. 1 and supplemental Tables 9–12). In other words, after experiencing the lockdown, subjects became more risk seeking in the pure gain lottery experiment under moderate risk and more risk averse in the pure loss lottery. This result supports our individual analysis of the changes in the risk attitudes of these 39 subjects. The results showed that in the MG experiment, 11 subjects changed their risk attitudes from risk aversion to risk seeking after the lockdown, and 1 subject changed from risk seeking to risk aversion (Fig. 4A). In the ML experiment, a total of 12 people changed from risk seeking to risk aversion, and 3 people changed from risk aversion to risk seeking (Fig. 4B). These results show that the changes in subjects’ risk attitudes are not chaotic but rather highly consistent with the direction of the overall level of risk attitude change. We also tested the number of subjects whose risk attitudes changed in the LG and LL experiments. In both groups of experiments, 2 people changed from risk-averse to risk-seeking, and 1 person changed from risk-seeking to risk-averse (Fig. 4C, D). This result shows that the risk attitudes of individual subjects in the LG and LL experiments are also highly stable. This result further verifies the findings of this study, namely, that lockdown has a significant effect on people’s risk preference at the moderate risk level, making individuals more risk-seeking in the gain domain and more risk-averse in the loss domain.

Fig. 4: Analysis of risk preference switch at individual level.
figure 4

A Number of risk preference switch in moderate probability pure gain lotteries (MG) B Number of risk preference switch in low probability pure gain lotteries (LG). C Number of risk preference switch in moderate probability pure loss lotteries (ML) D Number of risk preference switch in low probability pure loss lotteries (LL). Orange column represents number of individual switches from risk aversion to risk seeking, and blue column represents number of individual switches from risk seeking to risk aversion. N = 39.

Full size image

Influence of subject demographic characteristics on changes in risk attitudes

We used data from 39 subjects who participated across both the pre- and post-experiments to examine whether key demographic characteristics (gender, age, education, income) influenced individual-level changes in risk preference in the MG and ML domains. We conducted Pearson correlation analysis for statistical testing, and the results presented in Tables 2–4 show that the absolute values of all the Pearson correlations are less than 3, with corresponding significance levels greater than 0.1. This finding indicates that there is no significant correlation between demographic characteristics such as gender, age, education, income, and behavioural changes in our experiments. In other words, the changes in risk attitudes are not driven by specific demographic groups in our experiments.

Table 2 Descriptive statistics of the variables.
Full size table
Table 3 The influence of demographic differences on Changes in MG.
Full size table
Table 4 The influence of demographic differences on Changes in ML.
Full size table

Discussion

This study investigated whether and how strict epidemic prevention policies, such as the lockdown during the COVID-19 pandemic, affect people’s fourfold risk attitudes. The first-round experiment was carried out in 2018 before the COVID-19 outbreak. The results proved that the subjects’ risk preferences are in line with the theoretical expectations of PT; that is, in the gain domain, people show a risk-seeking attitude towards low-probability prospects and a risk-averse attitude towards moderate-probability prospects. In the loss domain, people are risk averse to small-probability prospects and are risk seeking to medium-probability prospects. This result also supports the PT’s reflection hypothesis; that is, people’s risk attitudes in the loss domain mirror their risk attitudes in the gain domain. We conducted the second round of experiments after a COVID-19 lockdown. Specifically, the subjects were tested immediately after completing their 14-day quarantine due to confirmed cases of COVID-19 in their city. The results revealed that people’s risk attitudes for moderate-probability prospects reversed. In the gain domain, the risk attitude changed from risk averse to risk seeking, whereas in the loss domain, the direction of the change was from risk seeking to risk averse. Notably, this trend was not only observed at the aggregate level but was also consistent at the individual level. Among individuals whose preferences changed across the two experiments, the vast majority (90%) exhibited changes in the same direction as the overall trend. This suggests that the alteration in risk attitudes induced by the lockdown was pervasive. Similarly, a study after the 2008 Wenchuan Earthquake in Sichuan also revealed that people are more risk seeking in the gain domain and more risk averse in the loss domain (Li et al. 2011). Notably, this finding also shows that not only do people’s risk attitudes in the gain and loss domains reflect phenomena, but the dynamic change in their attitudes also maintains this characteristic.

