Belgian Science Policy Office (Belspo)

Monitoring and evaluation of research and innovation (MERI)

Boulevard Simon Bolivar 30, 1000 Brussels, Belgium

Statistics on Business enterprise R&D (BERD) measure research and experimental development (R&D) performed in the business enterprise sector, i.e. R&D expenditure and R&D personnel. In line with this objective the target population for the national R&D survey of the business enterprise sector consist of all R&D performing enterprises (including all R&D performers – occasional and continuous, known and unknown - in all branches and size classes) belonging to this sector. The “enterprise” is defined in Council Regulation (EEC) No 1993/696 of 15 March 1993. The results are related to the population of all R&D performing enterprises classified in Sections A to U of the common statistical classification of economic activities as established by Regulation (EC) No 1893/2006 of the European Parliament and of the Council (NACE Rev.2).

The main concepts and definitions used for the production of R&D statistics are given by OECD (2015), Frascati Manual 2015: Guidelines for Collecting and Reporting Data on Research and Experimental Development, The Measurement of Scientific, Technological and Innovation Activities, which is the internationally recognised standard methodology for collecting R&D statistics and by Eurostat’s European Business Statistics Methodological Manual on R&D Statistics. (EBS Methodological Manual on R&D Statistics)

Since the beginning of 2021, the collection of R&D statistics is based on Commission Implementing Regulation (EU) No 2020/1197 of 30 July 2020. The Regulation sets the framework for the collection of R&D statistics and specifies the main variables of interest and their breakdowns at predefined level of detail. Statistics on science, technology and innovation were collected until the end of 2020 based on Commission Implementing  Regulation (EU) No 2012/995 concerning the production and development of Community statistics on science and technology.

• The distribution of principal economic activity and by product field are based on Statistical classification of economic activities in the European Community (NACE Rev. 2);
• The local unit for the statistics are compiled at regional level according to NUTS 2 – Nomenclature of Territorial Units for Statistics;
• The distribution by socioeconomic objectives (SEO) are based on Nomenclature for the Analysis and Comparisons of Scientific Programmes and Budgets (NABS);
• The fields of research and development are based on Classification and distribution by Fields of Research and Development (FORD).

See below.

According to the Frascati Manual, expenditure on extramural R&D (i.e. R&D performed outside the statistical unit enterprise) is not included in intramural R&D performance totals (FM, §4.12).

See below.

See below.

FTEs are collected directly in the questionnaire.

The statistical unit for BERD is the enterprise as defined by Council Regulation (EEC) No 1993/696 of 15 March 1993, – if there are deviations please explain.

 No deviation, reporting is done at enterprise level.

Sampling, data collection and data processing were however done at legal unit level.

See below.

The objective of the European R&D statistics is to cover all intramural R&D activities. In line with this objective the target population for the national R&D survey of the Business Enterprise Sector should consist of all R&D performing enterprises (including all R&D performers – occasional and continuous, known and unknown - in all branches and size classes) belonging to this sector. In practice however, countries in their R&D surveys might exclude some enterprises for which R&D activities are deemed to be non-existent or negligible, in order to limit the response burden or due to budgetary constraints.

The target population is the population for which inferences are made. The frame (or frames, as sometimes several frames are used) is a device that permits access to population units. The frame population is the set of population units which can be accessed through the frame and the survey data really refer to this population.

1)       i.e. enterprises previously not known or not supposed to perform R&D

Not requested. R&D statistics cover national and regional data.

Not requested. See point 3.4.

Not requested.

Thousand euros, %, FTEs, physical persons.

2021

See below.

- Frascati Manual 2015, Guidelines for Collecting and Reporting Data on Research and Experimental Development

- EBS Methodological Manual on R&D Statistics

Not requested.

Confidentiality, being one of the process quality components, concerns the privacy of data providers (households, enterprises, administrations and other respondents), the confidentiality of the information they provide and the extent of its use for statistical purposes.

