Abstract
This study underlines the limitations of commonly used proxies to measure value creation in interfirm alliances and addresses these limitations in two ways. First, this study adopts a co-opetition-based approach in theoretically conceptualizing value creation in interfirm alliances as a three-dimensional construct and argues that in addition to “common benefit” and “private benefit cooperation” (generally known as “private benefits”), a third dimension, namely “private benefit competition” should also be considered as an integral dimension of value creation. Second, by analyzing data collected from 155 firms of five high-technology research-intensive sectors in India that engaged in 288 alliances characterized by varying degree of co-opetition, this study empirically validates the distinctiveness of these three dimensions and presents a 17-item multidimensional scale of value creation.
A central concern in alliance research and practice is value creation (Bowman & Ambrosini, 2000; Lepak, Smith, & Taylor, 2007), which refers to the total sum of value that is created during alliance activities (Ritala & Hurmelinna-Laukkanen, 2009). With the increasing importance of interfirm alliances as a strategic instrument to achieve and maintain sustained competitive advantage, value creation in interfirm alliances has lately received considerable scholarly attention (Olk, 2006). Much of the attention, however, has been confined to the value created in interfirm alliances through collaborative behavior (Dyer & Singh, 1998; Khanna, Gulati, & Nohria, 1998; Madhok & Tallman, 1998). In contrast, the value created in interfirm alliances as a result of competition among the alliance partners has been viewed merely as opportunistic behavior, and it is argued that such behavior in the long run threatens the sustainability of the alliance (Hamel, 1991; Lavie, 2006). However, in reality, interfirm alliances are seldom purely collaborative in nature, and in fact they are characterized by the existence of simultaneous cooperation and competition (Bengtsson & Kock, 2000; Brandenburger & Nalebuff, 1996). Such interfirm alliances where cooperation and competition are present simultaneously are known as co-opetitive alliances (Bengtsson & Kock, 2000; Brandenburger & Nalebuff, 1996). 1 Evidence suggests that due to rapid advancements in technology and uncertainties in the business environment, a majority of the interfirm alliances today occur between competitors within the same industry and these alliances are largely co-opetitive (Han, Oh, Im, Oh, Pinsonneault, & Chang, 2012; Harbison & Pekar, 1998).
The increasing importance of co-opetitive alliances underlines the need to adopt the co-opetition-based approach for developing value creation measures. Extant literature has largely drawn from the collaborative paradigm to measure value creation and has neglected the value created as a result of competitive behavior. In the process, the strategic interdependence of alliance partners emphasized in the co-opetition paradigm has been overlooked (Dagnino & Padula, 2002). Another limitation is that although several studies (e.g., Dyer & Singh, 1998; Khanna et al., 1998; Madhok & Tallman, 1998) have theoretically conceptualized value creation as a multidimensional construct, so far it has not been operationalized as such. Furthermore, since the co-opetition-based approach in measuring value creation in interfirm alliances has been neglected, the dimension of value creation related to competitive behavior has also not yet been clearly delineated and operationalized (Pateli & Lioukas, 2012). Consequently, research in this stream has not been able to provide practical tools for managerial decision making.
Several approaches have been adopted to measure value creation in interfirm alliances: (a) financial measures and accounting indices indicating firms’ performance such as firms’ profitability, net income, return on investment or return on assets, and stock and market gains (Anand & Khanna, 2000; Kale, Dyer, & Singh, 2002; Pateli & Lioukas, 2012); (b) measures of alliance performance and success in terms of alliance duration, stability, termination, and assessment of partners’ ongoing relationship and fit (Pateli & Lioukas, 2012); (c) subjective measures such as managers’ satisfaction with alliance performance and the fulfillment of strategic goals (Anand & Khanna, 2000; Draulans, deMan, & Volberda, 2003; Mohr & Spekman, 1994; Parkhe, 1993; Simonin, 1997; Zollo, Reuer, & Singh, 2002); and (d) innovative performance measured by firms’ patenting activity, R&D expenditure, licenses, and so on (Ahuja, 2000; Hagedoorn & Schakenraad, 1994; Kim & Inkpen, 2005; Rothaermel & Deeds, 2006).
However, all these approaches fail to consider the increasingly important context of co-opetitive alliances and therefore suffer from several limitations. First, previous empirical research has largely concentrated on value created from cooperative behavior. Very little attention has been paid to value created from competitive behavior between the alliance partners, such as better differentiation of jointly developed products; these kinds of value are a principal feature of co-opetitive alliances. Second, these approaches have measured value creation as a unidimensional construct and neglected its theorization as a multidimensional construct comprising both private benefits and common benefits (Dyer & Singh, 1998; Lavie, 2006; Madhok & Tallman, 1998). Common benefits are those benefits “that are derived from the shared resources of alliance partners” (Lavie, 2006: 647), whereas private benefits are those benefits that are derived from a firm’s own resources by deploying knowledge and skills learned from the alliance activities (Khanna et al., 1998). In particular, measuring value creation as a unidimensional construct and ignoring private benefits and common benefits as its distinct dimensions neglects different forms of social capital engendered in interfirm alliances (Nahapiet & Ghoshal, 1998) even when researchers have identified social capital as existing both at public and private levels (Kemper, Schilke, & Brettel, 2013). Thus, viewing social capital generated in an interfirm alliance from a co-opetition-based approach allows us to understand value creation in interfirm alliances by simultaneous existence of cooperative and competitive behavior. Third, approaches employing firm performance measures ignore several intangible facets of value creation emerging from social capital dynamics (Kemper et al., 2013; Nahapiet & Ghoshal, 1998), such as enhancement of capabilities, improvement in organizational effectiveness and reputation, acquisition of new knowledge and skills, and so on (Brush, Bromiley, & Hendrickx, 2000; Müller, 2010). Last, these measures do not capture the value created from individual alliances; instead, they measure only the cumulative value created from all of the alliances. Moreover, these measures also include value created from other activities of the firm outside of such alliances, rendering them quite unreliable.
A more general concern regarding existing measures of value creation is their overlap with items measuring a conceptually distinct construct—value appropriation. Value appropriation is “the individual share of the value (created in an alliance) that a firm can capture” (Ritala & Hurmelinna-Laukkanen, 2009: 821). While operationalizations such as firm performance measures and stock market reactions claim to measure value creation, they actually end up measuring a different construct—value appropriation (Pateli & Lioukas, 2012). Generally, researchers have employed firm performance indicators to measure value creation and perception-based scales to measure value appropriation (Pateli & Lioukas, 2012). Consequently, ambiguities arise between the conceptualization and measurement of these two constructs. While value appropriation conceptually relates to firm performance indicators (Gulati & Wang, 2003), these indicators have been actually used to measure value creation (e.g., Kale et al., 2002). At the same time, some scholars argue that reliable indicators of firm performance are not available, and therefore there is a need to use perception-based scales to measure value appropriation (Gulati & Wang, 2003; Pateli & Lioukas, 2012). While issues of reliability arise in using firm performance indicators to measure value appropriation (Gulati & Wang, 2003; Pateli & Lioukas, 2012), a more serious issue arises when these indicators are used to measure value creation—the issue of validity. To address the issue of validity, it is necessary to develop a perception-based scale for value creation.
Thus, there is a lack of valid and reliable measures of value creation in interfirm alliances and this leads to the problem of comparability across different measures (Olk, 2002). This problem prevents researchers from drawing on results of previous studies to build a comprehensive theory (Yan & Zeng, 1999). Thus, research findings are highly static and idiosyncratic (Yan & Zeng, 1999). Also, they fail to take into account the co-opetition-based approach for measuring value creation, thus neglecting important features of interfirm alliances. This is a significant gap as over 50% of interfirm alliances occur between firms within the same industry or among competitors (Harbison & Pekar, 1998). I attempt to fill this gap by building a managerial perception-based scale for measuring value creation in interfirm alliance by adopting a co-opetition-based approach. The need for a perception-based scale stems from the inability of commonly used firm-performance-based indicators of value creation in measuring social-capital-based benefits like positive reputational effects, a broadened social network, and new knowledge (Barringer & Harrison, 2000; Brush et al., 2000; Hagedoorn, 1993). The advantage of a perception-based scale is that it takes us “beyond the black box approach” that characterizes the use of financial indicators (Venkatraman & Ramanujam, 1986: 804). Also, perception-based indicators provide a broader view of alliance effectiveness (Venkatraman & Ramanujam, 1986). The perception-based scale addresses many of the limitations of earlier approaches; in particular its main advantages are that (a) it is applicable for value creation in all interfirm alliance contexts, (b) it measures each dimension of value creation distinctly and helps in attributing value creation to both cooperative and competitive behavior, (c) it distinguishes measures of value creation from measures of value appropriation, and (d) it helps in measuring the value created from each alliance and differentiating it from the overall firm performance.
