Image Source: DepositPhotos
Data revisions or political data manipulation? When revised data support the in-group’s political policies, members of the out-group sometime allege that the numbers were cooked for partisan advantage. But the tension between speed and accuracy better explains data revisions than nefarious activity.The monthly employment report provides a good example. Market analysts and traders watch it closely, particularly the number of jobs added or lost. That data usually comes out the first Friday of the following month. So August 2024 data will be released September 6.Now consider that the United States has nearly 12 million establishments that employ workers, including large corporations, small businesses, government agencies and non-profit organizations. How does the data agency, the Bureau of Labor Statistics, get the data out just six days after month-end?
The true, hard data, won’t be collected for months. Every establishment submits a quarterly report on employment that is used to assess unemployment insurance taxes, but those reports are months away. So the BLS surveys establishments every month. Virtually all very large employers submit reports electronically. The August data actually refers to the mid-month pay period, so the figures for large employers are known and pretty accurate.
But BLS statisticians know they should also monitor mid-sized and small organizations. To do that, they have a sampling program that solicits information from small and medium organizations. Some of these establishments submit their information online, but some of the smallest submit paper reports. When it’s time to add up all the numbers, some of the numbers have not come in. Statisticians don’t simply write in zero. Instead they assess the changes where they do have comparable figures and apply those changes to all similar establishments.
Sometimes the data on a particular report look screwy. The company that last month reported 190 employees this month has only 19. Did they cut back severely, or did a clerk drop a zero? The computer may ask for human eyes to look at the report.
When the small and medium-sized businesses have been tabulated, BLS has another huge decision to make: guess how many total establishments there are. Let’s say they believe that small businesses added 0.21% to their payrolls. They apply that number to the total size of the small business sector. But what is the total size? How many new businesses were formed and how many old companies closed up? The idea of applying a sample to the whole universe of establishments becomes difficult when we don’t know the size of the universe. The size won’t be known until those quarterly reports are collected and tabulated, many months away. To solve this challenge, BLS created a “birth-death” model that projects the number of establishments created and closed, but it’s much less accurate when the economy is changing direction rather than continuing on a steady path.
In the two following monthly’ reports, BLS reports revisions to previously reported data. That’s the result of more information about those medium and small establishments. Then a year or more later, all of those quarterly tax reports have been tabulated and the previously reported data are “benchmarked” to the highly-accurate data.Our data collection system does not allow to have both rapid publication and high accuracy. An analyst who wants only accurate data cannot look at recent numbers, and the analyst who wants the latest information must settle for estimates that will be revised later.Similar processes are used for most economic data series, including gross domestic product. (The Consumer Price Index is not revised.)A cool thing about the U.S. data system is that the statisticians are very open about their methods, which are Bureau of Labor Statistics. (This article is just a rough summary of one report’s methodology.) Data from individual establishments is held confidentially, but everything else is in the open. The analysts take questions from the public, and I’ve spoken with a number of them over my years in the economics profession. They have all been happy to help. They seem delighted for someone to take an interest in their work.In the case of a recent allegation, an annual benchmarking was thought to be a leak, even though the Bureau of Labor Statistics the release date over a month earlier.Politicians of both parties will spin the official reports to look favorable to their policies, but the underlying data are probably the best that good, dedicated statisticians can come up with.More By This Author:Beyond Surge Pricing: How CEOs Can Navigate The Price Gouging DilemmaIgnore The Hype: A Targeted Approach To Economic News For ExecutivesFinancial Markets Have No Present Tense—Trades Are Gambles
No Comments