Upper echelons theory

Upper echelons theory (Hambrick, Reference Hambrick2007; Hambrick & Mason, Reference Hambrick and Mason1984) postulates that top managers’ characteristics influence their decision-making, thus shaping such organizational outcomes as innovation capability. It also acknowledges that a team of top managers, rather than a single top manager, typically makes strategic decisions (Hambrick, Reference Hambrick, Augier and Teece2018). Since TMTs must make strategic decisions under ambiguous circumstances and face more stimuli than they can process thoroughly (Hambrick, Reference Hambrick, Augier and Teece2018), their decisions are guided by their inherent characteristics instead of purely objective analysis (Hambrick & Mason, Reference Hambrick and Mason1984). Hambrick and Mason (Reference Hambrick and Mason1984) identify two clusters of characteristics that shape TMT member’s decision-making. Individual information about future events, alternatives, and consequences that is grounded in their educational, functional, and professional backgrounds (Hambrick & Mason, Reference Hambrick and Mason1984; Miller et al., Reference Miller, Chiu, Wesley, Vera and Avery2022) and individual values about what they want to achieve with the organization that derive from their attitudes, principles, and motivations (Finkelstein, Hambrick, & Canella, Reference Finkelstein, Hambrick and Canella2009; Li, Liu, & Wan, Reference Li, Liu and Wan2008).

A substantial body of literature examines how diversity in TMT members’ characteristics guides the strategic decisions they make in influencing organizational outcomes (e.g., Cuypers, Patel, Ertug, Li, & Cuypers, Reference Cuypers, Patel, Ertug, Li and Cuypers2022; Homberg & Bui, Reference Homberg and Bui2013; Ponomareva, Uman, Bodolica, & Wennberg, Reference Ponomareva, Uman, Bodolica and Wennberg2022). Building on the propositions of upper echelons theory, scholars argue that TMT members’ diverse information and values combine to the benefit of the organization (Hambrick & Mason, Reference Hambrick and Mason1984; Hemmert, Cho, & Lee, Reference Hemmert, Cho and Lee2024). For example, the literature identifies that variety in TMT members’ individual work experience, education, or functions expands the TMT’s pool of information from which its members can generate more ideas and spot additional opportunities, benefitting innovation (e.g., Belderbos et al., Reference Belderbos, Lokshin, Boone and Jacob2022; Qian, Cao, & Takeuchi, Reference Qian, Cao and Takeuchi2013; Schubert & Tavassoli, Reference Schubert and Tavassoli2020). Therefore, this literature conceptualizes TMT diversity from the variety perspective that benefits innovation.

Social categorization in dissimilar teams

The literature also proposes another perspective on team diversity, the separation perspective (Harrison & Klein, Reference Harrison and Klein2007). It contends that TMT members’ dissimilarity in terms of their information and values separates them from one another, inhibiting their ability to leverage information and values for decision-making (Harrison & Klein, Reference Harrison and Klein2007). Informational dissimilarity refers to perceiving differences in the information other team members hold because of differences in their education, functional, and professional backgrounds (Hobman, Bordia, & Gallois, Reference Hobman, Bordia and Gallois2004; Shemla, Meyer, Greer, & Jehn, Reference Shemla, Meyer, Greer and Jehn2016; Van der Vegt et al., Reference Van der Vegt, van de Vliert and Oosterhof2003). Value dissimilarity refers to perceiving differences in other team members’ perspectives because of differences in their principles that guide their work-related behavior, values, and motivation (Hobman et al., Reference Hobman, Bordia and Gallois2004; Shemla et al., Reference Shemla, Meyer, Greer and Jehn2016).