A property of data indicating the extent to which their unauthorised disclosure could be prejudicial or harmful to the interest of the source or other relevant parties.

a)       Confidentiality protection required by law:

The Belgian Interfederal Institute of Statistics (IIS) coordinates the statistics production at the regional and national level in Belgium.  It abides by the European Statistics Code of Practice, including its principle 5, on statistical confidentiality and data protection (https://www.iis-statistics.be/doc/CoC_fr.pdf).

b)       Confidentiality commitments of survey staff:

 All staff members of the regional and national offices are subjected to national statistical law. Data protection officers supervise the correct application of all relevant legal obligations.

Data cells compiled with data of less than 5 units are flagged confidential. In case the value of a cell corresponds for 80% or more with an observed response, these cells are also flagged as confidential.

Flanders:

-        core results documenting the extent to which Flanders approaches the 3% R&D target are published on June 30;

-        a more detailed report is published on September 30.

At the national level: data are not released at a fixed date. We try to publish them in July or August after the official transmission to Eurostat

In Flanders the release calendar for the 3% note and the more detailed report is publicly available (https://www.statistiekvlaanderen.be/nl/oo-intensiteit, https://www.statistiekvlaanderen.be/nl/oo-personeel, https://www.vlaamsindicatorenboek.be/nodes/publicatieagenda/nl).

At the national level: data are published at https://meri.belspo.be/

In Flanders there are strict guidelines for advance access to the core results published on June 30.  Advance access is limited in time (max. 24 hours in advance), and is made publicly known.  During the allotted advance access users are not allowed to publicly announce results.

Once results in Flanders are published on their dedicated websites, they are available to the general public.

Although the national level is not subjected to this policy, we explicitly ask all data producing organisations to respect the official transmission to Eurostat, which is on the same day (June 30). After this transmission all concerned parties are free to disclose aggregated data on their websites.

Results based on the R&D survey are published biennially, in uneven years.

In even years Flanders publishes a 3% note with core R&D numbers (R&D expenditure, R&D personnel) on June 30 based on responses obtained in the innovation survey (CIS).

See below.

1) Y - Yes, N – No

See below.

1) Y – Yes, N - No 

Not available.

Not requested.

See below.

See below.

Not requested.

1) Y – Yes, N - No 

We do not provide methodological documentation but we provide ad-hoc answers to users' requests.

Not requested.

See below.

We abide by the principles enumerated in the Qulaity Assurance Framework of the European STatistical System ( ESS-QAF-V2.0-final.pdf (europa.eu) ).

See our comments in the Section on accuracy. In the Flanders region university students are recruited and trained to follow up non-responding firms by phone, to motivate them to respond. In the data editing stage special attention is paid to the top R&D performers, given the skewed distribution of R&D expenditure and the impact these top performers have on the overall numbers. Even these top performers oftentimes need to be re-contacted to clarify inconsistencies in their responses. A similar methodology is applied in the Brussels Capitol Region and the Walloon Region but to a smaller extent because of a more limited availability of human resources (no students).

The R&D survey is voluntary in Belgium, which implies that respondents may not be too motivated to provide accurate, thoughtful responses, especially when it comes to consistency over time. We have been trying to make the survey mandatory, but, as this requires legal work and approval of employers' associations, this proves to be a difficult task.

Since 2005 Belgium has a tax incentive that reduces the wages for R&D employees when certain criteria are met. The introduction of this tax incentive had an impact on the willingness of some firms to respond, for example to define their work as R&D. A drawback, however, was that some respondents tend now to limit their numbers for R&D personnel to those employees for whom they obtained the fiscal incentive.

As in most surveys, the treatment of non-responses as well as the check of the quality of the information given by the enterprises answering the questionnaire are major issues in our data processing.

In many enterprises, R&D figures are not directly available, and even when available they might not be in conformity with the prescriptions of the Frascati Manual. So, despite guidelines, definitions and precisions given in the R&D questionnaire, and despite an extensive set of editing controls (to check the coherence and quality of the data, e.g. during the online coding of the data), we are well aware quality issues are an ongoing concern.

See below.

