In this study, we adopt a mixed-methods, four-stage approach. First stage is content analysis of open innovation literature resulting in 12 dimensions of types of open innovation strategies which are then trimmed to have 7 dimensions kept for this study. Second stage is in-depth interviews with 102 responsible individuals across 78 open innovation firms that show consistent differences in 5 out of the 7 dimensions arising from the first stage, and then after a series of further work such as checking with two academics and finding out inter-coder reliability 4 archetypes of open innovation strategies that can be described alongside the 5 dimensions have been developed. Third stage is cluster analysis resulting in four distinct clusters and the labelling of these four clusters as four archetypes of open innovation strategies. Fourth stage is large-scale survey to assess the relative impacts of the four archetypes of open innovation strategies on innovation performance with the moderating process variables in the analysis. Elaboration of these four stages is as follows.
First stage content analysis
We first followed previous researchers in innovation-related management (e.g., Liang, Li, Yang, Lin, & Zheng, 2013), using a configurational approach as a tool for constructing empirically-based typologies of open innovation strategies. Second, a content analysis of the open innovation literature identified the dimensions of types of open innovation strategies. As a result, 12 dimensions were identified.Footnote 4
The two co-authors of this study and two academics discussed which of the 12 dimensions derived from the literature could be used to differentiate persistent differences among different types of open innovation strategies. As a result of these deliberations, five dimensions were removed, leaving the other seven dimensions for further consideration.
Second stage in-depth interviews
We conducted 102 in-depth interviews to obtain empirical evidence from a case study-based investigation of 78 open innovation firms. After analyzing the interview data from these 78 open innovation firms, manually and electronically (NVivio 9), we agreed that we have arrived at theoretical saturation (Shiu, Hair, Bush, & Ortinau, 2009), as further collection and analysis of data from more open innovation firms was not likely to yield further insights. The analysis of the 78 open innovation firms established that, of the seven dimensions derived from the previous steps, two dimensions did not display strong enough variation in their manifestation levels across the 78 open innovation firms to ensure their further consideration for establishing whether types of open innovation strategies could be identified. The remaining five dimensions showed persistent differences among the 78 open innovation firms: knowledge linkages, knowledge complementarity, mutual trust, relational strength, and relational commitment.
To advance the results of the coding process in terms of reliability and validity, we asked another two academics, with backgrounds in qualitative research methods, to analyze interview results derived from the 78 open innovation firms. We then checked inter-coder reliability based on both the total number of units and the total number of exact matches in units coded by the two academics. The value of inter-coder reliability was acceptable (Cohen’s kappa = .95). The results suggest that knowledge linkages, knowledge complementarity, mutual trust, relational strength, and relational commitment are ideal dimensions to categorize different types of open innovation strategies.
Finally, we classified each of the 78 open innovation firms as a high/moderate/low profile across two knowledge dimensions and three relational dimensions. As a result, four archetypical open innovation strategies emerged along the five dimensions. Table 2 demonstrates the four archetypical profiles.
Third stage cluster analysis
We used a cluster analysis to validate the emergence of the four open innovation strategies. We hired three academics to study the 78 open innovation firms and rate them on a scale from 1 to 10 under the five dimensions. Based on guidelines in Milligan and Sokol (1980) and Punj and Stewart (1983), we used a hierarchical analysis followed by a k-means analysis to obtain taxonomies that are as stable and robust as possible. Regarding the hierarchical approach, we used Ward’s method based on squared Euclidian distances, which creates clusters of similar size.
We considered a range of initial solutions from the hierarchical analysis with either three, four, five, six, or seven groups, as suggested by the dendrogram. The number of clusters was selected based on the within-group sum of squares (Hair, Tatham, Anderson, & Black, 2013). To assess which solution was the most stable, we then computed kappa between each initial and final solution (Singh, 1990). The four-cluster solution appeared to best fit the data (k = .97, while k < .86 for other solutions). Three- and Five-cluster solutions were discarded based on the clustering statistics of the root-mean-square standard deviation (measuring the homogeneity of the cluster formed).
We also used a non-hierarchical analysis as a robustness check, and find that both hierarchical and non-hierarchical analyses gave the same results. Finally, we validated the results by using Kruskal–Wallis tests and found significant differences (p < .01) between the variables used to develop the clusters (Hair et al., 2013), suggesting that these results were robust.
In cluster 1, open innovation strategies are strongly involved in both high levels of two knowledge dimensions and three relational dimensions: high knowledge linkages, high knowledge complementarity, high mutual trust, high relational strength, high relational commitment. They are the strongest in both knowledge and relational dimensions as opposed to the other four clusters. Evidenced by their high scores in the four clustering variables overall, it is fair to suggest that this cluster of firms creates innovation through intensively interacting with external entities, as well as widely employing external knowledge. We therefore label them as full collaboration.
Firms in cluster 2 have moderate levels of two knowledge dimensions, with high levels of three relational dimensions: moderate knowledge linkages, moderate knowledge complementarity, high mutual trust, high relational strength, and high relational commitment. They focus more on relationship activities in open innovation activities. Therefore, we label them as relationship-centric.
