My Thoughts on the firing of the BLS Commissioner
I spent over 6 years working for the Census Bureau, of which three and a half were spent in Statistical Methods Division
Mason,
Your original post states that the relatively large revisions in stats for the past two months (and only the past two months) indicates there may be problems in the procedures used to generate the data and therefore the dismissal of the chief official may be justified.
Each of these revisions comes with detailed documentation explaining the procedures and sources of information for the revisions.
There is an overall summary as well as links to appendices and data tables.
The most recent one is publicly available here:
https://www.bls.gov/news.release/empsit....
Did you look at this before forming an opinion as to whether there are “problems that need to be fixed”?
If not, do you think there is a possibility that reviewing this material would better inform your opinion?
Mason,Your original post states that the relatively large revisions in stats for the past two months (and only the past two months) indicates there may be problems in the procedures used to generate the data and therefore the dismissal of the chief official may be justified. Each of these revisions comes with detailed documentation explaining the procedures and sources of infor
I think there have been significant revisions for many other recent months.
As for reviewing the published tables that you mention, I would need to know what the procedures are that produced those tables This would include things like such as non-response adjustments, independent estimate adjustments, how self-weighting the survey is (which affect variance) and whether there were any outliers in the sample that had to be accounted for, and some other stuff. Unless things have changed, none of this would be in what you read.
Again, I spent over three years in Statistical Methods Division at Census, and this is some of what we did. If there are flaws here that can be improved upon then this is what needs to be addressed.
By the way, if you were to read my Gambling Theory book, you'll see many ideas which were derived from my Census Bureau work.
I think there have been significant revisions for many other recent months.As for reviewing the published tables that you mention, I would need to know what the procedures are that produced those tables This would include things like such as non-response adjustments, independent estimate adjustments, how self-weighting the survey is (which affect variance) and whether there wer
The revision of these numbers is common though so what flaws are we improving? Sometimes data is reported ''late''. You revise your numbers because more data is collected, this is standard and they've been doing this forever.
I'm trying to understand why this is a problem other than that Trump doesn't like the numbers.
Mason,
Do you agree that if the reported job numbers had been strong -- even to the point of facial implausibility -- Trump would not have fired McEntarfer, and indeed would be vigorously defending her from criticism? In other words, do you agree that the best case scenario here is the happy accident of Trump firing an incompetent person for the wrong reason?
The revision of these numbers is common though so what flaws are we improving? Sometimes data is reported ''late''. You revise your numbers because more data is collected, this is standard and they've been doing this forever.
I'm trying to understand why this is a problem other than that Trump doesn't like the numbers.
If data is being reported late and that's the reason why the numbers are changing a lot, then it implies that the non-response adjustment is not doing a good job. This says to me that the data may be released too early.
I 1978 I wrote a paper that was given at the American Statistical Association Convention that summer titled "The Advantages and Disadvantages of Rotating the Annual Housing Survey National Sample from a Nonresponse Point of View." The reason for the paper was to address the rising non-response rate in AHS (at that time). So I have a lot of knowledge and experience in this area, and the paper should be somewhere on the Internet if you want to look for it. (I did find it on the Internet but it's on a site that's not secure. So, I don't want to post that link. But I can see it on my iphone.)
Mason,
Do you agree that if the reported job numbers had been strong -- even to the point of facial implausibility -- Trump would not have fired McEntarfer, and indeed would be vigorously defending her from criticism? In other words, do you agree that the best case scenario here is the happy accident of Trump firing an incompetent person for the wrong reason?
This is probably correct. But I also believe that Trump wouldn't have fired McEntarfer if this was the first time this happened. But before the election, didn't BLS put out numbers that were very favorable to Harris which then got revised downward by a lot once the election was over?
If data is being reported late and that's the reason why the numbers are changing a lot, then it implies that the non-response adjustment is not doing a good job. This says to me that the data may be released too early.I 1978 I wrote a paper that was given at the American Statistical Association Convention that summer titled "The Advantages and Disadvantages of Rotating the Ann
It says to me that they are giving ''estimates'' while they're waiting for more data to come in, revise and give more accurate numbers. I'm still looking for an explanation as to why this method is wrong other than that it shows that what Trump is doing is having the opposite effect. What people should be doing and perhaps the media can do a better job explaining this, is take the first number with a grain of salt.
