Serious Problems with the BLS Unemployment Numbers

By Dirk van Dijk

While the numbers that come out of the BLS on employment and unemployment are the gold standard -- they're the ones that everyone uses -- there are some serious problems with the numbers. The most significant of these, to my mind, is what is known as the Birth/Death adjustment.

This is an attempt by the BLS to correct for new companies being formed that provide jobs but which are not picked up in the official statistics, and also the firms that go out of business that are hidden from the official numbers. Here is how the BLS puts it:

The net birth/death model component figures are unique to each month and exhibit a seasonal pattern that can result in negative adjustments in some months. These models do not attempt to correct for any other potential error sources in the CES estimates such as sampling error or design limitations.

Note that the net birth/death figures are not seasonally adjusted, and are applied to not-seasonally-adjusted monthly employment links to determine the final estimate.

The most significant potential drawback to this or any model-based approach is that time-series modeling assumes a predictable continuation of historical patterns and relationships, and therefore is likely to have some difficulty producing reliable estimates at economic turning points or during periods when there are sudden changes in trend.

So given that the economy has been dropping jobs for 18 months and is in the middle of what is arguably the worst recession since WWII, is it rational to expect that the net effect would be for this adjustment to consistently add jobs? Here are the adjustments since last April 2008 from the BLS (by way of this blog):

So we are led to believe that in every month but one since then, the formation of new companies has added more jobs than companies going out of business have subtracted from the total job picture. These are not trivial numbers either, with 185,000 jobs added in June alone through this statistical magic.

The official numbers peg January as the worst month of job losses, but it is the only month where the B/D adjustment actually caused a net decline in jobs. The effect since the start of the year can be seen in the graph below (from here). The B/D adjustment over the last three months totals 631,000 jobs, or more than the official job losses in any of the last three months.

Now look at some of the individual line items in the tables. With housing construction falling off a cliff at a rate that has not been seen since the Great Depression, do they seriously expect us to think that with the exception of three months in the winter that new construction firms are being formed that are hiring thousands of people a month? Do you really think that enough bars, restaurants and hotels opened up in the last month to provide 87,000 uncounted jobs?

Remember that if McDonald's (MCD) or Yum Brands (YUM) open up a new store, those numbers would get into the official count. That 87,000 would have to entirely come from Mom & Pop-type operations.

Admittedly, over time it makes some sense to make an adjustment like the B/D model -- without one the numbers would become fairly seriously distorted over time. However right now, the adjustment looks very much wrong.

It serves as a reminder that the BLS numbers are estimates, and should be treated as such, even if they are generally the best estimates we have. It also means that as bad as the official numbers look, the reality is worse.

Comments

[Dave Wrixon:]

"The most significant potential drawback to this or any model-based approach is that time-series modeling assumes a predictable continuation of historical patterns and relationships, and therefore is likely to have some difficulty producing reliable estimates at economic turning points or during periods when there are sudden changes in trend."

Which makes it unfortunate that most pundits use them for exactly this purpose without questioning their fallibility.

Jul 05 05:32 AM

[CautiousInv...:]

The BLS has acknowledged that B/D adjustments are slow to respond to "turning points" such as that which occurred in December, 2007. Suffice to say that is an understatement.

The latest black box numbers suggest the B/D employment increases grew by 308,000 in 20082H and accelerated to 523,000 in 20091H as the recession was entering its job execution phase. Not only are the adjustments slow but they are directionally wrong.

In the month of June the black box determined that B/D adjustments increased employment in manufacturing and construction while employment in health and education declined. This is at odds with everything we know about employment trends.

I have no idea as to whether these adjustments are subject to political influence but it is a possibility I would not rule out given the current state of affairs within the financial sector. By comparison, tinkering with employment numbers would viewed as petty and pardonable.

Jul 05 06:40 AM

[Roger Knights:]

Surely these birth estimates must eventually be corrected, based on the dearth of income tax withholding deposits set up by these non-existent new businesses, right?

Right? (Hellooo?)

Jul 05 07:26 AM

[M-P:]

These are the type of pieces I love to see on SA because they educate and elucidate. The B/D adjustment is suspect right now as historicals are not that as relevant - frankly, BLS could do some sort of exponential smoothing to re-weight more recent numbers ... if you look at overall forecasting in the corporate sector, exponential smoothing has been found to be fairly accurate - more so than more complex models. I'm glad someone is unpacking to assumptions to see how tenuous they really are ... the true unemployment picture is a significant anchor to growing our GDP and improving consumer spending ... the new normal is still pruning.

Jul 05 10:05 AM

[Alan von Al...:]

The basis of Birth/Death is small business and self employment, which are unreported and immeasurable by BLS surveys. NFIB has a better handle on small business. seekingalpha.com/artic...

Anecdotally, my sources report that Shell and Conoco terminated all of their 1099 consultants in Houston, several hundred (perhaps as many as a thousand) highly paid engineers and geologists.

Jul 05 01:06 PM

[John Lounsbury:]

Dirk - - -

This is a very good article. I don't want you to take it as an attack on your fine analysis, but I need to develop some ideas that are outside of what you were addressing

Actually, the B/D adjustment is really just noise in the total unemployed number. This is based on the population (household survey) which has a measurement error uncertainty of +/- 320,000 (approximately, it varies from month to month, but is apparently always more than 300,000). The B/D ajustment to total employment/unemployment is (according to the BLS) less than the raw number (which recently has been as large as 200,000+). Therefore, although the B/D adjustment is suspect at extreme economic conditions, the inpact on the total unemployment rate is less than the fundamental measurement uncertainty.

