A couple of people emailed me last week wondering how the BLS birth/death model impacted last Friday’s payroll report.

Recall that the payroll report came in a blistering +288,000 but the household survey showed the economy lost 73,000.

For detailed analysis please see Nonfarm Payrolls +288,000, Unemployment Rate Drops to 6.3%; Household Survey Employment -73,000, Labor Force -806,000.

2014 Birth/Death Adjustment

There it is, highlighted in red. The BLS Birth/Death Model added 234,000 jobs to the non-seasonally adjusted jobs number.

But what does that mean, and is it significant? To address that question, we need to understand what the model does.

Birth/Death Model Description

The births and deaths in question are the births and deaths of businesses, not people. When businesses start, there is a lag before they are incorporated in any job surveys. When businesses go out of business, people who worked for them are unemployed.

Given that not every business reports statistics every month, the BLS uses models to estimate the net number businesses (and people employed in them) that go in and out of business every month.

Mathematical Mistakes

People take the headline number and compare it to the birth-death adjustment. This month, the headline number was +288,000 and the birth-death number was +234,000. Some conclude the adjustment added 234,000 out of the 280,000 headline number.

The problem is 234,000 is a seasonally unadjusted number and 288,000 is a seasonally adjusted number. It is mathematically invalid (and horribly so) to subtract one from the other.

Moreover, in CES Birth/Death Model Frequently Asked Questions, the BLS notes “Months with generally strong seasonal increases such as April, May and June generally have a relatively large positive factor. Conversely, months with overall strong seasonal decreases, such as January, generally have a relatively large negative factor.

The BLS takes all of the unadjusted data, combines it, then applies a seasonal adjustment to the total. There is no seasonally adjusted birth/death model.

Birth Death History

Year January April
2014 -307 +234
2013 -314 +236
2012 -367 +206
2011 -339 +172
2010 -427 +188
2009 -336 +126

January is uniformly negative, April uniformly positive. If one added back in the negative 307,000 results from the January data, jobs in January would have skyrocketed from the reported +113,000 to +420,000.

People seem to do these additions and subtractions only when the resultant number supports their point of view.

Realistic Way of Looking at Birth/Death Model

A more realistic way of thinking about adjustments is questioning if the adjustment is out of line with history or out of line with what one knows is going on.

The reported number this month is in line with what one would expect in an expanding economy. It is also in line with historical numbers. Unless this month was way out of line with what really happened, it did not impact the seasonally adjusted number in a major way, if at all.

Economic Turns

The BLS does admit that the birth/death model will be wrong at turns. And it was wrong in a big way in 2008, 2009, 2010.

Heading into the housing bust and recession, I was one of the biggest BLS critics you could find. It was clear their model was wrong. Housing died. So did commercial real estate. Yet, the BLS model added construction and financial activity businesses each month.

In October of 2009, the BLS finally admitted its birth/death model overstated jobs by about 800,000. That’s quite an error, and I felt vindicated (See Reader Emails on Birth/Death Model and Unemployment Rate).

The BLS has since changed its model, and it also does twice-annual adjustments as opposed to annual. I have no strong beliefs now that their model is incorrect. Actually, recent adjustments have been of a positive nature, not a negative one.

This can change if the economy is at a turning point and that turning point causes businesses to go out of business (or expand at a significantly reduced rate vs. model expectations). Both conditions must be true for the BLS model to be way out of whack.

Unless there is some reason to believe both of the above conditions are true, focus on the birth/death model as a source of error is likely to be wrong. Since I currently have no strong viewpoint on possible model errors, I stopped reporting on the BLS birth/death model in my monthly jobs report.

Reflections on BLS Competence

My opinion on the BLS has changed over the years. When I have called them on the phone, they are eager to help, and my perception is they are highly competent.

Yes, their model was horribly wrong in the housing bust. Is it equally wrong now? I rather doubt it, even if this is an economic turn.

The thing that most changed my mind has nothing to do with their model per se. In some recent conversations, I wanted to prove that a reason for the discrepancy between the employment survey strength relative to the household survey was due to Obamacare artifacts and part-time hiring.

The person I talked to wanted to help. She couldn’t help for the simple reason the BLS did not have the data. A brief conversation revealed the BLS does not have the data it wants or the data it needs to prove one way or another the question I asked (or many other equally important questions).

In defense of the BLS, some of the seemingly convoluted ways they do things (and the birth/net model is a prime example) is for the simple reason the BLS does not have the detailed data that analysts presume they do.

The person I talked to was desperate for more data, but for personal privacy reasons, and also because of burdens on businesses and various delays in reporting, they don’t have it. Even if one presumes the data is somewhere in the system, one should not presume the BLS has access to that data.

We can easily debate whether the BLS should have access to the data it wants. More importantly, we can step back and question whether those BLS jobs should exist at all! Most long-time readers probably know my point of view already.

Such debates aside, my current belief is the BLS does a good job with the data it has.

Mike “Mish” Shedlock