The History Of Bayesian Model Averaging

Published Date: 02 Nov 2017

In this article, we investigate how a company’s coopetition (collaboration with competitors), cooperation and competition strategies influence its usage of formal and informal intellectual property (IP) appropriation mechanisms. In addition to firm characteristics and sector variables as predictors for appropriation methods in previous studies, we disentangle coopetition, cooperation and competition as further determinants for a firm’s use of these instruments. Until now, research has not investigated the effects a company’s cooperation with its competitors has on the tendency to use formal (legally protected IP rights e.g., patents, trademarks, etc.) or informal (e.g., secrecy, lead time) appropriation mechanisms. Using Bayesian Model Averaging (BMA), we analyze survey data concerning the coopetition orientation of 1,879 German companies. We find that coopetition, cooperation breadth, and price and technology competition associate with informal appropriation strategies, while in contrast, cooperation depth and a competitive strategy relying on design associate with the use of formal appropriation mechanisms. Thus, we contribute to existing research by providing first exploratory results of coopetition, cooperation and corporate strategies as determinants for a firm’s use of specific appropriation mechanisms. Our findings have implications for management and research. We find that companies should strive for IP appropriation strategies that are well aligned with their choice of cooperation and competition strategies as this facilitates value capturing from innovation.

Keywords: Bayesian Model Averaging; Coopetition; Cooperation; Competition; Intellectual Property Rights

Introduction

Traditionally competition and cooperation are conceived as being at opposite ends of a single continuum {Quintana-García 2004 #3}. Recently the combination of cooperation and competition, vividly captured with the term ‘coopetition’, has received increasing awareness in academic literature (for a recent review see {Dunbar 2012 #144}) and in business practice alike {Dowling 1996 #62}{Hippel 1987 #142}{Chen 2011 #63}. More and more companies are organizing their resources and processes along the frontier of cooperating and competing with other companies. Thus, amongst competition and cooperation, coopetition has become another distinct strategic alternative as new and fast changing business environments require companies to become ambidextrous and pursue both competitive and cooperative strategies simultaneously {Lado 1997 #21}{Jorde 1989 #23}{Bengtsson 2000 #1}. Former and current competitors ever more team up or join forces, combine resources and cooperate on certain research projects, new product developments, or commercialization to exploit the partner firms’ resources, capabilities and know-how.

These relationships, in which firms can assume the role of partners, competitors, suppliers and customers for each other, create potential conflicts and tensions due to the risk of knowledge spillovers and appropriation of rents from joint R&D {Teece 1986 #158}{Pisano 2006 #474}. In this increasingly dynamic business environment, the ownership of the critical pieces of intellectual property (IP) is an important strategic source of a competitive advantage {Granstrand 2000 #18}. Thus, consistent with {Neuhäusler #145/persononly/nopar} {Neuhäusler #145/yearonly}, we define formal appropriation instruments as legally protected IP rights (e.g., patents, trademarks, etc.,) and informal appropriation instruments as methods to prevent involuntary spillovers (e.g., secrecy, lead time).

Moreover, prior literature has analyzed the relationship between the benefits and pitfalls of coopetition and innovation, but it remains unclear how coopetition explicitly influences a firm’s use of appropriation mechanisms. Research has identified several ways in which protecting IP caters to the corporate development of competitive advantage e.g., as an isolating mechanism to prevent imitation {Rumelt 1984 #50}{Mahoney 1992 #26}{Somaya 2012 #56}. Due to the importance of IP appropriation strategies for companies (e.g., {Blind 2006 #15}), it is an important and necessary challenge for research to put IP appropriation strategies in the context of coopetition, cooperation and competition, and to identify drivers for certain components of appropriation strategies {Rivette 2000 #39}.

The impact of coopetition, cooperation and competition on companies’ use of various knowledge and IP appropriation strategies – we differentiate between a formal component and an informal component of appropriation – has previously not been analyzed in ‘what is currently a somewhat disparate and fractured field of study within management’ {Somaya 2012 #56: p. 1084}. For the empirical analysis this entails that we do not only have to estimate the effects of coopetition, cooperation and competition on the implementation of the appropriation strategies, it also means that the structure of the regression model is essentially uncertain, as at the outset it is unclear which variables should be included in the regression model. This phenomenon offers an interesting application for Bayesian Model Averaging (BMA). Therefore, we empirically estimate the model structure and the parameters by means of Bayesian Model Averaging to account for this uncertainty (e.g., {Raftery 1995 #92}{Hoeting 1999 #107}).

The remainder of the paper is structured as follows. Chapter 2 provides an overview of the current literature on coopetition and factors influencing IP appropriation strategies. Next, we describe the methodology and the data set before we give a descriptive overview on the use of appropriation measures. Chapter 4 focuses on the factors explaining the use of property rights and appropriation mechanisms. In the final chapter 5, we give a comparative summary of the results and address and discuss potential challenges for management and research.

Conceptual Background

Coopetition

The concept of coopetition typically refers to the relationship between firms that simultaneously involves both competition and cooperation (e.g., {Brandenburger 1996 #59}). Thus, the concept of coopetition comprises a complex combination of two opposite logics of interaction: the competitive paradigm, assuming that companies interact based on conflicting interests, and the collaborative paradigm, asserting that companies interact based on common interests in a certain area {Dowling 1996 #62}{Bengtsson 2000 #1}{Cassiman 2009 #147}. Despite many risks and conflicts, cooperation with competitors is usually driven internally by the need to share R&D or production risks and costs, by the goal to pool resources, develop and expand markets, address major technological challenges, reduce costs and risks and realize synergistic effects {Das 2000 #67}{Tether 2002 #11}{Huang 2009 #68}, or externally by the requirement to comply with new regulations {Nakamura 2003 #71} or develop industry standards.

