4th Working Group Meeting

Report of Meeting

Monday 25 October 1999: 0930 to 1700

1.Introduction

The Director of the Statistical Office welcomed participants. He commented on current economic statistics: GDP had fallen by 5.8% in the first quarter of 1999, 4% in the second quarter and would as best be unchanged in the third quarter. There had been deflation in the previous five months and an increase of 0.7% in the most recent period. The statistical office was suffering from a cut in its budget, although the government agreed that statistics were important in the EU accession negotiations. The population census was being postponed from 2000 to 2001 and the next agricultural census would now be in 2002. The DOSME project was recognised as important, especially in the government programme for SMEs.

Document 43, the list of participants for the 4th meeting, was revised.

Document 44, the agenda for the 4th meeting, was agreed.

Information on related work

A TES course on statistical units is planned for May 2000. PHARE countries will soon receive an invitation to nominate staff. The details of the course are as they appear in the minutes of the Eurostat business registers committee. To date only Slovakia, FYROM and Bosnia have expressed an interest. The EU will fund all course members. Given the expected strong interest, there may be a need to restrict the number from each country and possible to arrange a second course. The annual TES course on business registers was held in Oslo early in October. There was a possibility of a TES course on use of administrative data in 2000 but the EU would not fund course members.

The Warsaw seminar on structural business statistics in June 1999 had attendance from all of the DOSME countries plus Bosnia and Cyprus. Some of the DOSME participants were also at the Warsaw seminar. The proceedings will be published soon. A further seminar is proposed on short-term statistics and there may be other meetings on definitions, units and administrative sources.

Eurostat Business Registers Committee and UNECE meetings June 1999, Eurostat web site). Additional documents are included on these topics for information. Also included is a single page summary of current Eurostat activity (annex 1). A specific event of interest is the meeting of EU business statistics directors on 27 October. Other meetings include a task force on enterprise group structures. Business demography was also being developed, including consideration of what is a real birth and death. On statistical units, a new regulation was not now being considered but there would be some work on secondary activity and on observation units for structural and short-period statistical surveys. A task force was planned before February 2000 and a seminar would be held in the second half of 2000. The NACE/CPA group is looking at rules for identifying principal activity, especially for the retail trades.

Latvia asked about the structural business statistics regulation requirements for branch statistics. The Czech Republic stressed the importance of statistical units to them.

The ETK seminar in Prague earlier in October covered items of relevance to the DOSME project. Both Ladislav Mezaros and Paul Smith had been at that seminar. The proceedings are expected to be on the Eurostat business methodology web site and therefore (presumably) soon on the CDROM version.

2. Progress on B4

Document 45, the Infostat report on the B4 survey, was supported by a set of slides (annex 2). Infostat had received first progress reports from Hungary and Latvia (and Eurostat had cleared payment of invoices for these). Latvia had also sent a second progress report.

A full summary of the progress is:

It was possible to demonstrate the DOSME web site locally. This was currently the only location of the latest version of DOSMEHELP. Infostat suggested including only the contact persons for each country on the COUNTRY part of the web site. They also suggested that it would be worth investigating permitting uploading of information to the site (under some form of password control). They also made a strong plea for countries to provide additional web links. Infostat will amend the site to include a “WHAT’S NEW” section. Access to the web was a problem for some participants. Production of a CDROM would be considered.

Infostat requested countries to provide an update of the hardware/software profiles.

CECs provided B4 reports:

Albania

Bulgaria

Czech Republic

Estonia

Hungary

Latvia

Lithuania

FYR of Macedonia

Poland

Romania

Slovakia

Slovenia

Infostat would check the operation of the data entry package.

Paul Smith presented the session on “non-ignorable non-response” (NIRO) modelling (document

50 with slides at annex 4), an important issue that was first raised briefly at the 3^rd working group meeting.

This session presented the issues that arise when you have a portion of the population that does not respond at all (or equivalently where the non-respondents have different characteristics from the respondents). In these cases some assumption must be made to enable estimates to be produced and these assumptions are not testable from the survey data. The main way to ensure that the assumptions are good ones to use is to vary them, recalculate the estimates and see whether they change; a method called sensitivity analysis since it tests the sensitivity of the estimates to the assumptions.

