Update on the Business Longitudinal Database
The December 2004 issue MD News announced an undertaking by the Australian Bureau of Statistics (ABS) to create a Business Longitudinal Database (BLD) as a response to strong external demand for longitudinal statistics on business characteristics and performance. This database would consist of directly collected survey data from a sample of businesses on the Australian Business Register (ABR) and linked administrative records from taxation and trade sources. The sample would be split into five panels, with units selected in each panel remaining in the sample for five years and one panel being completely refreshed each year. This ensures that for any five year period there is a longitudinal panel for analysis and for periods less than five years there will be multiple panels available. Each panel would be stratified using Australian and New Zealand Standard Industrial Classification (ANZSIC) division by employment size and sample would be allocated to ensure that at the end of five years there would be approximately equal numbers of live responding units in each stratum.
Since that issue much progress has been made. The ABS received funding in the 2005 Budget to pursue the BLD and an associated Food Industry Survey. The design was accepted with a modification to stratify businesses involved in the Food Industry at a finer level and to direct extra sample to these businesses. The scope of the database has been finalised. It includes non-employers and 12 ANZSIC divisions but is currently restricted to businesses with employment less than 200. The content of the direct collection is made up of a set of core business characteristic questions covering employment breakdowns; business location, ownership and age; cooperative arrangements; information technology; innovation; research and development; trade status; markets and competition; financial and government assistance; and performance measurement. There are also detailed Information Technology and Innovation questions asked on alternate years. In addition to the panel-based BLD discussed here, the ABS is developing a large/sparse BLD. This will essentially involve the integration of all available (mainly administrative) data for all businesses in Australia, and will be used to produce thematic outputs.
The first of the panel samples was selected for the 2004/5 financial year reference period. Data has been collected, edited and is now being prepared for the Remote Data Access Laboratory (RADL) environment. This sample contains around 2,300 unique live responding businesses. It is envisaged that future panels will contain around 3000 unique live responding businesses at the time of selection and the combined five panel sample will be around 12,000 live respondents.
The scope of the current Survey on Business Use of Information Technology (BUIT) and Innovation Survey overlap with the BLD and it has been decided to group these collections together under an Integrated Business Characteristics Strategy (IBCS). The BLD direct collect panel samples will provide the data for producing both the longitudinal database and the cross-sectional characteristic estimates currently output from BUIT and Innovation. In the initial years before all panels have been selected there will be a top-up sample to ensure high quality cross-sectional estimates can be produced. There will also be a sample of businesses with employment of 200 or more and businesses in ANZSIC divisions currently included by BUIT/Innovation but not in the scope of the panel samples. Once four panel samples are in operation it is expected a top-up sample will not be required to produce quality cross-sectional characteristic estimates but it may be necessary to include a sample of birth businesses.
The concept of a BLD is not new to the ABS. From 1994 to 1998 the ABS ran a Business Longitudinal Survey (BLS). The BLD will have learned from the BLS experience, particularly relating to the management of birth and death businesses. Restricting the longitudinal scope to businesses with less than 200 employees means many businesses with complex and fluid structures will be avoided. There is some challenging work underway to develop business rules for handling structural changes and cases where the unit on the ABR (Australian Business Number) does not have a one-to-one relationship with the data received from providers. Other methodological challenges that are currently being worked through in parallel with the 05/06 sample allocation and selection include:
For further information please contact Carl Mackin on (08) 9360 5250
- ensuring there is a large common sample between the historical BUIT/Innovation collections and the IBCS sample, without violating regulations over how long businesses can be in sample for;
- determining the requirements for a top up of birth businesses and the extent to which under-representation can be catered for by weighting techniques;
- investigating how best to combine the panel and top-up samples to produce cross-sectional estimates and standard errors
- gathering information on the ways the data will be used so appropriate support mechanisms can be put in place
- determining how to model and treat non-response over time and across panels