Discussion Week 4

Social Science & Medicine 282 (2021) 114171

Available online 21 June 2021 0277-9536/© 2021 Elsevier Ltd. All rights reserved.

Awareness of ethical dilemmas enhances public support for the principle of saving more lives in the United States: A survey experiment based on ethical allocation of scarce ventilators

Birendra Rai a, Liang Choon Wang a,*, Simone Pandit a, Toby Handfield b, Chiu Ki So a

a Department of Economics, Monash University, Clayton, Victoria, 3800, Australia b School of Philosophical, Historical, and International Studies, Monash University, Clayton, Victoria, 3800, Australia

A R T I C L E I N F O

Keywords: The principle of saving more lives Allocation of scarce medical resources Ethical dilemmas faced by public health experts Trust in science and experts

A B S T R A C T

Recommendations by health experts to deal with public health emergencies are primarily guided by the principle of “saving more lives”. It is unclear whether people perceive this principle as ethically more legitimate than some other principle such as “saving more life-years”. Understanding the answer to this question is particularly relevant to the allocation of scarce medical resources during public health emergencies. Different principles typically lead to different allocations, and consequently have dramatically different implications as to who survives and who dies. We fielded an online randomized controlled survey experiment in the context of scarce ventilator allocation with a demographically representative sample of US adults (n = 700) from October 22 to October 30, 2020. Participants faced hypothetical situations where they had to allocate few available ventilators among several needy patients. The experiment was designed such that the allocation decision made by a participant can be used to infer the principle in line with their personal ethical values. We interpret this inferred principle as the one that the participant perceives to be most legitimate. The treatment group, but not the control group, was provided balanced information that described the ethical dilemmas faced by experts in developing ventilator allocation guidelines. Nearly half of the participants in the control group perceive saving more lives the most legitimate principle. Despite the balanced nature of the information, the perceived legitimacy of saving more lives was 7⋅6 percentage points higher in the treatment group. The magnitude of this impact was partic- ularly strong among republican-leaning participants, a subgroup that has less trust in experts according to previous research. Our findings suggest that enhancing public awareness of ethical dilemmas faced by health experts can increase the perceived legitimacy of their proposed guidelines even among those with lower trust in experts.

1. Introduction

Recommendations by public health experts to deal with public health emergencies, including the Covid-19 pandemic, are primarily guided by the principle of saving more lives (SML). For example, social distancing guidelines, mandatory mask wearing, and lockdown mea- sures to “flatten the epidemic curve” are primarily motivated by SML (Sen-Crowe et al., 2020; Iacobucci, 2020). The guidelines for allocating scarce life-saving medical resources such as ventilators and intensive care beds also are primarily, if not exclusively, guided by SML (Piscitello et al., 2020; Emanuel et al., 2020; Truog et al., 2020; White and Lo, 2020). However, it is unclear whether the majority of the general public

perceives SML to be the most legitimate principle. One prominent alternative is the principle of saving more life-years (SMLY). For instance, there have been reports of hospitals refusing admission to elderly patients during the ongoing pandemic despite the availability of hospital beds, perhaps in anticipation that a younger patient might soon need one (Amnesty International, 2020). Such examples highlight the simple fact that differences in the principles that guide allocation of scarce medical resources create substantive differences in terms of who lives and who dies.

This study investigated which principle is perceived to be most legitimate by the general public for allocating scarce ventilators during public health emergencies. We focused on three principles – saving more

* Corresponding author. E-mail address: [email protected] (L.C. Wang).

Contents lists available at ScienceDirect

Social Science & Medicine

journal homepage: www.elsevier.com/locate/socscimed

https://doi.org/10.1016/j.socscimed.2021.114171 Received 4 March 2021; Received in revised form 13 June 2021; Accepted 19 June 2021

Social Science & Medicine 282 (2021) 114171

2

lives (SML), saving more life-years (SMLY), and saving healthier lives (SHL) – that are prominently discussed in ventilator allocation guide- lines (Piscitello et al., 2020; Emanuel et al., 2020; Truog et al., 2020; White and Lo, 2020). SML recommends allocating scarce resources in order to maximize the expected number of lives that can be saved. SMLY seeks to maximize the expected number of life-years that can be saved. SHL essentially recommends using SML as the primary principle; and, using SMLY as the subordinate tie-breaking principle if multiple de- cisions are consistent with SML. SHL is thus a refinement of SML. A decision consistent with SHL is consistent with SML, but typically inconsistent with using SMLY alone.

Ventilator allocation guidelines are formulated by expert committees after careful deliberation. The guidelines specify which patient charac- teristics (e.g., age, occupation, and comorbidities) should be used to determine priority orderings in cases of scarcity. The general public is likely unaware of the arguments and counterarguments regarding the use of these patient characteristics, and thus may not fully appreciate the ethical dilemmas that health experts have to grapple with in formulating the guidelines. Our experiment primarily focused on investigating the impact of balanced information about the ethical dilemmas faced by health experts on the perceived legitimacy of the underlying principles among the general public. We are not aware of any previous study aimed towards exploring this question.

During the main part of the experiment, participants faced hypo- thetical situations and had to decide who among several needy patients should receive the few available ventilators. The experiment was designed such that allocation decisions made by participants permit us to infer the principles behind their decisions. We assumed that a participant whose decisions are consistent with a particular principle perceives that principle to be more legitimate than other principles. As SML and SMLY are the two core principles, the situations presented to the participants were designed to ensure that decisions consistent with SML differ from those consistent with SMLY. The primary objective was to understand the impact of balanced information on the perceived legitimacy of these principles. The treatment group was provided this balanced information, while the control group was not.

Our interest in investigating the legitimacy of various principles is grounded in both pragmatic and principled reasons. For instance, the Massachusetts state guidelines for allocating critical care in response to Covid-19 that were first released on April 7, 2020 had to be revised within two weeks following a public outcry (After Uproar and Mass., 2020). Further, scholars across several academic disciplines have argued that the more the general public understands and accepts the arguments behind policies and guidelines, the greater their legitimacy among the public (JusticeHealthHealthcare, 2001; Rawls, 2005). In fact, some existing guidelines explicitly call for efforts to publicly justify the guidelines to enhance their perceived legitimacy among the general public. For instance, the New York state guidelines for the allocation of scarce ventilators state that “[I]mplementation of the guidelines re- quires clear communication to the public about the goals and steps of the clinical ventilator allocation protocol.”

