Deconstructing ShadowStats (Part 2): In Search of an Alternative Measure of Unemployment
A few weeks ago, I posted a critique of the alternate inflation measure devised by John Williams for his popular website ShadowStats.com. Several responders asked if I could also comment on Williams’ alternate unemployment rate.
Here, reproduced with permission is the latest unemployment chart from ShadowStats. Like William’s inflation rate, his unemployment numbers run far higher than the official data from the Bureau of Labor Statistics (BLS). This post addresses two questions: First, what is the intended purpose of the ShadowStats alternate unemployment indicator? Second, given what it tries to do, is it calculated in a reasonable way?
In search of the employment gap
The standard unemployment rate from the BLS, known as U-3, focuses on a very narrow segment of the labor market. It tells us how many people are not working but who have actively looked for work in the past four weeks, expressed as a percentage of the civilian labor force. (The civilian labor force includes all people 16 years and older who are not in prison or the armed services.)The ratio of active job seekers to the labor force is highly sensitive to ups and downs in the business cycle. For that reason, monetary and fiscal policymakers watch it closely in deciding whether the economy needs stimulus or restraint. It is not perfect, but if what we want is a short-term business cycle indicator, it is probably good enough.
Williams, however, is looking for something different. His alternate unemployment rate should be understood as a broad measure of the employment gap—the difference between the amount of work that people would like to do and the amount they actually do. Such a measure should include people who say they would like a job but who are not working, whether they are actively looking for work or not. It would also take into account people who are working, but who would like more hours.
The BLS has its own measure of the employment gap, U-6, often known as the broad unemployment rate. It begins from the standard unemployment rate, adds marginally attached persons and people working part time for economic reasons to the numerator, and divides the total by the labor force plus the marginally attached.
By “marginally attached,” the BLS means people who currently are neither working nor looking for work, but who indicate that they want and are available for a job and have looked for work sometime in the past 12 months. Discouraged workers, as the BLS officially defines them, are a subcategory of the marginally attached who report that they are not looking for work because they think no jobs are available, either because of the state of the economy or for reasons such as discrimination or their lack of qualifications. Discouraged workers currently make up about 25 percent of all marginally attached workers. The rest cite reasons like school, family responsibilities, ill health, or transportation problems as the reasons they have not recently looked for work.
People who work part-time for economic reasons, also known as involuntary part-time workers, are those who work from 1 to 34 hours per week and who would like to work more hours, but cannot do so either because they can only find part-time work, or because slack business conditions have forced their employers to cut their hours. It does not include people who work part time because that is what they prefer, or because of “noneconomic” reasons like family responsibilities or school.
Williams accepts the U-6 measure of the employment gap with one exception. As he explains, his ShadowStats alternate unemployment is equal to U-6 plus an estimate of what he calls long-term discouraged workers, or LTDWs, for short. I find that term a little misleading, since, as far as I can see, the people he counts as LTDWs include everyone who wants a job but has not looked within the past 12 months, regardless of reason. Strictly speaking, it would be better to call them “long-term marginally attached persons.” However, Williams is not alone in conflating the “discouraged” and “marginally attached” categories. There is no point in starting a dispute over terminology, so for the remainder of this post, I will stick with his term, LTDW.
Whatever terminology we use, Williams says that these long-term discouraged (or marginally attached) workers “were defined out of official existence in 1994.” To understand what he means by that, we need to look at what happened at the BLS in that year.
What happened in 1994
The changes in BLS methodology regarding discouraged workers, which Williams’ asserts to be the only source of difference between U-6 and his alternate unemployment measure, are described in detail in a 1998 paper by Monica D. Castillo, “Persons Outside the Labor Force Who Want a Job.” Here is her description of how the BLS identified discouraged workers before the 1994 changes:
Persons who were not in the labor force—that is, who were not working at all during the survey reference week and who had not looked for work during the prior 4-week period—were asked if they wanted a job “now,” either full or part time. Those with an affirmative answer were asked the reason they were not seeking a job. Persons classified as “discouraged workers” were those who responded that they did not think they could find a job, either for a job market-related reason, such as their belief that work was not available or, that they could not find any work for certain personal reasons, such as a perception that they were too young or too old, that they lacked education or training, or that they had some other personal handicap.
