Abstract:
Traditional face-to-face surveys are expensive, especially those designed to be representative for small subpopulations. They are thus infrequently carried out, limiting real-time awareness in epidemic settings. Lower cost alternatives such as online and mobile phone surveys can suffer from bias, but offer advantages in terms of cost and speed. We propose a new adaptive survey design strategy, based on an active learning approach. Our algorithm automatically oversamples hard-to-reach populations, adaptively optimizing for user-defined goals such as representativeness as the survey progresses, taking nonresponse into account. It has the potential to lower the cost of face-to-face surveys and reduce the bias of mobile phone surveys, by increasing representativeness at smaller sample sizes. We present experiments on simulated data and the results of a pilot telephone survey of food insecurity conducted jointly with the World Food Program in Zimbabwe in 2023.
Dr Flaxman is an associate professor in the Department of Computer Science at Oxford. His research is on scalable methods and flexible models for spatiotemporal statistics and Bayesian machine learning, applied to public policy and global health. He founded and co-leads the Machine Learning & Global Health network (www.MLGH.net).