At a more general but closely related level, prior studies across a variety of contexts have shown that people are more likely to comply with official guidelines if those recommendations are derived from principles that people perceive to be legitimate (Murphy, 2004; Jackson et al., 2012). Hence, while our experiment was conducted in the context of ventilator allocation, the findings may hint towards a broader mes- sage. It is possible that in many contexts increasing awareness of the ethical dilemmas that experts grapple with in formulating the relevant recommendations can increase compliance with those recommenda- tions. In the context of the Covid-19 pandemic, experts have largely relied on the unbiased communication of scientifically accurate infor- mation to justify their recommendations regarding social distancing and mask wearing. While our study is admittedly narrow, the broader motivation was to inquire whether ethical awareness can complement scientific information.

Before proceeding we would like to clarify that our investigation was largely exploratory. It would be internally inconsistent on our part to claim that (a) our goal is to assess the impact of balanced information, and yet (b) confidently hypothesize the direction of its impact. For instance, it is unclear whether support for saving more lives (SML) will be greater among older participants in the treatment (“Info”) group or in the control (“No Info”) group. Balanced information containing the competing ethical principles may lead some older participants in the treatment group to better appreciate that SML discriminates against the young. Other older participants in the treatment group may view the balanced information as providing, on the net, additional arguments as to why SML is the more legitimate principle. This argument applies for all population subgroups, and thus prevents us from hypothesizing about the impact of balanced information not only at the subgroup level but also at the aggregate level.

2. Methods

2.1. Overview

Ventilator allocation protocols typically have four components: (i) which patients are to be categorically excluded, (ii) how the eligible patients are to be prioritized, (iii) what tie-breaking criteria to use within the same priority ranking, and (iv) when to withdraw the ventilator from a patient (Piscitello et al., 2020; Emanuel et al., 2020; Truog et al., 2020; White and Lo, 2020). Our design implicitly accounted for the first three components and ignored withdrawal of ventilators.

All the 700 participants in our experiment were first asked a series of questions regarding their exposure to Covid-19, their willingness to accept a vaccine for Covid-19 that has been deemed safe and offered for free by health authorities, and familiarity with ventilator allocation guidelines. Then, the kind of situations they would face during the main part of the experiment was explained in ten steps, and an incentivized quiz question was included in each step. We are confident participants understood the situations they were going to face during the main experiment as the average number of correct quiz responses out of ten was 9.3 in the control group and 9.4 in the treatment group.

Participants that were randomly assigned to the treatment group (n = 355) were then provided balanced information about the competing ethical considerations before moving to the main part of the experiment. In contrast, participants that were randomly assigned to the control group (n = 345) did not receive this information, and moved to the main part of the experiment after the explanation stage.

In the main part of the experiment, participants faced hypothetical situations where several patients have contracted Covid-19 and need a ventilator to survive, but there is a shortage of ventilators. Participants had to choose a subset of patients who should receive the available ventilators. Finally, subjects answered a post-experiment survey.

2.2. Study sample

The experiment was conducted on the online platform Prolific that has approximately 36,000 USA-based participants who have registered to participate in research studies. At the time we conducted the exper- iment, there were approximately 25,000 eligible US participants who were deemed active (participated in a study in the past 90 days) and provided sufficient information about their characteristics for us to obtain a demographically representative sample. 722 respondents initiated our study, and 22 did not complete the study. While initiating the study, participants did not know whether they are in the treatment or control group. Incomplete responses are not due to assignment to the treatment or the control group. Our analysis is based on the 700 re- spondents who provided complete responses, and we refer to these re- spondents as the participants. The sample of participants was slightly more educated and slightly younger than the general US adult popula- tion but otherwise demographically similar. Further, there is no

B. Rai et al.

Social Science & Medicine 282 (2021) 114171

3

statistical difference between the participants in the control and treat- ment groups on a variety of demographic and socioeconomic di- mensions, prior familiarity with ventilator allocation guidelines, and attitudinal dimensions such as vaccine hesitancy and political orienta- tion (see Table S2 in Supplementary Material).

2.3. Information treatment

Before making decisions in the choice experiment, participants in the treatment group were informed about the competing ethical consider- ations that public health experts consider in developing guidelines for the allocation of scarce ventilators. This information contained argu- ments and counterarguments for using four patient characteristics – age, occupation, comorbidities, and survival chances conditional on receiving a ventilator – that are prominently discussed in ventilator allocation guidelines (Piscitello et al., 2020; Emanuel et al., 2020; Truog et al., 2020; White and Lo, 2020).

The arguments and counterarguments were not intended to prime the participants to favor any particular principle, but to instead increase their awareness of the ethical dilemmas that experts have to grapple with in formulating the guidelines. For instance, an argument for prioritizing younger patients over older patients is to provide them a fair chance to experience all stages of life. An obvious counterargument is the clear discrimination against the elderly. Table 1 provides the exact arguments and counterarguments provided to the treatment group regarding all the four patient characteristics.

2.4. The main choice experiment

The main experiment involved two stages: the “personal preference” stage and the “social agreement” stage. All participants faced the same five choice situations in both stages. The situations were presented in the same order across both stages to all participants.

In the personal preference stage, participants had to choose a subset of patients who should receive the scarce ventilators based on their own preferences. We refer to the subset of patients chosen by a participant in a situation as the participant’s “decision” in that situation. The personal decisions help identify which principle is personally most preferred. In the social agreement stage, participants played an incentivized

coordination game (Cooper et al., 1990). For each situation in the social agreement stage, a participant earned a monetary bonus if their patient choices in a situation matched the most frequent patient choices among all the participants in that situation. The coordination decisions help identify which principle is socially focal, i.e., the most likely principle upon which social agreement can emerge despite differences in personal preferences.

We designed the five choice situations to ensure that the decisions consistent with SML in any situation are different from the decisions consistent with SMLY in that situation. Each situation involved eight patients who had contracted Covid-19: two were 5-year-old children, two were 45-year-old doctors, two were 45-year-old general adults working in a sector other than healthcare, and two were 75-year-old elderly. It was explained to the participants that every patient needed a ventilator to survive Covid-19, but there was a shortage of ventilators. The participants were told to assume any patient who did not receive a ventilator was expected to die within a week.