After presentation, the countries were invited to describe similar examples from their own experiences. Poland talked about the problem of units where there is no contact at all; they had assumed both 25% dead and 50% dead to assess the impact. Poland had also been affected by flooding in some areas in 1998; a questionnaire had been used to assess the impact (to compensate directly for NINR). In the event, many enterprises did not respond (as they had too many troubles already!), and since the effect was small no adjustment was made.

In Albania the Kosovo troubles had caused some problems for the B3 survey, but were expected to have no effect on B4 results. However, for structural business statistics there are some regional response problems, especially in the south.

Latvia talked about some regional effect on non-response – in the country it is easier to track non-respondents through neighbours, whereas in the city where there is not the same sense of community, neighbours are less helpful.

FYR of Macedonia have a new law of registration, although most business were on the old system; where this causes duplicate registrations, there may be an example of NINR for companies not responding on the old registration number. This is a difficult register issue to avoid duplication.

3.Publications

Eurostat stated that the B3 publication would be on the shelves within a few weeks. All agreed that zipped versions of all publications should be held on the DOSME web site. Only a French version of the A publication was available in electronic form. Eurostat would check for the English version.

CECs gave the following information on their publication plans:

Albania

Bulgaria

Czech Republic

Estonia

Hungary

Latvia

Lithuania

Poland

Romania

Slovakia

Slovenia

A brief introduction was given regarding the plans for B4 publication (document 51 with slides at annex 5). The proposal is to produce the B4 publication along the lines of B1, B2 and B3.However, we wish to consider ways to accelerate the timetable. We propose specifically to ask countries to check for outliers at the data take-on stage, for the set of tables to be ready for the next working group and for CECs to consider the commentary supporting the tables at that meeting.

Tuesday 26 October 1999: 0900 to 1700

The problem of outliers was discussed. The impact is essentially on tables with employment as a basis Options included: give zero weight to all large units; change weight according to views about the unit (eg when a direct privatisation); and post-stratify by size. It was important to pick up outliers early in the processing. We needed to determine what was a large unit. A working definition is a unit with register or survey employment of 250 or more. A definition of micro, small, medium and large enterprises (in employment terms) would be circulated.

A major issue was the lack of employment indicators on the files (completely missing for three and partially missing for four countries). Hungary stated that employment was present. Infostat would check the register data files for all countries. Poland mentioned a confidentiality problem with large units and specifically that they had suppressed two or three large units in a previous survey (this is equivalent to giving it a zero weight). Bulgaria has a size class indicator but only after a delay of about two years. They will be able to supply class size for B5.

Release of data was covered in document 49 (with slides at annex 6). All agreed that the approach of a Eurostat contract was acceptable for research access to data. The contract should have a clause covering restrictions in the use of the data.

4. Classification

Feedback on the quality of coding of industrial classification (document 46 with slides at annex 7) was discussed. Poland stated that they had no quality problems because they had introduced NACE Rev 1 early. They had taken their sample of cases from the Warsaw office, which was stated to have the best quality staff. The classification questions would need careful examination for the B5/C2 surveys.

As for the check of distributive trade data, Hungary thought that the difference was due to the exclusion of sole proprietors on the structural business statistics estimates. Poland stated that their figures related to the number of registered, not just active, units. These comments highlighted the need for better co-ordination with the statistical offices.

5.Making population estimates

In document 47 (with slides at annex 8 and 9), Hans Velt and Poland presented draft final results from the project on longitudinal estimation. This document is essentially a replacement for document 36, which it updates. The problem to be solved is to make estimates of the business populations on 1 January 1995, 1996 and 1997 for the total and separately for legal and natural persons. To do this, Hans Velt had developed a methodology with the Netherlands National Economic Institute (NEI). The work had proved more complicated than expected and final results were not yet available. The work is based on a log-normal duration model using the variables unittype, pd2, pd6 and (the derived) catlry. A major constraint was the need to estimate for 1 January on the basis of an October survey date. As well as making estimates for Poland, validity checking had been completed with the Czech Republic, Slovakia and Slovenia. Preliminary estimates had also been prepared for the Czech Republic.

There were many missing cessation and birth dates, which needed to be estimated before making estimates. This affects the robustness of the estimates.