A footnote in the original text adds that “persons giving reasons other than or in addition to discouraged” were considered to be not available to take a job, and thus were not classified as discouraged workers. Castillo goes on to explain that critics of the pre-1994 procedure saw it as subjective and arbitrary. In response to the critics, the BLS made the following changes starting in 1994:
First, a question to determine whether a person had searched for a job within the prior 12 months was added. Second, a direct question on availability to take a job was added that replaced the inferred measure of availability used with the former definition. Under the new question design, individuals who are not searching for work but who want a job and are available for work are asked the main reason why they are not looking for work. (In the former scheme, individuals could provide any number of reasons for not seeking work. A “nondiscouraged” reason took precedence over a “discouraged” reason in the classification scheme.)
These changes both narrow and broaden the discouraged worker category. They narrow it in that they drop workers from the discouraged category if they have not looked for work within the previous 12 months. They broaden it in that they include workers who give both a “discouraged” reason and another reason, such as family responsibilities. The effect of asking directly whether a person is available for work rather than asking whether they “want a job now” is ambiguous.
According to Castillo, the net effect of these changes, as of 1994, was to cut the number of discouraged workers roughly in half. The 1994 changes also, for the first time, introduced the term “marginally attached” to refer to all people who want a job and have looked for work in the last year, but not the last four weeks, regardless of reason.
Why Williams’ estimate of long-term discouraged workers is implausibly high
Williams does not directly report his estimate of long-term discouraged workers on the public version of his website, but it is easy to reconstruct it from the data he does provide. For example, in May 2015, the BLS reported a U-6 broad unemployment rate of 10.4 percent, not seasonally adjusted. The numerator of U-6 consisted of 16,595,000 persons, of whom 8,370,000 were officially unemployed, 6,363,000 were working part-time for economic reasons, and 1,862,000 were marginally attached to the labor force. The denominator consisted of 157,719,000 members of the civilian labor force plus the 1,862,000 marginally attached, or 159,581,000 in all.
Williams’ adjusted rate for the same period is 23 percent. He says he gets that number by adding his estimate of long-term discouraged workers (to the numerator, and presumably, also to the denominator). Using x to represent LTDWs, we get the equation (16,595,000+x)/(159,581,000+x)=0.23. Solving that for x gives 26,114,100 as Williams’ estimate of the number of LTDWs.
I find that number to be implausibly high. Let let me explain why by comparing it with some other estimates.
BLS estimates of the impact of the 1994 changes. When the BLS changed its methodology in 1994, it did not neglect to look into the impact on reported numbers of discouraged workers. According to Castillo, the 1994 changes reduced the count of officially designated discouraged workers by half. It seems reasonable to suppose that the effect on marginally attached persons would be similar. If we apply the same ratio to recent data, the number of long-term marginally attached persons (those who Williams counts as LTDWs) would be closer to 2 million than to 26 million.
Current BLS estimates of long-term nonemployment. It is not quite true that long-term discouraged workers “were defined out of official existence” by the 1994 methodological changes, as Williams alleges. Rather, they were reclassified. The BLS continues to report the number of people who want a job but did not look in the past year in its monthly Table A-38, even though it does not use the terms “long-term discouraged worker” or “long-term marginally attached.”
However, the number of people that the BLS reports as wanting jobs but not looking for work in the past year is nowhere nearly as large as Williams’ estimate of LTDWs. In May 2015, the BLS reported that there were 92,025,000 people not in the labor force. Of those, an estimated 86,200,000 did not want a job now, leaving 6,536,000 who did want a job. Of those, 1,862,000 were classified as marginally attached, meaning that they were available for work and had searched for work in the last year, but not the last four weeks. Another 962,000 who had looked for work in the past year but not the past four weeks were not currently available for work. That left just 3,712,000 people who said they wanted a job but had not looked for work in the previous year—only one-seventh of Williams’ estimate for long-term discouraged workers.
Nongovernmental surveys. In private correspondence, Williams has told me that he considers the BLS data and survey methods to be “woefully inadequate.” He says that he has his own model for calculating LTDWs, based on “a variety of government and private surveys.” Unfortunately, he has not provided any detailed description of the model or cited his sources, so I took a brief look on my own at nongovernmental surveys of people who are not in the labor force.
Consider, for example, a poll conducted jointly by the Kaiser Family Foundation, the New York Times, and CBS News in late 2014. The survey covered people of prime working age (25 to 54 years), who constitute 52 percent of the total adult population of approximately 250,000,000. Three-quarters of people in that age group were employed. Of the roughly 32,500,000 people in that group who were not employed, about a third reported disabilities, leaving about 21,750,000 who were not employed but able to work.
