In response to ADP vs. BLS Job Reports – Who to Believe? I had an interesting Email discussion with Ernst Mayer who writes …

I think there is an important additional point that should be made: BLS’ birth-death model is supposed to be able to predict/model business creation/destruction “in real time”, i.e. without the lag that using actual payroll data imposes.

But here is the key point:

If the 2 datasets were accurately tracking long-term employment trends but one had a time lag which the other replaced with an inferential statistical model which was at least reasonably constructed, then any discrepancies should average out over time.

If some number of net businesses (and associated jobs) really was created or destroyed in a given month, that will show up in the actual-payroll numbers at some point in the not-too-distant future. On the other hand, a persistent systematic bias indicates that one (or both) of the 2 methodologies is flat-out getting it wrong.

My Reply …

The BLS models went haywire several years ago and in my opinion never recovered. Perhaps discrepancies before then did average out over time. However, revisions by the BLS (and perhaps ADP as well) make tracking these discrepancies difficult if not impossible as shown below.

Massive Birth/Death Model Revisions – April 2008 Thru March 2009

Here are some excerpts from my February 03, 2010 post 824,000 Will Disappear On February 5; BLS Admits Flawed Model But Plans No Changes.

Bloomberg has some nice interactive charts in an article Birth Death Model Insights.

click on any chart for sharper image

Originally the BLS said 4.8 million jobs were lost between April 2008 and March 2009. This is what it looks like now.

Birth Death Model Falls Out Of Bed

The labor department says there are flaws in its model but defends the process and says “no changes to the current modeling technique are scheduled at this time.” Please note that the birth death model has added 990,000 jobs since April. Those jobs are not reflected in the upcoming 824,000 revision.

Note: My above comment “the birth death model has added 990,000 jobs since April” Reflects the Period April 2009 through January 2010.

ADP vs. BLS Tracking

Please consider the annotated Chart of ADP vs. BLS Tracking

Serious Flaws in Tracking Comparison

The chart above has serious methodology flaws. For starters, the chart needs to reflect a private jobs vs. private jobs comparisons. A direct number to number comparison using the standard BLS report is inaccurate because ADP reports private nonfarm jobs while the BLS reports all nonfarm jobs. The latter is tremendously skewed this year by census hiring and firing. It is also skewed by normal government hiring and firing, and the BLS birth/death revisions.

It is interesting the charts merge closely in the revision period. They shouldn’t, at least if the chart is showing “BLS Overall” vs. “ADP Overall” because the BLS “Overall” numbers include government workers. Perhaps it does but that is not what is implied. I will ask ADP.

Once again, I point out the BLS should provide numbers with and without birth death adjustments but they don’t. Nor is it a simple matter to subtract the numbers. That does not work because the total numbers supplied by the BLS are seasonally adjusted while the birth/death reported revisions are not.

It’s as if the BLS makes it purposely difficult to see how screwed up (or accurate) their model is.

The second serious flaw is the chart implies that over time the BLS and ADP numbers merge. They most assuredly don’t. The only way they did merge is after huge BLS revisions.

I do not object to revisions made a month or two later. I do object to massive revisions made years after the fact that make the BLS look better than it is.

Widening Discrepancy Once Again

Now we see the same sort of “widening discrepancy” that we saw materialize starting in 2006. Part of that is census and/or government hiring but stripping that out, this is what it looks like.

Let’s go back to January and see what the data looks like year to date.

Private Nonfarm Jobs – BLS vs. ADP
Month BLS ADP
June +83,000 +19,000
May +33,000 +63,000
April +241,000 +65,000
March +158,000 +32,000
February +62,000 +3,000
January +16,000 -82,000
Total +593,000 +100,000
Average +98,833 +16,667

That difference is a result of BLS Birth/Death factors or possibly other errors in BLS data accumulation. Either way, I do not believe the BLS numbers. The BLS model is fatally flawed by the BLS assuming normal recovery patterns. Thus, massive over-reporting of “jobs gained or saved” is underway. More revisions are coming.

Mike “Mish” Shedlock
Click Here To Scroll Thru My Recent Post List