Based on their perceptions of dissimilarity, team members categorize others into subgroups that have high levels of similarity with and high levels of dissimilarity from themselves (Tajfel & Turner, Reference Tajfel, Turner, Austin and Worchel1986). According to the social categorization perspective (Hogg, Reference Hogg, Hogg and Tindale2001; Tajfel et al., Reference Tajfel, Billig, Bundy and Flament1971), the emergence of such subgroups leads to intergroup bias such that information from and the values of similar team members are integrated into decisions while those from dissimilar team members, i.e. from a different subgroup, are disregarded or devalued (Hogg, Reference Hogg, Hogg and Tindale2001). Consequently, team dissimilarity is associated with negative outcomes like a lack of helping behavior (Liao, Chuang, & Joshi, Reference Liao, Chuang and Joshi2008), empathy on the other team members’ viewpoints (Williams, Parker, & Turner, Reference Williams, Parker and Turner2007), and team turnover (Finkelstein et al., Reference Finkelstein, Hambrick and Canella2009; Jackson et al., Reference Jackson, Brett, Sessa, Cooper, Julin and Peyronnon1991). Specifically, research associates value dissimilarity in teams with such negative outcomes as low cohesiveness, low trust levels, poor communication (DeBode, Fox, & McSweeney, Reference DeBode, Fox and McSweeney2023; Dose & Klimoski, Reference Dose and Klimoski1999), poor team performance, poor team effectiveness (Jehn, Northcraft, & Neale, Reference Jehn, Northcraft and Neale1999), and low team involvement (Hobman et al., Reference Hobman, Bordia and Gallois2004).

Concerning, informational dissimilarity on a team, previous studies show inconsistent results. Pelled, Ledford, and Mohrman (Reference Pelled, Ledford and Mohrman1999)argue that educational dissimilarity is a sign of specialized knowledge that benefits perceived information access and decision-making influence, Adamovic (Reference Adamovic2020)and Chattopadhyay, Finn, and Ashkanasy (Reference Chattopadhyay, Finn and Ashkanasy2010) indicate positive and negative outcomes. Adamovic (Reference Adamovic2020) suggests that it can encourage information exchange and information elaboration but can also reduce team members’ willingness to contribute their information if it is unique to the team. Adamovic (Reference Adamovic2020) argues that this is because team members expect disadvantages upon perceiving informational dissimilarity between the team members. According to Chattopadhyay et al. (Reference Chattopadhyay, Finn and Ashkanasy2010), professional dissimilarity, one type of informational dissimilarity, decreases negative emotions and negative behaviors, such as yelling or making sarcastic comments toward other team members, by high-status team members, but increases these negative behaviors among low-status team members. Finally, Hobman et al. (Reference Hobman, Bordia and Gallois2004) find that informational dissimilarity decreases team involvement because team members overemphasize common information and neglect unique information, which reduces communication within the team. Further, Van der Vegt et al. (Reference Van der Vegt, van de Vliert and Oosterhof2003) find that team members’ categorization processes decrease members’ identification with the team.

Interestingly, research tends to focus on one dissimilarity characteristic at a time, such as educational background, professional background, or work behavior, although the reality is that team members simultaneously respond to different kinds of dissimilarity. For example, informational and value dissimilarity are characteristic of the specialized knowledge and perspectives in the TMT and are present within the TMT at the same time. This co-occurrence of different types of dissimilarity evokes simultaneous categorization processes (Grigoryan, Cohrs, Boehnke, van de Vijver, & Easterbrook, Reference Grigoryan, Cohrs, Boehnke, van de Vijver and Easterbrook2022). Considering the co-occurrence of different dissimilarities is necessary since categorization into subgroups is possible only when a team deals with a small number of dissimilarity characteristics (Crisp & Hewstone, Reference Crisp and Hewstone2007). Research shows that people cannot process categorization of more than four categories (Halford, Baker, McCredden, & Bain, Reference Halford, Baker, McCredden and Bain2005). If more than four types of dissimilarity co-occur, team members can no longer form clearly distinguishable subgroups (Macrae & Bodenhausen, Reference Macrae and Bodenhausen2001), so categorization processes will result in subgroups that may be dissimilar in one characteristic but similar in another (Chung et al., Reference Chung, Liao, Jackson, Subramony, Colakoglu and Jiang2015). For example, categorization may lead to forming subgroups that differ in values but share similar information. Multiple social categorization theory (Crisp & Hewstone, Reference Crisp and Hewstone2007) postulates that the more dissimilar a team, the less likely it is to form subgroups. As overlaps between the subgroups increase, no harmful categorization of team members occurs, and team members can leverage information and values from one another (Crisp & Hewstone, Reference Crisp and Hewstone2007). Overall, then, the social categorization perspective and multiple social categorization theory diverge: The former predicts that team dissimilarity leads to intergroup biases and negative team outcomes, for example low team effectiveness (Hogg, Reference Hogg, Hogg and Tindale2001; Tajfel et al., Reference Tajfel, Billig, Bundy and Flament1971). The latter predicts positive outcomes for highly dissimilar teams because no harmful categorization occurs, so no intergroup bias arises (Crisp & Hewstone, Reference Crisp and Hewstone2007). This controversy has not yet been solved in TMT diversity research. In fact, research around TMT diversity focuses on the variety perspective and neglects the separation perspective. Therefore, our study seeks to shed light on the role of the separation perspective in TMTs, thus contributing to clearing up the controversy of categorization in TMTs.

