Cambridge randomizer performance in a large scale, long-term experiment: Maintaining quality, reducing research costs

• In Event: Issues in the Use of Experimental Research Methods

Abstract

Large-scale field experiments have widely faced a dilemma with regard to the process of randomization. When practitioners control randomization, experiments risk an unknown and potentially damaging level of bias due to manipulation of random assignment. If the randomization process is removed from practitioners’ control, experiments must often accept the substantial cost of additional research staff.

The Cambridge randomizer was developed as an IT tool to reduce costs while ensuring quality, allowing practitioners to conduct random assignment themselves, but giving researchers control over the process. Using the tool, practitioners and researchers alike can log in and assess case eligibility, capture baseline information, automatically assign cases to treatment or control, and instantaneously disseminate necessary information. The Cambridge randomizer has now been trialed in a large-scale, multi-year experiment with a staff of over 100 practitioners entering cases for randomization in a city-wide 24/7 day-to-day police custody intake setting.

The implementation process and performance will be presented, including training/maintenance requirements and error rate.

Authors

• Peter Neyroud, University of Cambridge

• Barak Ariel, Cambridge University / Hebrew University

• Molly Slothower, University of Maryland