Retail Survey Redesigned to Reduce Sample; Time Series Implications Considered
The Retail Trade Survey collects data on a monthly basis from Australian businesses and estimates the total amount of retail sales at the Australia, state and industry levels. As part of the ABS budget savings announced earlier this year, it was decided that the sample size for the monthly Retail Survey be cut.
The initial proposal was to move to a quarterly survey. However, after consultation with key users, it was determined that a monthly indicator series was essential, although more detailed estimates were not required on a monthly basis. The goal of the new design is therefore to produce quarterly estimates at the industry group by state level, and monthly estimates at a broader level. There has also been a change in the scope of the survey: data will no longer be collected for the hotels and licensed clubs industry or the selected services (hairdressing and video hire) industry. Data for these industries are either not required by users or are available from other sources.
The new survey, introduced in July 2008, includes about 3,200 retail and selected service businesses each quarter. Of these, 500 are completely enumerated businesses that are contacted every month. The remaining sample of 2,700 businesses is split into three groups of 900 businesses, with each group being selected in a different month of the quarter. This is a 'one-in-two-out' strategy for collecting data from sampled units. Each business is required to provide a monthly estimate of turnover for the month of the quarter to which they have been allocated. They will then not be required to report data for the next two months.
Estimation for the new design is done in two phases. In the first phase, historical information that is known for the entire quarterly sample is used to estimate the proportion of the population benchmark that represents live in-scope units. In the second phase, the weights of units that have been contacted this month are calibrated to match this population benchmark.
The redesigning of the Retail survey has implications for time series continuity. The Time Series Analysis section (TSA) has been looking into this, and the section's assessment reveals that the 'one-in-two-out' sample rotation has a significant impact on the seasonal pattern of low level time series, such as state by stratification industries, and induces a different seasonal pattern under the new sample design.
Following TSA analysis and internal and external consultations, the ABS has decided to publish a monthly retail indicator at broad industry level, and at a state level where the sample design impact to seasonal pattern is less significant. Because of the changes in the scope, the publication groups for Retail Trade were also realigned to the remaining fifteen stratification industries. Australian, State and Total publication group are all to be directly seasonally adjusted. The increase in volatility that is expected due to the sampled units not overlapping between months has also led to Retail Trade adopting an end weight parameter of 3.5 for the asymmetric filters for all monthly trends.
Seasonal breaks were identified in a couple of series (NSW and Victorian Total), and evidence suggests that inclusion of a seasonal break would lead to improved revision properties for these series. There was also evidence of significant impacts on level, and backcasting of all series was required. Backcasting of the Retail Trade was done at the lowest level (State by Stratification industry) so that outputs could be produced on a quarterly basis. With an increased focus on ABS Trends, the trend revision analysis in the publication was updated.
The TSA is currently working through possible options for producing an independent, but consistent, quarterly adjusted Retail Trade series.
For more information about the sample redesign , please contact Amanda Norton on (02) 6252 5705 or firstname.lastname@example.org, and about the time series analysis, contact Kirk Hampel on (02) 6252 5659 or email@example.com.
This page last updated 12 December 2008