B3615 - Research-on-research Blinded data analysis to improve the robustness and reproducibility of health research - 08/10/2020
Bias in scientific research can lead to research waste and useless or even harmful health care and policy interventions. To reduce bias, researchers commonly employ experimental designs that blind both participants and outcome assessors. Data analysts, however, are rarely blinded. Here, we propose to test whether blinding data analysts improves the reliability of published research findings. To execute the study, we will randomize researchers who request the Avon Longitudinal Study of Parents and Children (ALSPAC) dataset, and consent to partake in the research, to receive either data from only 10% of participants, or data from all participants. Researchers who receive 10% of the data will develop their analysis based on this subset and be asked to register their analysis plan. After registering their analysis plan, the full dataset will be provided. We call this “Blind Access”. Researchers who receive the full dataset will not be required to register their analysis. We call this “Standard Access”. We will then compare the reliability of the published findings between the Blind Access and Standard Access groups.