Issues:

The approach adopted by Hans Velt was to impute start dates instead of using registration dates. Other administrative data might help with imputation. Also all missing and generate dates would be imputed. In Poland, the variable pd6 was dropped from the model as it was not significant. For the Czech Republic, unittype was not significant but pd6 was a reasonable explanatory variable. No explanation could be given for this difference. The Czech estimates were more stable than those for Poland.

For Poland, the resulting model did not fit well with expectations, as it showed a decrease in numbers registered in January 1996 compared with the previous year. Poland therefore generated a variation of the estimate. There was some external evidence to suggest a difference between registration and business start dates. Thus indirect control using registration date to replace start date was not reasonable. In its place, Poland calculated an interval value to be added to the registration date. Enterprises tended to start before the date of registration. Similarly death dates were earlier than administrative de-registration dates.

Jean Ritzen commented that there was a weakness in the method as it used estimates in the process of estimation. Could confidence intervals be produced? An alternative method was possible: using direct estimation of active units. This would involve studying the existing tabulation of active units by year started (table A of the B survey reports). The benefit of this approach was that almost all active units would provide dates for start of activity. Estimates made in this way would help to increase confidence in the Hans Velt (indirect) estimates.

It would also be helpful to validate against more current data, as well as in the other countries. As the DOSME results were now being used extensively (Eurostat receives requests from other directorates of the EU and from OECD), it was also important to ensure that all population estimates are consistent with those already in the project publications. If not, the difference would require explanation. Also more detail was required, especially a NACE breakdown.

What is the way forward?

Other issues relating to longitudinal databases included the changes needed for the introduction of the C2 survey and for the close-down of the second DOSME project. Infostat made some proposals (see slides at annex 10).

The first action was to change the country codes to international standards:

Infostat would hold old and new codes. Eurostat would have new codes and countries could, if they wished, retain the old codes for local files.

The logical design for the database is currently translated into a simple flat file structure. With the addition of C2 in particular, a relational database structure might be more appropriate. Infostat could convert easily to a relational structure under Oracle. Flat output files could be produced for users. To do this, the working group would need to agree a logical data structure.

A new Data Edit program has been written based on Delphi.

Some important C2 issues require urgent resolution: selection criteria and questionnaire types.

A final archive storage management system was required within the DOSME 2 project.

Year 2000 issues were covered in document 42. Infostat updated the working group (see slides at annex 11). All tests had been done to ensure the programs would work following 1999.All countries were urged to ensure that they also were prepared so that there was no delay or added risk to the final stages of the B4 survey processing (data entry and imputation). Date structures on the central database had been converted to be fully compliant. While it was not essential for the success of the project that all countries did the same, it would be desirable.

C2 and B5 questionnaires would need amending to include full year dates (instead of two digit years).

Wednesday 27 October 1999: 0900 to 1200

Document 48 (with slides at annex 12) covers planning for the DOSME 2 survey. A number of other issues had come up during the meeting and were fed into the discussion.

The important message for this session was that funding was not yet assured and would most likely be below the level needed to carry out the full B5/C2 survey programme. Beyond the DOSME part 2 survey there was a further issue of funding. Future surveys (B6+) could be built into national statistical programmes or funding could be sought from EU funds, for example for the development of structural business statistics. Documentation was also an important output from the project. It should also be possible for countries to run all of the processing stages without intervention by Infostat.

It is vital that DOSME 2 is seen as contributing to the statistical programme for each office and in particular relates to the structural business statistics system.

Although 8 Euros was seen by some as not now being enough to fund the surveys, all agreed to live with this amount. There may be an issue of funding national publications separately to keep within this limit. Eurostat would find out the exact rules for funding such publications.

The strength of the DOSME project is in the annual survey programme that it supports. Thus B5 was important and should not be dropped. C2 and B2 together would allow better population estimates for 1 January 2000 than C2 alone. The preferred option was to combine B5 and C2 for sampling purposes and then reduce the sample by the amount necessary to keep within budget. The reduction could be by industrial sector to ensure the quality of estimates for broad industry groups. This would affect the design of the B5 publication, which would report more on levels of enterprises rather than new registrations.

A more detailed question was the need for different questionnaire types for the C2 survey (first contact, first re-contact and second re-contact). Other questionnaire design issues covered the treatment of dead units, measurement of change of activity and additional questions (such as those asked by the Slovak Republic).