Of those surveyed who were not working but able to work, the Kaiser poll found that 57 percent had actively looked for work within the past 4 weeks, which would place them in the BLS category of officially unemployed. Another 21 percent, who had looked for work in the past year but not the past four weeks, would fall into the BLS category of marginally attached. One percent did not respond, leaving 21 percent, or about 4,568,000 individuals, who wanted a job, were able to work, but had not looked within the past year.
That estimate includes all people who were able to work, thus excluding people with disabilities, but not all of them were actually available for work. The Kaiser survey did not ask people in so many words whether they were available for work, but it did ask them to give their reason or reasons for nonemployment. Excluding respondents who reported a disability as their main reason for not working, 66 percent cited a lack of good jobs available and 42 percent a lack of education or skills—reasons similar to those that put people in the BLS discouraged worker category. However, 52 percent of the nonemployed cited family responsibilities, 32 percent cited health problems falling short of full disability, and 35 percent cited the fact that they did not need the income. Those answers suggest that at least some of the people who wanted a job, did not look in the past year, and were able to work were not actually available for work. It is hard to say exactly how many, since the Kaiser survey allowed people to give multiple reasons and did not ask which was the main reason.
The Kaiser study did not poll people in the half of the working age population who fall outside the prime working ages—those of high-school or college age, those of working age but older than 54 years, and the elderly. No doubt there are many people in those groups who also want to work, are able to work, and have not looked in the past year. That is likely to be particularly true among those in the 55 to 64-year age group, who account for about 20 percent of the entire population. However, it would presumably be less true of people of high-school or college age, or of the elderly. It seems like a reasonable guess that if the Kaiser survey were extended to all age groups, the number of people who wanted and were able to work, but who had not looked for work for more than a year, would roughly double, to around 9 million. Of those, a somewhat smaller number would actually be available for work.
On the whole, then, the Kaiser survey points to a significantly larger number of people who could potentially be classified as LTDW than we get from BLS Table A-38, but it is still far short of Williams’ estimate of more than 26 million.
Another approach to estimating long-term discouraged workers
Based on what he has told me, it seems that Williams does not rely solely on survey data for his LTDW estimate. Instead, he appears to supplement the survey data with a modeling approach. I have not been able to discover just what model he uses, but I have a conjecture, based on a remark made by Paul Craig Roberts, who is a fan of ShadowStats and has sometimes collaborated with Williams. Here is how Roberts characterized Williams’ approach in a post earlier this year:
When the headline unemployed [U.3 measure] become “discouraged,” they are rolled over from U.3 to U.6. As the short-term discouraged workers roll over into long-term discouraged status, they move into the ShadowStats measure, where they remain. Aside from attrition, they are not [sic; from context the “not” looks like a typo] defined out of existence for political convenience, hence the longer-term divergence between the various unemployment rates.
In an e-mail to me, Williams makes a somewhat cryptic statement that I construe as reflecting the same approach to modeling the problem:
After a year, the short-term discouraged are dropped from U-6 and labor force . . . The discouraged are not finding jobs and remain unemployed and should be counted as same and kept in the labor force of a broader unemployment measure.
I will call this the attrition model. This approach would estimate the number of LTDWs as the sum of a cohort who stopped looking sometime during year t, thus dropping out of the BLS U-6 indicator in year t+1; plus an older cohort who stopped looking in year t-1 and dropped out of U-6 in year t; plus a still older cohort who topped looking in t-2 and dropped out of U-6 in t-1, and so on. Once each cohort passes the 12-month limit, its members drop out of U-6 and into Williams’ category of long-term discouraged workers. However, they do not remain there indefinitely, since each cohort is subject to annual attrition because of death, finding work, starting to look and therefore becoming officially unemployed, or ceasing to want a job due to school, family responsibilities, retirement, or any other reason.
In practice, the size of the entering cohort and the attrition rate will vary from year to year depending on the state of the economy. However, we can approximate the total number of LTDWs for a steady-state economy by using the average values of the variables over the business cycle and the formula for the sum of a geometric progression. The variables we need are m, the average number of marginally attached workers in U-6; a, the attrition rate (or equivalently, the retention rate, r, which is equal to 1-a); and n, the number of years over which we sum the series. With those variables, the steady-state total is equal to m(1-rn)/(1-r), and the number of long-term members of the category will be m(1-rn)/(1-r)-m.