Incremental and radical innovation capability

Innovation capability is the ability to apply and integrate resources to generate innovations (Lawson & Samson, Reference Lawson and Samson2001) and refers to the use of resources rather than their availability (Menguc, Auh, & Yannopoulos, Reference Menguc, Auh and Yannopoulos2014; Teece, Pisano, & Shuen, Reference Teece, Pisano and Shuen1997). The innovation literature classifies innovation capability into incremental and radical (e.g., Lei, Ha, & Le, Reference Lei, Ha and Le2019; Slater, Mohr, & Sengupta, Reference Slater, Mohr and Sengupta2014; Wang, Lu, Zhao, Gong, & Li, Reference Wang, Lu, Zhao, Gong and Li2013). Incremental innovation capability is the ability to ‘generate innovations that refine and reinforce existing products and services’ (Subramaniam & Youndt, Reference Subramaniam and Youndt2005, p. 452). That is, to make moderate changes in existing services and products, as the underlying processes and knowledge are reinforced and do not significantly change (Menguc et al., Reference Menguc, Auh and Yannopoulos2014). In contrast, radical innovation capability is the ability to ‘generate innovations that significantly transform existing products and services’ (Subramaniam & Youndt, Reference Subramaniam and Youndt2005, p. 452). That is, the ability to make major changes to existing products, services, and processes or develop new ones by, for example, using breakthrough technology (Menguc et al., Reference Menguc, Auh and Yannopoulos2014). The capability to generate radical innovation requires a substantial change or transformation of prevailing processes and knowledge by, for example, using new resources or developing new capabilities (McGrath, Reference McGrath2001). Research shows that incremental innovation capability is of particular relevance to organizations when improving the consumer experience (Menguc et al., Reference Menguc, Auh and Yannopoulos2014), while radical innovation capability may be necessary to keep up with rapidly changing customer demands and environmental changes (Slater et al., Reference Slater, Mohr and Sengupta2014).

Increasingly, research distinguishes between incremental and radical innovation capability and identifies distinct antecedents for each type (e.g., Forés & Camisón, Reference Forés and Camisón2016; Gui et al., Reference Gui, Lei and Le2022; Han et al., Reference Han, Lyu, Ji, Zhu, Su and Bao2020; Slater et al., Reference Slater, Mohr and Sengupta2014), but most research on the implications of TMT diversity does not differentiate between the two (e.g., Azeem & Baker, Reference Azeem and Baker2020; Quintana-García & Benavides-Velasco, Reference Quintana-García and Benavides-Velasco2016; Song et al., Reference Song, He and Yan2022). Heyden, Sidhu, van den Bosch, and Volberda’s (Heyden, Sidhu, van den Bosch, & Volberda, Reference Heyden, Sidhu, van den Bosch and Volberda2012) study is an exception, as it argues that TMT members’ experience diversity increases the TMT’s pool of knowledge and, therefore, increases the set of feasible ideas for radical innovation. In contrast, little diversity in TMT members’ experience deepens the team’s knowledge but leads to compliance with the status quo and, therefore, to incremental innovation capability. These results suggest that the TMT’s composition in terms of diversity has distinct implications for the team’s type of innovation capability.

