Data-driven analysis is at the heart of economics and many other social sciences. The digitization of government records and social interactions has greatly expanded the set of questions which can be answered empirically.
Matthew Webb, an assistant professor in the Department of Economics has received $88,750 over four years for a SSHRC Insight grant project, “Bootstrap Methods for Cluster-Robust Inference.”
“While there has been a great deal of work on how to answer new questions, less attention has been paid to how precise these answers are,” says Webb. “With many new types of datasets, reliable inference can be a challenge.”
Webb says many of the newly available datasets are in the form of individual-level records from either government agencies or private companies. These datasets often involve clusters or groups of individuals, such as residents in a county or users of services in a metropolitan area.
“This grouping of observations means that the error terms in regression models will be correlated among observations within the same groups,” says Webb. “Many of the existing tools for inference will therefore be unreliable, resulting in too many false positives.
“Although the literature on cluster-robust inference has greatly expanded in recent years, there remain many statistical problems. Often empirical researchers are interested in analyzing specific public policies to see if they achieved their goals.
“When these analyses are conducted with flawed statistical tools, the conclusions reached can also be flawed. Because there can be far too many false positives, policies that in reality had no effect can often be determined to be effective. These problems can be compounded if flawed analyses of current policies are used to develop new policies.”
Friday, June 18, 2021 in Department of Economics, News, Research
Share: Twitter, Facebook
