Contextual framing effects on risk aversion assessed using the bomb risk elicitation task

Abstract

We examine the impact of framing on individuals’ risk-taking behavior in the context of health risks during the coronavirus outbreak. We elicit risk attitudes from a sample of 3,385 individuals across seven European countries using an incentivized decision-making task. Participants are randomly assigned to one of three versions of the task: one involving the risk of a bomb explosion, one involving the risk of contracting an infectious disease, and one involving opening an empty box. We find that the framing of the task significantly affects risk-taking behavior, with participants exhibiting greater risk aversion in the health task than in the bomb or neutral task. This framing effect is observed in the majority of the countries studied.

Graphical abstract

1. Introduction

The coronavirus outbreak has highlighted the crucial role of risk-taking behavior in the spread of infectious diseases. Understanding individuals’ attitudes towards risk is essential not only for gaining insights into their decision-making processes but also for designing interventions to limit the spread of disease and promote preventive measures and limit the spread of disease. An important question when designing public policies, however, is whether risk attitudes vary depending on the context in which they are considered. This is particularly important for decisions that have a health impact given their far-reaching consequences.

Attitudes towards risk have been shown to vary across different domains of life. While there is a high correlation between various risk domains, these are far from being perfectly correlated (Dohmen et al., 2011). In the field, Galizzi et al. (2016) found evidence that the risk attitudes of patients may differ between the health and financial domains. Likewise, various laboratory experiments have shown that the context presented to participants matters in decision-making. Health framing, in particular, has been found to affect medical decision-making among providers (e.g., Angerer et al., 2023, Kesternich et al., 2015, Ahlert et al., 2012, Ahlert et al., 2013), as well as the quality of health insurance choice (Kairies-Schwarz et al., 2017) and tax- and exit-rate decisions related to healthcare finance (Buckley et al., 2016) among consumers. These results contrast with classical theories, which assume that the level of risk aversion is stable across different contexts, implying that preferences for different options are unrelated to the underlying economic fundamentals (e.g., Hausman, 2011). Recent research, however, suggests that while risk aversion parameters do indeed appear to have a stable component, they are also dependent on context (Einav et al., 2012, Frey et al., 2017).

Our study aimed to explore whether there are differences in risk aversion between health-related risks and risks in a neutral context among the general adult population. To do so, we used an experimental approach, known as the bomb risk elicitation task (BRET), introduced by Crosetto and Filippin (2013), which is easy for participants to complete and does not require complex numerical skills. The task elicits attitudes based solely on the number of boxes that participants collect. Participants benefit financially from each collected box as long as they avoid selecting the box that contains the bomb, which results in a loss. Additionally, we introduced two other framing conditions to the BRET in order to test whether these would affect participants’ level of risk aversion.

In the first framing condition, we replaced the bomb with a person infected with the novel coronavirus and instructed participants to select people they wanted to meet while avoiding the person with the infection. In the second framing condition, we used a completely neutral frame, asking participants to collect boxes without any mention of bombs or infected people. We found that the health-related risk frame led to a higher degree of risk aversion than the bomb frame.

2. Experimental study

2.1. Risk-elicitation task

We employ a static version of the bomb risk elicitation task (Crosetto and Filippin, 2013) and adapted the approach used by Nielsen (2019). The task requires participants to decide on the number of boxes to collect, with a higher number of opened boxes corresponding to a greater potential financial benefit – but also to an increased risk of incurring a substantial loss. In our experiment, participants were confronted with a set of 25 boxes and received a starting balance of €0.20. For each box they opened without selecting the one containing the bomb, participants received an additional €0.04. However, opening the box containing the bomb resulted in a loss of €0.20.1

2.2. Treatments

In our experiment, we employed a between-subject design with three different treatments that varied in their contextual framing. Each treatment used identical probabilities and monetary incentives. Fig. 1 shows the user interface.

Fig. 1.

User interface: (a) neutral and bomb frame (b) virus frame before making a choice (c) virus frame with an illustrative example of choosing to meet nine people.

Bomb frame. We adapted the BRET applied by Nielsen (2019), which requires participants to select between 0 and 25 boxes while avoiding the box containing the bomb.