Measurement of Value Creation in Interfirm Alliances: A Review and Assessment
Several attempts have been made to measure value creation, but they have a few limitations. One of the methods used to measure value creation is to assess stock market reactions to alliance announcements (Anand & Khanna, 2000; Kale et al., 2002; Lavie, 2007). The limitations of this approach are that (a) it is concerned more with the expected future value rather than being an indicator of the actual value created in an alliance (Lane & Jacobson, 1995); (b) it is one of the indicators of overall firm performance and does not validly measure value created in a specific alliance (Müller, 2010); (c) since it is a measure of firm performance, it is actually measuring value appropriation and not value creation; (d) it is unable to distinguish between common benefits and private benefits; and (e) its applicability is limited by stock market reactions to key events such as alliance formation or alliance termination, and thus it cannot be used on a continuous basis to measure value creation during periods of stable functioning of alliance.
An approach that overcomes some of the limitations of use of stock market reactions as a measure of value creation is the use of financial indicators and accounting indices as proxies to measure value creation. For example, in some studies, return on investment (ROI), return on assets (ROA), return on expenditure (ROE), net income, and higher share prices have been used to measure value creation (e.g., Anand & Khanna, 2000; Combs & Ketchen, 1999; Gulati & Wang, 2003; Kale et al., 2002). To measure value creation at the alliance level, the indices of the partners are separately calculated and then aggregated (Pateli & Lioukas, 2012). Though this approach overcomes the limitation of being restricted to key events such as alliance formation and alliance termination, it also has several limitations: (a) it is unable to overcome the shortcoming of measuring value created from each alliance separately (Müller, 2010); (b) it is somewhat distant from alliance activities because of potential intervening and confounding firm-level factors, which may account for insignificant direct effects of alliances on firm profitability (Berg, Duncan, & Friedman, 1982; Hagedoorn, 1993; Lavie, 2007); (c) it is unable to account for long-term strategic implications beyond immediate financial results (Doz & Hamel, 1998); and (d) it is unable to distinguish between common benefits and private benefits (Dyer, Singh, & Kale, 2008).
Having realized that there is a need to also take into account intangible assets that alliance partners acquire from alliance activities (e.g., Brush et al., 2000; Hagedoorn, 1993) and their long-term strategic implications (Hagedoorn, 1993), besides measuring tangible benefits, some researchers have developed other measures of value creation, such as growth in sales, profitability, market share, market performance, and others (Brush et al., 2000; Delaney & Huselid, 1996; Hagedoorn, 1993; Simonin, 1997). For instance, Lunnan and Haugland’s (2008) measures of alliance performance include indicators such as the alliance’s duration, termination, and stability. Researchers have also measured value creation in terms of strategic, operational, and cultural compatibility (Futrell, Slugay, & Stephens, 2001), trust, conflict resolution, and opportunistic behavior (Olk, 2006). An important shortcoming of these measures is that they fail to distinguish between the indicators of value creation and its determinants, thereby creating problems of discriminant validity. Moreover, these measures are unable to measure common benefits and private benefits distinctly. Ariño (2003) attempted to overcome this limitation by developing two different scales—one for measuring satisfaction with overall performance and the other for measuring net spillover effects. However, the scale for measuring satisfaction for overall performance does not distinguish between common goals and private goals. Similarly, the scale for measuring net spillover effects does not distinguish between values created as a result of cooperative behavior and of competitive behavior.
Another approach to measure value creation is to use the Expected Alliance Value scale (Pateli & Giaglis, 2007). This scale consists of seven dimensions with 20 items focusing on alliance formation objectives (Contractor & Lorange, 1988; Hagedoorn, 1993; Vilkamo & Keil, 2003). Geringer and Hébert (1991) have developed another variation of this approach by evolving measures to assess overall performance of an alliance in terms of differences between the initial expectation and the actual performance. In practice, value may be created in interfirm alliances due to dynamic activities, which go beyond initial expectations. These dynamic aspects of value creation are not accounted for by these measurement approaches, as they remain anchored around initial expectation of value to be derived from the alliance. To account for dynamic alliance activities such as innovation, some researchers (e.g., Hagedoorn & Cloodt, 2003; Hagedoorn & Schakenraad, 1994; Kim & Inkpen, 2005; Rothaermel & Deeds, 2006) have measured value creation by using indicators such as the firm’s patenting activity, R&D expenditure, new products/services, licenses, and so on. Alegre and Chiva (2008) developed a scale to measure innovation in terms of a firm’s innovation in comparison with its competitors. Although this scale is a comprehensive measure of value creation for measuring innovative performance, its use may contribute to conflicting and misleading findings because firms differ in their orientation toward protection of their intellectual property. Thus, their patenting activity may be a misleading measure of value creation. Furthermore, these measures are unable to separate the innovation performance of a firm, which is purely on account of a firm’s internal resources and knowledge from performance on account of alliance-related activities.
An important aspect of scientific research is valid and reliable translation of different facets of theoretical constructs into empirically measureable variables (DeVellis, 2003; Netemeyer, Bearden, & Sharma, 2003). Thus, in the context of value creation in interfirm alliances, it is necessary to develop valid and reliable measures for both private benefits and common benefits. Moreover, since researchers have identified different antecedents for private benefits and for common benefits, separate measures are required for these constructs to empirically test and validate these linkages (Crook, Shook, Madden, & Morris, 2010). Separate measures for common and private benefits will also allow application of rigorous methodological procedures to build and test theory in alliance studies, and thus contribute to the legitimacy and advancement of this field (Slavec & Drnovsek, 2012).
Table 1 provides an illustration of different proxies that have been used to capture value creation in earlier studies.
Approaches to Measure Value Creation in Key Empirical Studies on Interfirm Alliances
Note: JV = joint venture; ROA = return on assets; ROE = return on expenditure; ROI = return on investment.
Co-opetition-Based Approach to Value Creation
Value creation as a concept has two dimensions—content and process (Bowman & Ambrosini, 2000; Lepak et al., 2007). Madhok and Tallman (1998: 328) define value as the net rent-earning capacity of tangible or intangible assets or resources. They further explain that in the context of interfirm alliances “such value can be conceptualized in terms of the ability of the partners to earn rents over and above what could have been achieved in the absence of the partnership, i.e., in alternative organizational arrangements.” Generally, value created in interfirm alliances is multidimensional and can be tangible as well as intangible. It includes common benefits and private benefits (Dyer & Singh, 1998).
