Cardiff, UK 16 - 21 October 2005
Session No 7
Mario Ménard, Statistics Canada
Convenors Report for Session 7: Implementing classification changes
1. Titles of paper & country:
- The Changeover from CMAP to NAICS: Implications for Production of Economic Stats -Mexico
- Implementation of a Statistical Business Register with a new Economic Activities Classification – a multi methods approach - Italy
- Implementation of a New Industry Classification on the ABS Business Register - Australia
2. Main themes of session:
Three countries presented their approaches in adapting their Business Register to a new industrial classification.Industrial classification is a key element in the operation of Business Registersas it underlies the gathering, analysing, and disseminating of business information and facilitates the evaluation and comparison of economic activity across industries. To stay relevant, industry classificationsmust be revised from time to time to reflect new and emerging economic activities. Introducing a new industry classification is often one of the most significant changes in economic statistics programs. Therefore, a detailed plan needs to be developed to ensure the success of such implementations.
Many factors must be considered, including the following:
- Parallel series: A new classification might introduce a shift in economic activity from one industry to another. To provide analysts with the necessary information to maintain the comparability of statistical information over time, it could be necessary to maintain parallel series for a pre-determined period.
- Changes in sample sizes: As a consequence units moving from one industry to another, it may be necessary to adjust sample size and stratification for economic surveys.
- Coding tools and mapping of old to new codes:Introducing a new classification requires the construction of correspondence tables that establish the relationships between classes in the old and new classifications. This makes it possible to transit from one classification to the other, placing economic units in the group to which they correspond according to the classification in question. These correspondence tables may present the following situations:
- 1 to 1 and many to 1 correspondence: These two cases are easy to solve; it is sufficient to know the old codes and to have a correct correspondence table to assign the correct new code.
- 1 to many or many to many correspondence: These cases are problematic because additional information beyond knowing the old code is required to assign the correct new code. Several methodologies to be combined in an optimized way. An automated coding tool could be used to assign a new code based on a text description of the economic activity, if one is already available. If not, a survey will be required to collect information on the specific economic activities carried out by the units.To obtain accurate and complete information on economic activity, the questionnaire must be carefully designed. A good practice is to organize the questionnaire with a filter system that guides the enterprises to identify first the general field of activity and then the specific activity. The questionnaire must be designed to guide the respondents to give an accurate description of their activity.
- Coordination of and cooperation with administrative data suppliers: In some countries, the role of other agencies is critical in the implementation of a new industrial classification. Therefore, it is necessary to ensure that they are part of the implementation of industrial classification changes. Collaboration could include planning, funding, training and implementation.
3. Discussion :
Concern was expressed by some delegates regarding the potentially high cost of maintaining parallel series. For example, the National Institute of Statistics, Mexico, plans to maintain their series for five years. It was clear that some countries have not dedicated any additional resources. Rather, they have reorganised workloads and staffing to ensure that the work is completed within agreed timetables. However, others, such as the Australia Bureau of Statistics (ABS), have been able to recruit additional staff. A successful bid by the ABS and their partners, the Australian Tax Office (ATO), enabled them to recruit up to 140 new staff as well as funding IT, accommodation and infrastructure costs. Additionally, a highly trained member of the ABS team has been co-located with ATO.
Some countries plan to run surveys to gather classification data. For some countries the estimated cost of such surveys could prove prohibitive. For example, for the National Institute of Statistics, Italy, the likely cost of running the Fiscal Survey will be in the region of £10M or 15m Euros. The survey will ask questions not only on classification but also income, technology and outputs, etc. Though acknowledged as expensive thedelegatesfelt it was necessary to ensure quality and guarantee success.
In the UK, the Office for National Statistics feels that consideration should be given to re-using existing data rather than surveying businesses to request classification details. Invariably this information will not have changed. Also, should consideration be given to more frequent updating of classification systems?
The National Institute of Statistics, Mexico, among others, are using equivalence tables to map old codes to new. The tables took some 2,560 person hours to create. This led to an increase in survey samples for some industries.
Several delegates reported difficulties in classifying agricultural units – not least the ABS. The Australian register has in excess of 360,000 units carrying the legal code for an agricultural unit. However, they are aware that in practical terms there are no more the 180,000 working farms. Hobby farms and multiple legal units are among the issues they have had to deal with. To overcome this they have created a new type of production unit – a farm unit. This has been done using tax and other administrative data and will be supplemented by an agricultural census in 2006.
The ABS learnt many lessons from their previous reclassification exercise back in 1993, primarily, the need to take a more strategic approach within a formal project management framework.
Close cooperation between ABS and the Tax Office, their primary supplier of administrative data, has been a key ingredient to the success of their work to-date and alsothe introduction of a formal training programme to ensure staff have the right skills.
Each country will be using automated coding tools to help with the changeover to the new classification system. For example, Italyis using Automated Coding by Text Recognition (ACTR) software, a generalized system developed by Statistics Canada. A cost implication of using a tool like ACTR, however, is the need to constantly update the dictionary. It is, essentially, a living document which requires constant attention.
There was discussion regarding the number of units that can besuccessfully coded using such tools as ACTR. For example, while some coding tools will successfully code 35% of units, how do users deal with the remainder? In the case of the ABS, the remainder are put through the same automated system with different parameters and then reported out for clerical intervention. The coding tool is used in a three step process, and uses key words, within indexes, to determine the most appropriate classification. In many cases the business will have to be contacted to ascertain the correct classification.
Other questions on this topic covered the use of decision trees or perhaps encouraging businesses to register on-line thereby reducing the volume of clerical work. However, as acknowledged by several delegates this would require better quality software than what is available presently including features such as drop down boxes based on business descriptions. Also, experience has taught some countries that where on-line registration is used, businesses are not overly concerned with ensuring that the classification is correct.
1 of 3