The patients differed in their chances of surviving Covid-19 condi- tional on receiving a ventilator. One patient in each age/occupation category had a relatively lower conditional survival chance (60%) and the other had a relatively higher conditional survival chance (90%). The conditional survival chances were used as a proxy for Sequential Organ failure Assessment scores and other such measures that are used in ventilator allocation guidelines (Piscitello et al., 2020). It was explained to the participants that these chances indicate the very near-term chances of surviving Covid-19 upon receiving a ventilator. Partici- pants were told to assume any patient who survived Covid-19 upon receiving a ventilator was expected to recover fully from Covid-19 and get discharged from the hospital within one month.

2.5. The choice situations

The five choice situations differed in the number of available venti- lators and which patients had comorbidities (Table 2). Incorporating decreased life-expectancy due to comorbidities is a challenging but important component of ventilator allocation guidelines (Piscitello et al., 2020; Emanuel et al., 2020; Truog et al., 2020; White and Lo, 2020). Presence of comorbidities was explained as affecting survival chances of patients beyond the immediate episode. While comorbidities

Table 1 Arguments for and counterarguments against using patient characteristics to prioritize the allocation of scarce ventilators.

Characteristic Argument Counter-argument

Young age An argument for prioritizing younger patients over older patients is that it allows the young to have a fair chance to experience all stages of life, and saves more life-years.

A counter-argument is that this clearly discriminates against the elderly and may not account for economic or public health considerations. It is debatable whether the right to life of the elderly can be ignored to satisfy the right to a long life of children and adults. Further, from an economic and public health perspective, general adults and healthcare workers may be more likely to keep the economy and the healthcare system functioning.

Healthcare worker

An argument for prioritizing healthcare workers is that society should provide some insurance to healthcare workers for the risks they face in caring for others. One way to do this would be to prioritize healthcare workers for receiving ventilators, if they contract Covid-19 while caring for others.

A counter-argument is that healthcare workers (i) should have priority in receiving Personal Protective Equipment so that they do not get sick, but (ii) if they get sick while working, then they should be regarded as a general member of the population and should not have priority in receiving ventilators. • This counter-argument highlights the difficulties in clearly distinguishing

between “critical” and “non-critical” healthcare workers, since the smooth operation of health services during a pandemic relies on a whole range of workers in the healthcare sector, not just doctors and nurses. In addition, one may ask, why not prioritize “critical” workers in occupations other than healthcare. For example, adults who are not healthcare workers may contribute significantly to keep the economy going during a pandemic.

No comorbidities

An argument for prioritizing patients without co-morbidities is that if ventilators are given to patients with co-morbidities, then scarce medical resources will get wasted because patients with co-morbidities are unlikely to live long even if they fully recover from Covid-19.

A counter-argument is that this discriminates against certain groups (e.g., some ethnic or racial groups, and some income groups) who may be systematically more likely to suffer from chronic illnesses.

High survival chances

An argument for allocating ventilators solely on the basis of who is more likely to survive Covid-19 if they receive a ventilator is that a greater number of people are likely to be saved.

A counter-argument is that this may implicitly discriminate against some groups of people. For example, people from low socio-economic backgrounds may be much more likely to get seriously sick if they contract Covid-19. This could be due to the poor environment they live in, the high-risk occupations they work in, and lack of nutrition or health services.

B. Rai et al.

Social Science & Medicine 282 (2021) 114171

4

and the immediate conditional survival chances are likely correlated, our experimental design manipulates them as independent factors.

In situation NONE, four ventilators were available and no patient had comorbidities. In each of the remaining four situations – which we label as ELDERLY, ADULT, DOCTOR, and CHILD – only three ventilators were available and both the patients in a particular age/occupation category had comorbidities. For instance, in situation ELDERLY the two elderly patients had comorbidities, while the remaining six patients did not have comorbidities.

Participants were told to assume that, conditional on receiving a ventilator and surviving Covid-19, a patient without comorbidities was expected to live for the remainder of their natural term of life (specified as 80 years of age). A patient with comorbidities was described as someone expected to die within two years even if they fully recovered from Covid-19 upon receiving a ventilator.

2.6. Key design choices

We made two key design choices. First, we ignored situations involving withdrawing a ventilator from a patient in order to allocate it

to another patient. Second, we specified the remaining life-expectancy of patients with comorbidities to be the relatively short period of two years. In the following, we describe the rationale behind these design choices.

Withdrawing a ventilator raises fundamentally different ethical questions relative to withholding (i.e., not allocating in the first place): withdrawal may be viewed as active killing, while withholding as pas- sive killing (Cartwright, 1996; Chu et al., 2020). The differences relate to acts of commission versus acts of omission. Previous research has documented that this difference substantially influences moral and legal judgments across a wide range of settings (Spranca et al., 1991; Feldman et al., 2020). We were concerned participants may be induced to think about both aspects – omission and commission – while making choices in a situation where only one is relevant.

In light of these considerations we chose to focus on withholding rather than withdrawing because withholding decisions have to be made much more often by healthcare professionals. Further, most guidelines view withdrawal as the more difficult decision. In fact, all publicly available guidelines contain details regarding withholding ventilators but some do not contain any information about withdrawing ventilators (Piscitello et al., 2020). Finally, we were concerned that situations involving withdrawal decisions might be more disturbing for the par- ticipants, and may thus cause greater attrition.

We specified the remaining life-expectancy of patients with comor- bidities to be the relatively short period of two years. There is significant variation in ventilator allocation guidelines across states in the USA on this matter. Four states in the USA use life-expectancy of less than 6 months as an exclusion criterion, but it can range up to fifteen years in some other states (Piscitello et al., 2020). A detailed explanation of this design choice is included in the Supplementary Material. Here we simply highlight that even a participant who finds SML the most legitimate principle may hesitate in allocating the ventilator to a patient with comorbidities if it can be allocated to a patient without comorbidities. Hence, the shorter the life-expectancy for patients with comorbidities, the greater the implicit force against SML. Consequently, if we find that balanced information increases support for SML despite this implicit force against SML, then we can be reasonably confident that balanced information indeed increases support for SML. We chose to trade off some realism for greater confidence in the potential impact of balanced information.

