Skip to main content

Human baselines

Simulated respondents are only worth anything if they reproduce what humans do. Each study below is a published human conjoint experiment, re-run on the platform with the same attributes, levels, and design, then compared effect by effect against the original results.

StudyDomainRank correlation
Adam (Patient Preferences in Complementary and Conventional Medicine)Comparison of our results to the Adam et al. 2019 paperr_{s} = .8277, p < .001
Adida (Immigration Policy)Comparison of our results to the Adida, Lo, and Platas 2019 paperr_{s} = .9593, p = .0002
Ares (Yogurt Consumer Choice)Comparison of our results to the Rao et al. 2009 paperr_{s} = .7723, p = .009
Bechtel (International Carbon Tax Policy for Environmental Mitigation)Comparison of our results to the Bechtel, Scheve, and van Lieshout 202r_{s} = .6711, p < .001
Claret (Consumer Choice for Fish)Comparison of our results to the Claret, Guerrero, and Aguirre 2012 par_{s} = .9442, p = .0002
Duch (COVID Vaccine Acceptance)Comparison of our results to the Duch et al. 2021 paperr_{s}=.7996, p < .001
Hainmueller (Immigration Policy)Comparison of our results to the Hainmueller and Hopkins 2015 paperr_{s} = .5406, p < .001
Kreps (COVID Vaccine Acceptance)Comparison of our results to the Kreps, Prasad, Brownstein et al. 2020r_{s} = .8734, p < .001
Luthi (Wind Energy Policy)Comparison of our results to the Luthi and Prassler 2011 paperr_{s} = .7884, p = .0004
Rao (Rural Clinician Scarcity and Job Preferences)Comparison of our results to the Rao et al. 2013 paperr_{s} = .7286, p < .001
Skreli (Organic Tomatoes Product Design)Comparison of our results to the Skreli et al. 2014 paperr_{s} = .5213, p=.1008
Wu (Subcompact Car Product Design)Comparison of our results to the Wu, Liao, and Chatwuthikrai 2014 paper_{s} = .7622, p = .006

Each page carries the comparison chart, a link to the original paper, and a link to the underlying run data.

note

Figures were produced when each replication was run and have not been re-verified against the current platform. Treat them as the record of that replication, not as a live benchmark.