Probability Samples, Nonprobability Samples, or Both?

Probability samples have been the foundation for producing unbiased statistics for at least 50 years. Yet, list frame coverage and response rates have been decreasing. At the same time, nonprobability samples, including those based on administrative, business, and remotely sensed data, have become increasingly available. Methods have been and are being developed for the principled use of nonprobability samples, often in combination with probability samples. With the tools that data science and AI are enabling, how should statisticians view this new landscape? Should nonprobability samples be used in the production of, say, official statistics only when a probability sample is available to correct for bias? Or, can valid statistics be produced with only a nonprobability sample? Or, should only probability samples be used? This panel will discuss the strengths and weaknesses of the various approaches.