2.7. Measures

A participant’s choice of patients in a situation allows us to infer whether the decision was consistent with SML, SMLY, or inconsistent with both SML and SMLY (see Table 2). For instance, 70 different de- cisions are feasible in situation NONE, and they correspond to the 70 different ways in which four people can be chosen out of eight to receive the available ventilators. Out of these 70 decisions, the unique decision consistent with SML involves choosing the four patients with higher conditional survival chance. Given that the maximum conditional sur- vival chance is 90%, the maximum expected number of lives saved is 3⋅6 in situation NONE and 2⋅7 in each of the four situations other than NONE. In contrast, the unique decision consistent with SMLY involves choosing both the children, and the doctor and the general adult with higher conditional survival chance, because the expected number of life- years saved corresponding to a decision is calculated by weighting the remaining life-years of the chosen patients with their conditional sur- vival chances.

Arguably, even people who are guided by SML may prefer to save “healthier lives” when multiple decisions are consistent with SML, especially under severe resource constraints. The ventilator allocation guidelines in some states (e.g., Colorado, Michigan and Pennsylvania) effectively suggest using SML as the primary principle, and SMLY as the subordinate principle if multiple decisions are consistent with SML (Piscitello et al., 2020; Subject Matter Experts Ad, 2020; Michigan

Table 2 Features of the five choice situations faced by the participants.

Feature Situation

NONE ELDERLY ADULT DOCTOR CHILD

No. of ventilators 4 3 3 3 3 No. of patients 8 8 8 8 8 Patients with 60%

conditional survival chance

E1, A1, D1, C1

E1, A1, D1, C1

E1, A1, D1, C1

E1, A1, D1, C1

E1, A1, D1, C1

Patients with 90% conditional survival chance

E2, A2, D2, C2

E2, A2, D2, C2

E2, A2, D2, C2

E2, A2, D2, C2

E2, A2, D2, C2

No. of feasible decisions

70 56 56 56 56

Patients with comorbidities

None E1 and E2 A1 and A2

D1 and D2 C1 and C2

No. of decisions consistent with SML

1 out of 70

4 out of 56

4 out of 56

4 out of 56

4 out of 56

Decisions consistent with SML

E2, A2, D2, C2

A2, D2, C2; A2, D2, C2;

A2, D2, C2;

A2, D2, C2;

E2, D2, C2; E2, D2, C2;

E2, D2, C2; E2, D2, C2;

E2, A2, C2; E2, A2, C2;

E2, A2, C2; E2, A2, C2;

E2, A2, D2 E2, A2, D2

E2, A2, D2 E2, A2, D2

No. of decisions consistent with SMLY

1 out of 70

2 out of 56

1 out of 56

1 out of 56

2 out of 56

Decisions consistent with SMLY

A2, D2, C1, C2

A2, C1, C2; D2, C1, C2

A2, C1, C2 A1, A2, D2;

D2, C1, C2 A2, D1, D2

Note. The easiest way to read this table is to fix attention on a particular column. For instance, consider the column labelled ELDERLY that refers to Situation ELDERLY presented to the participants. As we move down the ELDERLY column, we find information about the relevant features of Situation ELDERLY: 3 ven- tilators are available; there are 8 patients who need a ventilator to survive; 4 patients – E1, A1, D1, and C1 – have 60% survival chance if they receive a ventilator; the remaining 4 patients – E2, A2, D2, and C2 – have 90% survival chance if they receive a ventilator; two patients – E1 and E2 – have comorbidities; there exist a total of 56 different ways of allocating the 3 available ventilators among the 8 patients; only 4 out of these 56 decisions are consistent with the principle of saving more lives (SML); the 3 patients who receive the ventilators in the 4 decisions consistent with SML are reported and the 1 decision that is consistent with the principle of saving healthier lives (SHL) is italicized; only 2 out of the 56 decisions are consistent with the principle of saving more live-years (SMLY); the 3 patients who receive the ventilators in the 2 decisions consistent with SMLY are reported. All the other columns can be read analogously.

B. Rai et al.

Social Science & Medicine 282 (2021) 114171

5

Department of Community Health and Office of Public Health Pre- paredness, 2012; Pennsylvania Department of Health, 2020). The four situations other than situation NONE allowed us to investigate this possibility.

There are 56 feasible decisions in each of these situations, and they correspond to the 56 different ways in which three patients can be chosen out of eight. Four out of these 56 decisions are consistent with SML (see Table 2). These four decisions involve choosing any three out of the four patients with higher conditional survival chance. Among these four decisions consistent with SML, the decision that involves choosing the three patients without comorbidities is deemed consistent with “saving healthier lives” (SHL). We label the remaining three de- cisions that are consistent with SML but not with SHL as being consistent with “saving any lives” (SAL).

There is a fundamental difference between decisions guided by SHL versus those guided primarily by SMLY. For instance, SMLY permits categorical exclusion of patients with limited remaining years of life (e. g., the very elderly and patients with severe comorbidities) from receiving ventilators. In contrast, categorical exclusion is prohibited by SHL because it utilizes SML as the primary principle, and SMLY as the subordinate tie-breaking principle when multiple decisions are consis- tent with SML. Table 2 highlights that decisions consistent with SHL are different from the decisions consistent with SMLY in all the four situa- tions other than situation NONE. The decision consistent with SHL is the one that has been italicized among the decisions consistent with SML.

2.8. Statistical analyses

Our analysis focused on the principles described above: SML, SHL, SAL, and SMLY. In order to identify the causal effect of information about competing ethical considerations on the share of decisions consistent with a principle p, we estimated the following Ordinary Least Squares regression model (which was pre-specified in the analysis plan):

Yijp = a + b Informationi + εij (1)

where the outcome variable Yijp equals 1 if the decision by participant i in situation j is consistent with principle p and 0 otherwise; Informationi is an indicator variable for the participant being in the treatment group; a and b are coefficients to be estimated; and εij is a random error term.

The estimated value of a provides the share of decisions in the con- trol group that are consistent with principle p. The estimated value of b captures the impact of information about the competing ethical con- siderations by providing the difference in the share of decisions consistent with principle p between the treatment and the control groups. We estimated equation (1) separately for the personal decisions and the coordination decisions. Standard errors are clustered at the participant level since each participant made five personal decisions and five coordination decisions. When reporting the results, we also report the odds ratio between the treatment and the control groups that was estimated by a logistic regression.