Notably, we analysed natural attributes of our subjects, such as age and gender, and their social characteristics, including income, education, and employment. We found that they are generally representative of the urban employed population in China. This population constitutes the majority of the Chinese population and is the primary driver of social and economic development. Thus, our data can be considered reflective of the behavioural characteristics of this significant and mainstream population in China.

Specifically, in this study, we separated the effects of the pandemic and the lockdown on risk attitudes. Previous research has primarily discussed the impact of COVID-19 itself as a long-term stressor on individuals without meticulously analysing the immediate effects of the special policies and measures adopted during the pandemic period, which are brief but intense stressors. In this study, we focused on the immediate impact of lockdowns, which constitute a rare and strict COVID-19 policy. By comparing the experimental results of two groups—one that experienced quarantine during the pandemic and one that did not—we were able to distinguish between the effects of the pandemic and the lockdown. The experimental results indicate that the 14-day lockdown, rather than the pandemic itself, had a more significant effect on participants’ risk attitudes. This highlights the profound effects of strict COVID-19 prevention policies. In the past, research on the impact of disasters on personal risk attitudes focused mainly on more natural disasters, such as earthquakes and floods. Because large-scale pandemics and related lockdowns are very rare, few studies have evaluated whether they impact people’s risk attitudes. To date, existing studies on this issue have either focused on only gaining domain risk attitudes or used only survey data (Bu et al. 2020; Tsutsui and Tsutsui-Kimura 2022). Thus, this article is the first comprehensive experimental study of the impact of related strict epidemic prevention policies on individuals’ fourfold risk attitudes. Moreover, the results of this research not only provide further experimental support for PT’s fourfold risk preference theory but also prove that the COVID-19 lockdown has had a great effect on people’s risk preferences for moderate-probability events.

The results of research on changes in risk preference after disasters are not consistent. Many studies are consistent with the experimental results of this study; in the gain domain, people’s risk-chasing tendencies are intensified (Eckel et al. 2009; Page et al. 2014; Said et al., 2015; Hanaoka et al. 2018). However, some studies have shown that people are more risk averse in the gain domain after disasters. The conflicting results of the above experiments may be related to the time of the experiment; that is, if the experiment is carried out immediately after the occurrence of a disaster, people will often be more likely to seek risk for gain. For example, a few weeks after a flood, an experiment on affected subjects revealed that they were more risk seeking in the gain domain (Page et al. 2014). After the occurrence of Hurricane Katrina, the experiment was conducted immediately on the subjects who had evacuated to Houston, and it was found that the proportion of subjects who made risky choices increased significantly (Eckel et al. 2009). However, a longer duration between the disaster and the experiment could cause the opposite result. For example, 10–11 months after the Tohoku earthquake, the subjects were more tolerant of risk; that is, they showed stronger risk-seeking behaviours (Hanaoka et al. 2018). Four to five years after the flood in Thailand, experiments with disaster-stricken subjects revealed that they were more risk averse than unaffected subjects in nearby villages were (Cassar et al. 2017). Experiments on Indonesian individuals who have experienced floods or earthquakes in a three-year cycle have also revealed that disasters make people more risk averse (Berg et al. 2009). Dumm (2020) studied flood insurance purchases of homeowners in Florida and reported that subjects’ attitudes towards risk changed over time. In the event of a flood, the near-term effect suggests that homeowners may become more risk averse and adopt protective measures in the immediate aftermath of a flood; over time, if another flood does not occur, the emotional impact of the experience wears off, and homeowners may decide to take the risk and forgo insurance. Thus, it is worth examining whether the influence of disasters on people’s risk attitudes changes over time.

