Applied Economics: Which Confidence Interval?
Published: 11 February 2022
23 February. Professor John Mullahy, University of Wisconsin–Madison
Professor John Mullahy, University of Wisconsin–Madison
'Which confidence interval?'
Wednesday 23 February, 3pm - 4.30pm
Zoom online seminar
Abstract
As null hypothesis tests based on p-values lose favoUr in reports of empirical research findings there is a corresponding and widespread call to substitute confidence intervals (CIs) as measures of sampling variation. This call raises a centrally important question: Which CI should a researcher report? While it is conventional to report equal-tailed CIs with .95 coverage probability, defensible rationales for this convention are largely absent. Why might not a researcher instead reasonably report a CI having .90 or .667 or .50 coverage probability? Moreover, for any level of coverage probability there will typically be an infinite number of CIs corresponding to this coverage, so how does one sensibly choose amongst them?
This paper begins by reviewing the fundamental nature of CIs as interval or set estimators. It then describes traditional methods for CI construction and circumstances where differences among them are likely to emerge. The paper then develops a framework for defining optimal CIs, both in terms of their coverage probability as well as their length and location. Some numerical examples are presented to highlight key issues.
Biography
John Mullahy is a Professor of Health Economics at the University of Wisconsin–Madison in the Department of Population Health Sciences. At UW–Madison he is also an affiliate Professor of Risk and Insurance in the Wisconsin School of Business, and of Public Affairs in the La Follette School of Public Affairs. He is also an affiliate of the Center for Demography and Ecology, the Center for Demography of Health and Aging, and the Institute for Research on Poverty at UW–Madison. He is also an honorary professor of health economics at NUI Galway in Ireland and a research associate at the National Bureau of Economic Research.
Further information: business-events@glasgow.ac.uk
First published: 11 February 2022
<< 2022