8501.0.55.002 - Information Paper: Changes to the Retail Trade Series, Jul 2004  
ARCHIVED ISSUE Released at 11:30 AM (CANBERRA TIME) 02/08/2004   
   Page tools: Print Print Page Print all pages in this productPrint All


1. This paper describes changes to the Retail Trade series which will be implemented in the July 2004 issue of Retail Trade, Australia (cat. no. 8501.0). The changes will arise from the introduction of a new sample design and an improved estimation method. The changes will cause revisions to the historical level estimates but the effects of the changes on movements should be minimal.

2. A major aim of the ABS is to maintain or improve the accuracy of estimates while reducing the cost to the business community of providing data. A significant reduction in sample size, and hence provider load, for the Retail Business survey can be achieved by implementing a generalised regression estimation methodology with a new sample design utilising the Australian Taxation Office (ATO) Business Activity Statement (BAS) item "Total sales". The reduced sample size can be achieved without any deterioration in the quality of the estimates.


3. The introduction of The New Tax System (TNTS) in July 2000 provided the ABS with the opportunity to improve its economic statistics by introducing new statistical infrastructure and using data available from the taxation system to, among other things, improve sample design and estimation methodologies. These opportunities are described in two information papers released by the ABS, ABS Statistics and The New Tax System (cat. no. 1358.0) released in April 2000 and Improvements in ABS Economic Statistics (Arising from The New Tax System) (cat. no. 1372.0) released in May 2002.

4. This new statistical infrastructure was introduced into the Retail Trade series in the July 2002 issue. It included the use of a new economic statistics units model in which the unit registered for an Australian Business Number is used as the statistical unit for most businesses. For larger and/or more complex businesses, the ABS continues to maintain its own statistical unit. In addition, the ABS introduced a new sizing measure for businesses. For most business surveys, a sizing measure is used along with state and industry to stratify businesses (i.e. group businesses that have similar characteristics) on the survey frame so representative samples can be drawn.

5. An estimate of employment of a business was the size measure used to stratify the Retail Business survey when it was redesigned in 1994. In effect, it is a second-best measure of business size for this survey because a sales based measure would be preferred, but this was not available for all businesses.

6. The introduction of the TNTS in July 2000, with its reporting requirements via the BAS return, provided the ABS with the information to improve the size measure. The size measure introduced in July 2002, is a modelled employment size measure, based primarily on wages and salaries from the BAS or number of payees from the Australian Business Register. This size measure is also used in the ratio estimation methodology currently used to produce estimates for the Retail Trade series.

7. The rest of this paper describes how replacing the modelled employment size measure with a measure derived from BAS Total sales will improve the estimation methodology and sample design used to produce the Retail Trade series and how these changes will be managed.

Generalised regression estimation

8. The generalised regression estimation process has been developed to enable maximum use of observed linear relationships between survey data and auxiliary data (from BAS or other sources). When the auxiliary information is strongly correlated with a survey variable(s), the generalised regression estimation methodology will improve the accuracy of the estimates for a given sample size or enable a reduction in sample size with no reduction in the accuracy of the estimates.

9. The ABS uses BAS Total sales as reported to the ATO to create the auxiliary information for the Retail Business survey. An annualised value is created using BAS data at a point in time. This means that, in some cases, the ABS has to impute data to account for businesses that have not lodged BAS returns for the four quarters being used to create the annualised data. This imputation is primarily required for new businesses and business that report annually and also because of delays between returns being lodged with the ATO and the data being used by the ABS. However, in some cases this may result in the ABS imputing turnover for a business that subsequently has its Australian Business Number and/or Goods and Service Tax role cancelled on the Australian Business Register. Survey processing rules ensure that any imputation error in the stratification and estimation values does not impact on the resulting estimates. To reflect the imputation undertaken by the ABS, the auxiliary item based on BAS Total sales is hereafter described as "annualised BAS Turnover".

    10. Investigations show that Retail turnover is more strongly correlated to annualised BAS Turnover than it is to the current modelled employment size measure. This is demonstrated in the following charts which compare data reported in the Retail Business survey with the current modelled employment size measure (graph 1) and annualised BAS Turnover (graph 2).

    Graph 1: Retail Trade and Modelled Employment Size

    Graph 2: Retail Trade and annualised BAS Turnover

    11. However, for a minority of businesses the correlation between reported data and annualised BAS Turnover is not as strong. There are two main reasons for this. First, all units selected in the Retail Business survey have their industry classification checked and some are identified as being predominantly non-retail. Second, some businesses selected in the survey are discovered to have ceased trading or are seasonal operators. These non-retail and non-operating businesses contribute nil to the Retail Trade estimates even though they have an annualised BAS Turnover value.

    12. To further improve the efficiency of the estimation, businesses selected in the Retail Business survey will be post-stratified prior to estimation into those with reported data and those with nil data. Graph 3 illustrates the improved correlation when adjustments have been made for the issues discussed above. The key feature of the chart below in comparison to graph 2 is the removal of a number of units that have a high annualised BAS Turnover value and a zero Retail Trade estimate.

    Graph 3: Retail Trade and annualised BAS Turnover

    13. For the Retail Business survey, the generalised regression estimation model has been fitted at the state and publication industry level. This is the broadest level at which the model can be fitted to control estimates at the state and industry levels; it is also the most robust.