People’s greater risk aversion after a disaster seems to be more intuitively acceptable. Psychologists believe that disasters lead to emotional changes in people, mainly fear and anxiety (Lerner et al. 2003). Such emotional changes make people more averse to potential loss. Another explanation is that natural disasters cause people to suffer losses, both materially and emotionally, especially material losses, which might affect people’s reference point (status quo), thereby affecting their utility function and making them more averse to risky choices in the loss domain (Cameron and Shah 2015). These theories may explain our experimental results in the loss domain, which was influenced by the spread of COVID-19 and the strict quarantine policy, and the subjects became more risk averse in the loss domain.

Conclusion

Under the framework of the fourfold risk attitude theory proposed by prospect theory, this study investigates whether and how strict epidemic prevention policies, such as COVID-19 lockdowns, affect people’s fourfold risk attitudes. The first round of trials was conducted before the COVID-19 outbreak in 2018; the second round of experiments was conducted in November 2020, three days after the subjects were released from quarantine following the COVID-19 outbreak. The experimental findings are as follows:

Lockdown changed the subjects’ attitudes towards risk

We found that before the lockdown, our subjects’ risk attitudes were consistent with the prediction of the FFR of PT theory. They were risk-seeking in a low-probability high-payoff gain lottery and a moderate-probability low-payoff loss lottery but risk averse to a moderate-probability low-payoff gain lottery and a low-probability high-payoff gain lottery loss lottery. However, after the COVID-19 lockdown, their risk attitudes towards moderate-probability pure gain and loss lotteries reversed. This means that under the influence of the COVID-19 lockdown, our subjects became more risk seeking for gain and more risk averse for loss. Notably, our second-round experiment was conducted three days after the city’s lockdown was lifted, at which time the pressure of the pandemic had reached its peak. Thus, our experimental results show that major negative shocks such as a pandemic lockdown can reverse the risk preference of individuals in the medium-probability gain domain and loss domain, causing them to show risk-chasing in the gain domain and risk aversion in the loss domain.

On the basis of the fourfold risk preference theoretical framework of prospect theory, this change presents a more abundant face

The strict lockdown policy during the COVID-19 pandemic changed people’s risk attitudes under moderate probability, making them more risk seeking for gains and more risk averse for losses. In the first round of experiments in 2018, the subjects’ risk preferences are in line with the theoretical expectations of PT; that is, in the gain domain, people show a risk-seeking attitude towards low-probability prospects and a risk-averse attitude towards moderate-probability prospects. In the loss domain, people are risk averse to small-probability prospects and are risk seeking to medium-probability prospects. The results of the experiment in 2020 show that, in the gain experiment, the subjects showed risk-seeking characteristics in low-probability pure gain tasks, which was consistent with the risk attitudes before the lockdown policy. However, experiencing lockdown during the COVID-19 pandemic has changed people’s risky behaviours with moderate probability, increasing their risk seeking in pure gain tasks and risk aversion in pure loss tasks.

Our findings also have some management and policy implications. The findings that COVID-19 and its related lockdown policy strongly influence people’s risk attitudes have important reference value for public policy making in the post-COVID-19 era. For example, the specific actuarial policies of medical insurance and other social insurance may need to be adjusted to ensure that public welfare can be better improved under the condition of greater risk aversion in the public loss domain. Additionally, the changes observed in individual career preferences may be affected by changes in risk attitudes. In line with these changes, innovation and entrepreneurship support policies need to be adjusted. Moreover, in the postpandemic era, whether it is possible to return the public’s risk preference to a more conventional state through better community support policies also needs to be examined further.

Our research has several limitations, and some studies need to be further developed. In our study, the second round of the experiment was conducted immediately after the lifting of the lockdown for our subjects. Thus, we believe that the alteration of the risk attitudes of the participants was caused mainly by this strict intervention, which pushed the tension of COVID-19 to an extreme level. However, in this situation, we cannot rule out the chronic impact of the long-lasting pandemic. Future studies are needed to address this issue.