Since 2000 every month when it releases its employment situation report the Bureau of Labor Statistics applies an adjustment to the figures, presumably to make the estimate comport more closely with actual figures. The ratio used to make the adjustment is called the “birth/death ratio”. It’s based on the observation that new businesses opening their doors hire people (birth) and old businesses closing their doors fire people (death). Here’s the most recent set of adjustments caused by the birth/death model and here’s a detailed explanation of its methodology.
Increasingly, the birth/death adjustment, far from increasing the credibility of the employment statistics, is diminishing that credibility:
The unemployment numbers are worse than reported. Last year the Labor Department admitted it over-counted the number of jobs by 1.4 million. Why? Because they used a computer program that tries to extrapolate how many new companies are being created during each month and then estimates the number of jobs these firms should be creating. They were wrong.
Since April, the Labor Department has counted 550,000 nonexistent jobs under this so-called birth/death series. Without these phantom jobs, the economy this year created virtually no jobs—certainly not the 600,000 the administration has been touting.
A key problem with the birth/death model is that it is far too unsophisticated to provide a reliable gauge. It assumes consistent behavior over time and, unfortunately, the actual statistics (as reported by another government department, the Census Bureau), tell a different story. Business creation and, implicitly, business destruction varies based on demographic factors in the population. So, for example, it has been observed that whites and Asians tend to have higher rates of business formation and that the age of entrepeneurs tends to be older rather than younger. That in turn suggests that the rate at which businesses are opened or closed may depend on the actual composition of the population which changes over time.
I would further suggest that age cohort may be more important than simple age.
I suspect that the good people at the Bureau of Labor Statistics are hoping that incorporating the actual figures for each new month as they become available will help their model respond to these changing factors without actually utilizing them in their model. Quite to the contrary I think that by weighting the model towards past experience they’re actually obscuring trends that we might want to recognize.
Lest it be suggested that class, availability of capital, or educational attainment is a determining factor in entrepeneurialism the very highest level of business formation is seen among foreign-born Koreans. It’s no accident that the Korean-owned tiny inner city grocery or convenience store has become a cliche in television dramas.
I believe that the current birth/death model, insensitive as it is to the demographic changes in American society, will over-estimate the number of startups, under-estimate the number of failures and, consequently, will tend to overstate increases in employment and, as noted above, that’s exactly what’s happening. It’s time to dump the birth-death adjustment in favor of a more sophisticated tool. All other things are not equal.
Dump the birth-death adjustment? No way! It’s even better than picking the right technocrat. It’s a model, and one can’t argue with a model, can one? After all, that’s the same as agruing against science.
Other pluses of models over technocrats:
(a) They don’t make embarassing public statements.
(b) They don’t make embarassing private statements.
(c) They don’t even provide deep background to reporters.
(d) They never write autobiographies explaining themselves later.
(e) One never has to worry about the moral character of a model.
(f) Models are cheap, and they don’t require medical or pension benefits.
(g) It’s easier to fudge a model than to hire a new technocrat when new opinions/set of numbers are needed.
(h) Models never “shoot back” after being publically scape-goated.
Actually, I’d recommend working on the model and looking at the demographics effects Dave notes and possibly shifting to a Bayesian model that will “learn” given actual historical data.
Yes, that’s much what I have in mind. There’s an old Yiddish proverb: Az di bobe volt gehat beytsim volt zi geven mayn zeyde. (If my grandmother had balls she’d be my grandfather.) If the birth/death adjustment incorporated the measures you’ve suggested it wouldn’t be the birth/death adjustment any more. It would be another, better adjustment.
Since I’ve actually worked at the BLS, on the price index number side, I can say the following:
1. There are smart people at the BLS.
2. They are likely aware of this problem.
3. They are likely looking at these very things.
4. They will likely move slowly.
The last one for a number of reasons such as bureaucracy, it takes time to get this kind of work done (the smart people at the BLS would likely make use of the academic peer reviewed journals like Journal of Econometrics, etc.), they will want to get it right…or as close to right as they can.
For example, it was known for quite sometime the the CPI was biased upwards, but the work on fixing that bias was considerable and took years. Not only did it involve the people at BLS, but also researchers in academia.
When I wrote “the good people at the Bureau of Labor Statistics” I wasn’t being sarcastic. I think they probably are good people, doing their best. However, I don’t believe that translates into infallibility.
I also believe hat the people at the BLS are trying their best to do good work. And I also know they have an astoundingly difficult set of tasks to generate all those numbers. But at some point they have to defer to their political masters. And those guys are more interested in what will help them maintain their positions. There’s no reason to assume that the BLS is completely immune to political pressures.
You might be surprised. Back when the CPI bias was an issue Congress wanted the BLS to tell them how much to reduce various CPI linked things like Medicare and so forth, but the BLS stood firm saying, “Sorry, that is a political issue and outside the scope of our charter.” Now if they asked how much they thought the CPI was biased by, that is a different question than, “Should we reduce this programs annual COLA’s by that amount?”
Hence I don’t think the birth/death model is the result of answering to political masters, but it was at the time a sufficient fix to the problem they were seeing. Now, maybe not so much.
That’s actually my point. A lot of things have changed and the current approach to estimating the effect of company birth/death is among the things obscuring the change.