Big Data versus a survey

Economists are shifting resources from work on survey data to work involving “Big Data.” This analysis is an empirical exploration of the trade-offs this substitution requires. Parallel models are estimated using Equifax credit bureau data and Survey of Consumer Finances data. After adjustments to account for different variable definitions and sampled populations, it is possible to arrive at similar models of total household debt. However, the estimates are sensitive to the adjustments. In this example, some external education and income measures are successfully integrated with the big data, but other external aggregates fail to adequately substitute for survey responses.