Virus frame. Our experimental setting mirrored the decision situation faced by individuals during the coronavirus outbreak, in which meeting others entailed the risk of becoming infected. Participants were asked to choose the number of individuals they wished to meet from a group of 25 people, knowing that one person in the group had already contracted the virus. Meeting non-infected individuals allowed participants to derive financial benefits, but meeting the infected person posed the risk of infection, incurring a financial loss and nullifying any financial benefits gained from meeting the others. This decision scenario presented participants with a trade-off between the potential benefits of social interaction and the risk of contracting a potentially life-threatening disease.2

Neutral frame. As with the bomb frame, participants in this treatment were required to open boxes. In this case, however, one of the boxes was empty while the rest were full. Participants benefitted financially from opening full boxes but incurred a loss if they opened the empty one. We included this frame to ensure the validity of our findings, considering the potential for the word “bomb” to influence participants’ choices given the extensive media coverage of the Russian invasion of Ukraine during the survey period as well as the geographic proximity of Ukraine to the survey participants.

2.3. Procedure

The experiment was part of the European COVID Survey (ECOS), and the samples of participants were from Denmark, France, Germany, Italy, Netherlands, Portugal, and the United Kingdom. After preregistration (on aspredicted.org, #96 020), data collection for the experiment took place in May 2022 as part of the tenth wave of the survey. The experiment was programmed with Qualtrics. We restricted the age range to 18 to 54 years and aimed to recruit 160 participants per treatment condition and country. A total of 3385 participants (56.87% female) were recruited by the market research company Dynata and were randomly assigned to one of the three treatments. After the participants were provided with instructions, they were asked four comprehension questions to familiarize them with their task (see online Appendix). These questions were followed by the main task. The field with the bomb/infected person/empty box was randomly allocated. After the participants made their selections, they were promptly informed about the outcome and the resulting payment.

2.4. Risk attitudes

Risk attitudes can be assessed by examining the number of selected fields. Participants have a probability of $p=\frac{25-k}{25}$ to gain €$0.04\ast k$ for each selected field, where $k\in (0,1,\dots ,25)$. Conversely, they face the possibility of a loss of €0.20 with a probability of $(1-p)=\frac{k}{25}$. The expected value $E\left(k\right)=p\ast 0.04\ast k+(1-p)\ast (-0.2)$ reaches its maximum when $k=10$, indicating a risk-averse participant would select fewer than 10 fields.

2.5. Hypotheses

In contrast to classical theories, which assume that risk aversion remains stable across the various domains of life, we posit that health framing may have a different impact than a neutral one on participants’ decisions. This is because decisions made under risk in a neutral context can be expected to be based solely on the monetary payoff, whereas those made in a health context additionally involve potential (hypothetical) health consequences. This assertion is supported by previous experimental evidence (e.g., Ahlert et al., 2012, Ahlert et al., 2013, Angerer et al., 2023).

H1 Individuals exhibit a higher degree of risk aversion with virus framing compared to the neutral/bomb framing.

H2 Individuals exhibit a higher degree of risk aversion with bomb framing compared to neutral framing.

Considering the substantial impact of both the coronavirus outbreak and the Russian invasion of Ukraine on European nations during the relevant period, we expect:

H3 The framing effects are comparable across countries.

Overall, this gives us the following expectation:

$$\forall i:k_{virus,i}<k_{bomb,i}<k_{neutral,i}$$

with $k$ as the number of opened boxes/people met and $i$ as countries.

3. Results

Fig. 2 shows the mean number of fields selected by participants in each of the three different frames. In the virus frame, this corresponds to the number of people met, whereas the values for the bomb and neutral frames indicate the number of opened boxes. Overall, the mean number of fields selected by participants was below 10, which suggests risk-averse behavior across all three frames on the average.

Fig. 2.

Treatment comparison.

We observed significant treatment effects, with the number of selected fields differing across at least two frames (Kruskal–Wallis test, ${\chi }^2=34.628$, $p<0.001$). We found support for our first hypothesis (H1) because, on average, participants met fewer people ($k_{virus}=5.54$, $SD=5.43$) compared to the number of boxes they opened ($k_{bomb}=6.36$, $SD=5.33$). This difference was statistically significant ($p<0.001$).

Result 1 The virus frame elicited a higher degree of risk aversion than the bomb frame.

Although participants opened fewer boxes in the neutral frame than in the bomb frame ($k_{neutral}=6.46$, $SD=5.34$), this difference was not statistically significant ($p=0.648$).

Result 2 We did not find significant differences in risk-taking between the bomb and neutral frames.