For the subgroup analysis, we estimated equation (1) across popu- lation subgroups by focusing on three factors: (i) differential vulnera- bility to Covid-19, as captured by differences in age and race (Mueller et al., 2020; Hooper et al., 2020; Price-Haywood et al., 2020); (ii) po- tential differences in prior awareness of the competing ethical consid- erations, as captured by differences in educational attainment and prior familiarity with ventilator allocation guidelines; and, (iii) potential differences in sources of public health information as captured by dif- ferences in the willingness to take a freely available Covid-19 vaccine deemed safe by health experts and differences in political orientation (Motta et al., 2018; Figueiredo et al., 2020; Ecker and Ang, 2019; Mitchell et al., 2014).

3. Results

We first examine whether any particular principle received majority support. Each decision by every participant in all the five situations was categorized into one of three mutually exclusive and exhaustive cate- gories: consistent with SML, consistent with SMLY, and other (which was the residual category containing decisions inconsistent with both SML and SMLY). We shall refer to treatment and control groups as “Info” and “No Info” groups, respectively.

3.1. Saving lives versus saving life-years

We first discuss the findings in the No Info group. Fig. 1A shows the share of personal and coordination decisions consistent with SML and SMLY across all the five situations, and the impact of information about the competing ethical considerations. The underlying regression results are presented in Table 3A. The main finding is that decisions consistent with SML are the modal personal decision and the modal coordination decision among participants in the No Info group. The share of personal decisions consistent with SML is 0⋅50 (95% CI, 0⋅45 to 0⋅54) while the share of personal decisions consistent with SMLY is 0⋅31 (95% CI, 0⋅28 to 0⋅33).

For coordination decisions in the No Info group, the share of de- cisions consistent with SML is 0⋅47 (95% CI, 0⋅42 to 0⋅52), while the share of decisions consistent with SMLY is 0⋅35 (95% CI, 0⋅31 to 0⋅39). Overall, participants in the No Info group seem to perceive SML more legitimate than SMLY. However, for almost half of the participants SML is not the most legitimate principle.

3.2. Impact of balanced information

The left panel in Fig. 1A shows that balanced information about the competing ethical considerations increases personal decisions consistent with SML by 7⋅6 percentage-points (95% CI, 0⋅01 to 0⋅14; P = 0⋅02). This increase comes at the expense of personal decisions consistent with SMLY which decline by 6 percentage-points (95% CI, − 0⋅11 to − 0⋅01; P = 0⋅02).

The right panel in Fig. 1A shows that balanced information also in- creases the likelihood of SML being the socially focal principle. The share of coordination decisions consistent with SML in the Info group is 5⋅7 percentage-points higher than that in the No Info group (95% CI, − 0⋅01 to 0⋅12; P = 0⋅08).

Comparing the left and the right panels in Fig. 1A highlights that among the participants in the No Info group, the share of coordination decisions consistent with SML is 3⋅0 percentage-points lower (95% CI, − 0⋅06 to 0⋅001; P = 0⋅06) than the share of personal decisions consistent with SML. Similarly, among the participants in the Info group, the share of coordination decisions consistent with SML is 4⋅8 percentage-points lower (95% CI, − 0⋅08 to − 0⋅01; P = 0⋅005) than the share of personal decisions consistent with SML.

The comparison of personal and coordination decisions suggests as if some participants who personally perceive SML to be the most legiti- mate principle are doubtful whether others also perceive SML to be the most legitimate principle.

3.3. Saving healthier lives

To assess the support for SHL, SAL, SMLY, and other (i.e., the residual category), we next utilized the data from the four situations other than situation NONE. The underlying regression results are presented in Table 3B. Fig. 1B highlights that the shares of personal and coordination decisions consistent with SHL are significantly larger than the shares of corresponding decisions consistent with SMLY in both the No Info group and the Info group.

The comparison between the Info and No Info groups suggests that balanced information about competing ethical considerations increases

B. Rai et al.

Social Science & Medicine 282 (2021) 114171

6

the share of personal decisions consistent with SHL by 6⋅2 percentage- points (95% CI, 0⋅00 to 0⋅12; P = 0⋅05), and significantly decreases the share of personal decisions consistent with SMLY by 6 percentage- points (95% CI, − 0⋅11 to − 0⋅01; P = 0⋅03). The impact of balanced information on coordination decisions is qualitatively similar but quantitatively weaker.

Overall, balanced information increases support for SML at the

expense of support for SMLY. Also, the increased support for SML is largely driven by the increased support for SHL. These two findings suggest that balanced information induces some participants, that would have otherwise used SMLY, to use SHL (i.e., SML as the primary and SMLY as the subordinate principle).

All the findings reported above are robust to excluding the 64 par- ticipants with significantly less or significantly more response times

Fig. 1. Share of personal and coordination decisions consistent with various principles. Note. Mean differences between the No Info group and the Info group are shown. (A) Share of decisions consistent with SML, SMLY, and other principles pooled across all the five situations. (B) Share of decisions consistent with SHL, SAL, SMLY, and other principles pooled across the four situations other than NONE. The P values reported in square brackets are based on t tests of dif- ferences in means with standard errors clustered at the participant level. Error bars represent mean ± SEM.

B. Rai et al.

Social Science & Medicine 282 (2021) 114171

7

compared to the average response time (see Table S3 in the Supple- mentary Material). Also, we had included a quiz question in between the “personal preference” stage and the “social agreement” stage. It served as an attention-check, and also helped assess whether participants un- derstood the incentive mechanism used to determine their payment in the coordination games. Only about 4% of the participants (27 out of 700) answered this question incorrectly, and all the findings reported above hold upon excluding these 27 participants from the analysis.

3.4. Impact of balanced information on subgroups

Given that SML is the modal preferred principle at the aggregate level, we focus on investigating the support for SML in the subgroup analysis. Fig. 2 shows the share of personal decisions consistent with SML across the various subgroups categorized by differences in age, race, education, prior familiarity with ventilator allocation guidelines, vaccination attitude, and political orientation. The underlying regres- sion results are presented in Table S4 (see Table S5 and Fig. S1 for co- ordination decisions).

