Retail e-commerce sales are estimated from the same sample used for the Monthly Retail Trade Survey (MRTS) to estimate preliminary and final U.S. retail sales. Advance U.S. retail sales are estimated from a subsample of the MRTS sample that is not of adequate size to measure changes in retail e-commerce sales.
A stratified simple random sampling method is used to select approximately 10,000 retail firms whose sales are then weighted and benchmarked to represent the complete universe of retail firms. The MRTS sample is probability based and represents all employer firms engaged in retail activities as defined by the North American Industry Classification System (NAICS). Coverage includes all retailers whether or not they are engaged in e-commerce. Online travel services, financial brokers and dealers, and ticket sales agencies are not classified as retail and are not included in either the total retail or retail e-commerce sales estimates. Nonemployers are represented in the estimates through benchmarking to prior annual survey estimates that include nonemployer sales based on administrative records. E-commerce sales are included in the total monthly sales estimates.
The MRTS sample is updated on an ongoing basis to account for new retail employer businesses (including those selling via the Internet), business deaths, and other changes to the retail business universe. Firms are asked each month to report e-commerce sales separately. For each month of the quarter, data for nonresponding sampling units are imputed from responding sampling units falling within the same kind of business and sales size category. Responding firms account for approximately 79 percent of the e-commerce sales estimate and about 73 percent of the estimate of U.S. retail sales for any quarter.
For each month of the quarter, Horvitz-Thompson estimates are obtained by summing weighted e-commerce sales (either reported or imputed). Benchmarked monthly estimates are computed by multiplying each Horvitz-Thompson estimate by the carry-forward factor calculated during the most recent benchmarking. For more information on the benchmark process, click here. Estimates for the quarter are obtained by summing the monthly benchmarked estimates.The estimate for the most recent quarter is a preliminary estimate. Therefore, the estimate is subject to revision. Data users who create their own estimates using data from this report should cite the Census Bureau as the source of the input data only.
This report publishes estimates that have been adjusted for seasonal variation and holiday and trading-day differences, but not for price changes. We use quarterly e-commerce sales estimates for 4th quarter 1999 to the current quarter as input to the X-13ARIMA-SEATS program, using the X-11 filter-based adjustment procedure, to derive the adjusted quarterly e-commerce estimates. We derive quarterly adjusted sales estimates by summing the adjusted monthly sales estimates for each respective quarter from the Monthly Retail Trade Survey.
Because the estimates in this report are based on a sample survey, they may contain sampling error and nonsampling error.
Sampling error is the difference between the estimate and the result that would be obtained from a complete enumeration of the population conducted under the same survey conditions. This error occurs because only a subset of the entire population is measured in a sample survey. Standard errors and coefficients of variation, as given in Table 2 of the report, are estimated measures of sampling variation.
The margin of error gives a range about the estimate which is a 90 percent confidence interval. If, for example, the estimated percent change is -11.4% and its estimated standard error is 1.2%, then the margin of error is 1.753 x 1.2% or 2.1%, and the 90 percent confidence interval is -13.5% to -9.3%. Confidence intervals are computed based on the particular sample selected and canvassed. If one repeats the process of drawing all possible samples and forming all corresponding confidence intervals, approximately 90 percent of these individual confidence intervals would contain the estimate computed from a complete enumeration of all units on the sampling frame. If the confidence interval contains 0%, then one does not have sufficient evidence to conclude that at the 90 percent confidence level that the estimated change is different from zero.
Nonsampling error encompasses all other factors that contribute to the total error of a sample survey estimate. This type of error can occur because of nonresponse, insufficient coverage of the universe of retail businesses with e-commerce sales, mistakes in the recording and coding of data, and other errors of collection, response, coverage, or processing. Although not directly measured, precautionary steps are taken to minimize the effects of nonsampling error.