1)       Users' class codification

1- Institutions:
• European level: Commission (DGs, Secretariat General), Council, European Parliament, ECB, other European agencies etc.
• in Member States, at the national or regional level: Ministries of Economy or Finance, other ministries (for sectoral comparisons), National Statistical Institutes and other statistical agencies (norms, training, etc.), and
• International organisations: OECD, UN, IMF, ILO, etc.

2- Social actors: Employers’ associations, trade unions, lobbies, among others, at the European, national or regional level.

3- Media: International or regional media – specialized or for the general public – interested both in figures and analyses or comments. The media are the main channels of statistics to the general public.

4- Researchers and students (Researchers and students need statistics, analyses, ad hoc services, access to specific data.)

5- Enterprises or businesses (Either for their own market analysis, their marketing strategy (large enterprises) or because they offer consultancy services)

6- Other (User class defined for national purposes, different from the previous classes. )

To evaluate if users' needs have been satisfied, the best way is to use user satisfaction surveys.

See below.

Not available

Completeness is assessed via comparison of the data delivered against the requirements of Commission Implementing Regulation (EU) No 2020/1197.

Criteria:

A) Obligatory data. Only 'Very Good' = 100%, Poor' >95%; 'Very Poor' <100% apply.

B) Optional data. 'Very Good' = 100%; 'Good' = >75%; 'Satisfactory' 50 to 75%%; 'Poor' 25 to 50%; 'Very Poor' 0 to 25%.

See below.

1) Y-start year, N – data not available

1) Y-start year, N – data not available

1) Y-start year, N – data not available

1) This question is optional. It refers to variables and breakdowns NOT asked by the Commission Implementing Regulation (EU) No 2020/1197 (neither as 'optional'), if R&D data for BES are collected for additional breakdowns or/and at more detailed level than requested.

2) Y-start year

Accuracy in the statistical sense denotes the closeness of computations or estimates to the exact or true values. Statistics are not equal with the true values because of variability (the statistics change from implementation to implementation of the survey due to random effects) and bias (the average of the possible values of the statistics from implementation to implementation is not equal to the true value due to systematic effects).

Several types of statistical errors occur during the survey process. The following typology of errors has been adopted:

1. Sampling errors. These only affect sample surveys. They are due to the fact that only a subset of the population, usually randomly selected, is enumerated.

2. Non-sampling errors. Non-sampling errors affect sample surveys and complete enumerations alike and comprise:

a) Coverage errors,

b) Measurement errors,

c) Non response errors and

d) Processing errors.

Model assumption errors should be treated under the heading of the respective error they are trying to reduce.

1)  Ranking of the type(s) of errors that result in over/under-estimation, from the most important source of error (1) to the least important source of error (5). In the event that errors of a particular type do not exist, is used the sign ‘-‘.

2)  The perceived direction of the ‘overall’ error using the signs “+” for over estimation, “-” for under estimation and “+/-” when assumption of the direction of the error cannot be made for R&D.

1) 'Very Good' = High level of coverage (annual rate of substitution in the target population lower than 5%). High average rates of response (>80%) in census and sample surveys (BES R&D). Full data consistency with reference to totals and relationships between variables in the dataset sent to Eurostat.  

2) 'Good' = In the event that at least one out of the three criteria above described would not be fully met.

3) 'Satisfactory' = In the event that the average rate of response would be lower than 60% even by meeting the two remaining criteria.

4) 'Poor' = In the event that the average rate of response would be lower than 60% and at least one of the two remaining criteria would not be met.

5) 'Very Poor' = If all the three criteria are not met.

That part of the difference between a population value and an estimate thereof, derived from a random sample, which is due to the fact that only a subset of the population is enumerated.