Here, we use coopetition in the vein of {Bengtsson 2000 #1/persononly/nopar} {Bengtsson 2000 #1/yearonly}. According to these scholars, a firm is involved in coopetition if it carries out cooperative activities with other actors the focal firm itself classifies as competing, regardless of whether or not the competition is in the same product area or in the same industry. Although coopetitive activities can occur at multiple levels, such as at the firm level, at the industry level, at the level of strategic business units, the department level, or between teams {Luo 2006 #7}{Gnyawali 2011 #42}{Tsai 2002 #9}, we restrict our focus on coopetitive innovation activities at the firm level.

Different theories have been used to assess the value of coopetitive activities. Transaction cost theory focuses on the competitive dimension and therefore, pitfalls of this strategy. Reasoning based on a transaction cost rationale renders coopetition as a risky strategy because of the knowledge paradox [1] {MADHOK 1997 #136}{Nickerson 2004 #151} on the one hand and the involuntary leakage of tacit knowledge to the collaborating, yet competing, partners, on the other hand {Cassiman 2002 #122}. Incentives for opportunistic behavior originating from the competitive dimension of this strategy theoretically undermine the benefits of the cooperative dimension {Quintana-García 2004 #3}.

Arguments originating from the resource-based view {Barney 2001 #486}{Barney 1991 #485}{Teece 1997 #487} focus more on the cooperative dimension and thus, the benefits of coopetitive behavior. Firms gain a competitive advantage by absorption, assimilation and transformation of knowledge from different areas {Kessler 2000 #60}{Kogut 1996 #61}. The results of these activities accumulate as knowledge assets specific to the individual firm {DeSarbo 2007 #66}{Wang 2009 #65}. Competitors are valuable sources of complementary knowledge and resources, which can be accessed through cooperation {Grant 1995 #138}. The resource-based view, hence renders coopetitive activities as an important way to increase the innovation capabilities of firms. In a first step, coopetition can be interpreted as a collective effort in the form of cooperation leading to value creation, i.e., by creating new or improving current products or services as well as by establishing new or enlarging current markets. In contrast to the first step, the second rather focuses on individual firm aspects as it comprises a company’s competitive effort to appropriate value. How firms protect their intellectual assets and how they appropriate their returns, hence is largely contingent on firm specific cooperative and competitive strategies {Ritala 2009 #22}. In sum, different theories can explain advantages and disadvantages of a coopetition strategy.

Even though coopetition is challenging for the involved firms, it creates certain advantages such as a positive effect on new product development and innovation as it enhances the involved firms’ capacity to innovate {Ritala 2009 #22}{Gnyawali 2011 #42}. These effects exceed those generated by competitive relationships because partnering companies can control their competitors more effectively {Quintana-García 2004 #3}{Chen 2011 #63}. Despite the positive effects of coopetition on value creation, companies use this strategy as a means to imitate rather than to generate radical innovations due to opportunistic behavior and knowledge spillover {Mention 2011 #10}{Monjon 2003 #64}.

In sum, although earlier literature on coopetition has identified the related motives, unique potential and benefits (e.g., innovation activities), some studies on coopetition emphasize that it also comprises some major risks and drawbacks (e.g., {Hamel 1991 #467}{Park 1996 #470}{Oxley 2004 #468}{Ritala 2008 #472}) and thus, may not be desirable in certain cases.

Appropriation Strategies

In a review, {Somaya 2012 #56/persononly/nopar} {Somaya 2012 #56/yearonly} provides an overview of theoretical drivers of companies’ appropriation strategies. Using the special case of patent protection, he emphasizes the importance of integrating appropriation strategies into the company-level strategy. Although a protection use is often referred to as an isolating mechanism, i.e., to prevent imitation of the firm’s technological assets {Rumelt 1984 #50}{Mahoney 1992 #26}, there are other appropriation strategies that also support firm-level competitive advantage {Mansfield 1986 #55}. Among these strategies are blocking, building fences and thickets, earning licensing income, avoiding litigation by others, using IP in negotiation and exchange, motivating and rewarding R&D personnel, measuring performance, attracting investors, and forming image and reputation {Blind 2009 #14}{Cohen 2000 #57}. In general, appropriation strategies can be divided into two groups of measures (e.g., Cohen 2000 #57}{Neuhäusler #145}):

Formal appropriation instruments, such as patents, trademarks, utility patents or copyright, are state guaranteed legal instruments, which grant inventors and innovators an exclusive right to exclude others from the utilization of the protected subject matter.

Informal appropriation instruments encompass various measures on the part of companies to prevent spillovers of own innovation efforts and thus to safeguard the appropriation of one's own innovation returns. Typical forms are secrecy, lead time, complex design of new products or services, which make imitation more difficult, or an extremely rapid implementation of innovation projects to generate a lead time advantage.

Notwithstanding this terminology, formal appropriation mechanisms can also be used strategically as a quality signal to potential investors as well as to potential R&D, alliance or licensing partners {Gans 2008 #87}{Gick 2008 #88}{Somaya 2012 #56}. Firms may use patents and other formal IP to exhibit strategic commitment to a technological or research trajectory in order to drive competitors into exiting R&D competition, patent races or terminating their R&D efforts {Gill 2008 #89}{Somaya 2012 #56}, or to prevent patenting by others and guarantee freedom of operation for the filing company {Guellec 2011 #90}. Firms may also strategically employ formal appropriation measures to discourage competitors from further investments in the same technology domain {Somaya 2012 #56}{Agrawal 2007 #72}{Baker 2005 #85}. Formal IP can disclose information about a firm’s technologies and technical trajectory and competitors may use this information for future innovation competitions. Therefore, firms may strategically patent ‘poor’ inventions to misguide competitors {Langinier 2005 #28}.

