4 Empirical setting
4.1 Data description
The dataset used for the empirical model in the current thesis
is the German part of the Community Innovation Survey (CIS), also known as
Mannheim Innovation Panel (MIP). MIP is piloted by the Centre for European
Economic Research (Zentrum für Europäische Wirtschaftsforschung) in
Mannheim, on behalf of the German Federal Ministry of Education and Research.
It consists of a yearly mail survey, including an only response option.
Following the first contact by postal mail, if a firms does not answer it
receives a reminder by phone after six weeks with a second copy of the
questionnaire. After another six weeks, a second reminder follows. The sample
is constructed as a panel with lagged variables to allow the construction of
dynamic models. Considering the rather strict ER of Germany, the use of German
firms' data is ideal to test the PH (Rammer & Rexhauser, 2011).
The data used for the following model was collected in the
2009 MIP survey, particularly because it contains a set of relevant questions
on environmental innovations providing key variables for the model. Compared to
other CIS, the MIP has additional questions concerning firms' profitability and
other market structure information essential to build up a model with enough
control variables to avoid omitted variable bias (Rammer & Rexhauser,
2011).
The first wave in 1993 was only designed for the
manufacturing, mining, energy, water and construction sectors followed by
another wave in 1995 that included the service sector and more recently retail,
wholesale, telecommunication as well as consultancy firms. It is drawn from the
Creditreform database (a German credit-rating agency with the largest data base
on German firms) according to the following stratifying variables: firm size,
region, and industry. Every year the same set of firms are asked to participate
in the survey and to complete the questionnaire sent to them via mail. The
sample is updated every two years to account for exiting firms, newly founded
firms and firms that developed to satisfy the selection criteria of the sample.
Additionally a non-response analysis is performed via phone to check and
correct for non- response bias. The participation in the survey is voluntary
and the average response rate is about 25% (Vuong, 2011). According to Rammer
& Rexhauser (2011) «The survey adheres to the Oslo Manual which
provides guidelines for the definition, the classification and measurement of
innovation. The gross sample of the 2009 wave consists of 29,807 enterprises.
The sample is stratified by sector (56 sectors), size class (8 classes
according to the number of employees) and region (West Germany and East
Germany). The target population are enterprises with 5 or more
employees from most economic sectors excluding farming and forestry, hotels and
restaurants, public administration, health, education, and personal and
cultural services with German headquarters.»
In the 2009 wave the total number of companies that replied
with usable information was 7,657, equivalent of 26 % response rate which just
above the mean of similar for voluntary mail surveys of this scale in Germany
(Grimpe and Kaiser, 2010), especially because the questionnaire is considered
as relative long. The final sample is fairly representative of the gross one in
terms of sectoral composition and firms' size distribution of the whole German
companies' population. Rammer & Rexhauser (2011) provide more inside
information on the process of data collection and, eventually, how they
controlled to limit the «selection bias between responding and
non-responding firms in terms of their innovation status». To do so, they
conducted another non-response survey «surveying 4,829 enterprises by
telephone. This survey revealed a higher share of innovating firms among the
non- responding firms (63.1 %) compared to the net sample of responding firms
(54.3 %).» The sample size of the current model is 3,809 observations.
The main dissimilarity with several other CIS panel data sets
is the pattern of individual response. In fact, according to Peters (2008) MIP
«is not a typical unbalanced panel for which information on individuals is
available for a certain time period without gaps. Instead, one observes a lot
of firms which, for example, respond in a certain year but then refuse to
participate for one or more years, only to join in the survey again at a later
date. This means that the time span for firms under observation is marked with
gaps.» He further explains this phenomenon by the possible problem due to
link between firm closures and firms' innovation behaviour that could induce a
selection bias (Peters, 2008).
A last noteworthy drawback of the MIP is that the
eco-innovation data is collected each odd year and not every year.
| 