Research model

We use an exploratory study design to shed light on the association between TMT members’ informational dissimilarity/value dissimilarity and incremental innovation capability/radical innovation capability. Exploratory research is beneficial if the goal is to extend research by examining relationships when underlying information is scant (Swedberg, Reference Swedberg, Elman, Gerring and Mahoney2020). Although the body of literature concerning TMT diversity and innovation is substantial, it is deficient in several respects.

First, most upper echelons research uses objective proxies to measure the diversity of information and values. The proxy for diversity in values relies on donations to political parties (Narayan et al., Reference Narayan, Sidhu and Volberda2021), whereas studies often measure informational diversity using demographic proxies like educational background, education level, functional specialization, and experiences across industries (e.g., Azeem & Baker, Reference Azeem and Baker2020; Chemmanur, Kong, Krishnan, & Yu, Reference Chemmanur, Kong, Krishnan and Yu2019; Chung, Cho, & Kang, Reference Chung, Cho and Kang2018; Li & Huang, Reference Li and Huang2019; Li, Liu, Lin, & Ma, Reference Li, Liu, Lin and Ma2016; Qian et al., Reference Qian, Cao and Takeuchi2013; Wu & Park, Reference Wu and Park2019; Zhong, Song, & Chen, Reference Zhong, Song and Chen2021). However, Miller et al. (Reference Miller, Chiu, Wesley, Vera and Avery2022) claim that these demographic proxies perform poorly, which questions the predictions derived from them and the underlying mechanisms they help to identify. Consequently, the authors call for a more direct assessment of TMT characteristics and TMT diversity. A perception measure ensures that salient differences are measured, which may provide a more suitable proxy because team members act on their perceptions rather than reality (Lawrence, Reference Lawrence1997) and may not see all objective differences as salient (Randel, Reference Randel2002). However, few studies assess TMT diversity in terms of TMT members’ perceptions (Alexiev, Jansen, van den Bosch, & Volberda, Reference Alexiev, Jansen, van den Bosch and Volberda2010; Mehrabi, Coviello, & Ranaweera, Reference Mehrabi, Coviello and Ranaweera2021; Mihalache et al., Reference Mihalache, Jansen, van den Bosch and Volberda2012).

A second respect in which the literature concerning TMT diversity and innovation is deficient lies in distinguishing between informational diversity and value diversity in the TMT. Some studies (Alexiev et al., Reference Alexiev, Jansen, van den Bosch and Volberda2010; Mehrabi et al., Reference Mehrabi, Coviello and Ranaweera2021; Mihalache et al., Reference Mihalache, Jansen, van den Bosch and Volberda2012) use a direct measure from Campion, Medsker, and Higgs (Reference Campion, Medsker and Higgs1993), which asks TMT members about the TMT’s characteristics, but these studies focus on informational characteristics and do not address values. In fact, most studies rely on proxies that reflect informational differences like educational background, functional background, or expertise (e.g., Chung et al., Reference Chung, Cho and Kang2018; Li & Huang, Reference Li and Huang2019; Talke et al., Reference Talke, Salomo and Kock2011) and disregard diversity in values (Narayan et al., Reference Narayan, Sidhu and Volberda2021). Differentiating the two types of dissimilarity would allow for a more holistic assessment of the implications of TMT diversity, as values shape team processes (Chrobot-Mason, Ruderman, Weber, & Ernst, Reference Chrobot-Mason, Ruderman, Weber and Ernst2009).