Open innovation strategies in cluster 3 have high levels of two knowledge dimensions, with moderate levels of relational dimensions: high knowledge linkages, high knowledge complementarity, moderate mutual trust, moderate relational strength, and moderate relational commitment. Because they focus more on the transformation of knowledge into innovation processes than relational dimensions, we label them as knowledge-centric.Footnote 5
Finally, cluster 4 includes open innovation strategies that generally rely only slightly on two knowledge and three relational dimensions: low knowledge linkages, low knowledge complementarity, low mutual trust, low relational strength, and low relational commitment. These firms tend to be aware of the benefits of using open innovation activities, but they score weakly on the knowledge and relational dimensions overall when taking on new open innovation opportunities. They are definitely more restricted than other types of open innovation strategies. Therefore, we label them as minimal collaboration.
Fourth stage large-scale survey
Sample and data collection
Based on the commercial list available on subscription from China Credit Information Service (2013) in Taiwan, we developed a contact list of the top 1000 Taiwanese firms in terms of sales. To capture inter-industry variability, the sampling frame was not restricted to any given industry.
We contacted each firm by telephone to determine whether it had employed open innovation strategies in its innovation process. To avoid an arbitrary selection of a particular archetype of open innovation strategy, top senior managers were asked to identify a completed open innovation project, launched between 2010 and 2012, and to respond to the items as they are related to that particular open innovation project. As a result of this process, 392 firms were eligible and agreed to participate in this study.
As with previous studies in developing economies (e.g., De Luca & Atuahene-Gima, 2007), we collected the data on-site, so that we could clarify any questions that the respondents might have and ensure that the questionnaires collected were complete and usable. We recruited trained interviewers to conduct on-site surveys, who presented the questionnaires to the respondents. To reduce potential common method bias (Podsakoff, MacKenzie, & Podsakoff, 2012), we obtained different information from multiple sources in each firm. Specifically, data for open innovation strategies were collected from top senior managers, data for process interdependence, process complexity, and new product/service development speed were collected from new product/service development managers, and data for new product/service innovativeness and market performance were collected from marketing managers. All respondents were asked to fill in the questionnaire based on the open innovation project and to respond to the items as they are related to that particular open innovation project. This design ensures that the data are obtained from the most appropriate sources.
This leads to 248 sets of matched responses from top senior managers, new product/service development managers, and marketing managers, resulting in an effective response rate of 63.2% (248 out of 392). The sample represented six industries: information technology (22.5%), banking and insurance (21.3%), electronics (18.5%), telecom (17.3%), semiconductor (16.5%), and others (3.9%).
All items were measured on a seven-point scale. We used the double-translation method (Yang, Dess, & Robins, 2018) to translate the questionnaire from English into Mandarin (English-Mandarin-English). Specifically, we first developed an English version of the questionnaire, then used a double-translation procedure to translate it into Mandarin (English-Mandarin-English). This process included: (1) the authors initially translating the items into Mandarin; (2) three academics translating the Mandarin version back into English; and, (3) another two academics checked the translations to ensure conceptual equivalence. The same procedures are repeated for the questionnaire from Mandarin into English (Mandarin-English-Mandarin).
In addition, we conducted a face and content validity review (N = 18), the first pilot test (N = 69) to indicate any ambiguity respondents experienced when responding to the items, and the second pilot test (N = 93) to purify the scale and to obtain preliminary estimates of reliability. Next, we proposed a protocol on the basis of the interview results to describe high, moderate, and low levels of knowledge linkages, knowledge complementarity, mutual trust, relational strength, and relational commitment.Footnote 6
We developed the measure for assessing different archetypes of open innovation strategies based on the protocol used in the exploratory study. During the survey, top senior managers were asked to identify their most successful open innovation strategy between 2010 and 2012 and then to rate their strategy on a scale from 1 to 7 under the five dimensions (knowledge linkages, knowledge complementarity, mutual trust, relational strength, and relational commitment). Following the same procedure of the cluster analysis, our sample consisted of 70 full collaboration open innovation strategies (28.2%), 68 knowledge-centric open innovation strategies (27.4%), 65 relationship-centric open innovation strategies (26.2%), and 45 minimal collaboration open innovation strategies (18.1%).
Process interdependence was adapted from Fang (2011) with four items (α = .89). A sample item is “During the development process of the product/service, both external entities and we have to work very closely in each stage of the new product/service development project”. Process complexity was adapted from Griffin (1997) with three items (α = .86). A sample item is “Compared to other new product/service development process, the development process of this product/service is: Simple–Complex”.