Khanna et al. (1998) suggest that common benefits are created by the joint efforts of alliance partners within the alliance boundary. Dyer and Singh (1998) term common benefits as relational rents, which are realized from relation-specific assets, knowledge-sharing routines, complementary resources, and effective governance mechanisms. For the purpose of this study, I adopt the definition of common benefit given by Dyer and Singh (1998: 662), who define it as “a supernormal profit jointly generated in an exchange relationship that cannot be generated by either firm in isolation and can only be created through the joint contributions of the specific alliance partners.” Common benefits can be realized only from resources that are intentionally committed and jointly possessed by the alliance partners (Dyer & Singh, 1998; Khanna et al., 1998; Lavie, 2006; Madhok & Tallman, 1998). An example of the common benefit component of value creation in interfirm alliances can be understood by considering the co-opetitive joint venture (JV) between Samsung Electronics and Sony Corporation. This JV (S-LCD) was established in 2004 to develop liquid crystal display (LCD) panels for flat-screen TV sets. It created greater value for both alliance partners by winning the standardization battle between LCD and Matsushita’s plasma display panel (PDP) technology (Gnyawali & Park, 2011). Prior to the development of the LCD technology, PDP was the leading technology in the flat TV market segment. While LCD segment’s market share was 13% in the third quarter of 2005, it increased to 68.4% in the third quarter of 2009 (Gnyawali & Park, 2011). In the corresponding period, the market share of PDP segment increased from only 3.6% to 6.5% (Gnyawali & Park, 2011). Furthermore, LCD technology increased the market potential of flat TV market: The combined market share of Samsung and Sony was 18.4% in the third quarter of 2004. However, during the next 4 years, it increased to 40.9% (Gnyawali & Park, 2011).
While common benefits refer to mutual gains realized jointly by the alliance partners, private benefits refer to gains realized by each alliance partner individually. Khanna et al. (1998) define private benefits as values that a focal firm can earn unilaterally by acquiring skills and knowledge from its alliance partner and applying them to activities unrelated to the alliance activities and outside its boundary. These benefits result from cooperative behavior of the alliance partners (Dyer & Singh, 1998). In collaborative alliances, private benefits outside the alliance boundary are seen as positive spillovers (Lavie, 2006; Madhok & Tallman, 1998). Adopting a co-opetition-based approach to interfirm alliances, I term the private benefits that accrue to a firm as a result of cooperative behavior among the alliance partners as “private benefit cooperation.” The co-opetitive JV between Samsung and Sony provides an illustration of private benefit cooperation. In this JV, both Samsung and Sony created private benefit cooperation by utilizing the new knowledge developed from alliance-related activities in activities unrelated to the alliance and outside the alliance boundary. Samsung and Sony independently developed new technologies and products, such as LED TV and 3D TV (Gnyawali & Park, 2011). They also exploited this knowledge for independently developing organic LED TVs (Gnyawali & Park, 2011).
Alliance partners also generate private benefits by competing against each other. In the context of purely collaborative alliances, this behavior is seen as opportunistic behavior and such private benefits are viewed as negative spillovers (Hamel, 1991; Lavie, 2006; Porter & Stern, 2001). However, the co-opetition-based approach does not view these private benefits as negative spillovers. According to this approach, alliance partners willingly compete against each other and create value from the alliance-related activities that take place beyond the alliance boundaries (Burt, 1991; Kumar, 2010). Consequently, this approach recognizes that private benefits are created from both cooperative and competitive behaviors (Gnyawali, He, & Madhavan, 2006). Thus, adopting a co-opetition-based approach, there is a need to account systematically for these two distinct private benefits (Burt, 1991).
To differentiate between the above two types of private benefits in co-opetitive alliances, I term the private value created as result of competitive behavior as “private benefit competition” and draw from Burt (1991) and Kumar (2010) to define it as a value created by the focal firm by competing against its alliance partner to outperform it in activities, which are directly linked to the alliance-related activities but take place beyond the alliance boundary. This kind of benefit can be exemplified by considering the competitive behaviors of Sony and Samsung in the S-LCD JV. Sony and Samsung jointly developed the LCD panel technology for flat-screen TVs. The scope of this alliance was limited to the development of this technology. Later, by using this technology, Sony launched its Bravia series within one year and became the market leader in the LCD TV segment (Gnyawali & Park, 2011). However, Samsung launched its own Bordeaux series and soon overtook Sony in the LCD TV segment (Gnyawali & Park, 2011). Thus, due to the S-LCD JV, Sony could withstand the changes taking place in the flat-screen technology space and became one of the market leaders by combining the new technology with its already existing brand name and was able to overtake Sharp, its competitor, in terms of market share (Gnyawali & Park, 2011). On the other hand, Samsung created even greater value from the S-LCD JV by combining its better production capacity with the LCD technology. It not only acquired leadership in the large LCD panel segment but also created an advantage in establishing its leadership in technical standards (Gnyawali & Park, 2011). Since 2006, Samsung and Sony have been ranked first and second in the total TV market as well as in the LCD segment, respectively (Gnyawali & Park, 2011). According to the co-opetition-based approach, competition outside the alliance boundary in alliance-related activities is an important source of value (Gnyawali, He, & Madhavan, 2008). Such competition does not negatively affect the alliance and is part of the value creation process (Bengtsson & Kock, 2000; Brandenburger & Nalebuff, 1996). For instance, leveraging their respective competencies, both Sony and Samsung exploited each other’s complementary resources and capabilities and gained substantially from this alliance (Gnyawali & Park, 2011).
In summary, according to the co-opetition-based approach, value creation results from three types of activities: (a) “common benefits” jointly created by the alliance partners from the resources intentionally committed to the alliance, (b) “private benefit cooperation” unilaterally created by the focal firm as a result of cooperation between the alliance partners, and (c) “private benefit competition” unilaterally created by the focal firm as a result of competition against the alliance partner. While common benefits are created within the alliance boundary, the other two types of private benefits are created beyond the alliance boundary. The alliance between Sony and Samsung exemplifies these three components of value. The alliance boundary (S-LCD JV) was limited to the development of flat-screen LCD panel technology, and it did not extend to other activities, such as marketing, production, branding, and distribution. Thus, the common benefit that emerged from this alliance was the development of flat-screen LCD technology. The private benefit cooperation for Sony and Samsung was that they were able to use their learning from the alliance-related activities to build technologies in activities unrelated to the alliance such as 3D TV, LED TV, and organic LED TVs. Furthermore, Sony generated private benefit competition by combining the jointly created LCD technology with its brand value and reputation for high quality to compete against Samsung in the same technology space related to alliance activities but beyond the alliance boundary. It occupied a leading market position through such competitive behavior. Similarly, Samsung gained private benefit competition by leveraging its production capacity and technological standards in using the jointly created technology to compete with Sony in alliance-related activities beyond the alliance boundary.
Scale Development
The three dimensions of value creation in interfirm alliances have not yet been clearly identified and delineated. Value created in co-opetitive contexts has thus been ignored in existing measures. Hence, there is a need to develop psychometrically valid measures of value creation using a co-opetition-based approach. To this end, I employed a mixed-method approach using both qualitative and quantitative techniques for scale development (Hurmerinta-Peltomäki & Nummela, 2006; Jick, 1979; Reichardt & Rallis, 1994). I adopted a six-stage research design: (a) specification of construct definition and content domain (Churchill, 1979; Netemeyer et al., 2003), (b) item generation and assessment of face or content validity (Churchill, 1979; Hinkin, 1995; Netemeyer et al., 2003), (c) questionnaire design (Churchill, 1979; DeVellis, 2003), (d) questionnaire administration and data collection and purification (Hair, Anderson, Tatham, & Black, 1998), (e) scale construction and purification using exploratory factor analysis (EFA; Churchill, 1979; DeVellis, 2003; Hair et al., 1998; Netemeyer et al., 2003), and (f) scale validation using confirmatory factor analysis (CFA; Churchill, 1979; DeVellis, 2003; Hair et al., 1998; Netemeyer et al., 2003).
Using this research design, I conducted five studies to develop the value creation scale. Study 1 reports the process by which items with high content validity were developed. Study 2 assesses the content validity of the initial pool of items generated in Study 1 and describes the initial item reduction. In Study 3, the revised scale based on the results of Study 2 was administered to alliance managers and analyzed using EFA for scale construction and purification along with examination of scale’s dimensionality and reliability. In Study 4, based on the results of Study 3, the survey was revised and again administered to a new set of alliance managers, and convergent and discriminant validities of the scale were assessed using CFA. Finally, in Study 5, post hoc analysis was conducted to assess the validity of the scale by (a) developing an abbreviated version of the scale, (b) conducting tetrad analysis to confirm the reflective specification, and (c) examining the predictive validity.