    14. The ABS is currently exploring the possibility of introducing the generalised regression estimation methodology into other business surveys. Initial work indicates that the Quarterly Business Indicators survey is likely to benefit from this methodology. Users will be advised of this work in relevant publications.

    Sample design

    15. The current Retail Business survey sample design was last determined in 1994, with the allocation using employment as a sizing variable. Originally, this design was determined to meet target relative standard errors (RSEs) at the state, publication industry and Australian levels, using ratio estimation with employment as the estimation benchmark. The design had some minor adjustments in 2002 to account for the statistical infrastructure associated with the introduction of TNTS including the current modelled employment size measure replacing employment as a stratification variable. Currently, the target RSEs for all the publication industries and the smaller states are met but some inefficiencies have crept into the application of the design.

    16. New stratum boundaries based on annualised BAS Turnover have been determined which keep the completely enumerated sector (predominantly large or complex businesses) at a size similar to the current design. The completely enumerated sector contributes about 55% of the total estimate. This ensures a reliable total estimate. Additionally, production of state estimates are heavily dependent on survey responses from the completely enumerated sector. As such, any significant reduction in the coverage of these businesses would be problematic for the quality of the total estimates and the finer dissections. Consequently, most of the reductions in the number of respondents to the collection are in the sample sector.

    17. The new sample design will result in RSEs that are, on average, close to the design RSEs and similar to the RSEs currently achieved for level estimates. RSEs for movement estimates should be improved.

    18. As a consequence of the new design and the new estimation methodology, the sample size of the Retail Business survey will be reduced from about 5,300 businesses to about 4,350 businesses. The new sample of 4,350 businesses will include about 2,200 businesses newly selected in the Retail Business survey. This once-off high proportion of newly selected businesses compares to the 400 to 500 businesses that are normally newly selected in the collection each quarter. Subsequent to the introduction of the new sample design, the number of newly selected businesses each quarter is expected to be less than was the case with the previous design.

    Impact on the time series

    19. The changes to statistical infrastructure introduced in July 2002 had a significant impact on the Retail Trade series and this was managed by measuring the impact and revising the historical series to make them as continuous as possible.

    20. The changes currently being introduced should not have as significant an impact on the Retail Trade series because the statistical units and the population will not be changing as occurred in July 2002. However, the introduction of revised stratification, the use of a different estimation methodology and the impact of a large number of new businesses being included in the sample may result in a once-off change to the level of the series. The level of change in individual Retail Trade series will vary.

    21. For the June 2004 reference month both the old and the new sample designs will be used for the Retail Business survey (a parallel run), to determine any observed difference between the estimates compiled on the new and old basis. Estimates from the old sample design will be used in the June 2004 reference month issue of this publication. In the July 2004 reference month of this publication, the results from the new sample design will be used for both July 2004 and June 2004 estimates (i.e. movement estimates for July will be based on the new sample design). Estimates for historical periods will be revised by smoothing in the difference between the estimates from the old and new samples back to 1982. This will preserve the movement estimates.

    Change to industry detail

    22. Another change that will be occurring in the July reference month as a result of the new sample design is the combination of 'Toy and game retailing' and 'Sport, camping equipment and photographic equipment retailing' into a single industry for stratification purposes. These industries were previously stratified and seasonally adjusted separately and then combined onto 'Other recreational goods retailing' for publication. While this change has no impact on the range of data available in the publication, data for these two industries which were previously released separately as part of the special data service for Australia only (data at the state level were confidential), will no longer be separately available.

    Impact on the national accounts

    23. As historic Retail Trade estimates will change due to the smoothing process described above, there will be an impact on the quarterly chain volume measures, as well as the final household consumption expenditure component of GDP. The chain volume measures on the new basis will be released with the August 2004 issue of this publication, and the national accounts estimates on the new basis in the September quarter 2004 Australian National Accounts: National Income, Expenditure and Product (cat. no. 5206.0). The smoothing exercise should mean that historical movement estimates will be little changed.

    Seasonal adjustment

    24. The seasonal and trading day factors used in the Retail Trade series are reviewed annually at a more detailed level than possible in the monthly processing cycle. As concurrent seasonal adjustment is used, this annual reanalysis does not normally result in significant changes. For Retail Trade, the results of the latest review are usually shown in the July issue each year. Given that the difference between the new and old estimates will be backcast through the historical estimates, the annual seasonal reanalysis will be delayed a month and the results included in the August 2004 issue.

    25. The ABS is also currently investigating the feasibility of using improved methods to calculate seasonal factors. These investigations are focused on the potential application of ARIMA modelling techniques. The expectation is that these techniques will reduce revisions in the seasonally adjusted estimates at the current end of the series by the magnitude of six to seven percent. The magnitude of the revision may be different for individual series. It is expected that ARIMA modelling will be first introduced for the Retail Trade series, along with the annual seasonal reanalysis in the August 2004 issue. More detail on the use of ARIMA modelling for seasonal adjustment purposes can be supplied on request.

    Further information

    26. Further details on the impact of introducing the new sample design and estimation methodology will be included in the July 2004 issue. If you have any queries before then, please contact Michael Gurney on 02 6252 5487 or email m.gurney@abs.gov.au.