“Household Survey – The unemployment rate edged down to 5.4% in August from 5.5% in July. The decline … reflects a contraction in the labor force (-152,000) and a small increase in employment (+21,000)… The nation’s unemployment rate overstates the strength in the labor market. In August, the participation rate (civilian labor force as a percent of the civilian noninstitutional population) dropped to 66.0% from 66.2%. The unemployment rate does not capture the steady decline in the participation since the 2001 recession and the expansion thereafter. Typically, the participation rate increases as the economy recovers. By contrast, in the current expansion, the participation rate has shown a net decline in the entire expansion period…”

http://www.northerntrust.com/library/econ_research/weekly/

His view is that it is hard to imagine Greenspan continuing with the increases given this latest installment. Comparing this recovery to average recoveries since WWII, this one, in his view is a little anemic: nearly 10 Million jobs short of the average recovery 11 quarters after the trough.

However, like in the discussion on Daniel Pipes (!) here too I get the feeling there is too much focus on details. On my own site I put some graphs (made with the bureau of labor statistics). Developments for 1994-2004. The trend is similar on a lot of graphs. (you will have to look there, I can not post pictures here of course).
You really should not focus on the numbers for august (or september / october).

http://data.bls.gov/servlet/SurveyOutputServlet?data_tool=latest_numbers&series_id=LNS12000000&output_view=data|

Finally, the Bureau Of Labour Statistics also produced two tables with pre-2004 data “smoothed” for the population controls adjustion, for the total employment figures, see the last (fourth) page in this pdf: http://www.bls.gov/cps/cpspopsm.pdf

(I say apparent growth because it is unlikely either the payroll or the household survey are off by millions.)

The issue is not whether the household survey is inaccurate – all economic surveys are, up to a point. But … the household survey isn’t designed to tell us how many jobs the American economy creates each month. It’s purpose is to tell us what percentage of the American workforce is employed – using the special definitions applied by the Bureau of Labor Statistics.

The Cleveland Federal Reserve published a very clear, non-jargony paper on this topic just a couple of months ago, if you want chapter and verse. But the short form is roughly this:

Because the survey is based on a relatively small sample, the BLS uses a population estimate from the Census Bureau to extrapolate the survey into results for the national population. If the census estimate is wrong, the national results will be wrong, too.
Some results, however, will be more wrong than others. Because the unemployment rate (and a related measure, the employment-to-population ratio) have the Census Bureau’s population estimate on both sides of the equation (the numerator and the denominator) they’re less affected by an error in the census data. The estimate of total employed individuals, however, can be thrown be wildly off..

This, as it turns out, is exactly what happened during the 2001-2003 period, when the payroll survey was showing huge job losses, while the household survey was showing moderate job gains. As the Cleveland Fed explains:

In its most recent review of the population, the U.S. Department of Census determined that it had overestimated the U.S. population for the period from 2000 to 2003 primarily because of unanticipated changes in net international migration patterns.

As a result, the BLS notes that the upward trend in the employment estimates produced by the household survey since the end of the 2001 recession is largely a function of this overestimate.

In fact, through the end of 2003, the accumulated overcount of the estimate of employment in the household survey was nearly half a million workers. (emphasis added).

The household survey, in other words, was giving a reasonably accurate reading on the unemployment rate, a reasonably accurate reading on the employment-to-population ratio, and a completely bogus reading on employment growth…

…the accuracy of the household survey as a job creation measure has been disavowed even by Alan Greenspan… As the Great One himself told Congress last February:

“Having looked at both sets of data … it’s our judgment that as much as we would like the household data to be the more accurate, regrettably that turns out not to be the case.”

http://billmon.org/archives/001624.html

So I repeat: It seems we have a case of apples and oranges, you mix total employment with total non-farm employment – yet both the net change of 144,000 in August originally discussed, and the jobwatch.com figures refer to the latter. (Which should have already been pretty clear to you from the monthly data in my first data.bls.gov link: they mention those for June, July and August in the text.)

Below is again the link for the total non-farm data, in graphs and tables.

http://data.bls.gov/servlet/SurveyOutputServlet?data_tool=latest_numbers&series_id=CES0000000001&output_view=data|

There you have it, substract this August’s data of 131,475,000 from 132,507,000, the March 2001 data of jobwatch.com refers to, it is 1,032,000 (non-farm) jobs lost.

