What does the survey comprise?
In 2000 European government leaders set the joint goal that by 2010 the European Union (EU) would be the most important knowledge economy of the world. The Dutch government set the target that the Netherlands should be a country in the vanguard of Europe in terms of education, research and innovation. To be able to monitor the realisation of these goals, every two years the EU collects extensive information from Member States on private sector innovation. These surveys have been harmonised throughout Europe, and are called the Community Innovation Surveys (CIS). Furthermore, the EU collects annual information on Research & Development activities of companies and institutes and in higher education.
Statistics Netherlands conducts statistical surveys among enterprises based in the Netherlands with 10 or more employed persons. Not all sectors of the economy are examined to the same extent, as R&D and innovation are not equally relevant for different kinds of enterprises. Manufacturing and the services sector form the most important target population.
Statistics Netherlands maintains a register of enterprises used as statistical units. This General Business Register records autonomous enterprise units. The General Business Register of companies is the basis for population delineation and sampling for many business surveys, including those on Research & Development and Innovation.
Date/year survey started
Statistics Netherlands has conducted the innovation survey since 1994-1996 and the R&D surveys since 1970. Up to 2002, the population consisted of enterprises in the Netherlands with 10 or more employees. Enterprises with 1-9 employees were included in the surveys covering 1996-1998 and 1998-2000. Since 2002 the innovation survey covers enterprises with 10 or more employed persons.
Annually the surveys on Research & Development and Innovation measure the state of the Dutch knowledge economy. The surveys describe investments in R&D and Innovation and other innovative activities of Dutch enterprises and institutes. They also include knowledge flows and the results of the innovation process. In odd years Research & Development (R&D) is measured extensively, in even years the focus is more on other aspects of innovation (CIS).
The results for year t are calculated and published at least twice: in November of year t+1 a provisional R&D figure is published, in accordance with the European regulation. In July t+2 at the latest, definite R&D figures are published, including innovation figures for even years.
How is the survey conducted? '
Every year a stratified sample from the business units in the target population are surveyed for statistics on R&D and Innovation.
The survey is conducted both via the internet and using paper questionnaires.
Enterprises based in the Netherlands with 10 or more employed persons.
For even reference years (CIS) the total sample contains about 15,000 business units. For odd reference years (R&D survey) data are collected from approximately 1,500 units.
Checking and correction methods
First of all, the relationships between various questions are considered. If clear errors are found – e.g. a non-innovative company has innovation expenditure – a flag is placed. After studying the case, a decision is made about what actions must be undertaken to correct the information manually. In the next stage, a number of automatic corrections are made. The corrected fields are also given a flag value. Imputed missing values are labelled with a different flag value from the corrected values. The checks and corrections are prescribed by Eurostat, but Statistics Netherlands has added its own to these. If there is serious doubt about or ambiguity in the questionnaires, the scanned form is requested and if necessary checked with the respondent. The aim is to correct errors at Statistics Netherlands as much as possible. In addition to the form for the current survey year, the form for year (t – 1) is also available. Errors can often be corrected by comparing these. The remaining errors can be corrected automatically. If a critical company is missing, results from previous years are examined; if desired, the result of the previous year is duplicated. Eurostat prescribes that item non-response (partial non-response) be estimated. This means that missing values are estimated only for partially completed questionnaires. This applies to metric variables (expenditure and R&D staff), ordinal variables (strong/moderate/weak questions, etc.) and nominal variables (yes/no). An SPSS syntax has been developed, which works as follows. The file is divided into cells. These are combinations of company size and industry sector. For each company for which figures have to be estimated, a “donor” company is sought which has as many similar characteristics as possible as the company for which the estimate is made (the donor is therefore naturally found within the same cell). When a suitable donor is found, the figures are copied and used for the company to be estimated. A company can only be used once as a donor. The remaining non-respondents (and the population units not observed) can be estimated by determining a multiplication factor (stratum population/response – N/n) for the business units for each stratum (SIC 2-digit/company size combination).
Population estimation is done by determining a multiplication factor (stratum population / response – N/n) for the business units in each stratum (SIC 2-digit /company size combination).
Quality of the results
A large, representative and stratified sample is used for this survey. The population framework will always contain errors, such as over-coverage, under-coverage and classification errors. Over-coverage is a result of units wrongly included in the framework, e.g. those that no longer exist or have been taken over. Under-coverage is a result of companies not included in the framework as they are not yet registered. Classification errors arise when a company is included in an incorrect business sector or company size class. Between the moment the sample is drawn and the moment the results are published, a number of errors will be detected. These errors are processed in a co-ordinated way and results are made available to all other statistical processes.
In odd reference years the Research & Development (R&D) is measured very extensively, in even reference years other aspects of innovation (CIS) are also measured. Since 1996 the main variables in the two surveys are comparable.
First, figures are made consistent at the individual level. Then the aggregated data are compared with the figures for the previous year. Furthermore, the very elaborate set of questions and calculations guarantee the quality of the figures.