Study 1: Specification of Construct Definition, Content Domain, and Item Generation
Method
Following the procedure outlined by Churchill (1979) and Netemeyer et al. (2003), I conducted a detailed literature review to define value creation and outline its content domain in the context of co-opetitive alliances. Next, as recommended in the literature, I adopted a deductive approach to develop the initial set of items (Churchill, 1979; DeVellis, 2003; Molloy, Chadwik, Ployhart, & Golden, 2011; Netemeyer et al., 2003). A review of relevant and related literature was done by screening all articles published in the top 10 management journals from 1990 to 2011 to identify the body of knowledge in which the construct of value creation in interfirm alliances was situated. These journals were chosen based on the criteria outlined by Tahai and Meyer (1999). The time frame of 1990 to 2011 was chosen as the value creation in interfirm alliance research stream was introduced in a big way in the mid-1990s (e.g., Brandenburger & Nalebuff, 1996; Gulati, 1995). To ensure that important studies were not excluded, I also searched databases, such as Science Direct, JSTOR, Business Source Premier, and SSRN, for studies on the topic of value creation in interfirm alliances.
Results and Discussion
The process of specifying the definition and outlining the content domain is discussed above in the Co-opetition-Based Approach to Value Creation section. A review of the alliance literature reveals that value creation outcomes are commonly sourced from literature on alliance performance, alliance success, alliance effectiveness, common benefits, and private benefits (e.g., Dyer & Singh, 1998; Lavie, 2006; Madhok & Tallman, 1998; Pateli & Lioukas, 2012). I entered the above value-creation-related expressions as subject themes to search the databases (Feldvari, 2000; Knapp, 2000). Altogether, I shortlisted 372 studies, which were largely conceptual, case-based, or survey-based studies. Among them, 249 studies did not include any measurement instrument, while 123 studies made use of measurement instruments. Out of these 123 studies that had employed a measurement instrument, 76 studies did not have relevance pertaining to value creation in interfirm alliances. Thus, out of 372 studies, 325 studies were not useful for item generation as no item could be sourced from them. The remaining 47 studies were sources from which items could be drawn from definitions, qualitative quotes of respondents, direct scale items, or important explanations of value creation in interfirm alliances.
From these 47 studies, existing measures of value creation were examined, and where possible items from these scales were included in the initial item pool. Many of the existing items were modified or rewritten to ensure face validity and to establish consistency in tone and perspective across all the items in the initial item pool. However, proxies, such as ROI, ROE, ROA, R&D spending, and so on, were not included in the initial pool of items due to the inherent limitations associated with these measures, as discussed earlier. An initial pool of 38 items was developed to obtain as many different indicators of value creation that measure its three dimensions (Churchill, 1979). Out of these, 18 items described the common benefit dimension, 12 items described the private benefit cooperation dimension, and 8 described the private benefit competition dimension.
Study 2: Professional Review and Assessment of Content Validity
Method
Two pretests were conducted to improve and assess the content validity of the initial pool of items generated in Study 1. Pretest 1 was conducted to enhance the accuracy, comprehensiveness, and the quality of 38 items. I requested four practitioners who specialize in alliance management and four academics who have an interest in alliance research to review the initial pool of 38 items and opine about the appropriateness of these items for various dimensions of value creation (Netemeyer et al., 2003). These experts reviewed the items and classified them into three groups: (a) items that could be retained without change, (b) items that needed to be modified, and (c) items that needed to be deleted. Furthermore, they also suggested items that were not a part of the initial item pool but, according to them, captured the notion of value creation in the co-opetition-based context.
Next, Pretest 2 was conducted to assess the content validity of the items generated as a result of Pretest 1. Pretest 2 was conducted with a different set of four experts—two academics and two professionals. To assess the interrater agreement on content validity quantitatively, following Rungtusanatham, Anderson, and Dooley (1999), experts were requested to complete two tasks (see Figure 1). In Task 1, experts were provided with definitions of the three dimensions of value creation and were asked to assign the items uniquely to one of the three dimensions. The items were presented to them in random order. The percentage of experts who assigned an item to a particular dimension and Cohen’s (1960) kappa (κ) were computed. 2
Sample Score Sheet for Testing the Content/Face Validity
In Task 2, a variant of Zaichkowsky’s (1985) method was used where experts were asked to opine about the adequacy of each item by rating the degree to which an item adequately measured the construct on a 7-point Likert-type scale. A rating of 1 indicated very low adequacy, and a rating of 7 indicated very high adequacy. The average adequacy score for each item was computed as the mean of the scores provided by the raters. The standard deviation of the adequacy score was also calculated. The criteria for retaining items were set at the average adequacy score being greater than 4 and standard deviation being less than 1.
Results and Discussion
The review by experts in Pretest 1 resulted in the deletion of 11 items, the modification of 8 items, and the inclusion of 7 new items. This resulted in the number of items being reduced to 34 (14 for common benefit, 9 for private benefit cooperation, and 11 for private benefit competition). The list of 34 items resulting from Pretest 1 is provided in Table 2.
Item Pool and Sources of Value Creation After Pretest 1
Note: CB = common benefit; PB = private benefit.
Researchers propose a cutoff criterion ranging from 60% to 75% as a minimum index of agreement between the experts for retaining an item in the desired dimension (Hardesty & Bearden, 2004). I took the mean of this range, 67.5%, as the minimum criterion for an item to be representative of the underlying construct. The results of Task 1 of Pretest 2 indicated that eight items (CB1, 25.0%; CB2, 50.0%; CB3, 50.0%; CB4, 50.0%; CB6, 25%; PBCoop8, 50.0%; PBComp4, 50.0%; and PBComp9, 50.0%) failed to meet the criterion of more than 67.5% of experts assigning them to a particular dimension and were therefore discarded. These eight items were then excluded from the calculation of Cohen’s κ, which was .86, suggesting good interrater agreement. The standard deviation for Cohen’s κ, σκ, was .07, yielding a 95% confidence interval for κ of .75, .98. I also tested whether Cohen’s κ was significantly different from 0 (Rungtusanatham, 1998). This was necessary to ensure that the interrater agreement was not merely a matter of chance. The content validity of the items was established through this process.
The results of Task 2 of Pretest 2 indicated that 18 of the 26 selected items satisfied the criterion for retention (>4.00) with an acceptable standard deviation (≤1.00). Although standard deviations for eight items (CB9, CB11, CB13, CB14, PBCoop5, PBCoop6, PBComp6, and PBComp7) were marginally above the acceptable limit (≤1.00), I retained these items as they had good average adequacy scores between 4.50 and 6.00. Thus, in total, eight items (CB1, CB2, CB3, CB4, CB6, PBCoop8, PBComp4, and PBComp9) were removed during the Pretest 2 stage.
Figure 2 summarizes the steps undertaken in the pretesting phase.
Different Steps of Pretesting
Study 3: Scale Purification Along With Examination of Scale’s Dimensionality and Reliability
Method
Study 3 involved three steps: (a) designing of questionnaire, (b) administration of questionnaire and data collection, and (c) scale purification and assessment of dimensionality and reliability.
Questionnaire designing
A total of 26 items retained at the end of Study 2 were used to design the questionnaire. All the constructs were measured at the dyadic level by multiple items. There were 9 items for common benefit, 8 items for private benefit cooperation, and 9 items for private benefit competition. Following DeVellis’s (2003) suggestion for measuring opinions, beliefs, and attitudes, 7-point Likert-type scales were used to capture the extent of informant’s agreement, ranging from very strongly disagree to very strongly agree, regarding various facets of value creation in interfirm alliances. The 7-point scale anchor was considered appropriate since adding more response options might lead to respondents not being able to distinguish between the response points in the scale, thus reducing the validity of the response (Clark & Watson, 1995). On the other hand, reducing response options might lead to the informants giving a neutral response by choosing the midpoint of the scale (Prendergast & Huang, 2003).