The main indicator used to measure sampling errors is the coefficient of variation (CV).
Definition of coefficient of variation:
CV= (Square root of the estimate of the sampling variance) / (Estimated value)

None - we did not attempt to estimate variances and coefficients of variations (as 97.5% of R&D expenditure and 95%of  R&D personnel in FTE come from the census share of our surveyed set of enterprises)

1)        Industry sector (NACE Rev. 2: 01-03, 05-09, 10-33, 35, 36-39, 41-43)

2)        Services sector (NACE Rev 2.: 45-47,49-53,55-56,58-63,64-66 68,69-75,77-82,84,85,86-88,90-93,94-96,97-98,99)

Non-sampling errors occur in all phases of a survey. They add to the sampling errors (if present) and contribute to decreasing overall accuracy. It is important to assess their relative weight in the total error and devote appropriate resources for their control and assessment.

Coverage errors (or frame errors) are due to divergences between the target population and the frame population. The frame population is the set of target population members that has a chance to be selected into the survey sample. It is a listing of all items in the population from which the sample is drawn that contains contact details as well as sufficient information to perform stratification and sampling.

a)       Description/assessment of coverage errors:

Only 2.2% of the initial gross sample selected turned out to be non-eligible enterprises that had to be dropped from the net, post survey sample. 

b)       Measures taken to reduce their effect:

We use multiple sources of information to update the register of known or assumed R&D performers.  As they are the major contributors to the final R&D numbers, we feel confident in our coverage of the target population.  Moreover, we evaluate our sources for detecting newcomers in each survey wave. Newly detected enterprises generally only cover a limited share of R&D expenditure/personnel, implying our methods for updating our register of known or assumed R&D performers work well.

Magnitude of error (%) = (Observed Value-True Value)/ True Value (%) 

Not requested.

Misclassification rate measures the percentage of enterprises that changed stratum between the time the frame was last updated and the time the survey was carried out. It is defined as the number of enterprises that changed stratum divided by the number of enterprises which belong to the stratum, according to the frame. The rate can be estimated based on the characteristics of the surveyed enterprises.

Measurement errors occur during data collection and generate bias by recording values different than the true ones (e.g. difficulty to distinguish intramural from extramural R&D Expenditure). The survey questionnaire used for data collection may have led to the recording of wrong values, or there may be respondent or interviewer bias.

a)       Description/assessment of measurement errors:

We still are asked by firms whether the work they have been doing could be considered to be R&D or not.  As there are many more firms that do not contact us, we can assume that many of them are still unsure as to what should be considered to be R&D or not.

Unity measure errors (e.g., cases where time spent on R&D is given in hours rather than FTE, or where expenditure is reported in millions rather than in euros) occur occasionally.

Group level responses are given occasionally, when unusually high numbers are given for R&D, given a firm’s basic economic data (turnover, number of persons employed).

Some respondents limit their numbers for R&D personnel to those for whom they obtained the tax incentive for R&D, yielding un underestimate of overall R&D personnel.  Some firms had to return the tax incentives they initially had obtained for R&D personnel, definitely making them unwilling to respond to our R&D survey, even though their activities would still count towards R&D under Frascati Manual guidelines.

Several top R&D performers were unable to give head counts or FTE for consultants embedded in their own internal R&D projects, as they paid a flat fee for the work to be done (under their supervision).  Several of these firms included consultants abroad.  As those consultants they hired worked remotely, they were definitely not able to give more specifics such as gender and occupation of those consultants. Because of the difficulty in obtaining information on embedded consultants (“external R&D personnel” in the terminology of the 2015 Frascati Manual), we included in the statistics we reported for qualification of R&D personnel, only own, in-house R&D personnel, in line with what we have done in the past.

b)      Measures taken to reduce their effect:

We keep track of frequently made comments throughout the field phase, and try to take those comments into account when preparing the survey form for the next wave.  Whenever we revise or add new questions, we perform cognitive interviews to test them.

We provide a diagram to try to help respondents in distinguishing between in-house and external R&D, and to decide on how consultants embedded in own R&D projects fit into the reporting.

A definition of what is R&D and what should be excluded is included in the survey form.

We provide help to respondents by phone, and they do often contact this helpdesk.

The online version of the questionnaire form contains logical consistency checks, that produce error messages/warnings in case of inconsistencies. The error messages are not binding.