The table you refer to

http://data.bls.gov/servlet/SurveyOutputServlet?data_tool=latest_numbers&series_id=CES0000000001&output_view=net_1mth

is a table of one-month changes. It most definitely is not “total employment.”
If this is the source of the jobwatch data then the jobwatch table is
mislabeled.

Second if we suppose the jobwatch table was mislabeled and it should have read
one-month change in the 41st month after a recession began, then the sensible
question would be what such a comparison would mean. After all it tells us
nothing about the comparison in the 40th month, or the 39th, etcetera. In
other words it tells us nothing overall about how the different recessions
compare except the 41st month. And just why is that significant?

Third, you state: “jobwatch.org’s source is clearly the seasonally adjusted
monthly non-farm employment changes.” This is not clear to me at all. First
of all when do you think the recession began? And second does in fact the
job delta in the 41st month after match up to the jobwatch chart?

I don’t think it does. And then we have to do the same exercise for all the
other recessions being compared. Where is the data for that? Have you
done this comparison? (In other words I don’t quite understand your confidence
in asserting this is the source for the jobwatch data.)

Fourth, you state “From January 2001 to August 2004, the total change was
-966,000.”

What that means of course is that you are summing up all the one-month changes
for that interval to get a net change for that time. Now there are several
things about that exercise that puzzle me. First of all why is it necessary?
Why use a one-month delta table to reverse engineer a seasonally-adjusted
non-farm total employment table? Surely the Bureau of Labor Statistics prints
such a thing directly?

Such as the table I found.

See data.bls.gov/servlet/SurveyOutputServlet?request_action=wh&graph_name=LN_cpsbref1

And then that raises an even more perplexing question. When I did a sum of
changes calculation it does not match up to the total employment data that BLS gives.

So I started poking around further.

It turns out that BLS has two monthly surveys of employment: the household survey
and the establishment survey.

On the household survey, quote:

“The information is collected by trained interviewers from a sample of about 50,000
households located in 792 sample areas. These areas are chosen to represent all counties
and independent cities in the U.S., with coverage in 50 States and the District of
Columbia. The data collected are based on the activity or status reported for the
calendar week including the 12th of the month.”

On the establishment survey, quote:

“The employment, hours, and earnings data are based on payroll reports from a sample
of over 390,000 establishments employing over 47 million nonfarm wage and salary workers,
full or part time, who receive pay during the payroll period which includes the 12th
of the month.”

DoDo, since the data you cited is titled “employment, hours, and earnings” and since
it’s “nonfarm” I think we can conclude it’s from the establishment survey.

Since the data I cited has adjustments for census data, which obviously must occur
when you’re sampling the population, I think we can conclude it’s from the household
survey.

Here’s a further quote for the BLS website:

“The household and establishment data complement one another, each providing significant
types of information that the other cannot suitably supply. Population characteristics,
for example, are obtained only from the household survey, whereas detailed industrial
classifications are much more reliably derived from establishment reports.”

And that explains why when it comes to total employment all the BLS data will be derived
from the household survey (and census). The household survey after all covers the
establishments targeted by the establishment survey and for that matter everyone else, while
the establishment survey covers only the establishments.

If we assume the establishment survey is the source of the jobwatch data, we might guess
why they are so evasive about their source. For one their “total employment” is not
total employment but instead only employment at the large companies covered by the
establishment survey. That also explains why their chart was published with percent
changes in employment instead of employment numbers directly. After all many readers,
seeing a total employment figure of only 46 million, would think something was amiss.

Tidying up, DoDo you suggested: “If you factor in the change in population controls,
the figure would actually be *larger*: for 2003, and 2001 and 2002…”

No, it wouldn’t. The establishment survey has no population controls. The census doesn’t
enter into it.

Finally the fact that farm employment is not included in the establishment survey
is the least of what is not included. In fact I think that BLS needs to do a better
job of labeling their data. It may be obvious to a BLS employee which is establishment
survey and which is household survey without it being labeled directly but it’s hardly
surprising that others might be confused.

On the other hand it does seem a bit unnatural exercise to go to a one-month job change
chart and sum up to derive a total employment when there are already total employment
charts and data prominently labeled and displayed as such by the Bureau of Labor
Statistics.

http://data.bls.gov/servlet/SurveyOutputServlet?data_tool=latest_numbers&series_id=CES0000000001&output_view=data|

[Seasonally NOT adjusted: http://data.bls.gov/servlet/SurveyOutputServlet?data_tool=latest_numbers&series_id=CEU0000000001&output_view=data|]