{Brouwer 1999 #146/persononly/nopar} {Brouwer 1999 #146/yearonly} (also: {Levin 1987 #475}{Arora 1997 #129}) highlight that patenting is not the most important instrument for appropriation of innovation benefits. Companies rather find informal means crucial to capture value from invention. However, only the combination of formal and informal measures of appropriation constitutes a proper IP protection strategy {Hertzfeld 2006 #31}, allows for flexibility to adjust to different internal or external strategic requirements {Anton 2004 #76}, and hence determines the effectiveness of a firm’s IP strategy {Reitzig 2004 #131}{Reitzig 2009 #128}{Somaya 2012 #56}{Somaya 2010 #86}{Leiponen 2009 #132}. Research reveals that firms prevent imitation not only by using patents or secrecy, but by building on a full portfolio of appropriation mechanisms available to them and thus securing or developing competitive advantages {Somaya 2012 #56}. The interrelationship between patenting and secrecy follows a long research tradition describing patent rights and secrecy as natural substitutes (e.g., {Machlup 1962 #52}{Horstmann 1985 #53}{Kultti 2007 #32}{Arundel 2001 #81}) or strategic complements Anton 2004 #76}{Arundel 1998 #33}. Although {Schmoch 2003 #127/persononly/nopar} {Schmoch 2003 #127/yearonly} examines the relationship between trademarks and patents, empirical evidence on the association between copyright and patenting is still needed.

In a recent study, {Neuhäusler #145/persononly/nopar} {Neuhäusler #145/yearonly} (also: {Cohen 2000 #57}) examines and finds firm characteristics (e.g., R&D personnel, sales, size) and sectors that affect the decision for or against a specific appropriation method. In sum, the simultaneous use of formal and informal measures of a as part of a coherent protection strategy as well as their prevalance is – except for some contributions {Graham 2003 #75}{Cohen 2000 #57}{Neuhäusler #145}{Somaya 2011 #286} – still somewhat emerging in research {Somaya 2012 #56}.

Coopetition, Cooperation, Competition and Appropriation Strategies

A relatively new literature stream has begun to address the impacts of patents and patent strategy on firms’ value creation through innovation {David 2006 #80}. Particularly, strong IP portfolios and aggressive IP appropriation strategies serve two purposes when allying or licensing with other firms: as a deterrent of opportunistic behavior and as an enabler of value appropriation through commercializing inventions and R&D results {Oxley 1999 #77}{Arora 1995 #78}{Somaya 2012 #56}. This highlights that strategies to protect intellectual property eventually affect the competitive as well as cooperative structures (e.g., {Peeters 2006 #84}{Blind 2004 #30}{Hertzfeld 2006 #31}).

However, in a micro-perspective, the implementation of IP appropriation strategies also depends on the firm’s cooperative and competitive strategic options and the competitive environment it operates in. Only few studies deal with the relationship of coopetition and innovation (e.g., {Mention 2011 #10}{Ritala 2012 #149}{Belderbos 2004 #20}{Tether 2002 #11}), but in general these studies find coopetition to be beneficial for firms’ innovation activities or outcome. Moreover, prior literature has analyzed the relationship between coopetition and value capture (e.g., appropriation and imitation), but it remains unclear how coopetition explicitly influences a firm’s use of appropriation mechanisms.

With respect to the effects of competition on the appropriation strategies of companies, the evidence generally points towards a positive relationship. The more intense the competitive environment, the more intense the use of formal appropriation mechanisms becomes {Blind 2006 #15}{Peeters 2006 #84}{Hall 2001 #134}{Ziedonis 2004 #135}.

Regarding the strategic orientation of the firm, {Blind 2004 #30/persononly/nopar} {Blind 2004 #30/yearonly} find that protection activities are driven by technology protection motives and by the strategic rationale reflected in a defensive strategy. A firm specific strategic orientation strongly determines the composition and the value of the IP portfolio (e.g., {Blind 2009 #14}; also: {Anton 2004 #76}). {Peeters 2006 #84/persononly/nopar} {Peeters 2006 #84/yearonly} show that firms’ engagement in formal IP protection is more intense when innovation strategies are aligned with research intensive product innovation. Additionally, they find that broad collaboration activities, i.e., cooperative innovation activities with science and industry, increase the intensity of formal IP protection. Generally IP protection is an issue relevant for all partners involved in collaborative innovation activities {Hertzfeld 2006 #31}. {Blind 2006 #15/persononly/nopar} {Blind 2006 #15/yearonly} seem to contrast this finding, as they cannot identify cooperative innovation as a determinant of appropriation activities. However, these scholars find a strong effect of intensive technologically motivated collaboration captured by their co-patenting variable.

The conceptual framework for this study is summarized in Figure 1.

----------------------------------------------

Insert Figure 1 about here.

----------------------------------------------

Research Questions

Despite a considerable body of literature investigating the use of appropriation mechanisms, {Somaya 2012 #56/persononly/nopar} {Somaya 2012 #56/yearonly} argues on a general account that strategic and competitive determinants of appropriation strategies are still not fully explored. Previous research has analyzed IP as a facilitator or driver of coopetitive and cooperative relationships {Carayannis 1999 #148} {West 2006 #465}, but to our knowledge, no study has, as yet, explicitly analyzed the effect of coopetition on companies’ usage of IP appropriation instruments. This gives rise to the first research question targeted at filling this gap. Yet, at the outset it is unclear, on which model the estimation of these effects should be based. This means that the analysis faces model uncertainty: which variables should be included in the estimation of these effects? This leaves us with two research questions:

Which variables contribute to our understanding of the implementation of different types of appropriation strategies and hence should be included in the estimation?

How strong is the effect of coopetition, cooperation, competition strategies, and of the competitive environment on the implementation of the types of an appropriation strategy?

Methodology and Data

Generally when exploring the effect of an independent variable on a dependent variable one could suggest a single model including a number of control variables and estimate the effect based on the assumption that this selected model captures the true relationship. In our context, this means that for exploring the strategic and competitive determinants of appropriation strategies, we select and analyze one single model specified by a particular selection of exogenous variables. Yet, there may be two or more models – defined by two or more different sets of exogenous variables – that fit the data rather well. Each one of these models may conceptually be justified, but may at the same time lead to different conclusions. For policy conclusions and for corporate strategic decisions this is a rather unfortunate situation.