There exists significant heterogeneity in the extent of support for SML among the participants in the No Info group. For instance, the support for SML is highest among participants aged more than 47 years (mean, 0⋅55; 95% CI, 0⋅48 to 0⋅62), followed by non-white participants (mean, 0⋅54; 95% CI, 0⋅46 to 0⋅63). In comparison, the support for SML seems weaker among vaccine hesitant participants (mean, 0⋅46; 95% CI, 0⋅37 to 0⋅54), participants without a college degree (mean, 0⋅45; 95% CI, 0⋅39 to 0⋅51), and republican or republican-leaning (GOP-leaning) participants (mean, 0⋅43; 95% CI, 0⋅36 to 0⋅51).

Balanced information about the competing ethical considerations does not decrease the support for SML in any subgroup, and significantly increases the support for SML in several subgroups. The support for SML increases among Covid-19 vulnerable groups, such as older adults (mean, 0⋅09; 95% CI, − 0⋅003 to 0⋅17; P = 0⋅06) and non-white (mean, 0⋅13; 95% CI, 0⋅01 to 0⋅25; P = 0⋅03). It is perhaps more noteworthy that the support for SML increases by 16 percentage-points (95% CI, 0⋅04 to 0⋅28; P = 0⋅008) among vaccine hesitant participants, and 16 percentage-points (95% CI, 0⋅05 to 0⋅27; P = 0⋅003) among GOP-leaning participants. As in the aggregate level analysis, balanced information shifts some participants within a particular subgroup that would have used SMLY as the primary principle, to using SML as the primary and

SMLY as the subordinate principle. The estimates of the impact of balanced information in the subgroup analysis are similar when we use SHL, rather than SML, as the outcome measure (see Table S6 and Table S7).

Differences in political ideology predict differential levels of trust in experts, which in turn explains individual-level variation in vaccine hesitancy (Baumgaertner et al., 2018; Callaghan et al., 2021). While vaccine hesitancy seems positively correlated with negative views of experts, there is insufficient evidence to treat vaccine hesitancy as a proxy for lower trust in experts. However, there is broad agreement in the existing literature that political orientation is a proxy for lower trust in experts (Motta, 2018; Pennycook et al., 2020). Our finding for the GOP-leaning participants suggests that balanced information about the relevant ethical dilemmas can potentially increase support for SML even among those population subgroups that have lower trust in experts.

4. Discussion

We investigated whether the general public finds saving more lives (SML) the most legitimate principle in allocating scarce ventilators in a survey experiment. Participants who were not provided information about the underlying ethical dilemmas were split: half perceive SML is the most legitimate principle, while the other half do not. Among those who do not, the substantial majority seem to perceive saving more life- years (SMLY) as the most legitimate principle.

Lack of factually correct and scientifically accurate information is often viewed as an important driver of differences between expert views and public opinion. Provision of facts and scientific information is therefore often the primary tool to bridge such differences (Pennycook et al., 2020; Blastland et al., 2020; Islam et al., 2021). Our study focused on a largely neglected channel, i.e., provision of balanced information about the competing ethical considerations that health experts have to grapple with. Balanced information about the competing ethical con- siderations underlying allocation of scarce ventilators enhances the perceived legitimacy of SML at the expense of SMLY, at least to some extent, regardless of differences in age, race, education, attitudes to- wards vaccination, and political orientation. However, these findings are weaker for coordination decisions than for personal decisions. It seems people are doubtful whether others perceive SML as the most legitimate principle.

Table 3 Mean fraction of participant decisions consistent with various principles.

(1) (2) (3) (4) (5) (6) (7) (8)

———————— Personal decisions —————— —————— Coordination decisions ——————

No info Info Diff. Odds ratio No info Info Diff. Odds ratio

A. All 5 situations SML 0⋅50 0⋅57 0⋅08** 1⋅36** 0⋅47 0⋅53 0⋅06* 1⋅25*

(0⋅02) (0⋅02) (0⋅03) (0⋅17) (0⋅02) (0⋅02) (0⋅03) (0⋅16) SMLY 0⋅31 0⋅25 − 0⋅06** 0⋅74** 0⋅35 0⋅31 − 0⋅04 0⋅83

(0⋅02) (0⋅02) (0⋅03) (0⋅10) (0⋅02) (0⋅02) (0⋅03) (0⋅10) Other 0⋅20 0⋅18 − 0⋅02 0⋅91 0⋅18 0⋅16 − 0⋅02 0⋅89

(0⋅02) (0⋅02) (0⋅02) (0⋅13) (0⋅02) (0⋅01) (0⋅02) (0⋅14)

B. Situations except NONE SHL 0⋅44 0⋅51 0⋅06** 1⋅28** 0⋅44 0⋅48 0⋅05 1⋅20

(0⋅02) (0⋅02) (0⋅03) (0⋅16) (0⋅02) (0⋅02) (0⋅03) (0⋅16) SAL 0⋅06 0⋅07 0⋅02 1⋅30 0⋅04 0⋅05 0⋅01 1⋅30

(0⋅01) (0⋅01) (0⋅01) (0⋅25) (0⋅01) (0⋅01) (0⋅01) (0⋅30) SMLY 0⋅32 0⋅26 − 0⋅06** 0⋅75** 0⋅37 0⋅32 − 0⋅04 0⋅83

(0⋅02) (0⋅02) (0⋅03) (0⋅10) (0⋅02) (0⋅02) (0⋅03) (0⋅11) Other 0⋅18 0⋅16 − 0⋅02 0⋅88 0⋅16 0⋅14 − 0⋅02 0⋅88

(0⋅02) (0⋅02) (0⋅02) (0⋅14) (0⋅02) (0⋅01) (0⋅02) (0⋅15)

Note. In panel A, each mean estimate is based on 3500 observations made up of 700 participants across the five situations. In panel B, each mean estimate is based on 2800 observations made up of 700 participants across four situations, excluding situation NONE. Each mean estimate for the No Info group corresponds to a in regression equation (1). Each mean estimate for the Info group corresponds to a + b in regression equation (1). The odds ratios are estimated using logistic regressions. Standard errors clustered at the participant level are reported in parentheses. *, **, and *** indicate statistical significance for t tests at the 10, 5, and 1% levels, respectively.