I then invited eight senior alliance managers from four different firms—four from two different firms in the pharmaceutical sector, two from a firm in the power and energy sector, and two from a firm in the information technology sector—to pretest the refined questionnaire to assess the extent of the readability of the representative measurement items. None of them had participated in the professional review. I also conducted a face-to-face interview and/or discussion over the telephone to know their extent of understanding of the survey items and their suggestions for modifications of the survey items. A few modifications were made to improve the readability, format, and layout of the survey.
Questionnaire administration and data collection
This refined survey was administered to alliance managers of private and publicly listed firms in five high-technology research-intensive sectors: (a) information technology, (b) pharmaceuticals, (c) telecommunication, (d) power and energy, and (e) steel. I chose these sectors because co-opetitive alliances frequently occur in these sectors (Garrette, Castañer, & Dussauge, 2009; Gnyawali et al., 2006; Ketchen, Snow, & Hoover, 2004; Luo, 2007). Also, interfirm alliances in these sectors are formed for a variety of purposes, such as exploration, exchange of know-how/technology, R&D, production, marketing, sales, distribution, and so on (Lambe, Spekman, & Hunt, 2002), which provides more generalizability to the results. Following Das and Teng’s (2002) typology of interfirm alliances, the informants were also asked to categorize the form of their co-opetitive relationships as (a) equity JVs, (b) minority equity alliances, (c) bilateral contract-based alliances, and (d) unilateral contract-based alliances.
The Prowess Release 3.1 database from the Centre for Monitoring Indian Economy, which is being increasingly used by strategy researchers (e.g., Chacar & Vissa, 2005; Khanna & Palepu, 2000) for large-sample studies in India, was used to identify the firms in the five targeted sectors. Dyadic alliances were used as the unit of analysis as it allowed better capturing of the inherent tensions and complexities in the co-opetitive relationship (e.g., Bengtsson, Eriksson, & Wincent, 2010; Bengtsson & Kock, 2000).
Due to the nonavailability of a database for interfirm alliances and alliance managers (Lambe et al., 2002), it was difficult to construct a universe of alliance managers and draw a random sample from it. Therefore, I adopted the procedure recommended by Lambe et al. (2002) for data collection: I (a) identified a sample of managers, including CEOs, who were likely to participate in the survey and (b) prescreened them for alliance responsibilities such as experiences in initiating, negotiating, managing, or disengaging from alliances; that is, a key informant approach was used to collect data (Campbell, 1955; Philips, 1981). I used a random sample of these business executives as a seed sample. These business executives were further asked to provide names of three to five colleagues who also had alliance responsibilities and experience. Thus, a snowball method was used for identifying the informants. This approach enabled me to access respondents who I would have otherwise not been able to locate.
The data were collected from December 2011 to February 2012 using two methods—a web-enabled survey and e-mail. To situate the responses to particular alliances, the survey participants were guided to keep a particular dyadic relationship in mind while completing the survey. Furthermore, to increase the response rate, I requested the contact point in each queried firm to share the names and contact details of the executives who were asked to complete the survey. I called each executive to solicit her cooperation and participation in the study. I explained the purpose of the study to all those executives who could be reached and answered their follow-up questions. After each meeting or call, I also e-mailed a cover letter to each executive. Executives who were not eligible as appropriate informants forwarded the request to their colleagues who had experience of alliance activities. Two weeks after the first e-mail, a reminder e-mail was sent to nonrespondents requesting their cooperation. Phone calls were also made to solicit the cooperation of nonrespondents. Though this involved time and effort, it led to improvement in the response rate and the seriousness with which the informants responded to the survey. I promised to share an executive summary of the final results with the respondents and assured them of complete confidentiality.
A total of 488 surveys were released. Finally, 172 responses from 64 firms were received, with 156 responses found to be usable, yielding a response rate of approximately 32%. Out of these 156 responses, 18 were received by post, 36 via the web-based link (although 76 respondents tried to complete the survey through the web-based link, only 36 completed it), and 102 through soft copies on e-mail. Out of the 16 unusable questionnaires, 4 were discarded because they were not completed by informants having relevant alliance experience and knowledge, 6 were discarded because the age of the alliance was less than 2 years, 2 were discarded because the alliance had multiple partners, and 4 were discarded because there were too many missing values.
Data purification
Hair et al.’s (1998) univariate approach was adopted to detect outliers. Since the sample size was greater than 80, Hair et al.’s criteria of −4 ≤ Z ≤ 4 was adopted to determine the outliers. Results showed that no significant outliers existed. Little’s (1988) missing completely at random test showed that data were missing completely at random. Following Newman (2003), completed questionnaires with more than 10% missing values were excluded, and the remaining missing values were imputed using maximum likelihood estimation.
Following Armstrong and Overton’s (1977) approach, nonresponse bias among the informants was examined by comparing early informants with late informants assuming that late informants are more similar to noninformants. A 2-week period was used as the dividing point between early informants and late informants because the reminder e-mails were sent to noninformants 2 weeks after the initial e-mails. The results of a t test showed no significant differences in the means of all items between the two groups (p > .05), suggesting that there was no large nonresponse bias in the data.
Since the research design was such that responses were self-reported and collected through the same survey during the same period of time from a single informant, there was a likelihood that common method variance (CMV) may have caused systematic measurement error (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003; Podsakoff & Organ, 1986). Two statistical tests were conducted to check for CMV—the Harman one-factor test (Podsakoff & Organ, 1986) and common latent factor method (Williams & Anderson, 1994). In the Harman one-factor test, the results of the unrotated principal component analysis showed that the first factor explained only 37.56% (<50%) of the variance. Since no single dominant factor emerged, it could be assumed that CMV was not an issue in the data (Scott & Bruce, 1994). Furthermore, the results of a more robust common factor method test showed that the unstandardized regression coefficients from the common latent factor were 0.16. The square of these unstandardized regression coefficients from the common factor was approximately 2.56%, which was the common shared variance. Results of this test also indicated that the data did not suffer from CMV (Reinartz, Krafft, & Hoyer, 2004). The data were also assessed for normality, linearity, and multicollinearity assumptions, and they did not show any significant nonnormality.
Scale Construction and Purification
Following Churchill (1979), 156 cases were used as a calibration sample for scale construction and purification. Dimensionality and reliability were assessed using EFA and Cronbach’s alpha test, respectively. Worthington and Whittaker (2006) recommend that dimensionality of constructs should be established in two stages—EFA should be conducted in the first stage, followed by CFA. I used Hair et al.’s (1998) three rules for deciding the number of factors to be extracted. Netemeyer et al.’s (2003) two-step approach was used to test the unidimensionality of various constructs of value creation. First, I checked whether all dimensions had their designated items loading heavily (>0.5) on them (Netemeyer et al., 2003) and/or cross-loading difference (0.10; Kathuria, 2000). Each dimension of value creation was subjected to EFA separately. Each construct was factor analyzed by principal component analysis with oblique rotation (promax rotation) rather than the orthogonal method (varimax), as the latter assumes there is no intercorrelation between the dimensions (Lawley & Maxwell, 1971). Since I expected the dimensions to be correlated, promax rotation was chosen as it depicts the underlying factor structure with greater precision. Second, the retained measurement items were then aggregated and EFA was run again to check the dimensionality.
Results and Discussion
Before conducting EFA, the measurement items were purified by assessing their correlations with other items in the same construct. I decided to remove the items that were correlated negatively with one another (DeVellis, 2003) or were weakly correlated with other items of the same construct with correlation (<.20; Netemeyer et al., 2003). Since none of the items had negative Pearson coefficient values and were significantly correlated with other items of the same construct, I retained all 26 items for EFA.
The results of the EFA at the individual construct level are presented in Table 3. Six items (CB7, PBCoop7, PBCoop9, PBComp3, PBComp5, and PBComp7) were deleted because their factor loadings were less than 0.5. The percentages of variance explained by the three dimensions of value creation were all greater than 50% (Hair et al., 1998). The KMO estimates for all the three dimensions were larger than the suggested criterion of 0.60 (Worthington & Whittaker, 2006), indicating sample adequacy for running factor analysis. The reliability of all dimensions of value creation ranged from .88 to .94, which is larger than Nunnally’s (1978) suggested reliability criterion of Cronbach’s alpha (α > .70).
