No, I don’t like the latest employment numbers.
The Bureau of Labor Statistics reported today that the number of workers on U.S. nonfarm payrolls, as measured by their survey of establishments, increased by a seasonally adjusted 88,000 workers in April, of which 25,000 were government jobs. On Wednesday, Automatic Data Processing gave its estimate as 64,000 private sector jobs. The ADP estimate is based on actual payroll data, not a survey, and covers twice as many individual businesses as the BLS. When you add in the public employees (64 + 25 = 89), ADP called the nonfarm payroll number essentially spot on this month two days before we heard from BLS.
Unlike many analysts, I also pay attention to a separate employment estimate that comes from the BLS, which is based not on asking businesses how many people are working for them, but instead comes from sending interviewers to pre-selected residential addresses to ask how many individuals at that address are working. This BLS household survey claimed that the number of people working in America fell by 468,000 in April compared to March, on a seasonally adjusted basis.
I favor looking at all these numbers together because one survey can pick up things the other two might miss. One of the blind spots of the BLS and ADP establishment data is that, by definition, they cannot directly measure somebody who’s employed with a company that didn’t exist at the time the survey was set up. One of the ways that BLS tries to correct for this is with their CES Net Birth/Death Model, which tries to estimate what might be going on with new firms that aren’t in the sample on the basis of regular seasonal patterns and what is currently observed for those firms that are sampled. This imputation led BLS to suspect that there were 317,000 new jobs this April that they did not actually observe, which is three times the average absolute monthly adjustment for 2006.
Michael Shedlock and Barry Ritholtz
are deeply skeptical of this birth/death adjustment;
Dash of Insight offers a more balanced defense of the procedure. But I’m reminded of Tim Kane’s claim that the household survey can do a better job of catching turning points than payroll-based measures, for the reason that the birth/death imputations will systematically miss when things are dynamically changing. Yes, we usually see a large number of new jobs from new establishments in April. But did it happen again this year? No way to know from the establishment data alone.
I’ve been advocating combining the three separate estimates (BLS establishment, ADP, and BLS household) with the first regarded as 8 times as reliable than the other two. For April that calculation comes to
(0.8) x (88) + (0.1) x (89) + (0.1) x (-468) = 32,500 new jobs.
With 25,000 of those jobs coming from the government, I’m thinking we could have had essentially zero net job creation from the private sector in April.
Economic Policy Institute calls the April jobs report “a potentially ominous signal that the slowing economy is finally catching up to the job market.” Kash Mansori, William Polley, and Pro-Growth Liberal are all worried as well.
Ah, please, don’t make me put the sad little face back up so soon!
Technorati Tags: macroeconomics,
employment,
BLS,
BLS household survey,
ADP
Good overview/integration, Professor.
If April (real) retail sales come in flat or negative next Friday, I vote for moving to Mr. Frown emoticon.
Wow.
I don’t know when you’ve writ a more alarming note. 32,500 jobs…so sobering.
Time to resurrect the ‘This is but one month’s data, and not a trend.’ and cross your fingers, legs, eyes ,…everything.
As Barry Ritholtz points out, 49,000 of the 317,000 B/D adjustment is attributed to growth in the construction sector. That certainly seems suspect. The model probably automatically imputes an increase due to the spring weather, regardless of economic conditions in the housing market.
Ah, but it is a trend. Year-over-year figures show that employment is only 1.4% greater than last April. This is a smaller increase than, say, May 2006 when employment was 2% greater than the year before.
Manufacturing is worse. The Y-O-Y trend started declining as of last fall. April 2007 manufacturing employment was 1.1% lower than April of 2006.
Payroll Employment Report and the Birth/Death Model
How should we interpret today’s employment report, showing a growth in non-farm payrolls of 88K, a bit less than expected? There is also an increase in the unemployment rate and a different picture of job growth from the survey of
While at the door, for this household survey, why not ask how many are unemployed; if, of course, you wish to know how many are unemployed. If ‘household’ is to be used, in light of the evidence that nearly 10% of US jobs are held by undocumenteds, shouldn’t a distinction be made as to holds the jobs? I understand that social security knows that many of the payroll jobs use fake/stolen social security numbers. Household seems a deliberate inclusion of those known to be off the ‘books’. Seems to me, these things make a big difference when calculating E/P and unemployment. Take out undocs and E/P is ~.60 and unemployment more than 10%.
Of course they ask that, Ken, and a good deal more– that’s where our unemployment statistics come from. Everybody at the address is counted, and a representative sample of any addresses where people might be living is intended to be included, whether documented or not. For that matter, it is claimed that a large number of “undocumented” workers also get counted in the establishment survey, as this article in the Wall Street Journal discussed.
