October 16, 2003

A Genuinely Strange Way of Reporting the Numbers

A genuinely strange way of reporting the numbers:

Smartmoney.com: Breaking News: Weekly Jobless Claims Dip: The number of U.S. workers filing first-time applications for unemployment benefits dropped to an eight-month low last week, providing further evidence that the labor market is stabilizing. Initial jobless claims fell by 4,000 to 384,000 in the week that ended Oct. 11, marking the lowest level since the week of Feb. 8.... The Labor Department revised its preliminary estimate of initial claims for the week of Oct. 4, raising it by 6,000 to 388,000. The department routinely revises its initial estimates...

So last week they reported 382,000. This week they say that last week's number was really 388,000 and report 384,000. They managed to confuse me--and this is what I do for a living.

And I still cannot figure out why the press focuses on "below 400,000." If you want an unemployment claims number that corresponds to a stable unemployment rate, the number is probably not 400,000 a week but 360,000 a week.

Posted by DeLong at October 16, 2003 07:42 AM | TrackBack

Comments

I've been watching these reports since early summer and seems they're always understated in the Advance number. Is that usually the case?

Over the last 8 weeks the difference between the revised number and the advance has the following characteristics:

Mean Diff: 4875
Median DIff: 5500
Stddev: 1726
Min Diff: 1000
Max Diff: 6000

Posted by: section321 on October 16, 2003 08:16 AM

There's been much discussion of the Bushies cooking the data to make things look better than they are. I don't know much about the process in reporting these numbers, but it seems fishy to revise last week's number upward so as to be able to claim a decline this week and say things are improving.

Posted by: Dimmy Karras on October 16, 2003 08:23 AM

Think Dimmy hit this one. They revised last week's number upward so they could report a decline in this week.

Cooking the numbers? We report, you decide.

Posted by: Chuck Nolan on October 16, 2003 09:12 AM


The BLS is not cooking the numbers. If they were, why would they wait 2 1/2 years to do the cooking?

The claims numbers are volatile. It is stupid to look at week-to-week changes, but the media does it anyway. The four-week moving average of claims has been falling for the last four weeks, so hopefully things are improving. (But how many journalists covering the BLS claims numbers know what a moving average is?)

Posted by: dapple on October 16, 2003 09:31 AM

The BLS may not be cooking the numbers, but there's been a fairly long run now of preliminary estimates being revised only upwards. I can understand that preliminary estimates may contain errors, but I would expect that errors would go both ways. Some revisions would therefore be downwards, some upwards. That all revisions are upwards suggests systemic error. It seems to me that BLS ought to either be making great efforts to find and eliminate the systemic error before it does prelimiary reporting or else it ought to include in the announcement that there's a systemic error and these numbers are likely to be revised upwards next week (or both).

Posted by: jam on October 16, 2003 10:21 AM

The BLS may not be cooking the numbers, but there's been a fairly long run now of preliminary estimates being revised only upwards. I can understand that preliminary estimates may contain errors, but I would expect that errors would go both ways. Some revisions would therefore be downwards, some upwards. That all revisions are upwards suggests systemic error. It seems to me that BLS ought to either be making great efforts to find and eliminate the systemic error before it does prelimiary reporting or else it ought to include in the announcement that there's a systemic error and these numbers are likely to be revised upwards next week (or both).

Posted by: jam on October 16, 2003 10:24 AM

The BLS may not be cooking the numbers, but there's been a fairly long run now of preliminary estimates being revised only upwards. I can understand that preliminary estimates may contain errors, but I would expect that errors would go both ways. Some revisions would therefore be downwards, some upwards. That all revisions are upwards suggests systemic error. It seems to me that BLS ought to either be making great efforts to find and eliminate the systemic error before it does prelimiary reporting or else it ought to include in the announcement that there's a systemic error and these numbers are likely to be revised upwards next week (or both).

Posted by: jam on October 16, 2003 10:24 AM

The BLS may not be cooking the numbers, but there's been a fairly long run now of preliminary estimates being revised only upwards. I can understand that preliminary estimates may contain errors, but I would expect that errors would go both ways. Some revisions would therefore be downwards, some upwards. That all revisions are upwards suggests systemic error. It seems to me that BLS ought to either be making great efforts to find and eliminate the systemic error before it does prelimiary reporting or else it ought to include in the announcement that there's a systemic error and these numbers are likely to be revised upwards next week (or both).

Posted by: jam on October 16, 2003 10:27 AM

At least SmartMoney.com reported the revision to last week's numbers. The Reuters version of the story (http://www.nytimes.com/2003/10/16/business/16WIRE-JOBL.html) didn't even do that. It also includes a quote -- presumably referring more to the industrial production release rather than the CPI and unemployment claims data -- describing the data as showing a "strong economy." Not that the 0.4 percent increase is bad, but the industrial production story (http://www.nytimes.com/reuters/business/business-economy-production.html) naturally doesn't report that the revision to the August figure was down 0.2 percentage points. Could that be 'Why Oh Why Can't We Have a Better Press Corps part CCCCLXXV?'

