An Outside Look at Our Impact Modeling

Summary

• We commissioned Rethink Priorities to review the methodology behind our Newborn Care Foundations impact model. Check out their full report.

• They found that our model is internally consistent, credibly sourced, and generally conservative in its assumptions, tending toward underestimating our likely impact rather than overestimating it. 

• Alongside the steep drop in our cost per trainee over the past year, this assessment leaves us more confident that our program is cost-effective under a broad set of assumptions.

The claim we wanted tested

Our estimates place the Newborn Care Foundations online course well above GiveWell's bar for highly cost-effective programs. Such a major claim should not rest on our judgment alone, so we asked Rethink Priorities, an independent research organization, to critically assess our methodology. In addition to our cost-effectiveness analysis, they looked at our clinical practice evaluation and our desk research on the health impacts of the practices we teach.

What they found

Rethink Priorities concluded that our model suits the purpose we built it for, which is to test whether the Newborn Care Foundations course is plausibly cost-effective, even under conservative assumptions.

• The approach is conservative in scope. While our course teaches multiple life-saving techniques to facility-based health workers, they noted that our model focuses conservatively on only one outcome: early initiation of breastfeeding. It also leaves potential morbidity and spillover benefits of our work unmodeled.

• The model follows a logical chain. They found that the model’s three-step chain (from health worker training, to more mothers breastfeeding in the first hour, to reduced neonatal mortality) is logically constructed and well grounded in published research, and our cost-effectiveness ratios follow correctly from the model’s inputs.

• We make cautious assumptions. Key model inputs, such as mortality rates and relative risks, are drawn from credible sources such as UNICEF and peer-reviewed meta-analyses, generally taking the more cautious figure where several exist. To account for uncertainties in our model, we follow GiveWell’s approach of applying discount parameters to our calculations, which reduce our estimated impact by 91% compared to the unadjusted figures. Rethink Priorities found these discounts to be both reasonable and conservative based on the current evidence base.

Ways to strengthen our model

To strengthen our cost-effectiveness estimates even further, Rethink Priorities suggested several improvements, such as gathering data on the duration of improved health worker practices and better accounting for the possibility that multiple HealthLearn trainees from the same facilities may be attending the same births. Though these are uncertainties we’ve noted previously on our blog, we agree that they are critical questions, and we’ll aim to assess them in upcoming evaluations. 

Building from here

The cost-effectiveness of our program rests on two key parameters: impact per trainee and cost per trainee. Our work in 2025 demonstrates how scaling can drive down cost per trainee. Rethink Priorities' review supports our approach to modeling the impact side of that relationship. Lower per-learner costs and this endorsement of our impact modeling increase our confidence that the program is cost-effective across a wide range of plausible assumptions. We remain focused on improving our measurement of clinical practice change, since that is the main driver of impact per trainee. We look forward to sharing more on that work as it progresses!