Innovation success was measured by four indicators: financial performance, new product/service development speed, market performance, and new product/service innovativeness. Data for financial performance originated from the data reported by China Credit Information Service (2013). The firms’ financial performance was measured by the percentage of profits attributable to new products/services launched between 2010 and 2012. New product/service development speed was adapted from Rindfleisch and Moorman (2001) with four items (α = .88). A sample item is “Please evaluate the product/service launched to markets in the past three years on: Much slower than we expected–Much faster than we expected”. Market performance was adapted from Blazevic and Lievens (2004) with three items (α = .85). A sample item is “Relative to competing firms’ performance, your firm’s market performance is very successful in terms of Reputation”. New product/service innovativeness was adapted from Lee and Colarelli O'Connor (2003) with three items (α = .84). A sample item is “The benefits this new product/service offers are new to the customers”.
We controlled for firm industry type, firm size, firm age, and environmental turbulence. Based on the data from China Credit Information Service (2013), we determined each responding firm’s industry type, firm size, and firm age. For firm industry type, the sampling units were categorized into service and manufacturing firms, depending on whether the majority of a firm’s sales are derived from service or tangible products (Ettlie & Rosenthal, 2011). For firm size, we followed the definition of Small and Medium Enterprise Administration by Ministry of Economic Affairs, Taiwan, in which firms with fewer than 200 employees were classified as small firms, while firms with 200 employees or above were large firms. Firm age was measured as the number of years since the firm had started its business operation. Environmental turbulence was adapted from De Luca and Atuahene-Gima (2007) with five items (α = .93). A sample item is “In the markets in which the innovation operates, Customer’s preferences change rapidly over time”.
Analysis and results
The MPlus exploratory structural equation modeling (SEM) technique (Muthen & Muthén, 2010) was used to establish the internal consistency of our measures, because it combines exploratory and confirmatory factor analysis in one procedure, and avoids the problems associated with the traditional two-step process (Fornell & Yi, 1992). Based on the fit indexes, the model fit is satisfactory: χ2/d.f. = 1.92; comparative fit index (CFI) = .95, goodness of fit (GFI) = .94; Tucker-Lewis index (TLI) = .93; incremental fit index (IFI) = .95; root mean square error of approximation (RMSEA) = .04. In addition, the items load as expected and all factor loadings are significant (p < .01) on their latent factors.
Discriminant validity was assessed by using the Fornell and Larcker (1981) procedure and an alternative procedure that Anderson and Gerbing (1988) recommend. For each construct the value of the square root of each average variance extracted (AVE) is greater than the values of the inter-construct correlations. In addition, the confidence interval does not include 1.0 by plus or minus two standard errors around the correlation between the constructs (Anderson & Gerbing, 1988), and the Chi-square test between any two constructs is significant (p < .001). Table 3 shows a summary of respondent firms’ statistics and correlations.
To assess the differences in innovation success among the four archetypes of open innovation strategies, we followed the procedures of Cheng and Huizingh (2014) by using a SEM approach to assess their main effects.
As shown in Table 4, the R2 values of financial performance, new product/service development speed, market performance, and new product/service innovativeness are .403, .437, .584, and .562, respectively, all of which are between moderate and high, indicating that the model explains a substantial proportion of variance in the innovation success variables (Hair et al., 2013). In addition, the path estimate results with t-values of full collaboration, relationship-centric, knowledge-centric, and minimal collaboration open innovation strategies are significantly and positively related to all four indicators of innovation success. This suggests that the four archetypes of open innovation strategies examined in this study are positively associated with innovation success.
We then compared the relative strengths of the four open innovation strategies through a series of Chi-square difference tests. Specifically, to obtain the Chi-square value of the direct effect of full collaboration open innovation strategy, the first model includes the paths (1) full collaboration open innovation strategy—financial performance, (2) relationship-centric open innovation strategy—financial performance, (3) knowledge-centric open innovation strategy—financial performance, (4) minimal collaboration open innovation strategy—financial performance. The second model deletes the first path from the first model. We thus obtained a Chi-square difference value between the first and second models. Based on these Chi-square values, we again run Chi-square tests to compare the difference (Harmancioglu, 2009). The same procedures are repeated for the other archetypes of open innovation strategies.
As presented in Table 5, a significant positive Chi-square difference indicates that the selected archetype of open innovation firm has a stronger effect on innovation success, while a negative difference suggests a weaker performance. For example, in the first column of Table 5, which focuses on financial performance, the Chi-square difference of full collaboration vs. knowledge-centric is −4.85*, suggesting that full collaboration is less effective than relationship-centric.
Accordingly, the differences in innovation success among the four archetypes of open innovation strategies are all significant. Particularly, the results show that relationship-centric open innovation strategy has the strongest impact on innovation success, followed by full collaboration, knowledge-centric, and minimal collaboration open innovation strategies. Table 6 concludes the ranking of the relative effects of the four archetypes of open innovation strategies. The result suggests that Hypothesis 1 is supported.
We then used pairwise comparisons based on general linear model to test Hypotheses 2 and 3. The results shown in Table 7 indicate that Hypothesis 2 is supported, while Hypothesis 3 is partially supported at the statistical significance levels of p < .001. We discuss the results in more detail in the following sections.