We have an extensive data editing routine for detecting and correcting inconsistencies in survey responses.  Especially for major R&D performers we try to re-contact respondents in order to clarify and correct their responses.  We set up meetings with major R&D performers to make sure their responses are in line with Frascati Manual guidelines.

Non-response occurs when a survey failed to collect data on all survey variables from all the population units designated for data collection in a sample or complete enumeration.

There are two elements of non-response:

- Unit non-response, which occurs when no data (or so little as to be unusable) are collected on a designated population unit.

- Item non-response, which occurs when data only on some, but not all survey variables are collected on a designated population unit.

The extent of response (and accordingly of non response) is also measured with response rates.

The main interest is to judge if the response from the target population was satisfying by computing the weighted and un-weighted response rate.
Definition:
Eligible are the sample units which indeed belong to the target population. Frame imperfections always leave the possibility that some sampled units may not belong to the target population. Moreover, when there is no contact with sample units and no other way to establish their eligibility they are characterised as ‘unknown eligibility units’
Definition:
Un-weighted Unit Non- Response Rate = 1 - (Number of units with a response) / (Total number of eligible and unknown eligibility units in the survey) 
Weighted Unit Non- Response Rate = 1 - (Total weighted responding units) / (Total weighted number of eligible / unknown eligibility units in the sample)

1)        Industry (NACE Rev. 2: 01-03, 05-09,10-33,35,36-39,41-43)

2)        Services (NACE Rev 2.: 45-47,49-53,55-56,58-63,64-66 68,69-75,77-82,84,85,86-88,90-93,94-96,97-98,99)

See section 18.3, “incentives used for increasing response” for a description of reminders sent.

Definition:
Un-weighted Item non-Response Rate (%) = 1-(Number of units with a response for the item) / (Total number of eligible , for the item, units in the sample) * 100

Between data collection and the beginning of statistical analysis, data must undergo a certain processing: coding, data entry, data editing, imputation, etc. Errors introduced at these stages are called processing errors. Data editing identifies inconsistencies or errors in the data.

Not requested.

Timeliness and punctuality refer to time and dates, but in a different manner: the timeliness of statistics reflects the length of time between their availability and the event or phenomenon they describe. Punctuality refers to the time lag between the release date of the data and the target date on which they should have been delivered, with reference to dates announced in the official release calendar.

Time lag between the end of reference period and the release date of the results:
Indicator: (Release date of provisional/ first results) - (Date of reference for the data)

a) End of reference period: December 31, 2021

b) Date of first release of national data: Not applicable

c) Lag (days): Not applicable

a) End of reference period: December 31, 2021

b) Date of first release of national data: June 30, 2023

c) Lag (days): 546

Punctuality refers to the time lag between the release date of data and the target date on which they were scheduled for release as announced officially.

Punctuality of time schedule of data release = (Actual date of the data release) - (Scheduled date of the data release).

See below.

Not requested.

Not applicable

The following table lists a number of key survey concepts and conceptual issues; it gives reference to the Commission Implementing Regulation (EU) No 2020/1197 or Frascati manual and EBS Methodological Manual on R&D Statistics paragraphs with recommendations about these concepts / issues.

The following table lists a number of key methodological issues, which may affect the international comparability of national R&D statistics. The table gives the references in the Frascati manual, where related recommendations are made. Countries are asked to report on the existence of any deviations from existing recommendations and comment upon.

See below.

See below.

1)       Breaks years are years for which data are not fully comparable to the previous period.

Collection of data for even reference years (when the R&D survey is off; see also section 18.5.2): in one of the 3 NUTS 1 regions, numbers for intramural R&D expenditure and R&D personnel as FTE are obtained from the innovation survey (CIS) that surveys that reference year.  CIS includes detailed questions asking for these two aspects.  For the other two regions these two numbers are obtained from regressions (containing linear, quadratic and cubic terms) on numbers obtained from previous surveys at the region level.  For researchers as FTE the percentage of researchers obtained in the last survey is applied to the current number of total R&D personnel as FTE.  It should be noted, however, that these two regions that apply regressions, survey two years in their regular R&D survey (t and t-1), so the estimates obtained from regressions provided for the even years are updated using numbers obtained from an actual survey at t + 30.  Hence, at t + 30, numbers obtained from actual surveys are available for all 3 regions.