Comparing the seven countries (Fig. 3) revealed notable variations in the levels of risk aversion (${\chi }^2=64.080$, $p<0.001$), with Denmark exhibiting the lowest ($k_{DK}=7.44$, $SD=6.20$) and France the highest level ($k_{DE}=5.22$, $SD=5.23$) when considering all treatments together. As predicted, the majority of countries (France, Italy, Portugal, and the UK) displayed the expected pattern, with the highest risk aversion observed in the virus frame and the lowest risk aversion in the neutral frame. With the exception of Denmark, all countries exhibited the lowest risk aversion in the virus frame. In the case of Germany, the Netherlands, and the UK, we observed significantly more risk-averse behavior in the health frame compared to the bomb frame. In addition, France, Italy, the Netherlands, and the UK demonstrated a significantly lower risk aversion in the health frame compared to the neutral frame (see Table 1). However, no significant difference was observed between the bomb frame and the neutral frame for any country. Overall, it is evident that participants’ behavior varied across the countries.

Fig. 3.

Treatment and country comparison.

Result 3 Variation in framing effects is evident across the countries.

Table 1.

Risk taking by country and treatment.

Country	Virus	Bomb	Neutral	$p$-value	$p$-value	$p$-value	${\chi }_i^2$
$i$	$k_{v,i}$	$k_{b,i}$	$k_{n,i}$	$k_{v,i}=k_{b,i}$	$k_{v,i}=k_{n,i}$	$k_{b,i}=k_{n,i}$
DE	4.88	6.07	5.25	0.009	0.130	0.231	7.114
(N $=$ 481)	(4.90)	(5.05)	(4.23)				p $=$ 0.029
DK	7.49	7.51	7.33	0.626	0.662	0.988	0.284
(N $=$ 473)	(6.76)	(6.12)	(5.73)				p $=$ 0.868
FR	4.70	5.10	5.83	0.236	0.018	0.202	5.802
(N $=$ 484)	(5.13)	(5.06)	(5.46)				p $=$ 0.055
IT	5.11	5.98	6.53	0.134	0.007	0.183	7.529
(N $=$ 476)	(4.96)	(5.53)	(5.26)				p $=$ 0.023
NL	5.66	7.01	6.65	0.007	0.043	0.317	8.307
(N $=$ 480)	(5.19)	(5.23)	(5.35)				p $=$ 0.016
PT	6.04	7.02	7.12	0.058	0.063	0.960	4.696
(N $=$ 485)	(5.06)	(5.09)	(5.33)				p $=$ 0.096
UK	4.96	5.85	6.50	0.009	0.002	0.488	11.154
(N $=$ 506)	(5.40)	(4.84)	(5.70)				p $=$ 0.004
all	5.54	6.36	6.46	0.000	0.000	0.648	34.408
(N $=$ 3385)	(5.43)	(5.33)	(5.34)				p $=$ 0.000

Notes:$k$: Number of selected fields per treatment and country $i$; standard deviation in parentheses. The three columns of $p$-values indicate the results of the Mann–Whitney-U test. ${\chi }^2$ and the corresponding $p$-value indicate the results of the Kruskal–Wallis test, testing for differences between the three frames.

4. Discussion and conclusion

We conducted an online experiment using variations of the bomb risk elicitation task (BRET) to determine the effect of contextual framing on risk attitudes during the coronavirus outbreak.

Our results indicate that the virus frame elicited a higher level of risk aversion than the bomb frame, and that there was no significant difference in this regard between the bomb frame and the neutral frame. These findings suggest that contextual framing, particularly related to health outcomes and the pandemic, may influence individuals’ risk aversion. Our results are in concordance with a growing body of evidence demonstrating that risk aversion can vary depending on the specific domain, suggesting that researchers should try to calibrate their models by using preference estimates that are taken from similar contexts (e.g., Barseghyan et al., 2011, Einav et al., 2012).

Furthermore, our findings indicate that participants exhibited a level of risk aversion in the bomb treatment that is consistent with the findings of Nielsen (2019). Thus, it appears that the coronavirus outbreak did not affect the level of risk aversion in the classical bomb setting. These findings are in line with those of Angrisani et al. (2020), who also used the BRET as a measure of risk aversion and compared the behavior of professional traders before and during the pandemic.

Appendix A. Supplementary data

The following is the Supplementary material related to this article.

MMC S1

.

Data availability

Data will be made available on request.