B. Rai et al.

Social Science & Medicine 282 (2021) 114171

8

Fig. 2. Share of personal decisions consistent with SML across population subgroups. Note. Each panel shows the mean differences between the No Info group and the Info group pooled across all the five situations in the personal decision stage. (A) Participants aged 47 years or more were categorized as “Older” and the rest were categorized as “Younger”. (B) Participants who reported belonging to any race other than “White” were categorized as “Non-White”. (C) “No degree” refers to participants who reported not having a college degree. (D) Participants who reported they never heard of any guidelines for the allocation of ventilators or ICU beds were cate- gorized as “Never Heard”. (E) Willingness to take a Covid-19 vaccine was ascertained based on responses to the following question: “Suppose a vaccine is developed for Covid-19 that health authorities consider safe. If the government offers this vaccine for free, would you take it?” Participants who responded “No” were categorized as “Vaccine-Hesitant” and participants who responded “Yes” were categorized as “Vaccine-Confident” (F) Political orientation was ascertained based on responses to the following question: “In current politics, which category do you consider yourself more of?” Partic- ipants who responded “Republican” or “Independent, leaning to- wards Republican” were categorized as “GOP”. The remaining participants were categorized as “Non-GOP”. The P values reported in square brackets are based on t tests of differences in means with standard errors clustered at the participant level. Error bars repre- sent mean ± SEM.

B. Rai et al.

Social Science & Medicine 282 (2021) 114171

9

The increased support for SML due to provision of balanced infor- mation is largely driven by the increased support for the principle of saving healthier lives (SHL). This suggests that a significant fraction of participants that would have otherwise used SMLY as the primary principle, use SML as the primary and SMLY as the subordinate principle when they become aware of the underlying ethical dilemmas.

The finding at the aggregate level seem to hold for several population subgroups we analyzed. In particular, our subgroup analysis suggests that even the participants without a college degree and GOP-leaning participants are significantly more likely to perceive SML as the legiti- mate principle once they become aware of the ethical dilemmas. This is worth noting because previous research has shown that sources of in- formation and trust in science vary with educational attainment and political orientation (Kahan et al., 2017; Funk et al., 2019).

There are some noteworthy limitations of our study. First, our sample is not perfectly representative of the general US adult population. The average age of participants in our sample is slightly lower and the educational attainment is slightly higher than the corresponding actual means. Second, ignoring withdrawal of ventilators is a clear limitation of our study because withdrawal and (re)allocation decisions are not in- dependent of each other. Third, while the short life-expectancy of pa- tients with comorbidities (two years) offers greater confidence in the observed impact of balanced information, it suffers from the fact that life-expectancy of people with comorbidities is typically a function of their age.

The choice situations were presented to all participants in the treatment and control groups in a particular fixed order rather than a randomized order. We cannot be certain about the biases due to the particular order we used. However, since the same order was used for both the treatment and the control groups, such biases are unlikely to cause major concern about the estimates of the impact of balanced in- formation. Another potential limitation relates to our indirect inference method. We inferred the principle preferred by a participant via the allocation decisions. An alternative is to describe the principles, and have the subjects directly indicate their preferred principle. However, it is debatable whether the direct response approach or the indirect inference approach is more likely to suffer from responder bias.

As findings from survey experiments often suffer from generaliz- ability concerns, it is useful to clarify the particular sense in which these concerns are pertinent to our study. Most participants in our study were not healthcare professionals, and thus are unlikely to make ventilator allocation decisions in practice. Consequently, we cannot generalize the findings as to how ventilator allocation decisions by healthcare pro- fessionals might change if they are fully aware of the underlying ethical considerations. However, our interest was not in the decision-making of healthcare professionals. The central goal of our study was to investigate perceived legitimacy of the principles behind these guidelines among the general public because public opinion can also influence policies. Hence, given the nature of our question, external validity concerns are likely to be minimal.

Our work does not presuppose that SML is a more legitimate prin- ciple than the alternatives. Given that most public health experts and authorities do endorse SML as the primary principle, our findings sug- gest that, in addition to disseminating scientific facts and combating misinformation about scientific facts, promoting public awareness about the competing ethical considerations that public health experts grapple with can help increase the perceived legitimacy of SML, even among subgroups with lower trust in experts. Skeptics may erroneously but plausibly hold the belief that some factual or scientific information is produced and propagated by profit-seeking entities. Such a belief is harder to hold when the information is balanced, and relates to the underlying ethical dilemmas in the allocation of scarce medical re- sources. While further research is needed to better understand the behavioral mechanisms behind our findings, this difference between scientific and ethical information could be an important driver.

Credit author statement

BR and LCW conceived and designed the study with input from all the authors. CKS and LCW conducted the online experiment. LCW and SP analyzed the data with input from BR and TH. BR, SP, and LCW wrote the manuscript with input from TH.

Acknowledgements

We thank two anonymous reviewers for comments and suggestions that significantly improved this paper. We also thank Yee (Lisa) Chan, Gaurav Datt, Marwa El Zein, Vai-Lam Mui, and Richard Tee for useful comments. We gratefully acknowledge funding from the Centre for Development Economics and Sustainability and the Department of Economics at Monash University.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi. org/10.1016/j.socscimed.2021.114171.

References

WBUR. After Uproar, April 2020. Mass. Revises Guidelines on Who Gets an ICU Bed or Ventilator amid COVID-19 Surge. WBUR, USA. https://www.wbur.org/commonhea lth/2020/04/20/mass-guidelines-ventilator-covid-coronavirus.

Amnesty International, November 2020. The Rights of the Older People during the Pandemic. Belgium. Amnesty International Belgium. https://www.amnesty.be/cam pagne/discrimination/droits-agees-pandemie-covid/maison-repos-covid19.

Baumgaertner, B., Carlisle, J.E., Justwan, F., 2018. The influence of political ideology and trust on willingness to vaccinate. PloS One 13, e0191728.

Blastland, M., Freeman, A.L.J., Linden, S.V., Marteau, T.M., Spiegelhalter, D., 2020. Five rules for evidence communication. Nature 587, 362–364.

Callaghan, T., Moghtaderi, A., Lueck, J.A., Hotez, P., Strych, U., Dor, A., Fowler, E.F., Motta, M., 2021. Correlates and disparities of intention to vaccinate against COVID- 19. Soc. Sci. Med. 272, 113638.

Cartwright, W., 1996. Killing and letting die: a defensible distinction. Br. Med. Bull. 52, 354–361.