All that being said, I now have to ask: Why to they do it? Does it actually help offset some of the measurement error (polling error) in stable economic times? The B/D adjustment continues to be the subject of much discussion. However, with close to 15,000,000 unemployed, a measurement error of +/- 320,000 has an impact on the number employed of +/- 2.1% and a B/D adjustment less than 200,000 has an impact of less than 1.3%. Looking at the unemployment rate, +/- 2.1% of 9.5% means the unemployment rate is 9.5% +/- 0.2%; +/-1.3% of 9.5% means the unemployment rate is changed by less than 0.12%. If the B/D adjustment is accumulating errors as you suggest, perhaps the statement could be modified to a change of less than 0.3% or 0.4% in the unemployment rate. I am sorry that I have never been able to locate any specifics on what fraction of the B/D estimate actually does apply against the uemployment rate - the DOL documentation just says it impacts only fractionally. Fractionally could mean 0.99 or 0.01, or anywhere in between, without further definition.

Another factor to consider is that we have just gone through the seasonal maximum in B/D adjustment and it declines for the rest of the year. I believe it actually subtracts jobs in one month (December or January, I think).

There are much bigger factors to be examined concerning how unemployment rates are calculated. The biggest among these is the way that part-time employment and reduced weekly hours for full-timers is addressed. in my opinion.

Jul 05 02:20 PM

[John Lounsbury:]

Dirk - - -

Looking back at your data (last chart), I see the B/D number was very high for January, so the month that is usually subtractive must be December, assuming of course that my memory about the seasonality of B/D is correct. Somebody, somewhere, had a graphic on the B/D seasonality that I saw within the past week or so, but, unfortunately, I did not bookmark it.

Jul 05 02:24 PM

[Market Sniper:]

Good article and good comments. Special kudos to you John. You have explored this issue more throughly than any one else here at SA. From what I have been able to gather, the birth/death deflator may give a fairly accurate indicator during periods of economic growth. However, it horribly skews the numbers during a contraction. Shaddowstats estimtes that the under reported unemployed is 2.5 million in the last 12 months due to this factor alone.

Jul 05 02:28 PM

[Jeff Nielson:]

In FACT, given that the birth/death model ADDS far more jobs than it subtracts each year, the January adjustment is merely clawing-back some of the phony jobs from the PREVIOUS 12 months.

Defending this statistical abuse clearly demonstrates tunnel vision, or simply a lack of understanding.


On Jul 05 02:24 PM John Lounsbury wrote:

> Dirk - - -
>
> Looking back at your data (last chart), I see the B/D number was
> very high for January, so the month that is usually subtractive must
> be December, assuming of course that my memory about the seasonality
> of B/D is correct. Somebody, somewhere, had a graphic on the B/D
> seasonality that I saw within the past week or so, but, unfortunately,
> I did not bookmark it.

Jul 05 04:19 PM

[Jeff Nielson:]

When the author states that the birth/death adjustments "are not trivial numbers", in fact they are.

The BLS monthly jobs reports for at least the last 6 months have become TOTAL fiction, with the birth-death adjustments nothing more than minor "window-dressing". This is completely obvious merely by looking at the weekly lay-off numbers.

Last year, when weekly lay-offs were totally LESS than 1.5 million jobs per month, the U.S. economy was ALREADY losing significant numbers of jobs each month (on a net basis) - meaning that last year, when the U.S. economy was stronger, it was producing LESS than 1.5 million new positions per month (to offset some of the layoffs).

THIS YEAR, with weekly lay-offs totaling 2.5 to 3 MILLION - every month - and with the U.S. economy producing far FEWER jobs, we're supposed to believe the LIES that the U.S. economy is only losing 350,000 to 700,000 jobs per month.

Given the weekly lay-off numbers this year, the reality is that the U.S. economy is losing AT LEAST 1.5 million jobs EVERY month, and probably significantly more (see www.bullionbullscanada...).

Jul 05 04:25 PM

[437339:]

re

The BLS numbers are pure BS . They don't count self-employed-folks, or owners of small business . ADD to this the number of folks no-longer counted as unemployed as their benefits have run out . Add also to this the folks who are working part-time , needing full time work. NO accounts are made for folks working for a fraction of what they formerly made as well .

Jul 05 05:50 PM

[Jeff Miller:]

Dirk -- We all want the best available data. To be fair-minded, why not take a look at how the BLS responds to these criticisms?

Since we have actual data from state employment offices (with about a nine-month delay) we can now go back and evaluate the critics of the birth/death adjustment. oldprof.typepad.com/a_...

Put another way, the birth/death adjustment has improved the actual job count in every quarter since it has been implemented, including recessionary times.

The real question is not whether this is an effective method, but why it works so well. I am working on an article on this theme.

Jul 05 07:34 PM

[TeresaE:]

And a 1099 "employee' will NEVER be counted in employment/unemployment statistics.

Just like the owners of small businesses are never counted.

No unemployment check, and "magically" employed according to Washington.


On Jul 05 01:06 PM Alan von Altendorf wrote:

> The basis of Birth/Death is small business and self employment, which
> are unreported and immeasurable by BLS surveys. NFIB has a better
> handle on small business. seekingalpha.com/artic...
>
>
> Anecdotally, my sources report that Shell and Conoco terminated all
> of their 1099 consultants in Houston, several hundred (perhaps as
> many as a thousand) highly paid engineers and geologists.

Jul 05 10:02 PM

[Pax Americana:]

First class Post...and great reponse from everybody, especially
John lounsbury whom seems to have researched this problem with
care, personally I am in agrement with John .on attributing more importance to the "part time workers" aspect of this analysis.

Jul 06 05:33 AM