Advancing the measurement of beneficiary preferences through updated methods

What are moral weights — and why they matter for GiveWell

What is the value of saving the life of a child under five, compared to preventing a case of severe anemia? Or compared to increasing a poor family’s consumption for one year?

While attaching numbers to such outcomes can feel ethically fraught, these judgments are unavoidable. Every policymaker, donor, and foundation that allocates limited resources across programs is implicitly making trade-offs between competing goods: saving lives, reducing illness, and improving material well-being. In other words, they are already assigning values to these outcomes—just not explicitly.

To make these value judgments more transparent and accountable, GiveWell has chosen to confront them directly. Through moral weights, the organization explicitly quantifies the relative social value of different outcomes, enabling more consistent and open comparison across programs that improve health or increase consumption. Unlike a simple Value of a Statistical Life (VSL), which focuses narrowly on valuing mortality risk reduction, moral weights capture a broader ethical calculus—how much society should value different kinds of benefits, from extending life to improving quality of life or reducing poverty.

GiveWell’s journey with moral weights

GiveWell’s moral weights framework has evolved significantly over time. One of its first versions was derived from a staff survey, where team members were invited to provide their own values for trade-offs—either based on personal moral reasoning or by referencing empirical evidence on how others, including target populations, might view these trade-offs.

In 2019, GiveWell partnered with IDinsight to take this further. The goal was to move beyond staff opinions and gather data directly from people living in low-income settings of relevance for GiveWell. The study piloted stated-preference methods—asking respondents to make structured trade-offs between saving lives and increasing consumption. For instance, participants were asked how they would value saving the life of a five-year-old child compared to doubling one person’s consumption for a year.

While the study produced valuable insights, it also revealed challenges. Many respondents resisted the premise of putting a price on life. Even when presented with extreme numbers of cash-transfer beneficiaries, a substantial share refused to choose any option other than saving an additional life. This raised concerns about hypothetical bias—the difference between how people respond to hypothetical choices and how they might act in reality.

In light of the study and other work it had done, GiveWell subsequently revised its moral weights to value saving lives more highly than its previous moral weights had. 

Revisiting moral weights in 2025

In 2025,  GiveWell commissioned a new series of studies from IDinsight. These aimed to:

1. Re-test and refine the original 2019 protocols in light of new research on mitigating hypothetical bias;
2. Expand moral-weight estimation beyond mortality and consumption to include trade-offs involving anemia and access to family planning; and
3. Synthesize recent revealed-preference evidence—observed behaviors in real-world markets or policy contexts—to explore alternative ways of estimating these values in low-income settings, particularly in Africa.

For the stated-preference work on mortality and consumption, the team tested several refinements: enhanced comprehension checks using systematic “teach-backs” and certainty scales; cheap-talk scripts and time-to-think to reduce hypothetical bias; and incentive-compatible mechanisms including Bayesian Truth Serum and inferred-valuation designs.

For the new domains—anemia and family planning—the study piloted contingent valuation and discrete-choice experiments, focusing particularly on reducing anchoring effects, since participants were familiar with the market prices of these goods, and introducing mental visualization to help participants project themselves in the different scenarios.

Qualitative pilots were first conducted in Homa Bay, Kenya, in March 2025, followed by larger quantitative pilots in Machakos, Kenya and Bahraich, Uttar Pradesh, India in May and June 2025.

What we learned and what we recommend

First, strategies designed to reduce hypothetical bias or increase incentive compatibility did not improve the consistency of preferences compared to the 2019 study. Our Kenya sample produced results broadly similar to those observed six years earlier. Roughly one-quarter of respondents stated that there was no number of additional cash-transfer recipients that would convince them to choose a program providing cash instead of one that saves one more child’s life. This pattern mirrors 2019 findings and suggests that, whether because of deeply held moral convictions or a refusal to engage with the premise of such trade-offs, many respondents reject the idea of directly comparing lives saved to income gains. We therefore conclude that the contingent valuation protocol from 2019 remains the most appropriate approach for eliciting preferences on the mortality–consumption trade-off.

Second, when we extended the analysis to other domains such as anemia and family planning, we observed a similar pattern. Presenting trade-offs using discrete-choice experiment formats—where respondents chose between programs that reduce anemia or provide access to family planning versus those that increase consumption—led to widespread rejection of the scenarios. A large proportion of respondents declined to specify any number of cash-transfer recipients that would make them prefer a cash program over one that reduces anemia in children. And when the choice involved access to family planning for a young woman not yet pregnant, nearly all respondents stated that no amount of cash would convince them to select the option without family planning access.

These findings suggest that directly eliciting monetary values for certain health outcomes may be morally, cognitively, or ethically fraught for participants, especially when the trade-offs touch on sensitive or life-related domains. This reinforces the case for relying on contingent valuation approaches, which can provide clearer structure and more interpretable results when estimating moral weights across different program outcomes.

Future research directions for the field

Our results on the new trade-offs are still preliminary, and should therefore be interpreted with caution. Nonetheless, several promising directions for future work emerge.

First, there is scope to further refine methods to limit anchoring in willingness-to-pay (WTP) elicitation. This could include more iterative probing around WTP values and structured opportunities for respondents to reconsider their choices after reflection, helping to separate genuine preferences from reactions to the initial framing or price cues. Second, our experience suggests the potential value of integrating mental visualization and reflection exercises within elicitation protocols. 

More broadly, the limited consistency and framing sensitivity observed across stated-preference tasks highlight a need to move beyond standard survey formats. A key innovation for the field would be to design elicitation exercises that encourage deliberation and collective reflection, borrowing from participatory development methods. Such approaches could help participants articulate their priorities in ways that better reflect how real-world decisions and moral reasoning actually unfold.