Altering the Seasonal Adjustment Option For Producing Seasonally Adjusted Data For the September and October 2001 Release
Frequently Asked Questions
Was the option that was altered developed specifically to handle the impact of the terrorist attacks?
No. This option is used as a part of standard seasonal adjustment procedures at the Census Bureau but with a more conservative (restrictive) definition of what constitutes an untypical month. For these months, a definition of untypical was used that is customary in statistical practice when there is advance knowledge that the month’s survey value has a strong potential to be untypical.
What kind of test was used to identify untypical levels?
Each data series was analyzed by comparing the September and October 2001 levels to a range of typical levels as determined by a t-statistic. The option was applied to those data series with September or October levels that were outside of the typical range and for which there was an independent indication of an impact due to the September events.
How was the survey value for a kind of business determined to be out of the ordinary?
Each month, independent indications of how each kind of business performed are researched as part of our normal review of the estimates to detect reporting errors. This research includes, but is not restricted to, conversations with companies in our survey, a review of publicly available press releases from both companies and industry associations, and a review of other economic indicators.
The September or October value of a kind of business was considered to be out of the ordinary if it was found to be statistically untypical (as indicated by the appropriate t-statistics having a magnitude larger than a "standard" critical value) and independent sources indicated an impact from the September 11 events.
What was the impact of altering the seasonal adjustment option for the September and October 2001 data months?
The option ensured that the magnitudes and directions of the seasonal adjustments of September, October, and surrounding months were consistent with the magnitudes and directions of the adjustments for these same calendar months in the recent past.
Did you calculate the August to September and September to October percent changes without the alteration?
No. The alteration enabled the Census Bureau to produce the best estimates of adjusted sales values and percent changes for July, August, September, October, and future months.
Could the Census Bureau have overstated the decline?
The alteration affected the adjustments of only those kinds of business that had untypical results based on well-established statistical criteria. Its use minimized the chance of understating or overstating the decline.
How many kinds of business did the alteration affect?
The alteration affected kinds of business accounting for 60 percent of total retail sales for September and 9 percent for October. Note that the percent value changed for September because the t-statistics for September changed for two reasons. First, the data were revised to include results from the larger monthly retail trade survey sample. Second, new information from the October 2001 levels changes the range of typical levels for September.
Do you have an estimate of the impact on retail sales in New York?
No. Our sample is only designed to produce a national estimate.