The monthly jobs report is a flawed statistical series. Its just not flawed in the way you may think it is.

Monthly Non Farm Payroll Data is about a marginal change in measures of short term job changes — it is a tiny change in employment numbers relative to an enormous base of the people in the USA who are employed. What BLS actually measures is a single and relatively small data point — a very noisy series, subject to frequent revisions — against this much larger pool.

Here is what we wrote about this last year:

“To begin with, you need to understand the size and scope on the Labor market, and what is actually being mode led. There are [more than] 140 million Americans working full time in the country today. Another 15 million or so would like full time jobs, but don’t have one. They may be working part time or not at all.

What the monthly Employment Situation report measures — in near real time — is the net changes in that number. Take the total net number of new hires, subtract the job losses, and you get the marginal change in Employment.

Since our starting number is so big (140m+), and the monthly net changes are so small (200k), the overall change is a statistically small number. Typically, the net change is between one tenth [of a %] (140k) and one quarter [of a %] (350k). During the height of the 2008-09 crisis, the net change was approximately half a percent (700k).”

Now, this isn’t to suggest that the NFP data, however modest the monthly data series may be, is without value. There are lots of juicy morsels of statistical goodness buried within. However, one needs to tease it apart carefully and contextualize the number series relative to other economic information. What it is not is an end all bombshell number it is made out to be by the 24 hour news channels with lots of air time to fill. (And No, we don’t need a countdown ticker to the second as to when the jobs report comes  out).

People seem to get confused about two things: 1) The actual problems with the BLS model; 2) The value of this data series to investors.

The flaws in the series should be apparent: You are dealing with data that is cyclical, very difficult to measure in real time, based on statistically small changes of a very large number. No, the White House does not order BLS wonks to fake the data. Yes, Seasonal Adjustments are pretty standard statistical measures. No, the Birth Death adjustment is not the boogey man its described as. The grand conspiracists talk a good game, but have never proven their cases.

As to the value to investors, these days, NFP’s impact on Fed thinking is probably its most significant element.

I always find it pleasing when a meme gets pushed from the blogosphere into the MSM. The moderate statistical significance 0f NFP is the latest such example.  Here is the WSJ:

“The most-watched economic report of the month also is the most exasperating because it is hard to achieve precision when measuring changes in a very large number. For example, the consensus expectation for Friday’s nonfarm payrolls report is for growth of 163,000 jobs during April. But the two competing government surveys, payroll and household, are about 770,000 jobs apart in their estimates of jobs lost since the downturn began.

Even murkier is the official unemployment rate, which fell to 8.2% last month and is expected to hold steady in April. An alternate measure of unemployment known as U6, which counts discouraged workers, sits at 14.5%. The difference between the two gauges of unemployment is near a record.

All this imprecision should cause investors to take Friday’s report with a giant grain of salt.”

A giant grain of salt indeed.

As wonk statistician George Box stated 3 decades ago, “All models are wrong, but some are useful.” The key to NFP is understanding what aspects of the monthly release are useful. To me, that would be the overall trend, and three internal measures: Hours worked, temporary help, and wages paid.


BLS data released at 8:30am EST


Contextualizing the NFP Data (April 1st, 2011)

(Mis)Understanding the Birth Death Adjustment (July 3rd, 2010)

A Short NFP Q&A (November 4th, 2011)

For Job Gauges, Never the Twain Shall Meet
WSJ May 3, 2012

Category: Data Analysis, Employment

Please use the comments to demonstrate your own ignorance, unfamiliarity with empirical data and lack of respect for scientific knowledge. Be sure to create straw men and argue against things I have neither said nor implied. If you could repeat previously discredited memes or steer the conversation into irrelevant, off topic discussions, it would be appreciated. Lastly, kindly forgo all civility in your discourse . . . you are, after all, anonymous.

