Liza von Grafenstein
Bio
I am an Economist at IDinsight, based in Delhi, India.
In this role, I oversee the technical design and analysis of impact evaluations, representative sample surveys, and mixed-methods research. My portfolio covers diverse sectors and partners, including projects for government and non-government funders and implementers across nutrition, maternal and child health, public health, education, women’s empowerment, social protection, and socio-emotional learning. I have led work in the Democratic Republic of Congo, India, Nigeria, the Philippines, and Vietnam.
Before IDinsight, I used machine learning to predict real-time child malnutrition outcomes at the Indian district level as an Innovative Metrics and Measures for Agriculture and Nutrition Actions (IMMANA) Career Development Fellow at Georg-August-University Göttingen, Germany, the Center of Development Studies (ZEF), Germany, and the South Asia office of the International Food Policy Research Institute (IFPRI) in Delhi, India.
I completed my PhD in Economics at the Georg-August-University Göttingen. In addition, I hold an MPA with a focus on International Development from Cornell University, US, and a BA in Political Science and South Asian Studies from Ruprecht Karl University of Heidelberg, Germany. I speak English, Hindi, and German.
This exchange between Tony and Mark captures the diagnosis well, and GiveWell's concrete steps (the red-teaming exercise, the 14 independent monitoring grants, the coverage survey standards) are particularly encouraging. At IDinsight, we work at this interface regularly, alongside governments, NGOs, and funders including GiveWell and several of its grantees. I wanted to add texture to a couple of points here without restating what both of you have covered so eloquently.
On the incentive structure inside CEA models. Tony's observation is right that collapsing implementation quality into a single blended discount to the final cost-effectiveness output makes uncertainty invisible. One extension worth considering: publish the standards that would earn a grantee a lower discount factor. Right now the adjustment is applied after the fact. If implementers knew in advance that investing in independent monitoring (enumerator separation, neutral sampling frames, objective measurement) translated into a meaningfully higher cost-effectiveness rating, the incentive runs in the right direction from the start. It converts M&E quality from a compliance question into a competitive one.
On the practical side, IDinsight’s free M&E Health Check tool (built by Tony during his time at IDinsight), a structured self-assessment benchmarked against transparent criteria, could serve as a starting point for grantees wanting to improve their M&E systems. For higher-risk programs, self-assessment needs to be complemented by independent review, which is the kind of work we're already doing with GiveWell grantees in Northern Nigeria.
On Value of Information as a framework for monitoring investment. Tony and Mark are both right that the question is not whether to invest in independent monitoring but how much, and that Value of Information (VOI) thinking is the right lens for that decision. IDinsight's VOI work tries to bring this lens in at the outset of engagements precisely because the question of what to measure is easier to answer well before a monitoring system is built than after.
On field presence and psychological safety. There's a tension worth naming that isn't fully surfaced here. As funders rightly increase field presence and direct observation, the risk is that implementers become less likely to bring up problems candidly, particularly problems that reflect on their own monitoring design. External evaluation partners play a useful role here precisely because they create a channel for that information to travel without the distorting effect of a funder-implementer dynamic. GiveWell's instinct to commission independent surveys rather than rely solely on its own field visits seems right; the principle behind that instinct deserves to be made explicit in how monitoring partnerships are structured.
I agree with Tony that this should be a turning point rather than a one-off response. The steps shared by Mark here are substantive. The remaining question is whether the incentive architecture inside the cost-effectiveness model changes to make rigorous monitoring the rational choice for implementers, not just a funder requirement.