Studying the prominence effect amid the COVID-19 crisis: implications for public health policy decision-making.

Introduction

In just a few months, coronavirus disease 2019 (COVID-19) has spread globally, taking lives and undermining economies¹. Consequently, the World Health Organization announced a global emergency, and countries have taken numerous measures to fight it: isolation, quarantine, social distancing, etc.².

While these measures have clear health advantages, their effect on economies and the financial situation of individuals has been disastrous³: research has shown that social distancing measures, monetary and public health policy decisions, and the shutdown of international travel severely affected economic activity and the stock market⁴. This tension between health and financial interests has led to numerous debates surrounding one question: how to judge different alternatives (e.g., two different programs to fight the pandemic) when they vary across more than one criterion (e.g., their death toll vs. their financial impact)?

Research in behavioral decision-making has long shown that decision-makers diverge significantly from the rational “homo-economicus” of classic economics⁵. Researchers suggested that uncertainty is a breeding ground for irrational or biased judgments that are based on emotions and heuristics⁶. Such patterns persist even during crucial decisions (e.g., health issues⁷) and those made by professionals (e.g., courtroom judges⁸).

A relevant behavior is people’s tendency to overly focus on a single dominant attribute of objects they judge⁹. One particularly apt bias is the prominence effect - people’s tendency to focus on the most prominent attribute of judged objects instead of weighing all criteria¹⁰. Tversky et al. (1988) presented decision-makers a problem where health and economic considerations were mixed, a paradigm aptly suitable for research into the current situation. Using two decision modes (i.e., choice vs. matching), they showed that people tend to overly focus on the more salient attribute of casualties rather than costs. As detailed below, we used the same scenario amid the COVID-19 crisis with the following changes: the virus instead of traffic accidents and a town in the UK instead of Israel.

We hypothesized that unless decision mode (i.e., matching) or role manipulation deliberately made participants pay more attention to the financial criterion, they would overly emphasize casualties over costs.

Methods

Results and discussion

Choice condition. A binomial test indicated that the proportion of participants in the treasury group who chose program Y (69.4%) over program X (30.6%) was significantly greater than 50% (chance value), p = 0.001 (2-sided; see Figure 1¹¹). There was no significant difference in these proportions for participants in both the health (Y= 59.4%, X= 40.6%) and the control groups (Y= 52.9%, X= 47.1%). Additionally, Two chi-square tests examined the relationship between role and program-choice proportions. Whereas the proportions of health and control groups did not differ significantly, treasury and control groups were significantly different: χ² (1, N = 142) = 4.11, p = 0.043 (see Figure 1¹¹).

Figure 1. Program Choice Percentage.

Figure depicting the percentage of participants who chose each program within each role. *p<0.001, †p<0.05.

Matching condition. A binomial test indicated that the proportions of participants in the treasury (95%), health (97%), and control (98%) groups who thought that 55-million-pound sterling was too high a price were significantly greater than 50% (chance value), p < 0.001 (2-sided).

Triangulating lives saved (in comparison to the 600 default) with the cost of program Y, resulted in the appropriate cost of program X being 40-million-pound sterling. Three single sample t-tests with 40 million pounds as the criterion indicated that participants in all roles thought it too high a price: treasury (M= 15.87, SD= 9.44), health (M= 16.43, SD= 7.65), and control (M= 16.19, SD= 9.73) groups: t (65)= -20.74, p< 0.001; t (67)= -25.38, p< 0.001; and t (60)= -19.10, p< 0.001 respectively. The same analysis using 20 as a stricter criterion yielded similar results in the treasury (t (65)= -3.54, p= 0.001), health (t (67)= -3.84, p< 0.001), and control (t (60)= -3.05, p= 0.003) groups.

At first glance, the results of the choice condition seem to diverge from Tversky et al.’s prominence effect. There, participants – who were not given any roles - had significantly preferred program X (higher costs, fewer casualties), which implied the prominence of saving lives over costs. But, although our results do not show a significant preference for program X in the control group, 50% of participants still chose it – despite its disproportionate cost to the lives-saved ratio. This finding, coinciding with the beginning of the lockdown in the UK (which made the financial issues much more salient and relevant), lends support to the claim that health issues were still the most prominent. To claim that the financial matters were more salient, we would see a significant preference for program Y. Nevertheless, only the treasury condition yielded this result. Coupled with the results of the matching condition below, it seems like an illustration of the prominence of health issues over financial issues in the choice condition.

When matching, participants overwhelmingly preferred the more economical program. Indeed, they thought 55-million-pound sterling was too high of a matching price, and their proposed matching price was significantly lower than 40 and even 20. Comparing these results to the choice condition patterns suggests that “lives saved” was the more prominent attribute for participants who chose program X (approximately 50%). Only under the treasury condition did we see a difference in choice favoring the more economical program, demonstrating the prominence effect.

Taken together, our results support Tversky et al.’s prominence effect study. They suggest that decision-makers should match the options instead of choosing between them in such decision-making situations.

Nevertheless, our study is not without its limitations. First, the costs of both policies were high (in the tens of millions of £) and thus may have been unrelatable to the participants. Future studies should explore the consequences of this limitation by presenting choices with financial costs that are more familiar to participants (e.g. 0.1% of GDP). Second, participants did not have any special expertise in policymaking which, when coupled with the previous limitation, could have resulted in non-robust or generalizable results. Further augmenting this limitation was the fact that the roles (conditions) may have been unrelatable. Thus, futures studies should pool participants with prior professional experience in making these types of decisions in order to increase the generalizability of our results.

Conclusion

Given the highly consequential decisions policymakers must make in times of crisis, it is crucial to understand which decision-making paradigms might lead to better, less biased decisions. The prominence of health over financial issues may lead to irrational differences between choosing versus matching. Thus, changing the decision-making paradigm from simply choosing alternatives to a matching paradigm can aid policymakers to better weigh all attributes.

Data availability