For situations like this, {Raftery 1995 #92/persononly/nopar} {Raftery 1995 #92/yearonly} suggests that we face three options: first we pick one model, analyze the determinants of appropriation strategies and draw our conclusions based on this one model without further ado. This approach is arguably characterized by a considerable degree of arbitrariness. The second option is to base the analysis on all possible models without giving preference to one of these models. {Raftery 1995 #92/persononly/nopar} {Raftery 1995 #92/yearonly} argues that this is sounder, yet, not fully satisfactory. Bayesian model averaging (BMA) offers an appealing third option by accounting for model uncertainty, that is, that a priori we are not sure about the structure of the correct model. BMA accounts for this uncertainty without incurring the common problem of data mining, which is the search for and the selection of a single model without the presentation of the process leading to the selection {Brock 2007 #95}.

The BMA approach has recently attracted increasing attention in diverse areas of economic research such as macro-economic growth e.g., {Durlauf 2001 #106}{Magnus 2010 #93}{Crespo-Cuaresma 2010 #94}), forecasting (e.g., {Liu 2009 #99}{Wright 2009 #100}), or agricultural economics (e.g., {Balcombe 2010 #101}{Tiffin 2011 #102}). Micro-econometric applications of BMA in management research are still in their infancy with only a few applications in corporate finance (e.g., {Avramov 2002 #482}{Liu 2009 #99}{Pesaran 2009 #477}) and other areas of management (e.g., {Hansen 2004 #488}.

Bayesian Model Averaging

The identification of important determinants for the implementation of formal and informal appropriation strategies leads to a problem of model uncertainty, that is, which model of all models should be used for the estimation of the parameters. It is fundamentally unclear which variables out of the potential variables X should be included in a regression model (e.g., Hoeting 1999 #107}.

We assume the regression model to be linear, where y is the dependent variable, is the constant intercept, is the vector of coefficients and is the error:

and

In a BMA approach all possible models are generated by selecting all possible combinations of variables . Each of the models is subsequently estimated. The results of these estimations can be used to select the model with the highest posterior probability and to continue with the inference based on this single best model. If most of the posterior probability mass is not concentrated on this particular model, a lot of information would be lost in neglecting all the other models (e.g., {Raftery 1997 #483}. Hence, BMA constructs a weighted average of the coefficient estimates of all models, where the weights are derived from posterior model probabilities based on Bayes’ theorem:

This approach meets our empirical need as not only the parameters and have to be estimated; the model and in particular its specification captured by will also be estimated.

By means of BMA we can analyze two distinct questions {Montgomery 2010 #108}:

Research question 1:

Does the variable contribute to the explanatory power of the models explaining the dependent variable?

Our particular interest here is the posterior inclusion probability (PIP) of a variable , which is the sum of the probabilities of those models that include variable . We can interpret the posterior inclusion probability as the relative importance of this variable in explaining the dependent variable. In addition, we are interested in the weighted mean of the posterior distribution of this parameter estimate for this variable, which we denote as the posterior mean (PM).

The posterior variance of the parameter and hence its standard deviation PSD is given in a similar fashion.

Research question 2:

What is the size of the variable’s effect once it is included in the model?

This can be analyzed by the posterior distribution of the coefficient estimate conditional on inclusion, which is the distribution of the coefficient estimate of all models where is included, where weights are applied based on the posterior model probabilities.

By analyzing the distribution of the coefficient estimate conditional on inclusion we investigate question (ii) above.

In our case, we will have variables (see the data section below) which give rise to a set of different models. Instead of full enumeration of the model space we use a Monte Carlo Markov Chain sampler to sample the most important models from the posterior distribution of models. [2] The initial 1,200,000 burnin draws to find the most important region in the model space are discarded. We base our estimations on the subsequent 2,400,000 draws. The correlation between the iterations and the posterior model probability in our investigation is well above 0.999 (Table 2), which shows a sound degree of convergence of the algorithm.

For the prior error variance and for the constant intercept in the linear model, we impose an improper that is a non-informative prior. These priors capture a minimum of prior information. For the regression coefficients we integrate our prior belief in a normal distribution with mean zero and a variance structure of . This prior indicates a conservative belief that a given variable does not affect the appropriation strategies and that the variance-covariance structure is in accordance with the data. For our prior belief, we choose g = 1/N, which yields the same variance as the unit information prior, but ensures that the mean is zero rather than the maximum likelihood estimate {Eicher 2011 #479}{Fernández 2001 #484}{Fernández 2001 #96}.

We specify the prior model probabilities by a uniform prior . Essentially this implies a 50% probability that a given variable is part of the true model (e.g., {Eicher 2011 #479}). This implies that we treat the inclusion of the variables a priori as independent. The set of variables that will be included in the analysis will also include interactions. In the case of these interactions we deviate from this independence. In the case of interactions, say AB, between two so called parent variables A and B, we apply a strong heredity principle {Chipman 1996 #111}. This requires that whenever the interaction variable AB is included in the regression all parent variables, here A and B, have to be included as well. If this principle is violated the analysis of the interaction effects may under certain conditions lead to misleading findings, see the discussion in and between Masanjala 2008 #110/persononly/nopar} {Masanjala 2008 #110/yearonly}, {Crespo-Cuaresma 2010 #94/persononly/nopar} {Crespo-Cuaresma 2010 #94/yearonly}, and {Papageorgiou 2011 #109/persononly/nopar} {Papageorgiou 2011 #109/yearonly}. This also applies to squared variables.

To illustrate the distinct advantages of the BMA approach, we also report a baseline ordinary least squares (OLS) model and contrast the findings of both approaches.

As a caveat it has to be noted that generally it is not the case that model averaging "limits the effect of prior information" as {Sala-i-Martin 2004 #481/persononly/nopar} {Sala-i-Martin 2004 #481/yearonly, p. 815} suggest {Ley 2009 #480}{Eicher 2011 #479}. Hence, we conduct a robustness check to investigate how our findings depend on our selection of the model prior and on our specification of the g-prior. The robustness check bases on the preferred specification in {Ley 2009 #480/persononly/nopar} {Ley 2009 #480/yearonly, who impose a hierarchical model prior. They also use for the g-prior. The computations under these assumptions lead to findings that are structurally equivalent to the findings documented here. Detailed results of the robustness check are available from the authors upon request.