Chu, Q., Correa, R., Henry, T.L., et al., 2020. Reallocating ventilators during the coronavirus disease 2019 pandemic: is it ethical? Surgery 168, 388–391.

Cooper, R.W., DeJong, Forsythe R., Ross, T.W., 1990. Selection criteria in coordination games: some experimental results. Am. Econ. Rev. 80, 218–233.

Ecker, U.K.H., Ang, L.C., 2019. Political attitudes and the processing of misinformation corrections. Polit. Psychol. 40, 241–260.

Emanuel, E.J., Persad, G., Upshur, R., et al., 2020. Fair allocation of scarce medical resources in the time of Covid-19. N. Engl. J. Med. 382, 2049–2055.

Feldman, G., Kutscher, L., Yay, T., 2020. Omission and commission in judgement and decision making: understanding and linking action-inaction effects using the concept of normality. Soc Personal Psychol Compass 14, e12557.

Figueiredo, A.D., Simas, C., Karafillakis, E., Peterson, P., Larson, H.J., 2020. Mapping global trends in vaccine confidence and investigating barriers to vaccine uptake: a large-scale retrospective temporal modelling study. Lancet 396, 898–908.

Funk, C., Hefferon, M., Kennedy, B., Johnson, C., August 2019. Trust and Mistrust in Americans’ Views of Scientific Experts. Pew Research Center. https://www.pewrese arch.org/science/2019/08/02/trust-and-mistrust-in-americans-views-of-scientific-e xperts.

Hooper, M.W., Napoles, A.M., Perez-Stable, E.J., 2020. COVID-19 and racial/ethnic disparities. J. Am. Med. Assoc. 323, 2466–2467.

Iacobucci, G., 2020. Covid-19: UK lockdown is “crucial” to saving lives, say doctors and scientist. Br. Med. J. 378.

Islam, A., Pakrashi, D., Vlassopoulos, M., Wang, L.C., 2021. Stigma and misconceptions in the time of the COVID-19 pandemic: a field experiment in India. Soc. Sci. Med. 278, 113966.

Jackson, J., Bradford, B., Hough, M., Myhill, A., Quinton, P., Tyler, T.R., 2012. Why do people comply with the law? Legitimacy and the influence of legal institutions. Br. J. Criminol. 52, 1051–1071.

Justice, Daniels N., Health, Healthcare, 2001. Am. J. Bioeth. 1, 2–16. Kahan, D.M., Landrum, A., Carpenter, K., Helft, L., Jamieson, K.H., 2017. Science

curiosity and political information processing. Polit. Psychol. 38, 179–199. Michigan Department of Community Health, Office of Public Health Preparedness,

November 2012. Guidelines for Ethical Allocation of Scarce Medical Resources and Services during Public Health Emergencies in Michigan. MI.: Michigan Department of Community Health, Office of Public Health Preparedness, Version 2.0.. http:// www.mimedicalethics.org/Documentation/Michigan%20DCH%20Ethical%20Scar ce%20Resources%20Guidelines%20v2%20rev%20Nov%202012.0.pdf.

Mitchell, A., Gottfried, J., Kiley, J., Matsa, K.E., October 2014. Political Polarization & Media Habits. Pew Research Center. https://www.journalism.org/2014/10/21/poli tical-polarization-media-habits.

B. Rai et al.

Social Science & Medicine 282 (2021) 114171

10

Motta, M., 2018. The dynamics and political implications of anti-intellectualism in the United States. Am. Polit. Res. 46, 465–498.

Motta, M., Callaghan, T., Sylvester, S., 2018. Knowing less but presuming more: dunning- Kruger effects and the endorsement of anti-vaccine policy attitudes. Soc. Sci. Med. 211, 274–281.

Mueller, A.L., McNamara, M.S., Sinclair, D.A., 2020. Why does COVID-19 disproportionately affect older people? Aging 12, 9959–9981.

Murphy, K., 2004. The role of trust in nurturing compliance: a study of accused tax avoiders. Law Hum. Behav. 28, 187–209.

Pennsylvania Department of Health, April 2020. Interim Pennsylvania Standards of Care for Pandemic Guidelines. PA.: Pennsylvania Department of Health, Version 2. https://www.health.pa.gov/topics/Documents/Diseases%20and%20Conditions/C OVID-19%20Interim%20Crisis%20Standards%20of%20Care.pdf.

Pennycook, G., McPhetres, J., Zhang, Y., Lu, J.G., Rand, D.G., 2020. Fighting COVID-19 misinformation on social media: experimental evidence for a scalable accuracy- nudge intervention. Psychol. Sci. 31, 770–780.

Piscitello, G.M., Kapania, E.M., Miller, W.D., Rojas, J.C., Siegler, M., Parker, W.F., 2020. Variation in ventilator allocation guidelines by US State during the coronavirus disease 2019 pandemic: a systematic review. JAMA Netw Open 3, e2012606.

Price-Haywood, E.G., Burton, J., Fort, D., Seoane, L., 2020. Hospitalization and mortality among black patients and white patients with Covid-19. N. Engl. J. Med. 382, 2534–2543.

Rawls, J., 2005. Political Liberalism. Columbia University Press, New York, NY. Sen-Crowe, B., McKenney, M., Boneva, B., Elkbuli, A., 2020. A state overview of

COVID19 spread, interventions and preparedness. Am. J. Emerg. Med. 38, 1520–1523.

Spranca, M., Minsk, E., Baron, J., 1991. Omission and commission in judgment and choice. J. Exp. Soc. Psychol. 27, 76–105.

Subject Matter Experts Advisory Panel for the Governors Expert Emergency Epidemic Response Committee (GEEERC), April 2020. Crisis Standards of Care Guidelines for Hospitals for the COVID-19 Pandemic. CO.: Department of Public Health and Environment, Version 2.0. https://drive.google.com/open?id=1IllEDTwEP6HQx uzy5Az-9maYJxO7R6bp.

Truog, R.D., Mitchell, C., Daley, G.Q., 2020. The toughest triage—allocating ventilators in a pandemic. N. Engl. J. Med. 382, 1973–1975.

White, D.B., Lo, B., 2020. A framework for rationing ventilators and critical care beds during the Covid-19 pandemic. J. Am. Med. Assoc. 323, 1773–1774.

B. Rai et al.