I’m going to quibble with the terminology in your title. The Employment Situation report tells us nothing directly about new job creation. We may surmise new job creation has slowed, but that is only a guess (or perhaps it can be concluded by combining this report with other data). From this report alone, an equally strong possibility is that old job destruction has accelerated.
I agree completely, knzn. But I often try to keep one eye on Google when I come up with a title.
Don’t forget the 22,000 bartenders created.
If it were old jobs being destroyed wouldn’t we be seeing a rise in initial claims?
Initial claims looked like they were rising earlier in the month; then they came back down. I’m not sure. Right now I don’t have a model to tell me if the increase was enough to account for the slowing growth of employment. I would guess that new job creation has actually slowed. But I was just making a semantic point about what we can conclude from the Employment Situation report alone.
Claims fall at the end of the month, then rise toward the middle typically. So saying they “came down” is a little misleading. I don’t go by one week nor 4 week averages.
April’s monthly average will be around(the last week will be revised upward IMO) 325,000.
Claims usually react to recessionary pressures as layoffs mount. I don’t think we are at the point of layoff problems yet, but some shots are being thrown across the ship.
There is an inventory problem, but by historic norms it is quite minor. Moreover, much of the retail and wholesale problem seems to have been resolved with weak imports. Manufacturing still has a significant IS problem and it should be enough to keep industrial output weak but it does not seem to be big enough to generate significant lay offs.
Moreover, much of the retail and wholesale problem seems to have been resolved with weak imports
That is only if consumer spending stays strong. 2 more months of what we had in April, retailers will be laying off in droves.
Manufacturing will weaken more as the builder bust continues to deepen. We are only halfway down from that point.
How can the birth/death model be at all defensible when the government won’t let anyone audit it?
I don’t understand the rationale for mixing in the household survey number with the other two. As I understand the survey data would affect the total number of people seeking jobs, therefore the unemployment ratio through changing the denominator. What is the reason for adding it to the number of jobs created?
Sorry I see you are referring to a different number from what I was thinking?
The current uptick in unemployment, coupled with low job creation will further dampen the demand for housing, therefore making the current situation much worse, and deeper than expected.
Those in denial about the housing bust were clinging on ot the strong jobs reports as theri only hope of a rebound, but those hopes are now dashed
Hal, the household survey collects several different numbers. One is, as you say, the number of people seeking jobs, from which come our unemployment statistics. Another is the number of people who have jobs. That’s the number I’m comparing with the number of employees that establishments claim are working for them.
“Don’t forget the 22,000 bartenders created.”
Dr. Frankenfurter”s productivity has improved.
Me and my ‘undocumented immigrant’ fellow countrymen standing around Home Depot parking lots and on street corners waiting for a day job (on which we do not have to pay taxes or other payroll deductions) do not get counted in the government’s job numbers. Maybe we are part of the birth and death count that government conjurs up in order to jocky the jobs figures to fit political needs of the moment.
U.S. Government has found a way to dramatically increase the number of manufacturing jobs in U.S. Making burgers and fries is now to be considered manufacturing. However, there is a downside to this: eventually, all U.S. manufacturing jobs are exported.
In calculating the ADP ‘spot on’ number, where do you get the public employees from? (i.e. is it available on wednesday as well?!)
Secondly, what is the intuition behind the weightings of 80/10/10%? While your earlier post would suggest using approximate sample sizes for the two surveys, I’m guessing, given the round numbers, that this is not the case, at least not exactly!
Anonymous, yes, one has to wait until the Friday BLS report to get the true government employees number, but this is not that big and does not vary too much. As for the weights, the error of any of the three estimates is not one of sampling size– if you have ADP’s 22 million workers, you know for sure the population mean of that particular group. Instead, the error is one of specification– ADP is not a random sample from the population one is interested in (all the workers in the United States), nor is the BLS establishment survey, nor is the BLS household survey. The 80-10-10 is my own subjective assessment of the relative sizes of these specification errors– I’m guessing that the specification error in the BLS household survey, as measured by its average squared value, is 8 times the size of that for the BLS establishment survey. These numbers are rough, I admit. But one thing I’m confident about is that they are better weights to use than the 100-0-0 that seems in practice to be what is used by many other analysts.
Thank you for the reply. I know that this isn’t quite what you you were suggesting as an use for ADP, but I share my results anyway!
I tried using the ADP with the lagged (i.e. available as of ADP release date) Government figures as a simple predictor for the Friday release. (Jan 01- Present). The results aren’t great (compared to the Bloomberg mean survey):
Number of time sign called correctly:
ADP+GOV Estimate: 79%
Survey Estimate: 84%
ADP+GOV foresight: 83% (i.e. using the next GOV figure rather than just what was available on an ADP release date)
Charlie (Anonymous above!)