I've been wrestling with the BLS birth death model and am wondering if
anyone knows of a thorough critique of the model. Most criticisms I read
seem to have a mistaken notion of what the model does and how it is
constructed. However, the BLS's own publications on the model are
contradictory and may even contain simple arithmetic errors (unless I
misinterpret the purpose of one of the tables).

In "Impact of business births and deaths in the payroll survey", the author,
Kirk Mueller, claims that the imputation of business deaths is sensitive to
business cylces. But in "How the Business Birth/Death Model Improves Payroll
Employment Estimates, the unidentified author states that "the birth/death
model measures a component that does not reflect the business cycle and is
not intended to." How can it be both 'sensitive' and 'non-reflective'? Some
kind of anti-glare coating? ;-)

In the latter paper, Chart 1 compares "actual benchmark revision" with
"birth/death adjustment" and "simulated benchmark revision without
birth/death adjustment". The last element would appear to be the sum of the
first two and a table associated with Chart 1 confirms that for four of the
six years reported, the simulated revision equals the sum of the actual
revision and the birth/death adjustment. But two of the years don't add up.
Is this a simple error or were the four other sums merely coincidences?

It seems to me that the B/D model is a case of good intentions gone awry.
The establishment sample will systematically underestimate employment, so
the model is an attempt to compensate for that. But the model is based on
assumptions and projections from a very limited range of data points
(quarterly data from the 1990s on business openings and closings). The
theory behind it appears to be entirely that the data appear to show a
relatively stable relationship (over a relatively short and homogeneous
period of time). Is that all there is?

Is that all there is?

-- 
Sandwichman
_______________________________________________
pen-l mailing list
[email protected]
https://lists.csuchico.edu/mailman/listinfo/pen-l

Reply via email to