Data

Our analysis is based on the Mannheim Innovation Panel (MIP), ZEW, Mannheim, which includes the core Eurostat Community Innovation Survey (CIS) and additional topics for firms from Germany. The CIS, jointly launched by Eurostat and the Innovation and Small and Medium-sized Enterprise Program in 1991, aims at improving the empirical basis of innovation theory and innovation policy on the European level by surveying innovation activities on the company level in the member states' economies. The CIS surveys generate cross-sectional data on firm-level innovation activities across member states by means of largely harmonized questionnaires. The CIS closely reflects the definitions of the Oslo Manual {OECD 2005 #270} and hence provides a good coverage of the indicators for innovation input, innovation output, innovation strategy, and the use and appreciation of IP appropriation strategies employed by innovating companies. Initially, the CIS has been used to inform national and EU-level statistical analyses. In the past decade, the data have increasingly been used for scientific research on the micro-level in management (e.g., {Cassiman 2006 #121}{Belderbos 2004 #20}{Leiponen 2010 #113}{Ebersberger 2011 #119}{Grimpe 2010 #118}{Laursen 2006 #116} and in economics (e.g., {Cassiman 2002 #122}{Czarnitzki 2007 #98}{Ebersberger 2012 #120}). We use the German edition of the fourth (CIS4) covering the years 2002–2004. The dataset contains 1,879 companies which actively employ appropriation measures. These are the basis of the analysis below.

Dependent variables.

Appropriation strategies: The innovation survey inquires innovating companies about their usage of a set of measures to protect their IP: patents, utility model, trademarks, copyright, secrecy, complexity of design, and lead time advantage. The survey also investigates the appreciation of the used appropriation measures on a three-level Likert scale (high – medium – low). We use a factor analysis (principal component factors, varimax-rotated) to identify latent strategies in the responses (see Table 3 Panel –A– in the Appendix). We only extract the two factors with an eigenvalue larger than unity. The first factor bundles secrecy, complexity and lead time advantage. In accordance with the literature, we interpret this as an informal appropriation strategy (PROT_INF). The second factor bundles patent, utility models, design patents, trademarks, and copyrights. We interpret this factor as a formal appropriation strategy (PROT_FORM).

Independent variables.

Coopetitive strategy: A coopetitive strategy is approximated by a dichotomous variable that indicates that the firm collaborates in its innovation activities with a competing firm (COOPET).

Competitive strategy: To capture the competitive strategy of the firm, the survey asked the responding firms to assess the following factors and rank those according to their relevance for the firm’s competitiveness in its main market: price, quality, technological advantage, service and flexibility, variety of products, and design of the product and the marketing campaign. We capture the firm’s competitive strategy in six dichotomous variables. Each variable indicates that the respective factor has been ranked first or second by the firm. Some firms rank more than one factor first. These ties have not been resolved. The dummy variable STRT_PRCE hence captures a competitive strategy relying on price advantages, STRT_QUAL indicates a competitive strategy relying on quality advantages, STRT_TECH is a competitive strategy relying on technological advantage, STRT_SERV relates to a competitive strategy relying on service and flexibility, STRT_VARI reports a competitive strategy relying on variety of products, and finally STRT_DSGN designates a competitive strategy relying on design of the product and the marketing campaign.

Collaboration network: Analogous to the literature on innovation search {Laursen 2006 #116}, we capture innovation networks by their breadth (COOP_BR) and by their depth (COOP_DE). The former identifies the number of different collaboration partners, whereas the latter reports the fraction of collaboration partners with a high intensity of collaboration approximated by collaboration with this specific partner in more than one world region (Germany, Europe, US, other). For these indicators, we only consider the non-competitive collaborations.

Competitive environment: In a set of questions, the survey examines the competitive environment of the firms. We use the six items scaled with a four level Likert scale of agreement to construct latent dimensions of the competitive environment by means of a factor analysis. We only extract factors with an eigenvalue above unity and yield two factors (see Table 3 Panel –B– in the Appendix). In particular, we distinguish between environments where competition is driven by product and technology characteristics (COMP_PD) and environments where competition is driven by the behavior of competitors and markets (COMP_CO).

Controls.

The analysis also contains a number of control variables. Firm characteristics are controlled for by the size measured by the log of the number of employees (LEMP), by the research intensity measured as the sales share spent on R&D (RDINT), by the firm’s involvement in international trade measured by the sales share generated by exports (EXSHR) and by the firm’s location in Eastern Germany (EAST). Usually the nature of the knowledge a firm’s innovation activities builds on affects the way and intensity of protection {Norman 2002 #123}. To characterize the knowledge the firms rely on in their innovation activities, we build a dichotomous variable (ANALYT) indicating whether the firm’s innovation activities rely on an analytical knowledge base rather than a synthetic one ({Laestadius 2000 #154}{Asheim 2005 #155}{Asheim 2012 #157}). An additional dummy variable (CUM) indicates whether the innovation activities rely on a strong cumulativeness of the knowledge bases {Breschi 2000 #152}, which might be closely related to product sequencing {Helfat 2000 #153} and to related protection challenges. Issues hampering the firm’s innovation activities are bundled by a factor analysis (see Table 3 Panel –B– in the Appendix) and, in line with the findings in {Peeters 2006 #84/persononly/nopar} {Peeters 2006 #84/yearonly}, give rise to two factors: economic and financial constraints (HAMP_ECO) and internal or knowledge constraints (HAMP_INTERN). We use two sector controls, one capturing the overall sectoral affinity for employing formal or informal means of appropriation by the mean of the dependent variable broken down on NACE 3 digit sectors. In addition, we control for different propensities to employ appropriation mechanisms at all by the share of firms with a protection strategy in a NACE 2 digit sector and the respective size class. Table 1 summarizes the variables in the analysis. We standardize all variables for the analysis and report the descriptive statistics and the correlation table in Table 5 of the Appendix.

