Five years ago I published a paper on the link between inflation and labor force in the U.S. There was one stupid problem with the estimates of labor force level provided by the BLS. They competely ignored the changes in so called population controls after decennial censuses. Briefly, any census reveals the difference between projected and directly enumerated populations. The former figures are projected from the previous census using estimated birth and death rates and net immigration. After any census, this difference called "the error of closure" is proportionally distributed by the U.S. Census Bureau (CB) over the previous decade. This prodecure makes all population time series reported by the CB smooth.
The BLS does not address this problem at all. Therefore, its labor force time series has several "bumps" of a million an more people per one month. One should not use this time series as it is in statistical of econometric assessments. I had to redistribute all known bumps back into their past and obtained relatively smooth time series. This simple procedure did not work well in 1991 and additional investigation was needed to recover the reason of ~1,000,000 step in the labor force series.
Recently, I have found a paper written by Marisa Di Natale "Creating Comparability in CPS Employment Series" from the BLS (no publication date is specified). The author used the same method of the error of closure redistribution. It is a good paper with a simple but correct methodology. But do not believe it. The original time series has not been smoothed. I downloaded the most recent version of labor force time series )July 1, 2011) and found no changes in 1991 and 2001. Same sharp spikes:
Conclusion: Do not trust BLS!