Third, upper echelons research focuses on the variety perspective of team diversity, which conceptualizes team diversity as a process in which diversity in information and values are additive (Harrison & Klein, Reference Harrison and Klein2007), thus accelerating the generation of feasible ideas that benefit innovation (e.g., Belderbos et al., Reference Belderbos, Lokshin, Boone and Jacob2022; Boone, Lokshin, Guenter, & Belderbos, Reference Boone, Lokshin, Guenter and Belderbos2019; Mehrabi et al., Reference Mehrabi, Coviello and Ranaweera2021). However, team diversity also contributes to forming homogeneous subgroups that distinguish groups from each other (Hogg, Reference Hogg, Hogg and Tindale2001; Tajfel et al., Reference Tajfel, Billig, Bundy and Flament1971), separating team members and hindering them from working together effectively (Harrison & Klein, Reference Harrison and Klein2007). Research that examines TMT diversity from a separation perspective, that is, how diversity in information and values separate TMT members from each other, is lacking (Song et al., Reference Song, He and Yan2022). Therefore, whether the implications of the social categorization perspective (Hogg, Reference Hogg, Hogg and Tindale2001; Tajfel et al., Reference Tajfel, Billig, Bundy and Flament1971) or multiple social categorization theory (Crisp & Hewstone, Reference Crisp and Hewstone2007) apply in the context of informational and value dissimilarity between TMT members remains unexplored, as are whether and to what extent TMTs suffer from categorization processes and their detrimental effects on innovation capability.

Finally, scarce upper echelons research distinguishes between incremental and radical innovation capability. While the types of innovation capability suggest different strategic orientations (McDermott & O’Connor, Reference McDermott and O’Connor2002; Un, Reference Un2010), both types are suitable if organizations are to innovate (Gui et al., Reference Gui, Lei and Le2022; Sheng & Chien, Reference Sheng and Chien2016). In addition, the antecedents of the types of innovation capability are distinct (e.g., Forés & Camisón, Reference Forés and Camisón2016; Gui et al., Reference Gui, Lei and Le2022; Slater et al., Reference Slater, Mohr and Sengupta2014). Therefore, the ability to derive implications requires distinguishing between incremental and radical innovation capability, but few studies address how TMT diversity is associated with the types of innovation capability.

Taken together, the lack of literature that addresses the association between TMT diversity and incremental innovation capability/radical innovation capability and that takes the separation perspective in addressing informational and value dissimilarity results in scant information on the expected direction and magnitude of associations. Therefore, an exploratory study design in an initial examination of how the co-occurrence of the two types of dissimilarity are associated with incremental and radical innovation capability is required.

Figure 1 shows the research model of this study.

Figure 1.

Research model.

Data collection and sample

We use a quantitative approach and collect data through an online survey. We survey TMT members from organizations that have more than 250 full-time employees across industries in Germany. Table 1 provides details about the sample composition. A key advantage of using a survey is that it enables the respondents’ perceptions of their team members’ dissimilarity to be assessed. Respondents were recruited via an online panel provider. To remedy potential issues in data quality, we conducted multiple checks. We eliminated respondents who completed the survey too quickly, deleted surveys with inconsistencies in answers (e.g., indicating that more women were in the TMT than TMT members overall), and checked the respondents’ IP addresses to ensure that only unique IP addresses were included in the final sample. These procedures resulted in a final data set of 91 observations.

Table 1.

Sample composition

N = 91.

Measurement of constructs and variables

Weuse established scales to measure the latent variables of our research model and, since we surveyed top managers from German organizations, we translated the scales into German following Brislin’s (Reference Brislin1970) back-translation procedure to ensure equivalence between original and translated scale such that a third person conducted the back-translation. Unless stated otherwise, we measure all items on a seven-point Likert-type scale ranging from 1 = strongly disagree to 7 = strongly agree. Table 2 shows all items used in the study.

Table 2.

Measurement model

^a Item dropped due to low item reliability.

Common method bias

Since we collected the variables of interest from a single source, common method bias could be an issue. To address this problem, we ensured that the variables were captured in separate sections of the survey (Podsakoff, MacKenzie, & Podsakoff, Reference Podsakoff, MacKenzie and Podsakoff2003). We assessed common method bias with the marker variable approach (Lindell & Whitney, Reference Lindell and Whitney2001), using ‘having littered the environment’ as the marker variable, which is not theoretically related to the research model. The correlations do not substantially change after the marker variable’s effect is parceled out. Therefore, the results of neither the marker variable test nor Harman’s single factor test indicate an issue with common method bias.

