It’s Thursday, so … guess what?

Yep, the BLS announced this weeks initial unemployment claims, and you know what?

They revised last week’s headline number up.

Now we’re up to 81 out of 82 weeks — and, at least 22 election season weeks in a row — that the BLS’s “headline number” has under-reported the number of initial unemployment claims … and cast the jobs situation as brighter than it really is.

Based on Thursday’s BLS report, the number for the week ending Sept. 15 was revised upward from 359,000 to 363,000.

In itself, the 4,000 isn’t a big deal.

But, in context it is

Again, I ask: statistical bias or political bias?

I’m now starting to conclude the latter.

The BLS has plenty of statisticians on payroll … and this is an elementary stats problem

* * * * * *

Let’s try a new way of reporting … here’s a picture.


Note that the preliminary estimate (the blue line) is ALWAYS low … by a couple of thousand.

Hint to BLS: just add 2k or 3k … or .8% to your prelim forecast !

>> Latest Posts

Tags: ,

8 Responses to “It’s Thursday, so … guess what?”

  1. TK Says:

    Would still love to see an additional ten years of data. It would be interesting to see if economic contraction/expansion moves these numbers. Everything i can find online talks about the difficulty of compiling good data, the sophistication of the model, and the independence of the BLS from any branch of government.

    What is the word from the Stats department at g’town? They must have some thoughts. I need a little more before i buy into the conspiracy theory.

  2. Scott Allison Says:

    Read up on heteroscedasticity. In layman’s terms it is when the errors in a model are not at random but continually fall in one direction (above or below). It invalidates the model. I bet their model it pretty good, as this is pretty basic econometrics. They are just arbitrarily adjusting for political purposes….or they are just miserable statisticians. Regardless, the whole office should be poop-canned.

  3. TK Says:


    Ken is saying that they SHOULD adjust. You are saying they are corrupt since they ALREADY adjust. I am lost.

    Either way, how about some context? Why are you only showing us 5 months of data? Does the model typically get skewed in a recovery/collapse period (which is my understanding)?

    As an investor, I don’t want people monkeying with the model. I’d prefer the consistent, raw data so that I can understand why it has changed. At some point it may make sense to change the model, but in the short term it is likely to be seen as cherry picking.

    You are making some serious accusations with nothing more than your gut and fancy statistical terms. You “bet their model is pretty good”, but the model appears to be a combination of census data and anecdotal observations. There is a good argument for improving the model, but you need to do a little work and show a little evidence before I buy into the conspiracy theory.

  4. Scott Allison Says:

    My point was that the model is either: A. Miserable (model needs to be adjusted as Ken says) or B. Corrupt (the model is fine but they are adjusting the RESULTS for political gain)

    The results are NOT predictive, so there is no answer C. Since it is a weekly measurement, 82 weeks of data should be plenty to run a reasonable predictive model. Even a simple ARIMA model using only past unemployment data points would be much more predictive than the nonsense that they put out.

    Census data? Annectdotal observations? I sure hope those aren’t the inputs. That sounds like cocktail napkin math…which is fine in casual conversation, but not for a the Bureau of Labor STATISTICS.

    Note: my original “fancy term” was actually the wrong word (it’s serial correlation I was referring to, not heteroscedasticity)

  5. TK Says:

    Data for 2008. Tons of revisions. I have no problem if you think this needs investigating. You may be right that the model is miserable. But it is lazy and unfair to claim this is a corruption issue unless you can show the Obama team has somehow changed the way the numbers are compiled.

  6. TK Says:

    Data for 2000. It is more difficult than I expected to pull these numbers off of Bloomberg, but so far it seems clear that revisions are nothing new. Not much of a case for Pbama putting his finger on the scale.

  7. TK Says:

    Note: I was referring to the entire suite of statistics (unemployment, initial claims, etc) when I mentioned the difficulty of collecting data. Many of these statistics require BLS to estimate the size of the population and break-out natural retirement from quit searching and other (sometimes unknowable) inputs.

  8. TK Says:

    Finally – here a link to the BLS technical note about sample collection and employment revisions. Includes numbers going back to 1979.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: