by Calculated Risk on 9/02/2010 08:30:00 PM

1) The consensus is for a headline payroll number of minus 90,000 and for the unemployment rate to increase to 9.6% in August from 9.5% in July.

Goldman Sachs is forecasting a minus 125,000 headline payroll number, with no change in private employment and minus 115,000 decline in decennial Census employment. That gives a negative 10,000 ex-Census.

2) My estimate is the decennial Census workforce was reduced by 116,000 in August. This suggests a consensus headline payroll number of +26,000 ex-Census (see point 4). I'll take the under on payroll employment ex-Census.

3) The unemployment rate is dependent on both job creation and the participation rate (both numbers from the household survey - payroll employment is from the establishment survey).

Usually the participation rate - the percent of the civilian population in the labor force - falls when the job market is weak. And a decline in the participation rate puts downward pressure on the unemployment rate (and the opposite is true when the participation rate increases). For technical reasons, there is a possibility that the participation rate increased in August - even with weak job creation - putting upward pressure on the unemployment rate.

4) And here is an easy prediction: there will be some confusion about which payroll number to report!

Here is an excerpt from a employment report preview story from CNBC:

Investors are likely to focus on the private payrolls number, analysts said, given that overall payrolls data is expected to have been influenced by the loss of government census hiring, among other factors.

What "other factors"? The reason everyone has switched to the private payroll number is because of the hiring and layoffs associated with the decennial Census. But this misses any local and state government layoffs (kind of a big story right now).

Reporting that is consistent with non-decennial Census employment reports is to lead with the headline payroll number ex-Census. What has confused some people (I think) is that the Census hiring and layoffs is Not Seasonally Adjusted (NSA), and the headline number is SA. Usually it is not appropriate to mix NSA and SA numbers, but this is a rare exception.

I checked with the BLS, and I even submitted it as a question when the BLS had their first live chat back in March:

9:34 Michele Walker (BLS-CES) -
Submitted via email from Bill: Hi. The headline payroll number is seasonally adjusted, and the hiring for the 2010 Census is NSA. How would you suggest adjusting for the 2010 Census hiring to determine the underlying trend (not counting the snow storms!)?

Thanks for your question Bill.

There is an adjustment made for the 2010 Census. Before seasonally adjusting the estimates, BLS makes a special modification so that the Census workers do not influence the calculation of the seasonal factors. Specifically, BLS subtracts the Census workers from the not-seasonally adjusted estimates before running seasonal adjustment using X-12. After the estimates have been seasonally adjusted, BLS adds the Census workers to the seasonally adjusted totals. Therefore, to determine the underlying trend of the total nonfarm (TNF) employment estimates (minus the Census workers), simply subtract the Census employment from the seasonally adjusted TNF estimate.

Oh well ... this will be the last big change in decennial Census employment.