|Page tools: Print Page Print All|
DETERMINING OUT OF SCOPE RESPONSES FOR THE RESEARCH AND EXPERIMENTAL DEVELOPMENT BUSINESS SURVEY
The ABS produces statistics about business expenditure on research and development (R&D) via the Research and Experimental Development Business Survey (RDBS). RDBS is a sample survey that is conducted biennially, with approximately 4,000 businesses sampled.
Businesses are grouped into strata according to Industry division and then size. For the purposes of the RDBS, the size of a business is determined by its expected R&D expenditure, and this may come from several sources, for example, previous cycle’s frame or from auxiliary data sources. Units that report zero R&D expenditure are deemed as Out of Scope (OOS) and are subsequently cancelled. Reducing the number of OOS units contacted will result in reduced provider burden as well as realised benefits for the ABS through improved sample design.
A project was undertaken to develop a method to scope out likely OOS responders from the frame whilst keeping undercoverage low. Data from 2015/16 was used to devise the methods whilst 2013/14 data was used for testing. Data from 2015/16 was seen as the most up-to-date hence why we used these data to form the methods. A variety of different methods were investigated as part of this project. A decision was made to pursue the use of decision tree models given the relative ease of implementation.
Based on model covariates, the decision tree algorithm attempts to split the data into groups that are more or less likely to report OOS. This can be used to create a set of scoping out rules for the frame. The model covariates included size, final expenditure stream (which source was used as the expected R&D expenditure on the frame), ANZSIC division, and OOSLastCycle (taking a value of 1 if unit reported OOS last cycle, and 0 otherwise).
We found that units that sourced their expected expenditure from the previous frame were more likely to report OOS than those that didn’t use that particular expenditure source. Furthermore, units that didn’t use the previous frame as their expenditure source were further split with those that reported zero last cycle being more likely to report OOS.
The effectiveness of the scoping rules to reduce the number of OOS units was examined using both the 2015/16 and 2013/14 data. There were two sets of scoping rules tested, one that used the full decision tree and another that only used the first/main branch.
Main points to note:
1) The use of the full tree resulted in relatively more in-scope units removed from the frame as compared to the main branch.
2) The estimated undercoverage from the main branch was rather small and this approximately doubled when the full tree was used.
For the upcoming cycle of RDBS, we have partially applied the rules from the main branch by scoping out units that have expected R&D expenditure sourced from the 2011/12 frame. The effectiveness of the scoping exercise will be evaluated after the cycle. Future work is also planned to assess scoping out units that reported 0 in the previous cycle.
For more information, please contact Tom Davidson Tom.Davidson@abs.gov.au
These documents will be presented in a new window.