Key informant bias

To mitigate the potential for key informant bias associated with the self-reporting of incremental and radical innovation capability, we comply with Homburg, Klarmann, Reimann, and Schilke’s (Reference Homburg, Klarmann, Reimann and Schilke2012) suggestion to survey informants who hold high hierarchical positions, as those in higher positions tend to provide more accurate evaluations, than those in lower positions. We also asked the participants to rate their competence in answering the questions and eliminated four observations that indicated low competence. Then we follow Homburg and Wielgos (Reference Homburg and Wielgos2022) in rerunning the analysis by dropping seven observations that indicate moderate competence. The results did not change after dropping these observations, indicating that the participants who identified as moderately confident did not influence the results.

Reliability and validity of measures

Reflective measurement model

The measurement model contains four reflective measurement models: Informational dissimilarity, value dissimilarity, incremental innovation capability, and radical innovation capability. We assess the reliability and validity of the measures following the recommendation of Mehmetoglu and Venturini (Reference Mehmetoglu and Venturini2021). Table 2 shows the details about these assessments.

In assessing radical innovation capability, we found that one indicator loaded below the suggested threshold value of 0.708 (Hair, Risher, Sarstedt, & Ringle, Reference Hair, Risher, Sarstedt and Ringle2019) (see Table 2), and the AVE for the latent variable increased after we excluded it, so we eliminated it from the model. We also eliminated one indicator of the latent variable informational dissimilarity because of substantial cross-loading. In the final model, all standardized loadings are above the suggested threshold of 0.708. The AVE values exceed the recommended threshold level of 0.5. Dillon-Goldstein’s rho (D.G. rho), Rho (a) (ρ_A), and Cronbach’s α values are above the suggested level of 0.7 as well (Bagozzi, Yi, & Phillips, Reference Bagozzi, Yi and Phillips1991; Mehmetoglu & Venturini, Reference Mehmetoglu and Venturini2021) (see Table 2). These results indicate the model’s reliability and convergent validity.

We assess discriminant validity using the Fornell-Larcker criterion and the heterotrait monotrait ratio of correlations (HTMT). All AVE values of the model’s latent variables are larger than the squared correlations, indicating discriminant validity (Fornell & Larcker, Reference Fornell and Larcker1981) (see Table 3). We find that all values of the HTMT are below the threshold value of 0.9 for conceptually similar constructs (Hair et al., Reference Hair, Risher, Sarstedt and Ringle2019). We also calculate the HTMT2 since the HTMT tends to be upward biased for correlated latent variables (Henseler, Reference Henseler2021) and find that the HTMT2 is also below the threshold value of 0.9, indicating discriminant validity.

Table 3.

Discriminant validity

The table indicates the squared interfactor correlations. AVEs are in the diagonal in bold. Number 5 to 11 are measured through a single indicator, and therefore average variance extracted (AVE) cannot be computed.

Findings from the exploratory study

We assess multicollinearity of the structural model by calculating the VIF. All values are below 2.5, indicating no multicollinearity issues (Hair et al., Reference Hair, Risher, Sarstedt and Ringle2019). The R ² for incremental innovation capability is 44.0 percent and that for radical innovation capability is 42.1 percent. Table 4 presents the structural model that includes the direct associations. The results suggest a positive association between informational dissimilarity and incremental innovation capability (Informational dissimilarity → incremental innovation capability; β = .586, p < .001), so TMT members who perceive their team members as highly dissimilar from them in terms of information report higher incremental innovation capability than TMT members who perceive their team members as more homogeneous in terms of information. We also find that perception in terms of dissimilarity in values is negatively associated with incremental innovation capability (Value dissimilarity → incremental innovation capability; β = − .273, p < .05). Therefore, whereas TMT members who perceive their team members as dissimilar in terms of information associate that dissimilarity with higher incremental innovation capability, they associate perceiving their fellow TMT members as dissimilar in values is associated with lower incremental innovation capability.

Table 4.

Structural model – Direct associations

Fit statistics for the structural model: R ² for incremental innovation capability: 44.0%,.

R ² for radical innovation capability: 42.1%.

The findings show that informational dissimilarity is not associated with radical innovation capability (Informational dissimilarity → radical innovation capability; β = − .137, p > .1), but value dissimilarity has a positive and significant association with radical innovation capability (Value dissimilarity → radical innovation capability; β = .600, p < .001). In short, the TMT members in the sample who perceive a high level of value dissimilarity also report a high level of radical innovation capability.