For example, suppose a steady state economy in which the average number of people in U-6 is 1000 and the retention rate is .5. If so, then in 2015 there would be 1000 in the current cohort (that is, short-term marginally attached); plus 500 remaining in the cohort who first became marginally attached in 2014 and are in their first-year of long-term status as of 2015; plus 250 remaining in the cohort who first became marginally attached in 2013, and so on. If we carry the series forward for, say, 50 years, to allow for the possibility that a few people might spend their entire working-age life as marginally attached, that would give us a total of 1000 long-term marginally attached workers (or LTDWs in Williams’ terminology), in addition to the 1000 short-term marginally attached reported in U-6.
Over the 10 years from April 2005 to April 2015, the average number of marginally attached workers, m, in U-6 was 2,035,000 (minimum value of about 1.4 million and maximum of about 2.8 million). As in the above numerical example, I will set n at n=50.That leaves the attrition rate, a (or the retention rate r=(1-a), as the key variable that determines our estimate of long-term discouraged workers.
In a steady-state economy with m=2,035,000 and n=50, the attrition rate a needs to be approximately .07 to produce Williams’ LTDW estimate of 26 million. That is to say, each year each LTDW cohort needs to lose no more than about 7 percent of its members through death, entry into the labor force, or loss of the desire for a job.
Is an attrition rate as low as 7 percent per year plausible? I think not. One indication of more rapid turnover lies in the data that the BLS reports monthly on transitions into and out of various labor force categories. According to the report for May 2015, the category “not in the labor force” as a whole was losing almost 7 percent of its members per month through re-entry into the labor force or death. That category includes people who cannot work or do not want a job. The turnover rate for the subset of “not in the labor force” who want a job would certainly be even higher.
A less recent but more precise estimate of the annual attrition rates for marginally attached and discouraged workers can be found in the study by Castillo, cited earlier in this post. At the time the BLS made its methodological changes 1994, it tracked the results using a large sample of workers whom it surveyed in 1994 and again in 1995. It found the labor market status of the approximately 1 million persons who had been marginally attached in 1994 to be as follows in 1995:
- 30.8 percent were employed
- 16.7 percent were unemployed, that is, had requalified for “in the labor force” status by looking for work within four weeks of the 1995 survey date
- 9.3 percent had looked for work within a year of the survey date, thereby requalifying to remain in the short-term marginally attached category
- 34.6 percent no longer wanted a job
The total attrition rate, then, was 91.4 percent. Only 8.6 percent of those who had been in the short-term marginally attached category in 1994 rolled over into the long-term marginally attached category in 1995. If we look more narrowly at the 300,000 people who were officially classified as discouraged workers, the attrition rate was only slightly lower, 88.4 percent. Just 7.4 percent of official (that is, short-term) discouraged workers in 1994 rolled over to the long-term discouraged category in 1995. Another 4.2 percent were reclassified as marginally attached because they explained their failure to look for work for some reason other than discouragement with their job prospects.
If we apply an attrition rate of 91.4 percent (that is, a retention rate of 8.6 percent) to the 2,035,000 average marginally attached workers of recent years, we get an estimate for LTDW of just 191,000.
In short, I can find no way to get anywhere close to Williams’ estimate of 26 million long-term discouraged workers using any remotely believable parameters for an attrition model.
The bottom line
The bottom line here is that the ShadowStats alternate unemployment rate of 23 percent for May 2015, and similar rates for other recent months, are completely implausible. To get an unemployment rate that high, we have to include millions of people who do not even claim to want a job, let alone make any effort to look for one. We have to assume that the status of long-term discouraged worker is a trap from which almost no one escapes from year to year, by finding work, by making even a single unsuccessful attempt to find work during the year, by changing their mind about wanting work, or by death.
One thing is certain: We cannot, as Williams implies, get to the ShadowStats rate simply by going back to the methods used by the BLS before 1994. We can only get his numbers by using data sources and methods that the BLS never used in the past and does not use now.
Yes, there is something to be said for trying to constructing a measure of the employment gap that is even broader than the than the U-6 indicator offered by the BLS. Rather than relying on Williams’ opaque sources and methods, though, I suggest a new indicator, U-7, which would augment U-6 by including all people who want a job, rather than just those whom the BLS classifies as marginally attached. U-7 would stand at 12.9 percent for May 2015. That is 2.5 percentage points higher than U-6, but it is still a full 10 percentage points short of the ShadowStats alternative unemployment rate.