One role of DOSME 2 that was not covered was the measurement and management of business register quality. Decisions were needed on the design of the database for DOSME 2. Issues covered the files needed to produce publications, imputation and construction of the final database. Imputation for C2 will be relatively complex. The work could be simplified by not re-imputing for A and C(1) results.

There are some funding issues covering the closedown of DOSME 1. Poland offered to produce a publication for five pre-accession countries. They would produce a proposal for Eurostat. Invoices would need to be sent for this and for other work by mid-May 2000. Other suggestions were payment for translation (Eurostat would check the rules), a project on direct estimation, additional work on longitudinal estimates and a “task force” meeting covering Poland and Czech Republic (plus possibly Slovakia and Slovenia) to develop longitudinal estimates.

8.Next (5^th) meeting

The meeting was agreed provisionally for Riga, Latvia in the week commencing 10 April 2000. It would be held at the national statistical office building. The provisional agenda is to be:

The meeting would last 2.5 or 3 days.

Communication rules are needed. Infostat will draft some for comment. Essentially, those sending emails should request acknowledgement and those receiving should respond promptly (within 48 hours?). There are some uncertainties regarding the email addresses for Hungary. All contacts who change email address should notify the project co-ordinator and Infostat immediately. Romania does not have access to the Internet web site. Poland has problems from time to time with email and Internet access.

John Perry

Project co-ordinator

Document 1 List of participants (1^st meeting)

Document 2 Agenda (1^st meeting)

Document 3 Project description (this is a copy of Document 78 of the previous CEC panel project)

Document 4 Contents and planning of the project

Document 5 Summary of computer systems in CECs

Document 6 Survey B3: sampling and questionnaire

Document 7 Draft contracts for the B3 survey

Document 8 The seminar on the use of business registers

Document 9 Report of 1^st meeting

Document 10 B3 questionnaire

Document 11 Progress report to 1 August 1998

Document 12 Summary of register situation in CECs

Document 13 Update on technological issues

Document 14 Recommendations for further analysis of the data collected in the C survey

Document 15 Characteristics, maintenance and uses of the business register

Document 16 B3 summary tables

Document 17  B4 survey sample and questionnaire

Document 18 Proposals for project 2

Document 19 Enterprises in the CECs: initial longitudinal analysis 1995-7 - draft results

Document 20 Items and questions on the review of the use of business registers in CECs

Document 21 Current longitudinal database and future outlook

Document 22 Place in the national statistical office of the pilot project

Document 23 Agenda for 2^nd meeting

Document 24  List of participants for 2^nd meeting

Document 25 Imputation examples

Document 26 Report of 2^nd meeting (Tallinn, Estonia)

Document 27 List of participants 3^rd meeting

Document 28 Agenda 3^rd meeting

Document 29: B3 report by CECs (Document to be produced following meeting)

Document 30: Infostat report on B3

Document 31: B3 publication

Document 32: C survey – exploitation of data

Document 33: Eurostat questionnaire summary

Document 34: Planning for B4 survey

Document 35: Longitudinal database logical design issues

Document 36: Longitudinal estimates

Document 37: Imputation issues – B3 and B4 surveys

Document 38: Updated proposals for the second DOSME project

Document 39: Quality of coding of industrial classification

Document 40: Report from the 3^rd working group meeting

Document 41: Small and Medium Sized Enterprise Representative Bodies

Document 42: Year 2000 Issues

Document 43: List of participants for the 4^th meeting

Document 44: Agenda for the 4^th meeting

Document 45: Infostat report on B4

Document 46: Feedback on Quality of coding of industrial classification

Document 47: Longitudinal estimates – draft final results

Document 48: Planning for the DOSME 2 survey

Document 49: Release of survey data

Document 50: Non-ignorable non-response modelling

Document 51: Plans for B4 publication

Document 52: Report from the 4^th working group meeting

List of annexes

1. CES report on administrative records
2. Slides – document 45
3. Progress Report by Bulgaria
4. Slides – document 50
5. Slides – document 51
6. Slides – document 49
7. Slides – document 46
8. Slides – document 47
9. Slides – document 47p (Polish version of estimates)
10. Slides – longitudinal database presentation by Infostat
11. Slides – document 42
12. Slides – document 48
13. Slovak analysis of additional B3 questions

Please note that annexes 1, 3 and 13 are available only on paper (circulated to all participants at the meetings). Other annexes are MS Powerpoint slide shows.