Exploratory Factor Analyses at the Individual Construct Level
Note: CB = common benefit; PB = private benefit. Bold used to indicate that the factor loadings of the indicators were higher on the Factor 1 or 2.
The retained measurement items were then aggregated and EFA was run again. Before aggregating items for factor analysis, KMO was estimated for sample adequacy for 20 items using a sample size of 156. The results of the EFA of aggregated items are shown in Table 4. Two items (CB5 and PBCoop1) were deleted due to cross-loadings. The KMO index was 0.921, larger than the suggested criterion of 0.60. Thus, the unidimensionality of each dimension of the construct value creation was supported.
Exploratory Factor Analysis With All the Items Together
Note: CB = common benefit; PB = private benefit.
Study 4: Scale Validation
Method
Based on the results of Study 3, the survey was again modified and refined and floated in an online environment for collection of data for Study 4. Similar procedures as those employed in Study 3 were adopted for collection of data.
The data were collected between March and May 2012. The revised survey was released to 1,601 executives of 298 firms (102 pharmaceutical, 72 information technology, 46 telecommunication, 41 power and energy, and 37 steel). Since the unit of analysis was a dyadic alliance and not a firm, and a firm may have entered into multiple alliances with the same partner or with different partners, firms were eligible to participate in the survey for more than one alliance. Among 1,601 executives, 858 executives informed me that they were not eligible to respond to the survey, and the remaining 743 executives confirmed their willingness to participate in the study. However, among these 743 executives, 222 informants withdrew from the study for various reasons, which reduced the sample size to 521. Of the 521 completed surveys, 74 were unusable for the following reasons: (a) large amount of missing data (12), (b) age of alliance less than 2 years (19), (c) not adequately knowledgeable about the alliance (17), (d) multiple alliance partners (11), and (e) only a cooperative relationship existed between the alliance partners (15). Thus, 447 surveys were usable and the response rate was 27.92%.
A total of 155 firms participated in the study, which included 42 pharmaceutical firms (129 cases), 33 information technology firms (73 cases), 29 telecommunication firms (83 cases), 27 power and energy firms (86 cases), and 24 steel firms (76 cases). There was a wide variability in the net sales turnover of the participating firms, with a minimum of US$0.20 million to a maximum of US$66,800 million, a mean of US$5,268.12 million, and a standard deviation of US$12,819.1. Informants were the chief scientific officer/chief technical officer/chief operating officer (24.39%), director, alliance/alliance manager (23.71%), director/VP, R&D (18.12%), director/VP, marketing (13.65%), managing director/chief executive officer (9.40%), chief financial officer (6.26%), and others (4.47%).
The 447 cases collected in this study were used as a validation sample to validate the scale developed in Study 3 through CFA. CFA was performed using Amos 18 to assess whether the hypothesized measurement model fit the data and whether the indicators reflecting the various dimensions of value creation had good factor loadings without issues of cross-loadings. Furthermore, CFA also helped in assessing the construct validity through an examination of convergent and discriminant validities. Before CFA was conducted, procedures similar to those employed in Study 3 were conducted to examine missing values, identify outliers, estimate nonresponse bias, estimate CMV, and assess multivariate regression analysis assumptions.
Results and Discussion
The results of the tests conducted to examine the model fit of the measurement model, convergent and discriminant validities of the scale, and the competing model are discussed below.
Model fit of the measurement model
Model fit was assessed using the criteria of chi-square/degrees of freedom, goodness-of-fit index (GFI), normed fit index (NFI), Tucker–Lewis index (TLI), comparative fit index (CFI), root mean square error of approximation index (RMSEA), and standardized root mean square residual (SRMR; Hu & Bentler, 1995). The fit indices for all three dimensions were within the acceptable limits. I removed item CB8 to improve the model fitness of measurement model of common benefit.
Convergent and discriminant validity
Convergent validity is the extent to which an item correlates highly with other items designed to measure the same construct (Campbell & Fiske, 1959; Churchill, 1979). The nonnormed fit index (NNFI) provides evidence of convergent validity as it compares the hypothesized first order model to the null model. According to Bentler and Bonett (1980), an acceptable value for NNFI is greater than .90. The NNFI for all three factors was greater than .90, providing evidence of the convergence of the indicators of a factor. In addition, I adopted three other criteria to assess the convergent validity: (a) high factor loadings of measures on the same construct when using factor analysis with 0.40 as the threshold (DeVellis, 2003), (b) high composite reliability (CR) of the measures (>.60 as the threshold; Bagozzi & Yi, 1988), and (c) CR > AVE (average variance extracted; AVE > .5; Fornell & Larcker, 1981). Table 5 shows that the factor loadings of items to their underlying construct are greater than the threshold (>0.40; DeVellis, 2003) and the CRs lie between .88 and .94, which are greater than .60 (Bagozzi & Yi, 1988).
Confirmatory Factor Analysis
Note: CB = common benefit; PB = private benefit.
Discriminant validity is the extent to which the measures of a construct can be distinguished from another construct (Churchill, 1979). Discriminant validity is established when the AVE for each construct is larger than the square of the correlation between the construct and any other construct in the model (Fornell & Larcker, 1981). I adopted Hair et al.’s (1998) recommended thresholds for establishing discriminant validity: maximum shared squared variance (MSV) < AVE, average shared squared variance (ASV) < AVE.
Table 6 presents the validity and reliability along with the factor correlation matrix with the square root of the AVE on the diagonal. Table 6 shows that reliabilities of all reflective measures were above the .60 level suggested by Bagozzi and Yi (1988) and the .70 level suggested by Nunnally (1978; all values ≥ .88). Therefore, scales for each dimension of value creation show good internal reliability. The AVE for each reflectively measured construct is high (all values ≥ .58). The high AVE coupled with the strengths and significances of the parameter estimates of each of the measurement models establish convergent validity (Cannon & Perreault, 1999). Table 6 also reports the results of test of discriminant validity, which was assessed using Fornell and Larcker’s (1981) method. Results show that the AVE of each construct is greater than its MSV and ASV with other constructs (i.e., AVE > MSV > ASV), which provide evidence of discriminant validity.
Convergent and Discriminant Validities
Note: ASV = average shared squared variance; AVE = average variance extracted; CB = common benefit; CR = composite reliability; MSV = maximum shared squared variance; PB = private benefit.
Examining competing measurement models
I conceptualized that value creation has three distinct dimensions: (a) common benefit, (b) private benefit cooperation, and (c) private benefit competition. To establish the reliability and validity of measures of value creation in co-opetitive contexts, I assessed whether the hypothesized model was the most appropriate model. I assessed the fitness of the hypothesized model by comparing it with competing models. Three models were tested: (a) Model 1 conceptualizing that the three underlying dimensions of value creation are distinct and independent from each other (the hypothesized model), (b) Model 2 conceptualizing that the three underlying dimensions converge into one dimension, and (c) Model 3 conceptualizing that the private benefit cooperation and private benefit competition dimensions converge into one dimension and the common benefit dimension forms another dimension. This was done by fixing the paths PB to PBCoop and PB to PBComp at 1. The three models are represented in Figure 3.
Competing Measurement Models for Value Creation
I adopted a three-step approach to compare the competing models. In Step 1, the model fitness was assessed based on the value of fit indices including chi-square/df, p value, GFI, NFI, TLI, CFI, RMSEA, SRMR, Akaike information criterion (AIC), and Bayesian information criterion (BIC; Bollen, 1989; Hu & Bentler, 1995). In Step 2, to determine whether the hypothesized model achieved a better fit regardless of the probably acceptable overall fit in Step 1, the chi-square difference test was used to compare them (Bollen, 1989; Hair et al., 1998). In Step 3, the component fit (factor loadings) of measurement models was examined (Bollen, 1989).