----------------------------------------------

Insert Table 1 about here.

----------------------------------------------

Results

Table 2 summarizes the results of the BMA analysis. It contains the posterior inclusion probability (PIP) for each of the variables, which is the sum of the posterior model probabilities (PMP) of all those models that include the variable. Table 2 also contains the mean of the posterior distribution (PM) and its standard deviation (PSD). The PIP can be interpreted as the probability of a variable being part of the true model ({Cuaresma 2010 #124}, p. 289). It is the probability of the variable to be included in a model ({Koop 2003 #125}, p. 267). Hence, the PIP gives the importance of the variable for explaining the dependent variable. With a posterior inclusion probability larger than 50% we regard the associated variable as a robust determinant of the appropriation measures (e.g., {Raftery 1995 #92}). This threshold value is approximately equivalent to PM/PSD=1; in a frequentist interpretation this means that the variable increases the power of the {Masanjala 2008 #110}.

----------------------------------------------

Insert Table 2 about here.

----------------------------------------------

Both in the analysis of formal appropriation mechanisms and in the analysis of informal appropriation mechanisms, we observe that differences between the sectors (SEC_MEAN) are robust. Moreover, the data reveal strong differences between the determinants for the dimensions of appropriation strategies for companies actively pursuing an appropriation strategy. We also observe that the implementation of appropriation measures is positively determined by the firm size (LEMP), where larger firms exhibit more intensive protection through formal and informal measures. Yet, the positive effect on the implementation of informal measures with a posterior inclusion probability of 38.3% is below the robustness threshold of 50%.

Firm location in Eastern Germany (EAST) with posterior inclusion probability of 91.6% is found to be a robust determinant of formal measures of appropriation, where firms in Eastern Germany exhibit a lower level of formal appropriation strategies. Informal measures are negatively affected by the firms’ environment. Firms in sectors and in size classes with a lower share of protecting companies (PROT_PROB) exhibit a higher usage level of informal measures.

After this brief overview of the control variables’ effects, the discussion now focuses on coopetition, cooperation and competition as determinants of appropriation strategies.

Determinants of Formal Appropriation Strategies

Aside from the control variables discussed above, the use of formal appropriation mechanisms is determined by two factors. First, a firm’s collaboration strategy strongly affects its implementation of formal appropriation methods. In particular, it is the depth of the collaboration network (COOP_DE) that exerts the univocally positive effect on the use of the formal appropriation mechanisms. 92.4% of the probability mass of the models rests on models containing the depth of the network.

We also find that corporate strategy matters. A competitive strategy, which strongly builds on design (STRT_DSGN), is also a robust determinant of formal appropriation mechanisms, as 69.4% of the probability mass rests on models including this variable, all of these models indicate that a competitive strategy that heavily relies on design has a positive effect on the implementation of formal measures of appropriation.

To investigate the effects of the robust determinants in more detail and to shed light onto the question about the size of the effects (see research question 2 above), we graph the posterior coefficient distributions of the robust determinants in Figure 2. The distributions are conditional on inclusion, and hence the area under the density equals the probability of inclusion (PIP) reported in Table 2. We also report the mean, the standard deviation and the 95%-HDI (highest density interval) of the parameter estimates conditional on inclusion. A baseline OLS regression explaining PROT_FORM and PROT_INF including all exogenous variables is documented in Table 4 of the Appendix. For reference, it is also represented in the diagrams. The solid vertical line gives the mean of the distribution. For reference, the dash dotted lines indicate the parameter estimate of the OLS regression and the dotted line indicates the 2 SD interval on each side of this parameter.

We observe that cooperation depth (COOP_DE) has a mean effect of 0.094 and that at least 95% of the probability mass of the models including cooperation depth as an explanatory variable yield positive parameter estimates between 0.043 and 0.140. For the competitive strategy focusing on design (STRAT_DSGN), the analysis produces a distribution of the estimated coefficient with a mean of 0.064, conditional on the variables inclusion in the regressions. The 2.5% quantile of this distribution is clearly larger than zero, which indicates a consistently positive effect of a design focused competitive strategy on the implementation of formal appropriation strategies.

----------------------------------------------

Insert Figure 2 about here.

----------------------------------------------

Determinants of Informal Appropriation Strategies

The analysis reveals a number of robust determinants of informal appropriation strategies that are related to coopetition, cooperation and competition strategies. In contrast to the determinants of the formal appropriation mechanisms, the informal appropriation mechanisms are robustly and positively affected by the breadth of the company’s cooperation network. 99.8% of the probability mass of all models rest on models including COOP_BR. Two dimensions of the firm’s competitive strategy robustly determine the implementation of informal appropriation methods. First, a strategic orientation of the firm focusing on price advantages (STRAT_PRCE) reduces the implementation of informal appropriation measures. Second, strategic focus on competitiveness through technological advantage (STRAT_TECH) is a strong determinant of informal appropriation mechanisms. Coopetition (COOPET) is also found to affect the implementation of informal appropriation measures robustly.

----------------------------------------------

Insert Figure 3 about here.

----------------------------------------------

In order to address research question 2 above and analogous to Figure 2 for the formal appropriation measures, we investigate the effects of the robust determinants on informal measures in more detail by displaying the posterior coefficient distributions conditional on inclusion in Figure 3. For coopetitive innovation strategies (COOPET), we find a distribution of the parameter estimates conditional on inclusion with a mean of -0.068. The quantiles reported in Figure 3 show that the estimates are clearly below zero indicating a negative effect of coopetitive strategies on the implementation of informal measures of appropriation. Also the distribution of the parameter estimates for a strategic focus on price advantages (STRAT_PRCE) reveals that a mean of the estimated parameters of -0.085 when this strategic option is part of the model. Furthermore, the interval between the 2.5% quantile and the 97.5% quantile indicates that the parameters are precisely estimated. The two robust determinants of informal appropriation methods – collaboration breadth (COOP_BR) and a competitive strategy relying on technological advantage (STRAT_TECH) – are also rather precisely estimated. The distribution of the estimate conditional on inclusion shows a mean of 0.120 and 0.106 respectively. For both distributions, the given quantiles indicate a strongly positive mean of the parameter estimates.

