NonSeasonally Adjusted Data?

In the comments of an earlier post, Roman asks "If you are going to not consider the birth/death adjustment, then its only fair to not count the seasonal adjustment. If you take that out, then the report shows a large gain of jobs. Why does no one here want to talk about that?"

Okay, its a question we might as well address (those of you with statistics or applied mathematics backgrounds please bear with us).

I criticize the B/D model because it has changed what was primarily a measured count (via tax withholding data at established firms) into something more theoretically based. Hence, what was once actual measurement is now primarily a form of modeling. We know that at this late stage of the economic cycle, the modeling creates these bizarre aberrations, such as +45,000 new construction jobs and +8,000 new financial activities jobs in April 2008.

Those data points aren’t merely wrong, they are patently absurd.

But what of seasonal adjustments? What they do is attempt to smooth out or reduce the effect of the regular seasonal patterns that tell us nothing about the economy, and everything about calendar effects: winter weather, school years, planting seasons, holidays, etc.

But since you asked . . .

Let’s do a month-to-month comparison of the non-seasonally adjusted data:

From the CES establishment data, table B-1,we learn that the 2008 numbers were 137,019 in March and in 137,722 in April, for a total nonseasonal adjusted change of +703k. Last year, in 2007, the March was 136,835 versus April 137,668, for nonseasonal adjusted change of +833,000.

So before seasonal adjustments, April 2008 created 130,000 less jobs than April 2007.

Here’s a graphic depiction of the year over year, non-seasonally adjusted NFP:


Chart courtesy of Brian Jacobs

For those of you wonky enough to care, here is the BLS’ explanation as to their Seasonal adjustments:

Over the course of a year, the size of the nation’s labor force and the levels of employment and unemployment undergo sharp fluctuations due to such seasonal events as changes in weather, reduced or expanded production, harvests, major holidays, and the opening and closing of schools.  The effect of such seasonal variation can  be  very large; seasonal fluctuations may account for as much as 95 percent of the month-to-month changes in unemployment.   (emphasis added)

That’s a rather substantial seasonal impact. What do you suppose we should do about it? 

Because these seasonal events follow a more or less regular pattern each year, their influence on statistical trends can be eliminated by adjusting the statistics from month to month. These adjustments make non-seasonal developments, such as declines in economic activity or increases in the participation of women in the labor force, easier to spot.  For example, the large number of youth entering the labor force each June is likely to obscure any other changes that have taken place relative to May, making it difficult to determine if the level of economic activity has risen or declined. However, because the effect of students finishing school in previous years is known, the statistics for the current year can be adjusted to allow for a comparable change. 

Insofar as the seasonal adjustment is made correctly, the adjusted figure provides a more useful tool with which to analyze changes in economic activity . . ."  (emphasis added)

Now you know . . .  Aren’t you glad you asked?




Table B-1  Employees on nonfarm payrolls by industry sector and selected industry detail 

Employees on nonfarm payrolls by industry sector and selected industry detail CURRENT
BLS, Table B-1.

Print Friendly, PDF & Email

What's been said:

Discussions found on the web:
  1. Rich Shinnick commented on May 3


    Terrific response-taking ownerhip of this blog is what sets you apart, the readers actually feel that YOU read their comments.


  2. John commented on May 3

    If I am understanding this correctly, these seasonal adjustments are all a bunch of statistical hand waving that can be avoided by comparing monthly data to the previous year’s monthly data instead of the previous month. For example, compare April 2008 to April 2007 instead of March 2008. Am I wrong?

  3. Alan von Altendorf commented on May 3

    Thanks for the link to BLS establishment data. The way I read it 4 out of 10 U.S. workers are direct government employees or funded/mandated/contracted by state, local, and Federal government, not including all government consumption of oil. Could be that we’ve tripped over the 50% of GDP that was maximum war effort in 1943, an absolute ceiling on ‘public goods’ consumption.

  4. m3 commented on May 4

    courtesy Dennis Gartman, via John Maudlin:

    “And even the BLS says that the birth/death numbers have little statistical meaning. The following is from their own website (courtesy of Dennis Gartman) [emphasis obviously mine]:

    Birth/death factors are a component of the not seasonally adjusted estimate and therefore are not directly comparable to the seasonally adjusted monthly changes. Instead, the birth/death factor should be assessed in the context of its effect on the not seasonally adjusted estimate… The components are not seasonally adjusted separately because they do not have particular economic meaning in and of themselves.”

    so I’m confused by this post. the BLS says in plain english that the seasonal adjustment is useless (4th question on the page):

    i may be wrong, but barry is defending something he has no reason to defend.

  5. KJ Foehr commented on May 4

    Bush Says Americans Face a `Tough Economic Period’ (Update1)
    By Holly Rosenkrantz
    May 3 (Bloomberg) — President George W. Bush said Americans are “facing a tough economic period” after reports this week showed the economy lost jobs for a fourth straight month and economic growth was close to a standstill.
    “My administration has been clear and candid on the state of the economy,” Bush said today in his weekly radio address. “We saw the economic slowdown coming, we were up front about these concerns with the American people, and we’ve been taking decisive action.”

    See! They’ve been clear and candid. Now aren’t you guys ashamed of yourselves, questioning the BLS data like that? And that guy from Barron’s, well he must be part of the liberal MSM that’s trying to talk the country into a recession for political gain!

    You guys probably think the GDP and the CPI numbers are phony too I guess! What a laugh! W said they ain’t hiding nothing, and they already fixed it all anyway.

    You got to get your mind right!

  6. Jay Weinstein commented on May 4

    I think we would be best off with the government just reporting the raw data–no seasonal adjustments, and certainly no B/D adjustments. As we know, even the raw data has issues of collection, timing, and accuracy.

    It is our job to interpret/massage the data, we don’t need help from the bureaucracy.

    And maybe we could save some tax dollars by firing the BLS guys that come up with this garbage…

  7. SI commented on May 4

    Shouldn’t one be adjusting not with addition/ subtraction of jobs, but rather addition / subtraction of percentages, to take into account the trend for growth.

    ie 30k jobs added on 100k base this year vs, 50k jobs added on 200k base next year..

    By add/subtract, this would be an adjusted 20k gain next year.

    By percentage adjustment, this would be a 10k LOSS next year. (30is 30% of 100k, 30% of 200k = 60k, observed = 50k)

    I am exaggerating the numbers for effect, because real growth is just a few percent, not a doubling from 100k to 200k, but you get the point.

  8. SI commented on May 4

    PS: In what I posted above, de-trending could be done with population data, no?

  9. Walt French commented on May 5

    @John, seasonal adjustments go back to the early 20th century and so therefore have been used for your entire adult lifetime. S.A. isn’t some conspiracy to deceive you and it isn’t some untested theory developed by a pointy-headed, self-declared genius at BLS.

    Go with the flow.

    That history also indicates that many efforts pre-dated computers, although the most modern “X-12 ARIMA” requires something a bit savvier than simple Excel. The concepts are important to anybody who looks at data — trend, cycles, seasonality, noise, spike events — and the approach ATTEMPTS to handle them all.

    Couldn’t you just compare to year-ago figures? Yes, if Easter fell in the same week each year; if Thanksgiving-to-Christmas had the same number of weekends; if Katrina-type disruptions happened predictably each year; etc.

    Oh, and IF you were happy to leave ambiguous how a growth spurt of 6 months ago was feeding thru to the current conditions and the “momentum” of business activity.

  10. Kevin commented on May 12

    Mr. Ritholtz,
    Thanks for taking the time to explain this topic in layman’s terms.

Posted Under