The comparison of fit indices between the rival models presented in Table 7 shows that the three-dimensional model fits the data better than the unidimensional and two-dimensional models. Based on Browne and Cudeck’s (1993) cutoff standard, the RMSEA for the three-dimensional model indicates a reasonably good fit, while RMSEA for the unidimensional and two-dimensional models indicates unacceptable fit. I then used the chi-square difference test to compare the overall fits between Models 1, 2, and 3. The results of the chi-square difference test presented in Table 7 show that Model 1 demonstrates a significantly better fit than Model 2 and Model 3. This shows that the three-dimensional model is more valid than the unidimensional model and the two-dimensional model.
Comparison Among the Competing Models
Note: AIC = Akaike information criterion; BIC = Bayesian information criterion; CB = common benefit; CFI = comparative fit index; GFI = goodness-of-fit index; NFI = normed fit index; PB = private benefit; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual; TLI = Tucker–Lewis index.
Significant at the .001 level.
Second, factor loadings of the items are significant for all the competing measurement models. Thus, the criteria of significance of factor loadings were not useful in assessing validity of one model over the other. Finally, while the constructs in the three-dimensional model are correlated with each other (∅12 = .27, ∅23 = .45, ∅13 = .20), the magnitude of unshared variance (1 – 0.272 = 0.93, 1 – 0.452 = 0.80, 1 – 0.202 = 0.96) remains very large. This indicates that the three dimensions of value creation in co-opetition context are distinct.
In addition, a second-order CFA was also conducted to compare the relative loadings of the three dimensions of value creation (Bagozzi, 1994). CFA was conducted using three second-order models—Models 4, 5, and 6, corresponding to the competing Models 1, 2, and 3, respectively. In Model 4, the path from the second-order latent variable, value creation, to the first-order latent variable, common benefit, was set to 1. In Model 5, the paths from the second-order latent variable, value creation, to the first-order latent variables—common benefit, private benefit cooperation, and private benefit competition—were restricted to 1. In Model 6, the paths from the second-order latent variable, value creation, to the first-order latent variables—private benefit cooperation and private benefit competition—were set to 1, while the path from value creation to common benefit was not restricted. Model 4 is identical to the basic first-order model and provides the target or optimum fit for other models (Marsh & Hocevar, 1985). The target coefficient index (TCI) was calculated as a ratio of χ2 of Model 4 to the χ2 of the more restrictive models (Models 5 and 6; Marsh & Hocevar, 1985). Since Model 4 is identical to the basic first-order factor model, TCI is 1. In Model 4, private benefit cooperation has the highest factor loading of 0.96, while common benefit and private benefit competition have factor loadings of 0.60 and 0.67, respectively. In Model 5, the TCI is 0.53 and the factor loadings for common benefit, private benefit cooperation, and private benefit competition are 0.70, 0.85, and 0.72, respectively. In Model 6, the TCI is 0.64 and the factor loadings of common benefit, private benefit cooperation, and private benefit competition are 0.62, 0.88, and 0.75, respectively. Since the TCI of both Model 5 and Model 6 is much lower than 1, Model 4 is the best fitting model.
Study 5: Post Hoc Analysis
Method
To assess the validity of the scale, I conducted four post hoc tests. First, I assessed the validity of the scale by verifying whether it held in different contexts by conducting a CFA for Model 1 using different data subsets. In addition, to compare the relative factor loadings of the different dimensions of value creation, I conducted a CFA for Model 4 using different data subsets. Second, to assess whether there was a significant difference in the response pattern, I administered an abbreviated version of the scale developed in Study 4 to a subset of the same informants to whom the full-scale was administered 4 months previous. Third, to verify whether the conceptualization of value creation scale as reflective scale was valid, I conducted a tetrad analysis. Fourth, to assess the predictive validity of the scale, I examined the significance of the relationship between the three dimensions of value creation and a common antecedent variable.
Results and Discussion
I divided the data into different subsets based on different criteria, such as size of the firm, type, and form of alliances, and industry sector, because each of these criteria might influence value creation differently. For instance, firm size may influence value creation as capital may be available at a lesser cost to larger firms, which also have the added advantage of economies of scale (Gulati, 1995). On the other hand, smaller firms may benefit more from interfirm alliances than larger firms due to the lower costs of integrating resources and knowledge as it is easier for them to substitute older routines with new ones (Koh & Venkatraman, 1991). Similarly, sector-specific characteristics may also influence alliance performance (Gulati, 1995; Krishnan, Martin, & Noorderhaven, 2006). Some authors argue that characteristics of international alliances vary from those of domestic alliances (e.g., Harrigan, 1985; Kogut & Singh, 1988; Parkhe, 1993; Saxton, 1997). Therefore, the type of alliance partner—domestic or international—may also influence value creation. Similarly, since the nature of governance structure may also influence value creation, I repeated the analysis for subsamples of types of alliances—equity JV, bilateral-contractual-based alliance, and unilateral-contractual-based analysis. 3 Likewise, I also conducted similar analysis for a subsample of alliances which differed from each other in terms of intensity of interaction between the cooperative and competitive behavior: 4 high cooperation–low competition and high cooperation–high competition. Table 8 shows that the scales developed for the dimensions of value creation provide psychometrically valid results across the different subgroups. Thus, the scale development process has not been adversely affected by any kind of bias. Table 8 also presents the results of the second-order CFA in terms of the relative factor loadings of the three dimensions of value creation.
Post Hoc Analysis for Subsamples
Note: CB = common benefit; CFI = comparative fit index; GFI = goodness-of-fit index; NFI = normed fit index; PB = private benefit; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual; TLI = Tucker–Lewis index.
A second-order latent variable value creation was constructed with first-order latent variables CB, PBCoop, and PBComp being its reflective indicators.
I constructed an abbreviated version of the scale developed in Study 4. This scale included four items for common benefit (CB10, CB11, CB12, CB13), three items for private benefit cooperation (PBCoop3, PBCoop4, PBCoop6), and four items for private benefit competition (PBComp1, PBComp2, PBComp8, PBComp11). Since according to Kline (2010: 138) “it is better to have at least three to four indicators per factor to prevent” identification issues in structural equation modeling (SEM), I decided to have at least three items for each dimension of value creation in the abbreviated scale. The criterion for choosing the items for the abbreviated scale was that they should have the highest factor loadings on the respective dimension of the value creation scale (Table 5). Four months later, I administered this abbreviated version of the scale to 198 informants from among the 447 informants who had participated in Study 4. No significant differences were detected (using t tests) in the means of designation of informants, alliance age, or alliance experience between those who participated only in the Study 4 survey (n = 249) and those who participated in both surveys (n = 198). The fit measures of the three-dimensional abbreviated scale were within the acceptable limits (χ2/df = 4.27, p = 0, GFI = .93, NFI = .95, TLI = .95, CFI = .96, RMSEA = .08, SRMR = .05, AIC = 225.20, BIC = 327.76). In addition, the abbreviated scale shows adequate convergent validity for the three value creation dimensions (CB: CR = .91, AVE = .72; PBCoop: CR = .85, AVE = .66; PBComp: CR = .91, AVE = .73).
Based on literature review and Jarvis, MacKenzie, Podsakoff, Mick, and Bearden (2003), the developed scales were conceptualized and operationalized as reflective measurement models. Jarvis et al. use the following four criteria to distinguish reflective measures from formative measures: (a) the latent variable is the determinant of the observed indicator items, (b) the observed indicator items are interchangeable among each other, (c) the correlation between observed indicator items is high, and (d) the observed indicator items share the same antecedents and consequences as the latent variable. The conceptualization of the measurement model as reflective was validated using the procedure for tetrad analysis given by Bollen and Ting (2000). The CTA-PLS approach was used for deciding the choice between reflective and formative measurement models according to the procedure outlined by Gudergan, Ringle, Wende, and Will (2008). First, the vanishing tetrads for each dimension of value creation were identified. Second, the model-implied vanishing tetrads were identified and the redundant model-implied vanishing tetrads were eliminated. Third, tests were performed using bootstrapping approach and Student’s t distribution to examine whether the tetrads were significantly different from 0. Fourth, to overcome the multiple testing problems of CTA-PLS, the significance levels were adjusted using Bonferroni adjustments. The results of this analysis showed that for all the model implied nonredundant vanishing tetrads, the parameter values were within the 90% one-tailed Bonferroni-adjusted confidence interval. Therefore, the null hypothesis that the model implied nonredundant vanishing tetrads are equal to 0 could not be rejected. The sample estimate for the model implied nonredundant vanishing tetrad ranged from −0.54 to 0.86. The sample mean estimates ranged from −0.56 to 0.82. While the confidence interval for the largest negative sample mean ranged from −0.85 to −0.23, the confidence interval for the largest positive sample mean ranged from 0.21 to 1.09. Thus, the tetrad analysis confirmed that the reflective indicator specification was more appropriate than the formative indicator specification for the three dimensions of value creation.
