The statistics producers' corner: improved estimates of advance monthly retail sales

Business Economics, Jan, 1997

Effective with the advance retail sales estimates for May 1996 (published in mid-June), the Census Bureau introduced a new sample and modified some of the statistical procedures used to estimate "advance" estimates of monthly retail sales. These improvements included the use of a new procedure to estimate nonemployer businesses, adoption of more efficient sample selection procedures, and real-location of businesses sampled by kind of business. Initial results indicated that the accuracy of advance estimates has improved in months since these changes were implemented, and research is underway to confirm the improvement.

The Census Bureau publishes advance estimates of retail sales within two weeks after each calendar month. The advance estimates are in current dollars, are seasonally adjusted, and feature totals for the United States, subtotals for durable and nondurable goods, and detail for sixteen kinds of retail business. These early estimates are based on "flash" reports from a small but representative sample of U.S. retail business.

Advance estimates are the first of three monthly retail sales estimates. "Preliminary" estimates are published about six weeks after each month, and "final" estimates are published after about ten weeks. Differences in successive estimates for the same month reflect the incorporation of reports from many more retailers and information more often drawn from actual business records. In addition, these monthly estimates, which are reported voluntarily, are subject to annual benchmark revisions based on data from mandatory annual surveys and quinquennial censuses.

One improvement was to change the procedure for estimating advance sales for nonemployer businesses (i.e., those with no paid employees) by adopting the same method used to estimate subsequent preliminary and final sales. This procedure attributes the month-to-month trend for businesses with employees to all retail businesses.(1) This method, which has been used in the full monthly retail survey since 1993, improves the consistency of sales of nonemployer businesses for all three monthly estimates. With the old sample, about 500 retail nonemployers, identified by a special "area" survey, were included in the monthly advance sample and their responses were used to develop estimates for nonemployers.

Another improvement was to increase advanced retail survey sampling efficiency so that the businesses selected for this survey are more likely to be actual participants. Under the new sampling procedures, businesses are not included in the advance retail mail survey if they have not responded to previous information requests. Excluding these businesses helps encourage higher levels of useful responses.

A final improvement was to reallocate the number of businesses selected by kind of business to estimate better sales that vary widely and tend to have the largest revisions. The sample allocation for automotive dealers was doubled to nearly one-third of all sampled businesses, because this industry is volatile and a major component of all retail sales. This increase did not require a larger overall sample because of sample "savings" from the other procedural changes and reduced allocations for other kinds of retail business.

Early indications are that these improvements were effective to enhance or maintain the quality of the advance estimates for each kind of retail business, although more experience is needed for a statistically valid evaluation. As detailed in the table below, the median advance-to-preliminary revision for estimated automotive dealer retail sales was 0.4 percentage point for the first four months of the "new" procedures (May through August 1996), compared with a median revision of 1 percentage point for the past twelve months under the "old" procedures (May 1995 through April 1996).

Estimated Retail Sales

Median absolute percentage point revision to advance estimates

                             New Procedures      Old Procedures
                            (First 4 months)    (Last 12 months)

All U.S. Retailers                 0.2                 0.2
Automotive dealers                 0.4                 1.0
Other Retailers                    0.2                 0.3

An additional benefit of the new sample was to redistribute the small business burden of participating in the advance retail survey. The total survey sample size remained the same at nearly 3,400 businesses. However, the 2,400 small businesses included in the old sample were dropped from further advance retail survey participation, and their counterparts in the new sample are not likely to continue in this survey when the sample is updated next time.

Monthly advance retail sales estimates are available on the day of release at the Census Bureau worldwide web site (http://www.census.gov), and in published press releases. For additional information about these data, monthly retail surveys, and recent statistical changes, contact Ron Piencykoski, at (voice) 301-457-2779, or (fax) 301-457-3685.