#  "Selecting Experimental Sites for External Validity" 

Michael Gechter, Keisuke Hirano, Jean Lee, Mahreen Mahmud, Jonathan Morduch, Orville Mondal, Saravana Ravindran, and Abu Shonchoy

September 26, 2026

The ability to generalize experimental findings usually depends on where the research took place and where the evidence is being applied. By choosing where to experiment, researchers influence the external validity of estimated treatment effects and, ultimately, the quality of decisions informed by them. We characterize the optimal selection of a small number of experimental sites to maximize the value of evidence for policymaking across a wider range of settings. We develop a Bayesian decision-theoretic framework that links sites through two sources of information: a smoothing prior based on proximal similarity and a structural prior derived from a micro-econometric model. Using data from South Asia, we show that experimentation alone does not necessarily improve policy choices: depending on how the estimates are extrapolated, some experimental evidence can be so unrepresentative that policy decisions generate large expected welfare losses. We quantify the welfare gains from Bayesian updating of evidence from optimally-selected sites, and we identify when replicating experiments in multiple sites can increase welfare gains within a given budget.



File

 [Gechter-et-al-Selecting-Experimental-Sites-for-External-Validity-9-26-26.pdf](/files/faculty/publications/Gechter-et-al-Selecting-Experimental-Sites-for-External-Validity-9-26-26.pdf)

 

 

Wagner Faculty

[Jonathan J. Morduch](/community/faculty/jonathan-j-morduch)

 

 

Financial Access Initiative

[Finance](/impact/research/topics/finance)

[International Development](/impact/research/topics/international-development)

[Program Evaluation](/taxonomy/term/8785376)