The findings show that informational dissimilarity is not associated with radical innovation capability (Informational dissimilarity → radical innovation capability; β = − .137, p > .1), but value dissimilarity has a positive and significant association with radical innovation capability (Value dissimilarity → radical innovation capability; β = .600, p < .001). In short, the TMT members in the sample who perceive a high level of value dissimilarity also report a high level of radical innovation capability.

We follow Chin, Marcolin, and Newsted’s (Chin, Marcolin, & Newsted, Reference Chin, Marcolin and Newsted2003) two-stage approach to test for interaction effects of informational and value dissimilarity. The first stage estimates the latent variables and builds the main effects, as reported in Table 4. The second stage uses the latent variable scores from the first stage to build the interactive model (Mehmetoglu & Venturini, Reference Mehmetoglu and Venturini2021). One disadvantage of the two-stage approach is the possibility of multicollinearity issues, so we calculate the VIF to assess multicollinearity in the interactive model. All values are below 2.5, indicating no multicollinearity issues (Hair et al., Reference Hair, Risher, Sarstedt and Ringle2019). The R ² value for incremental innovation capability is 40.6 percent for incremental innovation capability and that for radical innovation capability is 47.6 percent. Table 5 shows the structural model with the interaction effects included.

Table 5.

Structural model – Interactive associations

Fit statistics for the structural model: R ² for incremental innovation capability: 40.6%,.

R ² for radical innovation capability: 47.6%.

† < p < .05; ** p < .01; *** p < .001.

The findings indicate that the interaction between informational dissimilarity and value dissimilarity is not associated with incremental innovation capability (Informational dissimilarity × value dissimilarity → incremental innovation capability; β = − .025, p > .1). The findings also show no significant interaction between informational dissimilarity and value dissimilarity on radical innovation capability (Informational dissimilarity × value dissimilarity → radical innovation capability; β = .167, p> .1). Hence, the associations between informational dissimilarity and incremental and radical innovation are not contingent on the level of value dissimilarity, and the associations between value dissimilarity and incremental and radical innovation are not contingent on the level of informational dissimilarity.

Sensitivity analysis

To determine whether the reported findings are robust to alternative explanations, we follow Hair et al.’s (Reference Hair, Risher, Sarstedt and Ringle2019) recommendation. To check for unobserved heterogeneity, we employ a response-based unit segmentation in the model that explores the direct associations.Footnote ¹ The algorithm uses the residuals of the PLS-SEM analysis to segment the data and analyzes it for each segment. If unobserved heterogeneity is present, the segmented model’s results differ from those of the overall model (Mehmetoglu & Venturini, Reference Mehmetoglu and Venturini2021). We use the group quality index and the permutation test to assess the results of the response-based unit segmentation, and these tests show no indication of unobserved heterogeneity.

To check for nonlinear associations, we first run Ramsey’s (Reference Ramsey1969) regression equation specification test and find no indication of nonlinear associations. Then we follow Svensson et al. (Reference Svensson, Ferro, Høgevold, Padin, Carlos Sosa Varela and Sarstedt2018) in creating a quadratic term of the exogenous variables like an interaction and rerun the model to check for a U-shaped association. The results show a significant and positive quadratic term of value dissimilarity in association with incremental innovation capability (Value dissimilarity × value dissimilarity → incremental innovation capability; β = .324, p < .05). A significant and positive coefficient of the quadratic term is indicative of a U-shaped association (Haans & Pieters, Reference Haans and Pieters2016), so we check whether the turning point lies within the data by setting the partial derivative of the model equation to 0 and solving for it (Obadia & Robson, Reference Obadia and Robson2021). The calculated turning point is 0.200 and is within the data range, indicating a U-shaped association. However, running further analyses by splitting the sample at the estimated turning point or specifying the model such that only a quadratic term of one exogenous variable is included provides no indication of a U-shaped association between value dissimilarity and incremental innovation capability. Therefore, we conclude that value dissimilarity and incremental innovation capability are negatively associated.