Measuring the internet economy in The Netherlands a big data analysis 2016 | 14

Retail Survey
The retail survey provides data on a sample of businesses with more than 10 employees
in the retail sector. The size of the sample is 7,456 BUs and the data refers to 2015. If the
business operates an online store then the turnover from that online store is reported
separately, next to the turnover for the whole business. However, turnover from online stores
is not reported for businesses of a certain size classes and for certain SIC codes (retail and
wholesale). This data in the retail survey is used in the categorisation of online stores.
ICT Use Survey
This survey contains annual data on automation and the use of information and
communication technology (ICT) in companies in the Netherlands. The results describe,
among other things, the use of computers, the internet, electronic buying and selling,
software and ICT applications and show the trends in these phenomena for the period since
2003. The survey is carried out on a sample of roughly 11,000 businesses from the population
of businesses with at least 10 employees (which consists of approximately
60,000 businesses). We used data from this survey to cross-validate our results and improve
our methodology.
Other sources
Finally, we make use of several other publicly available sources of data from the internet.
For example, where lists of online stores are available online, we use this to check that our
method to identify websites is capturing all the most important online stores (see section
5.1.1). We also use lists of ICT businesses for a similar purpose (see section 5.1.2). These
sources allow us to check the plausibility of our results and to complement the other data
sources as necessary.
5. Methodology
This section describes the steps to create a database from which all the results are derived.
Figure 5.1 shows a schematic summary of all the research steps.
We begin with the Dataprovider data on Dutch websites and allocate each website to a
category. We explain our method for this in section 5.1. After websites have been allocated
to given categories, we link the websites to the GBR. The method for linking to the GBR is
quite complex and employs diverse methods. For some linking methods more than others,
we can be more confident that the link between the website and the BU is correct. Therefore,
each linking method is described in detail in section 5.2. At this stage, we have a database
in which a BU can have 1) no website, 2) one website or 3) multiple websites allocated to it.
This means that the database is not unique at the BU level. In order to accurately represent
the economy, this database needs to be unique at the BU level. Further, we need to translate
the categorisation of the websites attached to each BU to the classification of the BU. Thus if,
for example, one BU has two websites, one belonging to category B and the other belonging
to category C, then the BU could be categorised as B or C. To deal with this, we develop a
series of decision rules which allow us to create a database which is unique at the BU level
and for which all BUs are allocated to a category. These decision rules are explained in
CBS | Discussion Paper, 2016 | 14 16