Related BLS programs | Related articles
September 2005, Vol. 128, No. 9
Model-based seasonally adjusted estimates and sampling error
Richard Tiller and Marisa Di Natale
The Current Population Survey (CPS) is the source of the Nation’s official estimates of total employment and unemployment. The CPS is a nationally representative, scientifically selected monthly sample survey of approximately 60,000 households. The survey yields data that are rich in demographic detail, including such characteristics as age, sex, race, and Hispanic or Latino ethnicity. Estimates from the survey are published monthly in the BLS news release Employment Situation and in the BLS publication Employment and Earnings.
In order to make the time-series data collected from the CPS more useful to analysts and policy-makers, the monthly data from the survey are adjusted for seasonal fluctuations. As is well known, the purpose of seasonally adjusting a series is to remove seasonal fluctuations in the data so that users can more easily observe fundamental changes in the level or trend of the series that are associated with business cycle contractions and expansions. Approximately 116 time series from the CPS are directly seasonally adjusted, and many more are indirectly seasonally adjusted, as sums or ratios of the original 116.
There is, however, a source of spurious random fluctuations in the CPS data that arises because the CPS samples only a fraction—1 in 2,200, on average, of the working-age population each month: sampling error—the difference between the survey estimates and the values that would be produced by a complete census of the population.
Simultaneously removing both seasonality in the data and noise due to sampling error can prove quite challenging. The monthly estimates produced for the national aggregated series, such as total employment and total unemployment, are highly reliable relative to smaller, disaggregated series. Many of the more detailed demographic series, such as employment and unemployment for blacks, are based on relatively small sample sizes, so that survey error dominates movements in the underlying level of the series. The standard error for a (not seasonally adjusted) month-to-month change based on the CPS can be quite high for some of these series. For example, the standard error for a change in the unemployment rate of adult black males can be as large as 0.8 to 0.9 percentage point, compared with 0.2 percentage point for the unemployment rate for all persons aged 16 years and older. As a result, drawing meaningful conclusions about trends or month-to-month changes is difficult, even after the data have been adjusted for seasonal movements.
This excerpt is from an article published in the September 2005 issue of the Monthly Labor Review. The full text of the article is available in Adobe Acrobat's Portable Document Format (PDF). See How to view a PDF file for more information.
Read abstract
Download full article in PDF
(98K)
Related Monthly Labor Review articles
A state space
model-based method of seasonal adjustment.—July
2001.
Seasonality:
economic data and model estimation.—Dec.
1994.
Within Monthly Labor Review Online:
Welcome | Current Issue | Index | Subscribe | Archives
Exit Monthly Labor Review Online:
BLS Home | Publications & Research Papers