The state’s Employment Security Department (ESD) releases monthly estimates of the number of nonfarm jobs in the state. (The schedule for this year’s releases is here.) These estimates are prepared in conjunction with the federal Bureau of Labor Statistics (BLS) Current Employment Statistics program (CES). By their nature, these survey-based estimates are subject to considerable error, although this error is eventually removed when employment counts become available from the administrative records of the unemployment insurance program.

CES conducts a monthly survey of 141,000 employers, with approximately 486,000 individual work-sites (the payroll or establishment survey), covering the pay period that includes the 12^th day of the month. From this survey BLS estimates the month-over-month change in the number of nonfarm jobs, and then calculates an estimate of the level of employment in the survey month by adding the estimated month-over-month change to the estimate of employment for the previous month. Preliminary results for the nation as a whole are released by BLS three weeks after the reference week, generally on the first Friday of the month. State and local estimates follow from BLS three to four weeks later, although state ESD’s often release state and local estimates before they are available from BLS.

Revisions. Revised sample-based estimates are published one month and two months following the release of the initial preliminary estimate. Once a year, BLS releases a comprehensive revision, benchmarking employment to unemployment insurance (UI) tax records. The most recent such revision occurred this February, benchmarked to March 2011 UI records.  For more on BLS methods see this web page.

For many years individual state ESDs were responsible for producing state and local estimates in partnership with BLS. Beginning in March 2011, responsibility for producing these estimates shifted to BLS. Washington’s ESD continues to produce its own employment estimates, however. These estimates are more accurate than BLS’s because ESD benchmarks to UI tax records four times per year, rather than just once. Current ESD employment estimates are benchmarked to September 2011.

The sample size of the CES is large for the nation as a whole, and so it produces fairly precise estimates of national employment. For a state of Washington’s size, however, the CES sample size is not so large, and employment estimates are subject to significant sampling error. Revisions from the initial preliminary estimate of the month-over-month employment change are often large, as shown by the following three charts, which cover the 48 months from October 2007 to September 2011. (These are the most recent months for which ESD has produced benchmarked job estimates.)

The first chart compares the initial preliminary estimate of the gain or loss in seasonally adjusted Washington State nonfarm employment for the month to ESD’s current estimate of the gain or loss for the month. The blue line represents the initial estimate, while the red columns represent the current estimate. (The nearly 35,000 jobs loss in October 2008 includes roughly 24,000 machinists on strike against Boeing. The end of the strike explains the next month’s employment gain.)

The second chart is a scatter plot. Each point pairs the initial estimate (horizontal axis) and current estimate (vertical axis) of the job gain/loss for a specific month. The correlation between the initial and current estimates is 0.76. The regression line through the points is shown in black; the equation of this line is shown on the chart. In the equation, x stands for the initial estimate; y stands for the current estimate. The slope coefficient 0.79 indicates that on average the current estimate is 21 percent smaller (in absolute value) that the initial estimate.

The third chart shows a histogram of the absolute value of the difference between the current estimate and the initial estimate. In nine cases the initial estimate is within 1,000 of the current estimate. In eight cases the difference between the initial and current estimates is greater than 10,000.

Because of the errors inherent in the initial estimates of state nonfarm employment, one should be a bit skeptical when the preliminary estimate show large gains (as on February 29)  or large losses (as on January 20).

The errors in the initial estimates of employment create problems for those whose job it is to forecast the future path of the state’s economy. The staff of the Economic and Revenue Forecast Council use a little bit of rocket science (the Kalman filter) to reduce the impact of these errors, as detailed here.