14 Responses to “Wise Up to the Proper Flaws of Monthly NFP Data”

  1. The Ryan Budget May Cut Economic Data
    May 03, 2012

    Starting in the early 1990s, the U.S. Census Bureau asked Congress for extra funding each year so it could better analyze the services sector, which was quickly replacing industrial activity as the biggest driver of the U.S. economy. In 2003 the bureau requested more funding to survey financial, real estate, and other companies on a quarterly basis, rather than wait to take their pulse with its Economic Census, which gathers data on business every five years. Census data are funneled to the Bureau of Economic Analysis, which shares its conclusions with the president’s Council of Economic Advisers, the Federal Reserve Board, and Congress. Every year, Census asked for the extra funds; every year, Congress denied them the money, leaving the Census Bureau largely blind to the health of a sector that made up more than half the total economy.

    Finally, in early 2009, after the real estate-fueled financial crisis, Congress gave Census what it had been asking for—an extra $8.1 million. In the view of many, it was too late. “That’s a grand example of how nickel-and-diming statistics agencies can screw up the economy,” says Andrew Reamer, a research professor at the George Washington University Institute of Public Policy and a member of the BEA’s advisory committee. “The government saved $8 million, but how many trillions were lost as a result of not being able to see the crisis coming?”

    That extra data, says Reamer, would’ve revealed just how quickly certain parts of the economy were slowing down. For example, in April 2008 the BEA, with no quarterly data to work with, estimated that finance and insurance sector activity fell 0.3 percent in 2007. In July 2011, the BEA recrunched those numbers using quarterly data and showed declines of 2.2 percent, 5.3 percent, and 9.9 percent for those sectors in the last three quarters of 2007.

    Most U.S. economic data come from three federal agencies: the Census Bureau, the BEA, and the Bureau of Labor Statistics. They have a combined budget of $1.6 billion, less than 0.05 percent of President Barack Obama’s $3.7 trillion proposed budget. These agencies have always had to fight for more funding. Now they may have to fight just to keep their budgets intact. As part of $19 billion in nondefense discretionary cuts in Paul Ryan’s (R-Wis.) budget—recently passed by the House of Representatives—the agencies are likely to get less funding.

  2. Dave J says:

    Barry, I believe you said that B/D overstates job growth when employment is falling. Is it understating job growth now, or is it just too noisy to provide a useful indication?

  3. B/D overstates Job Creation later in the cycle — think 2003-07.

    By the time we hit 2007, nearly all of the new jobs BLS reported were B/D related . . .

  4. mgkurilla says:


    Regardless of the flaws, the real problem is that this data point factors into substantial confirmation bias. It works like this: if you want to argue the market is going higher, then with a good jobs report, you say the economy is humming along nicely justifying your risk on. If you feel the market will decline, then you conclude that with the economy doing better, the Fed will rasie rates and that will hurt. Conversely, with a bad jobs report, either it justifys your market fall position, or if you want risk on, the Fed will lower rates (or QE, or Twist) to support the market. This is where all the misrepresentation comes in because people will explain the same data set with divergent interpretations which highlights and accentuates the flaws.

  5. rktbrkr says:

    It’s the jobs, stupid…

    Meh jobs continues, should be touch and go for the Prez vote although job market dropouts should be counted as losses when figuring the voter effect. Then less than meh.

    The nation’s employers added 115,000 positions on net, after adding 154,000 in March. April’s job growth was less than what economists had been predicting. The unemployment rate ticked down to 8.1 percent in April, from 8.2 percent, partly because workers dropped out of the labor force. .

  6. mathman says:

    Is it true that using their model, if everyone were unemployed there would be no unemployment?

    All this economic nonsense reminds me of that goofy movie Brazil.

  7. AHodge says:

    all dead on
    the array of jobs data is by its nature a complex sorting out, not a headline
    you get paid in the dealing room when you do it in 45 seconds
    and the others take 2 minutes

  8. ElSid says:

    Mathman: The answer is yes, as long as they hadn’t looked for a job in the last 4 weeks. For instance, the move from 8.2% down to 8.1% was likely due to the 522,000 who dropped from the “labor force” this month.

