The uncertainty that exists in the release of important US economic numbers as point estimates without their accompanying error measurements should be communicated. The author believes the communication would clarify the data and help end users, such as policymakers, make better-informed decisions.
What’s Inside?
Many US federal statistical agencies report official economic statistics as point estimates without releasing the accompanying measures of uncertainty around those numbers. The author believes that the agencies should show the range of uncertainties along with the point estimates. He also thinks that agencies that conduct surveys should address the issue of nonresponse rather than make assumptions about the missing data.
The author identifies three concepts of sampling error—transitory, permanent, and conceptual statistical uncertainty—and then provides examples of these three concepts by examining GDP data released by the Bureau of Economic Analysis, household income survey data from the Census Bureau, and employment numbers from the Bureau of Labor Statistics. Finally, he provides some solutions to help clarify the uncertainty around the data that will better inform end users.
How Is This Research Useful to Practitioners?
Many practitioners, such as government agencies, companies, and individuals, make important decisions based on economic releases. Providing more clarity around the possible uncertainties in the data would allow practitioners to be better informed. The author highlights quarterly GDP numbers and their subsequent monthly and yearly revisions as an example of transitory statistical uncertainty. Clarity could be improved by showing the measure of errors around those estimates.
Permanent uncertainty is exhibited by the nonresponses in surveys used to assess current employment levels by the Bureau of Labor Statistics every month. Employment information is gathered via surveys sent to households and employers, and a percentage of these surveys are not completed. The bureau’s solution is to assume a number for the incomplete surveys based on an average answer from completed surveys with similar demographics. The author is critical of this method and believes that more research should be spent on creating more insightful analysis of this uncertainty rather than on making assumptions.
The author defines conceptual error as the incomplete understanding of the information that the official statistics provide. He highlights the seasonal adjustments to the employment data as well as the periodic changes to the definition of GDP as an example of this error. A sophisticated algorithm is used to adjust seasonal employment data. The author believes that the employment data should be presented without adjustments rather than applying a black box approach. This raw number could be more useful to end users when comparing trends in the data, especially with year-over-year comparisons.
How Did the Author Conduct This Research?
The author builds on the work of Oskar Morgenstern, who argued forcefully for a change in reporting standards by federal agencies. Morgenstern stated in his book On the Accuracy of Economic Observations (1963) that “perhaps the greatest step forward that can be taken, even at short notice, is to insist that economic statistics be only published together with an estimate of their error.” A method to clarify the errors surrounding GDP data might be to follow the lead of the Bank of England, which uses fan charts when reporting GDP growth. Point estimates, along with the range of errors, are shown in an easy-to-read graphical fan chart issued by the UK Office for National Statistics.
Morgenstern briefly mentioned nonresponse rate in his book, but the author argues that this problem is a bigger issue with current survey data than it was in the past. A suggestion would be to show all of the possible ranges for the missing data rather than to use assumptions. And although the intervals might be quite large, the author believes that it would create a more accurate understanding of the uncertainty around the estimates. He also suggests that surveys could be improved in order to lower the nonresponse rate. He believes that communication about the uncertainty of nonresponses should be addressed.
Abstractor’s Viewpoint
Issuing point estimates without the accompanying range of uncertainty could cause end users to assume that these estimates are more accurate than they truly are. Policymakers and others who make important decisions based on these numbers need to be aware of the margin of error in the data. Presenting a range of uncertainties along with the estimates could help improve decisions by clarifying the data.