The unemployment rate has reached its lowest level of the last four years, and yet some
economists still wring their hands in despair over the anemic job situation. I’m having some
trouble following their reasoning on this one.
The Bureau of Labor Statistics
today reported that 146,000 more people were employed during the month of June than in the
previous month, according to its establishment survey, and 163,000 more people, according to its
household survey. The unemployment rate edged down to 5.0%, its lowest level in 4 years, and in
fact a lower value than seen any time between 1990 and 1996.
That sounded like good news to Outside the Beltway, and Economic
Policy Institute has a nice objective analysis. But General
Glut is generally glum, and Angry Bear is,
well, angry and bearish. The primary concern seems to be that, although the unemployment rate
has come down nicely from its peak during the last recession, the fraction of the population
that is employed has failed to return to its values of 2000.
The difference between these two measures has to do with whether unemployed individuals, when
interviewed by the BLS, describe themselves as actively looking for a job. If they say that
they have been actively looking for a job, they are included in the unemployment rate. If they
say they have not been actively looking for a job, they’re not counted as unemployed, but
obviously can’t be counted as employed either. The BLS designates such people as not being in
the labor force. Angry Bear’s idea is that, although the individuals did not say they were
looking for a job, in a stronger economy, they would have looked and would have succeeded in
obtaining one.
The group that’s exhibited the sharpest drop in labor force participation since 2000 has been
teenagers. As the graph at the right shows, for this group the drop in the participation rate
is actually an acceleration of a downward trend that began in 1978. An analysis of this longer
run trend by the Bureau of Labor
Statistics in 2002 concluded that an important explanation for this trend was increased high
school and college enrollment rates and attention to school work among this group. Rising
affluence of a partner is another factor that could account for a drop in the labor force
participation rates among some adults. Observations like these make it clear that, although a
low employment to population ratio could be symptomatic of economic problems, it need not be,
and certainly should not be viewed in and of itself as an indicator of a weak job market.
Those who suggest that the economy currently is displaying weakness must be basing that
conclusion on something other than the jobs figures alone.
I do grant that the job market is doing much better in some parts of the country than in
others, as the latest BLS map of county-by-county unemployment rates displays.
 td> |
James,
My concerns re the US job market:
1. Slowing job growth;
2. Overly concentrated in real-estate related sectors;
3. Stagnant real income combined with growing real consumption, and thus near-zero savings from labor income.
Gen’l Glut
So, why have the household survey statistics, and in particular the unemployment number, regained such respect recently?
Didn’t most economists dismiss them years ago because of fundamental flaws?
And U6 unemployment gained. That shouldn’t have happened.
Tim, I think that the question you’re raising, which relates to the differences between the household- and establishment-based measures of employment, is separate from the issue I’m discussing here, which is how to interpret the employment/population ratio.
But, since you raise the question, let me go ahead and offer my opinion. I believe that the household and establishment surveys have some nonoverlapping strengths and weaknesses, and making a correct assessment about where the economy is requires you to take a look at both. On occasions where they seem to be giving conflicting indications, I would have expected a majority of economists to take the view that the truth is probably somewhere in between the two.
Its probably been said somewherelse, but what sort of job growth does the economy need to generate to keep up with population growth and immigration??
About 150,000 per month, Harold, or pretty much what we had in June. The concerns are from people who think the employment/population ratio should be going up rather than holding steady.
I’m not the only economists who have suggested a discouraged worker effect. OK, the Krugman “only” seemed to suggest that all of the decline in the participation rate was due to discouraged workers – but you seem to argue that little or none is. I suspect the truth is somewhere in between. This is a bit of a long post not to have even mentioned the discouraged wroker effect as a serious concern.
Larry Kudlow thinks the break even number is only 98,000, which makes the 146,000 number Fabulous!
If you look at the June data you see that the unemployment rate went down as a result of people moving from “unemployed” to “not in the labor force”, rather than from “unemployed” to “employed”.
This is one of the reasons why I asked the original question about why anyone looks at the household survey at all anymore.
Thanks for the reply.
Let me add one small qualification to your observation, Tim. What the BLS data show is that: (1) there were 163,000 more people working in June than there had been in May; (2) there were 161,000 fewer people unemployed in June than there had been in May; and (3) there were 240,000 more people “not in the labor force” in June than in May. Now, one story that would be consistent with those numbers is the one that a careless reader of your statement might think you were suggesting, namely, that 161,000 people who used to be unemployed just gave up trying, and went into the “not in the labor force” category, with all new 163,000 jobs going to people who weren’t previously in the labor force. Another story, which I think pretty obviously isn’t correct either, would be that all of those 161,000 unemployed people found jobs, and none of the new jobs went to people who formerly weren’t in the labor force. To me, the accurate way to present these numbers is that, the unemployment rate went down even while the employment/population ratio stayed the same. I don’t see anything about the combination of those two facts that should lead you to conclude that one or the other statement must be in error.
Could it be that the “not in labor force” number is slower to change?
One possible reason: Individuals who enrolled in a multi-year course of study during the recession are not in the job market yet.
Just a guess.
So, how many individuals are unemployed based on the latest BLS data?
I ask because it’s one of the interesting questions that few, including economists, can answer without digging.
Let’s see if others know the answer…
James – new post up at Angrybear.
Cheers
PGL
A big problem is the quality of jobs, as measured by real incomes, security, longevity, etc.
We have traded millions of high value manufacturing jobs for millions of low value service jobs, so even if the unemployment number is down there is “underemployment” in terms of job quality.