The findings indicate so far, that the strongest determinants of the composition of appropriation strategies can be found among the firm specific cooperation strategies, among the firm specific competitive strategies and in the coopetitive strategy.

In contrast to the frequentist approach of reporting p-values, BMA can also shed light on the factors which do not determine the appropriation strategies. These are those variables with a posterior inclusion probability close to zero {Hoeting 1999 #107}. In particular, we find two groups of variables which fall into this group. First, the external competition environment of the companies – be it caused by product and technology characteristics (COMP_PD) or be it caused by the behavior of competing companies (COMP_CO) – reveals a posterior inclusion probability close to zero. The second group of variables comprises the pairwise interactions of competitive strategy, competitive environment, cooperation and coopetitive strategy. All the interactions generate no effect on the implementation of formal or informal appropriation strategies. Hence, the effects of the robust determinants indicated above are not moderated by any other competitive or cooperative strategy.

Does Model Uncertainty Matter?

In this brief analysis, we will investigate whether model uncertainty matters in the given context. We do so, by briefly highlighting the differences in the findings between an OLS regression and our BMA analysis, where both analyses include the same set of initial variables. We plot the p-values of the OLS regression against the posterior inclusion probabilities in Figure 4.

----------------------------------------------

Insert Figure 4 about here.

----------------------------------------------

In Figure 4, we identify robust determinants of the formal or informal appropriation strategies in the BMA analysis by their location above the horizontal line indicating a PIP threshold of 0.50. We can also classify significant determinants in the OLS regression by their location on the left of the vertical line indicating a significance of 10%.

Both the BMA analysis and the OLS regression agree on a number of determinants and their high relevance for the implementation of either the formal or the informal appropriation measures. These determinants are plotted in the upper left quadrant of the panels in Figure 4. In particular, for formal appropriation measures these are the firm size (LEMP), the sector control (SECT_MEAN), the location in East Germany (EAST) and a competitive strategy relying on design advantages (STRAT_DSGN). For the informal measures, these are the sector control (SECT_MEAN), the probability of protection (PROT_PROB), the competitive strategies relying on technology and price advantages (STRAT_TECH, STRAT_PRCE), the breadth of the collaboration network (COOP_BR), and a coopetitive strategy (COOPET).

The BMA analysis and the OLS regression also concur in assigning negligible relevance to a number of potential determinants. These are plotted in the lower right quadrant. They are not labeled.

The BMA analysis and the OLS do not match in their assessment of relevance concerning the depth of the cooperation network (COOP_DE) for the formal appropriation measures. It is clearly identified as a robust determinant in the BMA analysis, whereas the OLS does not find this variable to be significant.

More importantly, and certainly more striking due to the number of involved variables, the OLS regression generally finds more determinants of both the formal and the informal appropriation measures as it neglects the model uncertainty as an important source of uncertainty in the estimation process. The variables that are found significant in the OLS regression, but that turn out to be not robust in the BMA analysis, are located in the lower left quadrant. In particular, this relates to the competitive strategy based on variety and services (STRAT_VARI, STRAT_SERV), on R&D intensity (RDINT) and on the interaction between cooperation network and a coopetitive strategy (COOP_DE*COOPET) in the case of the formal appropriation mechanisms. In the case of informal measures of appropriation, this relates to perceived hampering factors (HAMP_ECO, HAMP_INTERN), to the firm size (LEMP), to various interactions between the collaboration network, the competitive environment and a coopetitive strategy (COOP_BR*COOPET, COMP_PD*COOP_BR, COMP_CO*COOPET). The relevance of the determinants plotted in this lower left quadrant is contingent on the choice of the structure of the linear regression model. Taking these determinants as relevant may lead to governmental policy conclusions or to strategic corporate decisions that cannot be maintained when a different structure of the model is used for the analysis.

Discussion and Conclusions

Consistent with {Arora 2006 #79 /persononly/nopar} {Arora 2006 #79/yearonly} and {Arora 1997 #129/persononly/nopar} {Arora 1997 #129/yearonly}, this study examines the interplay between different formal and informal IP instruments available to firms in their quest to appropriate rents from innovation. From a theoretical perspective, this paper sheds some light on the strategic and competitive determinants of appropriation strategies that have not been fully explored, as yet. Moreover, prior literature shows ambiguous results regarding companies’ prevalence for informal versus formal IP instruments.

Research has remained largely focused on firm characteristics as drivers of companies’ use of various knowledge and IP appropriation strategies while the impact of coopetition, cooperation and competition on remains underexplored {Somaya 2012 #56: p. 1084}. Therefore, the structure of the regression model is essentially uncertain, as at the outset it is unclear which variables should be included in the regression model. This phenomenon offers an interesting application for Bayesian Model Averaging as this method empirically estimates the model structure and the parameters to account for this uncertainty. In this paper, we use BMA analysis to disentangle determinants of companies’ use of specific appropriation strategies. Interestingly, the competitive environment does not have a robust effect on the implementation of formal or informal measures of appropriation. Only firm level strategies can be identified as robust determinants of appropriation strategies. Either the appropriation strategies are simultaneously developed with corporate strategies (cooperation or competitive strategies) or they are the result of these.

Following a coopetitive strategy is not a robust predictor for formal appropriation methods, but for the use of informal appropriation instruments. This is a rather surprising result as one would assume enforceable IP rights as well as informal appropriation mechanisms to be associated with a coopetition strategy. Due to the nature of the phenomenon coopetition, informal appropriation instruments should go hand in hand with formal appropriation instruments in order to mitigate the risk of involuntary knowledge spillover and imitation. Moreover, the depth of cooperation relations is a very robust parameter explaining formal appropriation instruments. Obviously, intensive cooperation requires the use of formal rights in order to manage and control critical knowledge assets. In contrast, the breadth of cooperation requires the use of informal appropriation measures due to the heterogeneity of their features, e.g., using either lead time advantage, secrecy, or other instruments. A company sharing knowledge with many different partners in a broad open innovation setting creates more potential imitators. Thus, strengthening imitators and new competitors can be mitigated by employing informal appropriation methods. Prior research also has revealed that companies rely more heavily on informal appropriation instruments when protecting critical knowledge and capabilities {Cohen 2000 #57}{Levin 1987 #475}{Neuhäusler #145}.

