BLS or Just BS?

A map’s accuracy is directly linked to its scale. The more detail one wants in their map, the larger it will need to be. At some point, a tradeoff must be made between a map’s portability and its accuracy. Said differently, the map is not the terrain. But maps are still useful even if they don’t include every inch of the area represented; they manage to help us get a sense of direction.

Statistics face similar limitations. Done correctly, they can infer characteristics about the population they represent. They cannot, however, fully map each component of the population or they would require too much time in curation to be timely. Think about the data from yesterday’s note on the employment situation. The Bureau of Labor Statistics (BLS) made its annual revision, which, while not as large as the prior year, was significant.

Each month the BLS surveys about 119,000 businesses and government agencies covering roughly 629,000 worksites. While a large undertaking, this is obviously not a full census but a sample of the labor market instead. Other parts of the economy that it does not see directly (e.g., small businesses) must be modeled. Things like business creations and closings (they call it the birth-death model) are estimated, and that estimation can be off the mark and require sizable corrections when more complete data are available.

Once a year, the BLS “benchmarks” their survey to the Quarterly Census of Employment and Wages (QCEW). This is built from unemployment-insurance tax records covering more than 95% of U.S. payroll jobs. The QCEW is much more complete but arrives with considerable lag, rendering it useless for monthly estimates. Instead, they wait for the previous year’s March data to be released and then revise their estimates for the year around that more complete figure0401.

Could the statisticians at the BLS do better? Technically yes, but improvement would come at a cost. Trade-offs include financial burdens, timeliness, and potentially legal constraints (e.g., access to private information). Could the map in your glove box or smartphone be more accurate? Technically yes, but improvements there would require similar constraints be overcome. The BLS continues to be one of the most respected groups of statisticians and that’s no BS.