Jobs reporting…gained and lost confusions
I noticed in comments on Spencer England’s post for this month’s Employment Situation that the 1.2 million jobs lost number was put forth as fact and a reason to dismiss BLS numbers outright. Links to a more careful and complete look at the numbers follows:
Brad Plummer at the WaPo discusses two issues with employment and seasonal factors: Wait, the U.S. economy actually lost 1.2 million jobs in July?
Bill McBride at Calculated Risk adds his own analysis and charts in this post…Payroll Employment and Seasonal Factors
Take the time to note limits as well as information gained in this very general number used for many different reasons….there is also a lot of data available to see trends within the aggregated number.
as you know, i took on that seasonal adjustment problem in my emailed commentary, which is now posted; i explain here why i think july’s report overstates employment, & why last septembers & this coming septembers will understate it..
i believe using the year over year number, which is that 1,830,000 jobs were added, may make more sense than trying to outguess an imperfect seasonal statistical process…
Lots of folks try to draw conclusions about the bias of monthly jobs numbers based on various adjustment factors. That is generally a mistake, though in the case of the “Lehman bias” of strong winter numbers and weak spring numbers, taking seasonal bias into account does seem to help.
What does make sense, I think, is to view data from months with large adjustment factors as more prone to error than months with small adjustment factors. If the adjustment is much larger than the reported job count, then the reported jobs count is much more likely to be overwhelmed by the adjustment. In July, a 5% error in the adjustment factor is the difference between the expected result and the actual result – about 65K jobs. In January, a 5% error in the seasonal factor would have been something like 6k instead of 65k.
So the reasonable thing to do is to give less weight to January and July seasonally adjusted job counts, especially if they deviate from other recent results, without a clear reason.
The real lesson we should take from this is that we shouldn’t really worry about the monthly numbers at all.
For the number reported is the “net balance” between jobs created and jobs destroyed.
The US economy destroys some 15 – 20 million jobs each and every year (I’m projecting from UK numbers but that’s in the right region).
The US economy also creates some 15-20 million jobs a year.
What we see as the “job creation number” is not in fact the job creation number at all. It’s the netting off of those two much larger numbers.
Which is what leads to the huge error bars of course. If we were actually measuring 100,000, or 150,000 jobs created then an error bar of 100,000 would be appalling. But we’re not. We’re measuring 1 million, 1.5 million, jobs created and destroyed each month and so a 5% error bar in either is decent enough statistics. But it leads to the error bar of the net number being huge: often about the same size either way as that net number.
Yes, I know, net monthly job creation numbers are terribly important politically. But I would argue that they shouldn’t be. Yearly ones more so perhaps….
mr worstall’s comment brought to mind something else ive posted on:
BLS indicates in its technical notes that due to possible sampling errors “BLS analyses are generally conducted at the 90-percent level of confidence” and “the confidence interval for the monthly change in total nonfarm employment from the establishment survey is on the order of plus or minus 100,000”
so it seems fairly likely that one report a year will be off by more than 100K anyhow…
Uh, wrong on the magnitudes by a fair bit. On a gross basis, the US sheds something like 75-80 mln jobs a year, not 15-20 mln. In June, for instance, there were 6.4 mln jobs lost through layoffs and quits.
And I’m afraid you have done a better job of convincing yourself than convincing me of you point that the monthly net should be ignored. The monthly net is the way you know whether things are going well or badly, and policy makers need to know whether things are going well or badly. If you just stare at the gross figures, you can know if they are going up or down, but not whether they amount to a labor market absorbing most new entrants. You’d end up doing the mental netting calculation if you couldn’t get a look at the formal net job change figure.
Your view seems to be another plea to consider less information, rather along the lines of people who don’t like core inflation measures or ask which inflation measure is best, so they can ignore the others. Differences and trends, alternate definitions and quantifications all work to inform us better. We need to understand the strengths and weaknesses of each, understand how data series are designed and collected to make good use of them, but in this case, more is surely better.