Furthermore, companies using both price and technology in their competition strategy prefer implementing informal than formal appropriation instruments. Pricing obviously reflects setting rather low prices, which contradicts the focus of premium price segments characterized by valuable patents and trademarks. In contrast, achieving lead time advantage may require sophisticated pricing strategies to either enter the market or to exploit pioneering market positions. Following a technology-based competition strategy may increase the likelihood of a lead time strategy on the one hand and hence render secrecy much more effective than trying to enter long patent application processes on the other hand. Furthermore, a complex technology often is an effective appropriation instrument in itself. Finally, companies, which rely on and derive a competitive advantage from a unique product design as part of their competitive strategy, are also more inclined to use formal appropriation instruments. For example, Apple regularly defends and litigates its iconic product design against its rival Samsung. As product design is widely visible, and hence informal appropriation is difficult to enforce, it is more connected to formal appropriation methods (e.g., registered design, design patents).

We find that among other firm level characteristics, size stands out as a robust determinant of formal appropriation strategies. The use of formal appropriation instruments is connected with rather high fix costs (e.g., setting up an IPR department). For informal appropriation measures, we do not find this relationship as all these instruments usually comprise less fix costs (e.g., secrecy might be more difficult to implement in large companies). Additionally, we find that sectoral conditions determine both appropriation strategies robustly, which has been confirmed in previous studies {Cohen 2000 #57}.

Informal instruments are less intensively used in a competitive environment, where product and technology characteristics play an important role because here formal instruments (e.g., patents and trademarks) may be more effective.

In general, the OLS regression confirms the variables of the BMA except for cooperation depth. In addition, the use of formal instruments is positively explained by competitive strategies focusing on product variety (e.g., leveraging brands), but negatively by competitive strategies focusing on quality (less importance of signaling) and services (difficult to use formal instruments). In addition, the interaction between coopetition and cooperation depth is negatively related to formal instruments, which is difficult to explain. Finally, R&D intensity is a robust predictor of firms’ use of formal instruments.

In contrast to the variables identified by BMA, the additional explaining factors by the OLS regression reveal, with a few exceptions (e.g., R&D intensity), rather puzzling results especially regarding informal appropriation instruments. However, these puzzles, including the large number of insignificant factors, can be the starting point for future research. We encourage continuing the research on the general influence of the competitive environment on the use of formal and informal appropriation instruments. For example, the competition in the information and communication technology (ICT) markets is dominated by patents. However, insights into further industries and sectors, with the exception of the patent dominated pharmaceutical industry, are missing. Furthermore, the various dimensions of competition deserve in-depth research as the current approach does only reveal few robust determinants. Finally, the relevance of company size for the use of informal instruments promises further interesting research questions, especially since non-linear relationships might play an important role.

Prior research suggests that management of the new strategic option of coopetition remains unclear. In this paper, we contribute to resolve this issue by linking a firm’s coopetition, cooperation and competition strategies to its usage of IP appropriation instruments. By doing so, we hope managers will gain a better understanding of the impact of exercising these different strategies. Thus, we expect managers to develop better knowledge and expertise of when and how the use of either a coopetition, cooperation or competition strategy is an appropriate measure to capture value from an innovation. Our analysis provides two main insights. First, innovating companies must assess the benefits and drawbacks of the different organizational strategies early on, and then decide for the most effective appropriation strategy for the given context. Second, the innovating company must decide on appropriation strategies that are well aligned with its choice of cooperation and corporate competition strategies and goals as companies can only capture value if they understand the importance of IP when commercializing an innovation. In sum, appropriation strategies should provide enough incentives to perform R&D and innovation activities in the first place. Thus, innovators should be able to appropriate sufficient rents from the innovation in compensation for their initial investments {Somaya 2011 #286}. Consistent with {Somaya 2011 #286/persononly/nopar} {Somaya 2011 #286/yearonly}, we emphasize the importance of an IP appropriation strategy as an essential part of the firm's business strategy and not just as an ‘afterthought’.

Moreover, the results have implications for using patents and other formal appropriation instruments to measure innovative activities. As firms in the sample seem to rely on both formal and informal appropriation methods, we argue consistent with {Neuhäusler #145/persononly/nopar} {Neuhäusler #145/yearonly} that relying on formal appropriation instruments exclusively could lead to a truncated picture when measuring the innovation activities and outcomes of companies. Particularly, formal appropriation mechanisms may only reflect and report an incomplete, and hence underestimated picture of these activities.

Our contribution has some limitations. The number of robust variables identified by the BMA explaining the factor scores related to the use of formal appropriation instruments is quite limited. Furthermore, we certainly face an endogeneity problem because appropriation strategies, especially large patent portfolios, might influence the competitive environment, as we currently see in the markets for smart phones. In this context, the life cycle of industries and products have to be taken into account, i.e., this might also change the use and relevance of appropriation strategies over time, e.g., first starting with patents and then using more informal instruments or the other way round. The limited significance of the competitive environment raises questions about the survey approach, which is based on subjective assessments of innovation managers about their competitive environment. In addition to the sector dummies, concentration indices might be used for further robustness checks. Finally, the company as a unit of observation might also be challenged because both the competitive environment and also the use and relevance of formal and informal appropriation instruments may differ from product to product, i.e., a company enjoying a monopoly position supported by strong patent portfolios in one market can generate significant profits, whereas in another market segment its products are challenged by numerous competitors, which might make patenting rather ineffective and superfluous.

Acknowledgements

The authors thank the ZEW Mannheim for sharing the data. The usual disclaimer applies.

References

FOOTNOTES