Virtually all scientists would agree that, everything else being equal, randomized study designs have stronger internal validity than comparable observational study designs. But does it necessarily follow that results from observational studies are unreliable in guiding treatment decisions and policy choices? This plenary session will address how big data is changing the strength of evidence in observational studies and how this is influencing the types of studies where randomization is necessary. How does the ability to link data from multiple sources (such as medical claims, electronic medical records, sociodemographic data, etc.) affect our confidence in drawing causal inferences from observational data? Does it help to reduce missing variable bias? Or does it introduce bias because the people that link are atypical? What is the evidence that observational studies can replicate the average treatment effects from randomized studies conducted on similar populations to answer similar questions? How do considerations of study cost, speed to answer, feasibility, sample size, and other factors influence study design? This panel of leading researchers and policy experts will provide state-of-the-art viewpoints on the usefulness of big data for health care policy decisions and discuss the challenges remaining.