Posted by: Tom Bozzo on October 16, 2003 10:32 AM

I doubt there is any systematic problem in the initial unemployment numbers. Or if there is a systematic problem, it is unlikely that the Department of Labor could fix it easily.

The DoL simply tabulates numbers provided to them by the individual states that participate in the UI program. The "advance" numbers (those reported in the press, and revised a week later) are probably based on incomplete state reports. Unless the "incompleteness" of the state reports is the same every week, there is no way to control for any seemingly persistant bias in the advance numbers.

Unless state labor departments are participating in the conspiracy, I doubt that either the advance or revised UI numbers are begin cooked. In any case, there certainly is a long list of Bush administration lies that are far worse than cheating on the UI numbers by a few thousand.

(Note: The BLS does not tabulate the UI numbers, as I stated before. They come from the Employment & Training Administration. However, the sesaonal adjustment procedure used with the UI numbers was created by the BLS and the Federal Reserve Board of Governors.)

Posted by: dapple on October 16, 2003 10:55 AM

I would suspect that virtually all reports of this number would be revised upward. After all, it is probable that the states will send in a few late reports of further claims. It is rather unlikely that they will send in late reports saying "Hey, those unemployment claims we sent in? We didn't get them." The data is probably garbage anyway, but consistent garbage - we should be comparing the unadjusted numbers to unadjusted numbers week to week. They went up last week. The adjusted numbers are probably consistent compared only with adjusted numbers.

Posted by: rvman on October 16, 2003 11:24 AM

The problem is still a deteriorating labor market in what is finally a rapidly growing economy. We had better hope for radip growth for quite a few quarters or the employment data is going to stay weak. Now, we are not yet adding enough jobs to keep up with population driven labor force growth.

Posted by: anne on October 16, 2003 11:58 AM

A systematic error in early estimates is not necessarily a sign of intentional error. It may well be the result of systematic problems that can only be corrected over time. One of the things that the pros at BLS and other data collection agencies know is that you don't go jiggering the system every time some new problem shows up. Each jigger raises its own problems. If the problem proves short lived, then you've gone and changed everything just in time to have to change it back.

Just for the sake of example, let's assume it costs money to count jobless claims. Next let's assume states are short of money...

In any case, the revisions are small, relative to the level of overall claims and the difference between the recent level of claims and the level in a period of labor market health. Why do something that many government employees might be able to find out about, when it doesn't really work its way into the public impression of the job market.

The right thing to ask, I think, is just what the present level of new claims might mean. The last prior to the recent recession that claims were around this level (4-week average just around 400k) was in 1992, when the US was recovering from recession. Employment growth was a bit better than it is now, so something is going on beyond a simple-minded equilibration of a certain level of jobless claims with a certain pace of employment growth. The jobless rate, on the other hand, was higher, over 7%, the last time claims were around these levels. Again, you have to go beyond simple-minded equivalences to make sense of that.

Finally, Bloomberg notes today that claims have fallen 13% since the official start of recovery 98 weeks ago. Ford and Big Bush both lost their bids for another term after presiding over percentage reductions in jobless claims that were at least twice that size. Is that a worthwhile comparison? Don't know.

Posted by: K Harris on October 16, 2003 12:20 PM

Well then, the question is can we grow above 5% for the coming year. If so, there will be decent job creation finally.

Posted by: anne on October 16, 2003 01:30 PM

A problem (related to the cooked data theory) is that the headline numbers clearly aren't telling the full story on employment. Paul Krugman said essentially that in the Q/A session at a book signing on Tuesday (in Madison WI), when someone asked him whether he thought the unemployment data were cooked.

PK's answer was structured as an extended 'no' followed by an explanation of a couple ways that the headline numbers mask the poor labor market conditions. Before he got to the latter part of the answer, another guy angrily stormed out muttering something to himself about PK's gift for understatement (of all things).

So there seems to be an undercurrent of mistrust on this front. But it may just reflect the observational near-equivalence, to a casual observer, of data that are falsified and data that, because of their incompleteness, just happen to be self-spinning from the administration's perspective.

Posted by: Tom Bozzo on October 16, 2003 02:06 PM

I've never understood why people bother to read the new-unemployment index. First, rumors were, back in the Clinton day, Fed's reported index and labor specialists' own hiring benchmarks began to diverge. A lot. Second, in a stable economy, it's presumed that after six weeks, or six months, all of this week's unemployed have been re-employed, and the index is merely a healthy turnover in search of better wages and conditions. Well, now there are 3,000,000 people who have lost their jobs and not got them back. Do you see any omen of this in the last years' numbers? I sure don't. And what of all those who gave up, went SOHO, filled a slot cheap with no bene's, and then got laid off again. They can't file for unemployment, they're invisible. And I'll tell you, there are a lot of SOHO's, many millions of underemployed's, housekeeping illegals and outsourced overseas workers, you name it, none of them showing up on the charts. But scratch ANYWHERE, you'll find hard times stories of layoffs with no callbacks.

Posted by: Sidi Ifni on October 17, 2003 12:04 AM
Post a comment