CES estimates are considered preliminary when first published each month because not all respondents report their payroll data by the initial release of employment, hours, and earnings. BLS continues to collect payroll data and revises estimates twice before the annual benchmark update (see benchmark revisions section below
It says to me that they are giving ''estimates'' while they're waiting for more data to come in, revise and give more accurate numbers. I'm still looking for an explanation as to why this method is wrong other than that it shows that what Trump is doing is having the opposite effect. What people should be doing and perhaps the media can do a better job explaining this, is take
I've already given an explanation. My explanation has to do with the non-response adjustment that is applied to the weighting of the data not doing a good job, and if this is the case, the people putting these estimates together should know this and they should wait for more data to come in before they put the estimates out.
The final number after the Massive post-election revision was still ~8x higher than the current numbers.
Its just as likely or, if we were to make inferences based off Trumps personality, even more likely that Trump directed the BLS commissioner to lie about job reports in his favor and the revisions are the actual numbers and thats why Trump is mad but I'm sure that possibility didn't even cross many Trumper's minds
This is probably correct. But I also believe that Trump wouldn't have fired McEntarfer if this was the first time this happened. But before the election, didn't BLS put out numbers that were very favorable to Harris which then got revised downward by a lot once the election was over?
No. That's just a lie you read on Trump's Truth Social account. According to CEI (a think tank devoted to promoting deregulation):
Last November, in a report just days before the fall election, BLS found that a mere 12,000 jobs were created in October and revised downward the gains for September and August by a combined 112,000. It was a gut punch to the faltering campaign of Democratic presidential nominee Kamala Harris and likely contributed to Trump’s eventual win. That followed an August 2024 BLS report that found 818,000 fewer jobs had been created between March 2023 and March 2024 than previously determined.
I've already given an explanation. My explanation has to do with the non-response adjustment that is applied to the weighting of the data not doing a good job, and if this is the case, the people putting these estimates together should know this and they should wait for more data to come in before they put the estimates out.
They are revisions of -0.08% for May and -0.16% for June, wouldn't these numbers imply they are doing a good job?
Might be my own limitation on this subject but again other than Trump not liking the number I don't think anyone(including you) has done a good job explaining why any of this is wrong or a problem.
Its just as likely or, if we were to make inferences based off Trumps personality, even more likely that Trump directed the BLS commissioner to lie about job reports in his favor and the revisions are the actual numbers and thats why Trump is mad but I'm sure that possibility didn't even cross many Trumper's minds
The one thing Trump has achieved with all this nonsense is that moving forward these numbers can't be trusted anymore.
The CEI article appears to be correct.
A month before the election, the BLS jobs announcement, on Oct. 4, showed growth of 254,000 jobs for September. But “days before the election,” as Trump put it, the BLS announcement on Nov. 1 showed weak growth of 12,000 jobs in October.
The press release noted that strike activity had led to a loss of jobs, and hurricanes that hit the country also may have had an effect on estimates in some industries. BLS also said the August and September figures were revised downward by 112,000. A Fox News headline called the report “dismal” and said it “gives Trump last-minute political ammunition to fire at Harris.” The Trump campaign at the time said the jobs report was “a catastrophe” that showed how Harris “broke our economy.”
A month after the election, the October figure was revised upward.
As for Trump’s reference to an 800,000 or 900,000 reduction, in late August 2024 — well before the election — BLS announced a preliminary estimate that the number of jobs created over the 12 months ending in March 2024 would be adjusted downward by 818,000.
Its just as likely or, if we were to make inferences based off Trumps personality, even more likely that Trump directed the BLS commissioner to lie about job reports in his favor and the revisions are the actual numbers and thats why Trump is mad but I'm sure that possibility didn't even cross many Trumper's minds
You can speculate lots of reasons, and there may be a chance that you're right.
However, I'm looking at this stuff as a statistician who actually has experience in this area, something that I think no one else on this website has. If you don't like what I'm saying because it includes the possibility that Trump may be right, that's your issue, not mine.
