2. Champions of Digital Transformation?

Innovation broadly refers to the, at least attempted, implementation of an idea (Harhoff 2008; Christensen et al. 2015). Baregheh et al. (2009) have identified over 60 different definitions and typologies of innovations. However, in this paper it encompasses major improvements or novel approaches in the context of a firm’s processes, products, or its business model (OECD 2018a, 2018b). Digital innovations specifically refer to “the carrying out of new combinations of digital and physical components [in a layered modular architecture] to produce novel products” (Yoo et al. 2010, p. 725).

According to Simon (2012), who coined the term hidden champion in the early 1990s, they are defined as firms that, in general, generate below 5 billion Euros in revenues, belong to the top three in their respective market globally or are market leading on their continent, and are relatively unknown to the broader public. Estimates of the number of HCs range from approximately 1,300 (Simon 2018) up to around 1,700 (Rammer & Spielkamp 2019).

Mittelstand broadly refers to mid-sized manufacturers, sometimes referred to as the ‘backbone’ of Germany’s economy. They tend to be family-owned and usually sell specialized machineries and components in B2B markets (The Economist 2014).

Combinatorial effect refers to the complementary character of many digital technologies. The transformative effect of combinations of digital technologies is often of far greater impact than that of individual technologies alone (e.g., WEF 2020).

Network effects as described in economics and management are the effects in which the number of users of a product or network positively affects its value, or in the case of indirect network effects, the value of a complementary product or network. A frequently used example in this context is the invention and expansion of the telephone (Katz & Shapiro 1994).

Some argue that there might be a direct link (Teece et al. 1997; Griffith & Harvey 2001; Lee et al. 2002) while others emphasize an indirect relationship and argue that they enable resource and competence reconfigurations that create competitive advantages (Eisenhardt & Martin 2000; Zott 2003).

Firms in markets that are characterized by low dynamism and rates of change, however, should focus on efficiency enhancing ordinary capabilities (Drnevich & Kriauciunas 2011).

Dyer and Shafer (1998, p. 6) define organizational agility as “the capacity to be infinitely adaptable without having to change. It is viewed as a necessary core competence for organizations operating in dynamic external environments”.

The cost of a patent is defined as the total sum of costs beginning with the idea stage until a successful patent grant. According to Simon (2012), HCs have costs of approx. 0.5 million Euros per patent, whereas large corporations spent 2.7 billion Euros per successful patent filing. The average number of patents per employee is 31 for HCs and 6 for large corporations.

The MIP is a yearly survey conducted by the Center for European Economic Research (ZEW), the Fraunhofer Institute for Systems and Innovation Research (ISI), and the Institute for Applied Social Sciences (infas). It is commissioned by the German Federal Ministry of Education and Research (BMBF) and gathers information on the innovation activities of German firms. The survey allows for extrapolation of the survey results to the total firm population. Every two years, the MIP is part of the Community Innovation Survey (CIS) of the Statistical Office of the European Commission (EUROSTAT) which serves as the basis for the European Innovation Statistic (ZEW 2019). See, e.g., Rammer and Peters (2013) or visit ‘https://​www.​zew.​de/​en/​research-at-zew/​mannheim-innovation-panel-innovation-activities-of-german-enterprises’ for general information about the MIP. See Appendix B.1.1 for the complete survey of 2016.

See Chapter 1 of this dissertation for a detailed overview of the identification steps and descriptive statistics for HCs based on the MIP.

See Table B.3.1 in the appendix for the detailed industry distribution. As most HCs operate in manufacturing, I apply a more fine-grained industry classification, which allows to distinguish between several manufacturing-related sectors, for the analysis. The table also presents the corresponding NACE2 codes.

Hainmueller and Xu (2013) provide additional information about the approach and its advantages over earlier control group methods, such as nearest neighbor or propensity score matching techniques that often yield rather low levels of covariate balance.

See Table B.3.2 in the appendix for the covariate balance table of all applied balancing variables of HCs and control group firms before and after the reweighting by Stata’s ebalance command (Hainmueller & Xu 2013). The table also includes the mean, variance, and skewness for all variables as well as the standardized differences after the entropy balancing.