    The good news is, we could be down to 4% unemployment again soon, if most of these unemployed would just stop looking for work!

    Also, the thing that’s not mentioned above is that the BLS numbers are also +/- 100,000, which is significant on a 115,000 number, which was the actual print today. And the birth/death adjustment added 206k and the seasonal adjustment was +22k which of course makes that hypothetical 115,o00 about -111,000. Some reporting system, huh?

  9. Julia Chestnut says:

    I find the number practically worthless for a number of reasons. First, it is based on a model: it doesn’t actually measure much in the real world, because it makes a set of assumptions meant to be maximally useful at a specific point on a bell curve. If I make decisions in the real world, why do I particularly care about a number based on a set of nesting models like matrushka dolls with rosy painted cheeks? Second, to a large degree, this is often reported as a change from the prior month, itself a construct of nesting models that WILL be revised. Even if you are merely looking for a trendline, it’s practically worthless in real time.

    It’s kind of like diagnosing baldness by looking at the number of hairs you lost this morning in your comb: the set of assumptions you use is going to dictate your answer, and you may or may not be right.

    But what I definitely find most disturbing is the amplification of this little piece of data through the “news” cycle and how it is used in the Great Casino. Those seem to be much more important factors than reality these days.

  10. AHodge says:

    so this time the jobs itself not so bad
    there were 53 k of upward revisions
    per Julias point it may be better to look at forecast levels incl revisions. which came in about concensus forecast. and the Urate dropped.
    but the other stuff like wages and hrs were bad.
    julia partly right,
    but if you are trading it doesnt matter
    this is partly a beauty contest or a reason to trade
    it sure is that
    hundreds of billions can ride on this # in a day, better have a view and fast,
    and a view on how the market will trade it

  11. biscuits says:

    To me the real story is in the increasing number of people being dropped out of the labor force, not the marginal improvement in unemployment. This number needs some explaining for me. Why have these people dropped out, have they given up? If there were jobs available, would they be working? Where are the entry level jobs? What does this imply for a growing need for social safety nets like foodstamps, housing and medical care, not to mention a cheap education?
    The only bright spot on this labor report is that the questionable unemployment number gives the Bernank no reason to commence with QE3.

  12. mark says:

    biscuits is on the right track – the employment to population ratio and the participation rate (especially the 25 – 55 yr old participation) is largely immune to all the statistical machinations needed to give the real time snapshot of jobs known as the NFP. Both the E-to-PR and Pop. Ratio have consistently given a picture of a labor market and economy that is stagnant at best following the financial crash.

    Here is my context for today’s NFP – interest rates, inflation and commodities like oil and copper are telling us that Mr. Market’s reaction to today’s snapshot of the labor market is not unwarranted.

  13. Sunny129 says:

    ‘Mr. Market’s reaction to today’s snapshot of the labor market is not unwarranted’

    Mr. Market never accommodates the majority unless Ben’s helicop takes off! investors are addicted to QE ever since 2009. ZRP has distorted along intentional mis-pricing of risk and the credit.

    Economy is very little to do with the market where M to make believe is the accounting standard. Less than expected is bad and more than….. bad!

    All the usual feeders or the monitoring system in ‘free market capitalism’ has been distorted by crony capitalism.

    TRUTH is force of nature and ultimately assert itself. Mr. Market is getting slowly sober from all cool aid he drank for the past 3.5 years!

    No country in history has prospered by spending more debt over debt. As long as the question DEBT is NOT addressed effectively nothing has changed.

  14. Despite the faux rage over two soft releases, U-3 is on a glide path to 7.0% by 2013Q3 and the natural unemployment rate (6.0%) a year later. The first event will induce the FOMC to its first interest rate increase of the this business cycle.