←back to thread

280 points dargscisyhp | 1 comments | | HN request time: 0s | source
Show context
padjo ◴[] No.44765718[source]
It’s pretty clear that the only numbers this administration are interested in are ones that support the narrative that the great leader is infallible.
replies(2): >>44765768 #>>44765910 #
exe34 ◴[] No.44765768[source]
They just fired the commissioner of Labour Statistics. The great thing about autocrats is that they neuter their own country pretty quickly. When you make it risky for people to give you bad news, you end up with missiles that don't work and capital ships that sink.
replies(4): >>44765787 #>>44765836 #>>44766036 #>>44775268 #
roenxi ◴[] No.44765836[source]
The US BLS does seem to have a bit of a history [0] with their job reporting though. The process they've been using appears biased to over report initially and then get revised down over time. I'm sure there are a lot of political considerations, but from a raw statistical perspective there is a pretty easy path to getting better results. They could eliminate the optimistic bias and aim for accuracy.

If it were me I'd be sacking people until they started getting a mean adjustment somewhere around 0. I doubt that is what Trump is doing, but the managers left themselves vulnerable to technical criticism.

[0] https://mishtalk.com/economics/in-honor-of-labor-day-lets-re...

replies(4): >>44766006 #>>44766042 #>>44766083 #>>44766383 #
blackbear_ ◴[] No.44766083[source]
The "history" you cite only goes back three years. Meanwhile, the BLS publishes the monthly corrections since 1979, and the average correction since 2003 is +9k between first and third estimates [1].

Moreover, do note that all published numbers come with standard errors [2] and 90% confidence intervals, which did include the corrections of -133k and -120k that were made for May and June. The current interval for July is -63k to +209k [3]. Anybody who understood high school stats knows the meaning and implications of this.

[1] https://www.bls.gov/web/empsit/cesnaicsrev.htm#Summary

[2] https://www.bls.gov/web/empsit/cesvarae.htm

[3] https://www.bls.gov/ces/

replies(1): >>44766185 #
roenxi ◴[] No.44766185[source]
McEntarfer [0] had only been commissioner for around 18 months. The performance of the BLS in 1979 probably isn't reflective of her skills and talents.

And I'm not going to bother digging through the manuals to figure out how the BLS is calculating their standard errors, but there is a pretty decent chance they've been calculated assuming that the error mean is 0 when in fact it appears to be biased.

[0] https://en.wikipedia.org/wiki/Erika_McEntarfer

replies(1): >>44769017 #
exe34 ◴[] No.44769017[source]
> but there is a pretty decent chance they've been calculated assuming that the error mean is 0 when in fact it appears to be biased.

Could you explain a bit how you arrive at this conclusion?

replies(1): >>44773297 #
1. roenxi ◴[] No.44773297[source]
I can't really call it a conclusion, I just don't know if the residuals are assumed to have a mean of 0 in their model or having a mean of 0 is simply standard practice. Alternatively if you mean the appearance of bias, you can plot the cumulative sum of the revisions (first chart in the mishtalk.com link) vs. a cumulative sum of a normal variable with the same standard deviation. A clear trend emerges because the mean adjustment isn't 0.

In fairness though, Trump's decision is clearly political, these sort of technical factors aren't important enough to rate an official press release and there isn't a cut and dried case that there is anything wrong with the BLS's methods looking in from the sidelines. But for the last few years they have been too optimistic with their estimates and that does strengthen Trump's case.