This is eating the heart out of the Midwest (Rustbelt) states (ironically Bush’s economic policies have been best for the Blue states plus Texas and Florida).
The Carnival of the Capitalists for the week of July 11, 2005
Economics. Business. Marketing. Investing. Technology. Real Estate. That and more in the July 11, 2005 Carnival of the Capitalists, which sets up its tents at Multiple Mentality this week.
The assertion that the truth lies somewhere in between the payroll and household survey results is an odd one. There is really only one major series in which the two surveys overlap – the count of those employed. The general impression of the two reports may be sort of averaged, in a rough, kinda none mathematical, touchy-feely way, I guess. When it comes down to the question of “which to believe” as regards the job tally, “somewhere in the middle” is the wrong anser. This is what is known as the fallacy of the middle ground – assuming that, when there is disagreement between two results, averaging those results is likely to give a better approximation of reality. Consider, for instance, the possibility that a third survey found job losses over the past 3 years. Should we toss that one into the mix and take the average, without knowing much about what goes into it? No. We should investigate the characteristics of the new survey, to see how reliable it is.
Between the payroll and household surveys, there is fairly good agreement that the household survey is not as reliable as the payroll survey when it comes to the employment count. Chairman Greenspan need not be bowed to as an unassailable authority, but it is worth looking up his answer when asked whether the payroll or household survey did a better job of capturing net job growth. Unambiguously, the answer is the payroll survey.
Kharris, I could give you a quite rigorous mathematical derivation of the principle that when you have two different measures of the same underlying quantity, the best inference will come from using a weighted average of the two. Under the view of the relative merits of the household and establishment surveys that you espouse, the weight coming out of that mathematical calulation that you would want to place on the household survey would be significantly less than 1/2. However, I doubt that you can defend a weight of zero on scientific grounds.
If you had three measures, the optimal inference would come from using a weighted average of all three.
Here is the actual BLS release, which cites the discouraged worker breakouts along with everything else. Unemployment is down across all measures.
http://www.bls.gov/news.release/empsit.t12.htm
“Scientific grounds” is rather broad. Logical grounds? Yes. If, year-in and year-out, the payroll survey produces more reliable estimates, I prefer it. If the payroll survey uses a substantially larger sample, I prefer it. If self-report of emotion-laden circumstances like employment is notoriously suspect, I avoid it. So I prefer the payroll job tally, and avoid the household job tally.
I’m aware that, given two different measurements drawn from the same population, there is support for combining the results in the way you suggest. The two series in question use different definitions and are drawn from surveys of different entities. There is an argument to be made that these surveys are from different, though overlapping, pulations. It is also not clear what the result would mean, if you claimed that the average of two figures, defined differently, came to a certain number. It is not as simple as offering a “simple mathematical derivation” in this case.
The question of whether to believe the household or payroll survey is laden with political overtones, after partisans skipped from citing which ever result favored their position in a particular month, then off to the other when it suited better. The argument itself is now quite tainted. There is very good reason to think that, over the course of several months, the payroll survey will give a better indication of how many new jobs have been created than will the household survey. How much better? Enough better that, to avoid confusing the issue, I’ll allow the household survey zero weight in any given month. It sparks my curiousity when the two diverge for an extended period, but if I cannot devise any method better than guessing to figure out why they have diverged, I’ll prefer the superior series.
These are two different series. We should treat them as such, recognize which has historically been more reliable, then move on to the rest of the report.
JDH,
I’m going to make an argument about methodolgy I’m fairly confident about, but I would really like to be convinced that I’m wrong.
I gather that the use of multiple, independent measures to increase accuracy of estimates is NOT a mathematical truism, and cannot be applied to the case of household and establishment estimates of employment.
Let’s make an analogy to the introductory chemistry laboratory:
If you’re making 100 observations of the weight of a substance, using the same scale (with a predetermined bias), under substantially invariant weight and pressure conditions, then yes, the sample mean beats the individual esimates.
But if you’re making 2 observations of the same substance, even under substantially invariant weight and pressure conditions, but using two independent and different scales, each with different unknown biases and variances, it’s not clear that aggregation yields a more reliable result.
I think the employment numbers reflect the second case because we do not have a benchmark for measuring the bias of either series. If you add a largely biased measure to an largely unbiased one, the mean is going to be more biased than the consistently best estimate…
Not at all, Kevin. I’m talking precisely about the statistical theory and practical experience on the benefits that result from combining different measures or forecasts that may have been constructed using possibly completely different methodologies or data sets. For a brief survey of this literature you can look at the chapter that my colleague Allan Timmermann has written for the forthcoming “Handbook of Economic Forecasting”: http://www.econ.ucsd.edu/%7Eatimmerm/combine.pdf
May was the 32 nd straight month in which more than one out of every five unemployed workers had been looking for work for more than six months, the longest such period on record. June saw some improvement, with the share of long-term unemployed dropping by 2.3 percentage-points to 17.8 percent. However, such a high long-term unemployment level is unusual when the unemployment rate is low; historically, 5 percent unemployment rates have been accompanied by average long-term jobless rates of 10.7 percent.
http://www.workingforamerica.org/econupdate.htm
How shaky is the US labour market?
The short answer is: shakier than the headline job numbers suggest. Now here is
Since the oldest baby boomers are now entering their 60’s, it seems we would expect there to be a big reduction in the fraction of the population that is in the workforce over the next 20 years.
Of course, I think that “letting” people retire because they hit some magical number that we came up with a generation ago is nuts, and we need to get those folks working.