See Appendix B.1.1 for the complete MIP survey of 2016.

For more information about the concept of industry 4.0 see, e.g., BMWi (2017a).

These may include a firm’s strategy, leadership, culture, customer experience, products and processes, and technologies (e.g., WiWo & Neuland 2015; Berghaus et al. 2017; Capgemini Consulting & MIT Sloan Management 2017; Microsoft 2017; IW Consult 2018). IW Consult (2016) provide a good overview of studies that have measured digitalization in the context of SMEs.

See Table B.3.3 in the appendix for the individual alpha values.

See Table B.3.3 in the appendix for the individual KMO values.

The Bartlett test shows a Chi-square of 19,018.94 and a corresponding p-value of 0.000. Therefore, the null hypothesis that variables are not intercorrelated can be rejected.

PFA is frequently applied to identify higher order factors that explain individual items. I applied this method instead of a principal component analysis, as the latter requires the strict assumption that all variance in the data is common variance. PFA on the other hand allows for unique variances and attempts to partition the common variance (Cleff 2015; Mooi et al. 2018). See Table B.3.4 in the appendix for an overview of the results of the factor analysis.

Figure B.2.4 in the appendix shows the respective scree plot.

See Figure B.2.5 in the appendix for the distribution of the digital readiness variable.

Research emphasizes that incremental innovation activities are much more predictable and less risky than radical innovation efforts which may explain why they account for most innovation activities (e.g., Leifer et al. 2000).

For additional information about the database visit www.​inkar.​de.

One could argue that this indicator may not perfectly describe the actual broadband access for every firm. In particular, larger firms may enforce contracts to ensure broadband access in otherwise isolated areas. Despite this potential issues, the broadband access indicator should still provide a suitable measure for the availability of fast Internet for the on average rather small firms in the sample.

See Table B.3.5 in the appendix for the OLS and IV estimates for the effect of digital readiness on revenue from all types of innovation for the HC and control group firm subsamples.

See Tables B.3.5 in the appendix for the corresponding results of the ‘Chow test’ (Chow 1960) of equality of coefficients between models.

See Tables B.3.6 and B.3.7 in the appendix for the OLS and IV estimates for the effect of digital readiness on the share of revenues based on incremental innovation and radical innovation, respectively, for the HC and control group firm subsamples. The tables also provide the results of the ‘Chow test’ of equality of coefficients between models.

The coefficient of digital readiness in model (1b) is 2.219. As the variable ranges from 0 to 1, a 10 percentage points higher digital readiness increases the logarithm of productivity by 0.2219. Accordingly, the exponentiated value equals 1.2484 indicating a 24.8% increase in productivity.

The coefficient of digital readiness in model (2b) is 1.946. As the variable ranges from 0 to 1, a 10 percentage points higher digital readiness increases the logarithm of productivity by 0.1946. Accordingly, the exponentiated value equals 1.2148 indicating a 21.5% increase in productivity.

The coefficient of HC dummy in model (2b) is 0.223. Accordingly, its exponentiated value equals 1.250 indicating a 25% increase in productivity.

Chapter 1 of this dissertation provides a detailed panel investigation of HC productivity, amongst others, and suggests a similar HC productivity premium.

See Table B.3.8 in the appendix for the OLS and 2SLS regressions of the effect of digital readiness on productivity for the control group firm and HC subsamples.

See Table B.3.9 in the appendix for the OLS and 2SLS regressions of the effect of digital readiness on revenue growth rate for the control group firm and HC subsamples.

This information is provided in the description of the digital readiness index in Section 2.4.2.

See Table B.3.15 in the appendix for an overview and description of the included regional control variables.

See Table B.3.16 in the appendix for a detailed overview of the individual item scores for HCs, non-HCs, and control group firms.

See Table B.3.15 in the appendix for an overview and description of the included regional control variables.

A general limitation of the IV-setup is that if the untestable exchangeability and exclusion restrictions do not hold, the estimations of the 2SLS are not consistent. Moreover, the LATE measures the effect of digital readiness on performance for an unknown subgroup which may limit generalizability of the results (Becker 2016).