September 24, 2003

The Employment News Is a Bit Better Than We Think

The extremely keen-eyed and keen-witted John Kitchen thinks that employment has not fallen as much in the past three years as we economists think that it has: John Kitchen, an economist with the House Budget Committee... studied the performance of payroll employment during the first 20 months of five previous economic recoveries, and found that in each case, the government had to revise up its initial payroll estimates.... During the 20 months after the 1991 recession ended, for instance, the government... bumped up [employment growth by 330,000].... "It might be that we're not seeing as much of an ongoing decline as is suggested in the official numbers," Mr. Kitchen says....

But... Labor Department officials, are skeptical.... One reason is that the department's Bureau of Labor Statistics has been aware of the establishment survey's flaws for many years and has tried to do something about it. Three years ago, it began phasing in a new system for surveying businesses that is more aggressive.... Ms. Getz notes that because Mr. Kitchen's study was based on events that happened before this redesign, it might not apply anymore.... However, she acknowledges that the redesigned survey is going through a trial by fire, because the recession and recovery marks the first test of the new procedures.

That's not the only reason some economists are reluctant to put added weight on the household survey.... "The household survey is completely misleading," says Mark Zandi, chief economist with, a West Chester, Pa., consulting firm...

I've seen John Kitchen work. I'd bet on Kitchen.

Posted by DeLong at September 24, 2003 07:11 AM | TrackBack


> I've seen John Kitchen work. I'd bet on Kitchen.

I've seen a lot of my talented, educated co-workers and friends at work. Now a lot of us are out-of-work, or underemployed. I'd bet on the skeptics, but there's no money in that, and I'm too broke anyway. ;-)

Posted by: William in Beijing on September 24, 2003 08:11 AM

Slippery Data on the Job Market

ESSENTIALLY unchanged" that is how Kathleen Utgoff, commissioner of the Bureau of Labor Statistics, described the unemployment rate in August compared with July. The secretary of labor, Elaine Chao, however, hailed the news, saying, "I'm pleased that the unemployment rate has dropped."

Why the different interpretations? Who is right?

In a sense, both are. The official estimate did fall by a tenth of a point, from 6.18 percent in July to 6.08 percent in August.

But the unemployment rate is just an estimate based on a sample of the population, and like all estimates, there is no guarantee it is exactly right. It could be off because of sampling errors (results differ from sample to sample) and nonsampling errors (respondents may not answer the questions correctly).

Sampling errors alone are enough to lead one to doubt whether unemployment actually fell in August.

If everyone in the country were included in the labor force survey, the unemployment rate would be unaffected by sampling errors. But interviewing everyone every month is not feasible. Instead, the survey is given only to a representative sample of households.

Granted, the number of households is large about 60,000 but if the same survey were put to a different 60,000 households, the rate would probably come out slightly different.

"Sampling variability" is the name that statisticians give to differences in estimates that arise by chance because of the particular sample surveyed. This variability can be reduced by drawing a larger or more efficient sample, but not eliminated....

Posted by: anne on September 24, 2003 08:46 AM

Princeton's Alan Krueger wrote an excellent column on the unemployment and job loss calculations that have been made since the end of the recession in November 2001. Essentially, the household and business surveys are shown to agree on job losses after statistical adjustments are made.

My bet is that we have been and are shedding jobs, and will likely continue to do so this month.

Posted by: anne on September 24, 2003 08:51 AM

The Dept of Labor statistics are hideously bad. One of their monthly unemployment reports completely and consistently contradicts the other (payroll survey vs household survey) and I have yet to see them seriously address the anomaly. We all can objectively see the job market sucks so we choose to put more faith in the payroll survey which is directionally consistent with what we know to be true. I agree with you that Kitchen is probably right and that payroll survey is overstating the number of jobs lost and I also seriously doubt that productivity is increasing at a 7% annual rate. Is Elaine Chao a pathetic Secretary of Labor or does it just appear that way?

Posted by: Joe Blog on September 24, 2003 01:36 PM

"Is Elaine Chao a pathetic Secretary of Labor or does it just appear that way?" -- These are mutually incompatible choices-?

Posted by: joe on September 24, 2003 03:50 PM

Any idea where John Kitchen's original study can be found? No sign on the House Budget Committee site.

Posted by: Terry on September 25, 2003 05:19 AM

Slippery Data on the Job Market

Jog Blog - Here is the answer to your question!

Posted by: Anne on September 25, 2003 11:41 AM

There is other evidence to support the general picture of a jobless (or should it be job-killing) recovery. If employment were actually growing, we should be seeing an improvement in applications for jobless benefit claims. But that series has been stuck near 400,000 claims per week, consistent with not a lot of job creation going on. Also, anecdotal scanning of the scene, of the type conducted for the Fed's Beige book, has been consistently reporting soft labor markets.
It may well be that, down the road, the payrolls data will be shifted up a bit. But it wouldn't really alter what is relevant: productivity has escalated, raising the bar for what constitutes sufficient GDP growth to absorb labor market slack. While B. Delong wants to bet on the Kitchen, he's also been betting on employment being enough in the Basement to justify recommending very aggressive monetary ease.

Posted by: Avery Shenfeld on September 25, 2003 04:52 PM

Krueger's calculations appear roughly correct for one 12-month period -- for August 2002 to August 2003. But the data move around and the issue covers a broader time period than just that one 12-month horizon. If one looks, for example, at the 12-month horizon from March 2002 to March 2003 (the benchmark months for the establishment data), then the difference between the payroll jobs change and the household employment change after the "bridge" calculations like Krueger performs is about 500,000. A 3-month moving average of the difference re the 12-month comparison runs roughly in the 400,000 to 700,000 range over the February to July period. That interestingly corresponds with the rough range estimated based on the historical bias in the real-time data.

Next Friday (10/3) BLS will announce a preliminary estimate of the amount of the payroll jobs benchmark revision for March 2003; it may or may not confirm the view of bias. Even so, the jury will still be out as current months' job growth data (e.g. post-March 2003) will yet be subject to revision from the benchmark revision that will occur next year -- and we won't get a preliminary view of that til October 2004 and the final data til Feb. 2005.

Regardless, even if one makes the suggested adjustments to account for the suggested possible bias, payroll jobs may still be at best flat over the past year. There are increases in the number of folks working for themselves rather than taking a paycheck and a bunch of reservists are getting paid by Uncle Sam rather than their employers (and some other adjustments like the "census" adjustment in January). The unemployment rate appears to have started trending down -- and its movements tend to be preserved through revisions. Some good signs exist -- but not good enough yet if you are a payroll worker without a job. For economic analysts and the media, though, we've tended to treat the flow of payroll jobs data as a clean, real-time series with a clear signal. But there's a lot more noise in the series than we've typically thought, especially near the turning points of the business cycle -- and we need to be careful how much weight we place on any given month's -- or set of months' -- payroll jobs data.

Posted by: JK on September 26, 2003 02:15 AM

Congressional report on the two surveys:

Posted by: Larry on September 26, 2003 12:52 PM
Post a comment