This part deals with any national coherence assessments which may have been undertaken. It reports results for variables which are the same or relevant to R&D statistics, from other national surveys and / or administrative sources and explains and comments on their degree of agreement with R&D statistics.  Intramural R & D expenditure (code 230101 in the Commission Implementing Regulation (EU) 2020/1197) and R & D personnel (code 230201) are surveyed also in foreign-controlled EU enterprises statistics (inward FATS).

The Community innovation survey 2020 (CIS2020) (inn_cis12) (europa.eu) also collects the R&D expenditure of enterprises that form the coverage of the CIS2020 survey. 

Not requested.

Not available.

To determine foreign-controlled, multinational or domestic status of enterprises we use multiple sources at our disposal: information from a commercial firm-level database (Bel-first), firms’ balance sheets, firm websites, information from the Euro Groups Register (EGR), information from colleagues of the Belgian National Bank who provide FATS. 

Our impression is that FATS, EGR Global Decision Center (GDC) and EGR Global Group Head (GGH) data still contain information on intermediate group members rather than the Ultimate Controlling Institutionnal Unit (UCI).

See below.

This part compares key R&D variables as preliminary and final data.

(1)    Calculate the average remuneration (cost) of individuals belonging to the internal R&D personnel, excluding those who are only formally ‘employees’ (university students, grant holders, etc.).

(2)    Calculate the average remuneration (cost) of individuals belonging to the external R&D personnel (FTEs/other current R&D costs for external R&D personnel).

The assessment of costs associated with a statistical product is a rather complicated task since there must exist a mechanism for appointing portions of shared costs (for instance shared IT resources and dissemination channels) and overheads (office space, utility bills etc). The assessment must become detailed and clear enough so that international comparisons among agencies of different structures are feasible. 

1)       The shares of the figures given in the first column that are accounted for by payments to private firms or other Government agencies.

1)        T = the time required to provide the information, including time spent assembling information prior to completing a form or taking part in interview and the time taken up by any subsequent contacts after receipt of the questionnaire (‘Re-contact time’)

Not requested.

Not requested.

Not requested.

Several separate activities are used for the collection of raw data or pre-compiled administrative data and statistics related to R&D. For simplicity, we call them surveys irrespective of whether they are sample surveys, censuses, collections of administrative data/pre-compiled statistics. This section presents the names of the surveys by sector of performance as well as methodological information for each survey. Depending on the type of survey and sector of performance, only the sections corresponding to that survey and sector are filled in.

We run surveys for RD in the BES sector every two years (see 12.3.3. for more details).

See below.

See section 13.3.2:

The online version of the questionnaire form contains logical consistency checks, that produce error messages/warnings in case of inconsistencies. The error messages are not binding.

We have an extensive data editing routine for detecting and correcting inconsistencies in survey responses.  R&D numbers given by respondents are also compared to numbers for turnover, personnel and to responses given to previous R&D or innovation surveys, to detect outliers or major changes.  Especially for major R&D performers we try to re-contact respondents in order to clarify and correct their responses in cases of inconsistencies or major changes.  We set up meetings with major R&D performers to make sure their responses are in line with Frascati Manual guidelines.

See below.

Imputation is the method of creating plausible (but artificial) substitute values for all those missing.
Definition:
Imputation rate (for the variable x) % = (Number of imputed records for the variable x) / (Total number of possible records for x)*100

1)       Industry (NACE Rev. 2: 01-03, 05-09,10-33,35,36-39,41-43)

2)       Services (NACE Rev 2.: 45-47, 49-53, 55-56, 58-63, 64-66 68, 69-75, 77-82, 84, 85, 86-88, 90-93, 94-96, 97-98, 99)

Not requested.

Not requested.

The documents in Annex shows how se compute the various response rates presented in this document.