Following Hinkin (1998) and Way et al. (2012), I assessed predictive validity of the newly developed scale of value creation by examining the relationship between value creation dimensions and a variable, which could be hypothesized to predict them. I hypothesized that top management commitment would be related to common benefits, private benefit cooperation, and private benefit competition. In the alliance literature, top management commitment has been regarded as a driver of value creation (Lambe et al., 2002; Spekman, Isabella, & MacAvoy, 1999), and it was measured using four items of Lambe et al.’s (2002) joint senior management commitment scale. Using SEM, the relationship between top management commitment and the different dimensions of value creation was assessed. In all, 36.8% of the variance of common benefit (β = .61, p < .01), 36.4% of the variance of private benefit cooperation (β = .60, p < .01), and 48.5% of the variance of private benefit competition (β = .49, p < .01) was explained by top management commitment.
Table 9 presents the validated scale items for value creation in interfirm alliances using a co-opetition-based approach.
Final Value Creation Scale
Note: CB = common benefit; PB = private benefit.
These scale items were used in the abbreviated 11-item scale.
General Discussion
This study was designed to construct a new scale that measures value creation in interfirm alliances using a co-opetition-based approach. The results of Studies 1 through 5 provide empirical evidence supporting the validity of value creation as a three-dimensional construct. This research suggests that the scale developed through these studies demonstrates acceptable levels of reliability and validity. The 17-item scale is advantageous over existing approaches in that it offers researchers an instrument for measuring value creation that (a) is applicable to all interfirm alliance contexts, (b) clearly identifies the different kinds of value created in such alliances, (c) distinguishes between value created due to cooperative and competitive behaviors in interfirm alliances, (d) distinguishes between value jointly created by alliance partners and value privately created by individual firms in the alliance, and (e) enables the measurement of value created from each alliance separately. Another important objective was to sharply define the measures of value creation so that they were not confused with the measures of value appropriation. An additional strength of the scale is that it was developed using diverse samples from multiple sectors (overall N = 603) differing in terms of degree of cooperation and competition, nature and purpose of alliance, age of alliance, and type of alliance partners. Another advantage of this study is that an abbreviated scale containing 11 items for the three dimensions of value creation has been developed so that researchers can use this abbreviated scale if they are working with complex models and it is difficult for them to use all the 17 items of value creation.
Since enhancement of value creation is the primary objective of firms entering into interfirm alliances (Bowman & Ambrosini, 2000; Lepak et al., 2007), various dimensions of value creation need to be understood. Researchers have conceptualized and measured value creation in interfirm alliances in many different ways, and thus findings from these studies can not be effectively compared (Olk, 2002). By developing a new scale of value creation by adopting a co-opetition-based approach, this study addresses this gap in literature and advances the knowledge of value creation in several ways. First, this study develops psychometrically valid scales for the three components of value creation. Second, this study empirically validates that private benefit competition is a benefit that is distinct from common benefit and private benefit cooperation. Third, this study shows that value creation is not a unidimensional construct but comprises three different components. Thus, contrary to the conceptualization by several scholars (e.g., Dyer & Singh, 1998; Madhok & Tallman, 1998) that value creation in interfirm alliances is a two-dimensional construct, this study shows that a three-dimensional value creation model provides a better explanation of value creation in interfirm alliances. Fourth, there is a conceptual similarity between private benefits and private capital on one hand and common benefits and social capital on the other. Common benefits comprise several intangible aspects of value, which are collectively generated by the alliance partners on account of development of social capital at the alliance level due to the harmonious integration of process, managerial networks, and tacit knowledge undertaken by the alliance partners (Kemper et al., 2013; Nahapiet & Ghoshal, 1998). Private benefits also include several intangible features, which reflect the exploitation of private capital as alliance partners institute mechanisms to use knowledge developed within the alliance boundaries for meeting their private objectives (Kemper et al., 2013; Nahapiet & Ghoshal, 1998). While earlier operationalizations of value creation, which relied on objective indicators such as financial indices and stock market returns, overlooked intangible sources of value (Brush et al., 2000; Hagedoorn, 1993), the newly developed scale in this study is able to capture the intangible aspects of value that are derived from the exploitation of both private capital and social capital.
This study encourages alliance managers to take a holistic view toward value creation. It provides alliance managers with an instrument through which they can estimate the magnitude of value created across the three dimensions. This can help them in identifying whether the alliance is meeting its objectives and whether the gap between potential value and realized value has been bridged through alliance activities (Madhok & Tallman, 1998). Managers can effectively enhance a firm’s value creation by exploring ways to enhance the three dimensions of value creation. Managers may also adopt the scales to assess the continuity of their value creation performance. They can leverage their firm’s value creation capability by monitoring the realized value along the three dimensions over a period of time and take corrective measures to achieve their strategic goals.
The findings of this study are limited by its attention to the high-tech knowledge-intensive industrial sectors in India. Therefore, it is necessary to examine whether the validity of the scale would hold for other industrial contexts in developed economies. The scale may provide different results with a variation in cultural variables such as power, distance, or individualistic and collectivist cultural orientations (Hofstede, 2001). Therefore, new or additional items might be needed for different cultural contexts to account for indicators of value creation that may be specific to some contexts; for instance, if a particular industrial context is technologically less developed, acquiring existing technologies for large-scale production might be an important indicator of value creation. On the other hand, in industrial contexts with very rapid technological changes, innovations and novel products and services might be the appropriate indicators of value creation. Therefore, the value creation items might need to be revisited for use in industrial contexts that depart from the Indian setting.
Another limitation of this study is the use of snowball sampling. The sample was not drawn at random and is prone to community bias as the initial respondents influence the final sample (Atkinson & Flint, 2004). The sample may not accurately reflect the population, and it may be difficult to estimate the required sample size as the population size is not known (Morgan, 2008). Furthermore, this is a cross-sectional study that does not address the evolution of value creation in interfirm alliances over time. In the future, longitudinal studies may be conducted to investigate whether and how transitions occur in value creation. This would also address the need to pay significantly more attention to the processes through which value is created rather than the tangible outcomes alone (Yan & Zeng, 1999). In addition, while focusing on dynamic processes, value judgments about emphasizing either stability or instability of alliance need to be avoided (Yan & Zeng, 1999). Sometimes competitive behaviors may create tension, yet these tensions are a part of the value creation process in a co-opetitive alliance. Another limitation of the study is that the scale items do not take into account the threats to value creation, which may lead to erosion of value and thus help in arriving at a net measure of value creation (Ariño, 2003; Madhok & Tallman, 1998). These items can be incorporated in future studies to increase the applicability and robustness of the scale.
Footnotes
Acknowledgements
This article was accepted under the editorship of Deborah E. Rupp. This article is based on my PhD dissertation titled “A Study of Value Creation and Value Appropriation in Inter-firm Alliances of Simultaneous Cooperation and Competition” (Indian Institute of Management, Ahmedabad, India). I thank Himanshu Bhatt, Srinath Jagannathan, Jeremy Short (Action Editor), and two anonymous reviewers for their constructive feedback and suggestions, which greatly improved the quality of this article. I am also grateful to Vatsala Vasudeva for her painstaking efforts in reading through the earlier drafts of this article. The article reads far better due to her efforts, and all errors and omissions that remain are of course only mine.