But notice I used the word "possibility." I didn't say that this was the case, and obviously I don't know.
By the way, I'm sure there are other statisticians around, some of whom are probably more knowledgeable than I ever was. But this stuff comes out of finite sampling theory, something that very few statisticians study. And here's the book I used:
However, I'm looking at this stuff as a statistician who actually has experience in this area, something that I think no one else on this website has. If you don't like what I'm saying because it includes the possibility that Trump may be right, that's your issue, not mine.
Trump claimed that the Commissioner was deliberately manipulating the numbers. You haven't said anything that supports even a possibility that Trump's actual claim was correct.
The one thing Trump has achieved with all this nonsense is that moving forward these numbers can't be trusted anymore.
Which probably means we're moving to donny's old valuation method of whatever I feel like it is determines the number 😀
Trump claimed that the Commissioner was deliberately manipulating the numbers. You haven't said anything that supports Trump's actual claim.
I would read again what I have said which clearly talks about possible problems with the numbers and/or their revisions. As for what Trump says or what the commissioner may have or may not have done is not where I'm coming from.
I think you need to get over your Trump hatred and reread exactly what I have written.
I know I’m fairly new around here, but this has to be a pretty incredible troll job by Mason. I’ve never met an actual person who argues as erratically as this.
-We aren’t talking about Trump’s patterns of firing experts he disagrees with
-We aren’t talking about the importance of central banking and economic data independence
-We aren’t talking about how a sampling error of 0.03% is entirely normal and well within normal revisions we’ve seen for decades
-We aren’t talking about how BLS reports their data and how these are standard practices that have been in place for decades before the former commissioner was appointed
We are talking about a paper OP wrote half a century ago about survey non-response rates from an entirely different department.
Any attempts to point out the logical in-congruencies of any of the above led to another massive leap in logic further down a seemingly endless wormhole that would rather discuss sampling error in a paper from 1978 than consider that Trump maybe, just maybe, has a poor relationship with the truth.
You can speculate lots of reasons, and there may be a chance that you're right.However, I'm looking at this stuff as a statistician who actually has experience in this area, something that I think no one else on this website has. If you don't like what I'm saying because it includes the possibility that Trump may be right, that's your issue, not mine. But notice I used the word
A little side story. Years ago, gambling writer Stanford Wong, who holds a Ph.D, (I think in business) from Stanford claimed that there was no way I was a real mathematician or statistician because of my use of the term "self-weighting." He challenged anyone to show him a statistics book where this term was used. I called him and this was one of the books I told him to look at.
I
We are talking about a paper OP wrote half a century ago about survey non-response rates from an entirely different department.
This statement is probably not accurate. Unless there have been a lot of changes, it's the Census Bureau, on behalf of other government agencies, that collects the data. If you want to spend some times reading about Census, you'll see that they do close to 100 different surveys for different government agencies.
I think there have been significant revisions for many other recent months.As for reviewing the published tables that you mention, I would need to know what the procedures are that produced those tables This would include things like such as non-response adjustments, independent estimate adjustments, how self-weighting the survey is (which affect variance) and whether there wer
I’m sure in the last 40 years after u worked there , nothing has changed.
It’s just not possible that something involving government could possibly get better or at the very least not get worst in the long run.
Wild how Mason thinks experience from his pre-internet job over 45 years ago is in any way relevant today.
Wild how Mason thinks experience from his pre-internet job over 45 years ago is in any way relevant today.
Has the statistical theory changed? If it has, then you're certainly correct.
Guy caught on camera cheating at golf multiple times just this past week insists he's giving it to you straight 😀
They are revisions of -0.08% for May and -0.16% for June, wouldn't these numbers imply they are doing a good job?
I'm looking at this stuff as a statistician who actually has experience in this area
Would you mind responding to the above as a statistician?
Which probably means we're moving to donny's old valuation method of whatever I feel like it is determines the number 😀
Just like the Supreme leader said in North-Korea. one + one isn't two. It's bigger one. Next month the report will show the US has gained 300m new jobs and everything is great and big and fantastic and huge.
