When the September employment data were released by the Bureau of Labor Statistics (BLS), depending on political persuasion, the news was either excellent or it was a sham. We saw reactions like those former General Electric CEO Jack Welch, who tweeted a suggestion of manipulation. On the other hand, the Obama administration has made political hay with the rapid fall in the unemployment rate in August and September.
Every month the BLS takes two surveys relative to employment, the Household Survey (officially titled The Current Population Survey), and the Establishment Survey (The Current Employment Statistics Survey). Both surveys have acknowledged flaws and both have a significant bias that pushes the number of jobs upward and the unemployment rate lower. To correctly interpret the data, one must understand how the statistics are calculated, how the biases are imparted, and the magnitude of those biases.
The Household Survey is used to calculate the various employment and unemployment indexes and rates. There are several of these indexes. Most of the public only hears about one of them, the one the BLS refers to as U-3 (7.80% seasonally adjusted (SA) for September). The public may be vaguely aware of one other one, the U-6. The numbers are produced from a monthly survey of 60,000 households. Here are some of the flaws:
•Because the sample of households is small relative to the total number of households, the series is notoriously volatile. In August, for example, the raw data (Not Seasonally Adjusted (NSA)) showed the number of jobs fell by 568,000. In September, that same number showed an increase of 775,000 jobs (NSA). The BLS reported this as 873,000 SA which is the number that the media got all excited about. Using the NSA data, over the two months, 207,000 jobs were created, or 103,500 per month on average. This leads to a very different conclusion from a single 873,000 data point.
•In 1994, the BLS changed the way in which it counts “discouraged” workers for the U-3 index. If one is unemployed for more than 52 weeks, even if one continues to look for employment, one is dropped from the labor force. A smaller denominator with the same number employed leads to a higher employment rate and a lower unemployment rate. Ask yourself how much sense this makes in today’s world where the average unemployment duration is 40 weeks and there have been several years where unemployment benefits last for 99 weeks.
•The definition of employment is biased. If one worked part-time in the last 30 days, even baby sitting for a few hours one time, one is counted as employed. There is no weighting for part-time work in the U-3 index.
•The biggest issue with the Household Survey is the seasonal adjustment (SA) process itself. Theoretically, for the year as a whole, the changes in employment by month should add up to the same number, i.e., the monthly SA and NSA changes should each add up to the same amount. And, theoretically, the SA should be calculated once at the beginning of the year. But, for the last few years, the BLS has adopted what they call a “Concurrent” SA process in which they recalculate the seasonal factors every month. The practical result of this method is that every month, all of the 12 seasonal factors change, which means that all of the year to date monthly SA data also changes. As a result, by December, the January number has changed 11 times, the February number 10 times, the March number 9 times, etc. Here’s the rub. The BLS will not publish the changed monthly data on the grounds that they don’t want to “confuse” the data users. Because they do this, the monthly change in the unemployment rate is not meaningful because the number it is being compared to has changed, but the BLS won’t tell us what it has changed to. The September 7.8% SA unemployment rate (U-3) as reported in early October is being compared to August’s 8.1% SA rate (reported in early September) despite the fact that August’s unemployment rate has likely changed due to the calculation of new seasonal factors. The BLS knows what the changed August number is, but they won’t publish it until January, 2013.
All in all, the U-3 unemployment number is deeply flawed and should not be relied on as the business media and even the capital markets do. A better (though still flawed) indicator of labor market conditions is the U-6 measure. For both August and September, U-6 showed an unemployment rate of 14.7%. Unlike U-3, U-6 adds back to both the labor force and to the unemployed “discouraged’ and “marginally attached” workers, i.e., those who have stopped looking for work but still want a job, and accounts for part-time workers who want full time employment. The flaw is that U-6 removes the long-term discouraged worker after 52 weeks of unemployment. Nevertheless, it is still a much better indicator than U-3. John Williams estimates that if U-6 counted the long-term discouraged workers, the unemployment rate would be 22.8%.
The Establishment Survey collects data from more than 141,000 businesses and government agencies. The sample is about one-third of all nonfarm payroll employees in the U.S., and, as such, it is much less volatile than the Household Survey. Normally, the business media concentrates on this survey. This survey suffers from the same seasonal adjustment issues as the Household Survey except that BLS reports the current number (141,000 SA for September) and the revised data from the immediate past two months. It does not report the changes from earlier months, so it is possible that jobs reported in the current month were “borrowed” from earlier months, which aren’t reported until the next January. In fact, Mr. Williams contends that this is precisely what happens in the second half of each year.
Besides the transparency issue in the SA process (which can lead some to the manipulation conclusion) which the BLS could easily remedy simply by publishing the changed data on a monthly basis, the Establishment Survey suffers from a significant upward bias, known as the Birth-Death model. In the 80s, the BLS was constantly embarrassed that it was under reporting the number of jobs in the Establishment Survey by approximately 50,000 jobs per month. That occurred because more small businesses were being established than were being closed. And, one could probably argue that this was also true in the 90s during the tech boom. As a result, BLS adds approximately 50,000 jobs per month to the Establishment Survey report. That seems inappropriate in today’s world.
From all of this, it is clear that the U-6 measure is a lot more reliable than the U-3, the one that is most widely reported. In addition, when dealing with the Establishment Survey, be wary of the 50,000 jobs bias.
When I began work on this paper in early October, I was skeptical that there could be actual manipulation of the data. Mr. Williams has documented at least three cases of manipulation which he says have been confirmed by employees or former employees going all the way back to the 1960s. That is not a lot. Yet, one must worry about the lack of transparency in BLS’s reporting. After all, for the years 2010 and 2011 for which we have final numbers released in January of the following year, there were much lower levels of job creation than originally reported. Unfortunately, the media pays no attention to such revisions, and the bias goes unnoticed.
In an October 9 Wall Street Journal op-ed, Jack Welch defended his tweet, indicating that the economy would need to be growing at breakneck speed for unemployment to drop from 8.3% to 7.8% over two months. While this is quite different from the “manipulation” charge, it does make sense. The fact is, almost all other underlying data point to weaker, not stronger jobs numbers. New part-time jobs dominated the Household Survey data in September. Goods producing jobs actually fell. The National Federation of Independent Businesses index of employment softened in September as did Monster’s employment index. All of this seems to be in direct conflict with a SA increase of 873,000 jobs in September (Household Survey), the largest increase since 1983. The data also show that in August and September, governments added 602,000 new employees. Anyone following state and local government finances knows that number has to be far from accurate.
While there is no direct proof of manipulation, there are a lot of sound reasons, based on flawed methodologies, and based on nearly every other underlying employment data series, not to trust the headline making unemployment data.
Robert Barone (Ph.D., Economics, Georgetown University) is a Principal of Universal Value Advisors (UVA), Reno, NV, a Registered Investment Advisor. Dr. Barone is a former Director of the Federal Home Loan Bank of San Francisco, and is currently a Director of Allied Mineral Products, Columbus, Ohio, AAA Northern California, Nevada, Utah Auto Club, and the associated AAA Insurance Company where he chairs the Investment Committee.
Information cited has been compiled from various sources which UVA believes to be accurate and credible but makes no guarantee as to its accuracy. A more detailed description of the company, its management and practices is contained in its “Firm Brochure” (Form ADV, Part 2A) which may be obtained by contacting UVA at: 9222 Prototype Dr., Reno, NV 89521.
Ph: (775) 284-7778.