The concept of ACAP was defined by Cohen and Levinthal (1989, 1990) as a firm’s ability to identify relevant external knowledge, assimilate it, and apply it commercially. Initially, this ability was seen as a by-product of internal R&D efforts. However, in a later study, Cohen and Levinthal (1990) employed Nelson and Winter’s (2005) concept of organizational capability to elaborate on the intra-firm elements that support the development of ACAP. Among these elements are the gatekeepers, interfaces between the firm’s areas and external agents, and actions that favor the dissemination of intra-firm knowledge. These actions allow for connecting a mosaic of individual abilities (Cohen; Levinthal, 1990) in favor of building a capability at the organizational level. After this initial work, different reinterpretations have been carried out, and three of these can be highlighted.

Zahra and George (2002) redefine the ACAP as a dynamic capability composed of potential and realized capabilities. Potential ACAP is composed of acquisition and assimilation capabilities. The first is related to the firms’ capabilities to recognize the value of new external knowledge, acquire it, and add it to their knowledge base, while the second is related to the routines and processes that allow the company to analyze, process, and interpret external knowledge, given the existing knowledge. The second set - realized ACAP - reflects the capabilities for transforming and exploiting external knowledge, where the former refers to the processes for combining old knowledge with new knowledge, and the latter is associated with the ability to apply knowledge commercially, aiming at raising or building competencies. For the authors, the higher the ratio between realized ACAP and potential ACAP, the greater the efficiency in absorption, that is, in applying the assimilated knowledge in innovations and competitive advantages. This “efficiency factor” would be determined by “social integration mechanisms” (Zahra; George, 2002) to be discussed in the next subsection.

Criticizing the “static” characteristic of this “efficiency factor”, Lane, Koka, and Pathak (2006) redefine ACAP as a sequential process composed of three forms of learning: exploratory, transformative, and exploitative learning. The first learning allows the company to recognize and understand the potential of external knowledge, which is related to the first dimension of ACAP proposed by Cohen and Levinthal (1990). Transformative learning, on the other hand, refers to processes for assimilating external knowledge and combining it with existing knowledge, encompassing the assimilation and transformation capabilities exposed by Zahra and George (2002). It is responsible for making a connection between exploratory and exploitative learning. The latter, in turn, is related to the application of this new external knowledge. Using March (1991) and Bishop D’Este and Neely (2011), it can be said that exploratory learning contributes especially to obtaining long-term benefits by expanding and diversifying the internal knowledge base, giving greater flexibility to the company; whereas exploitative learning is more focused on the application of knowledge in products and processes, focusing on short-term results and more direct appropriability.

Lane, Koka and Pathak (2006) also contribute to the literature by decomposing ACAP into two parts: (1) funnel; (2) pipelines. The first would allow the company to select which are the relevant knowledge for its strategies, approaching Cohen and Levinthal’s (1990) “recognition” dimension. This funnel would be determined by the breadth of the company’s previous knowledge base, built especially through exploratory learning. The pipelines, on the other hand, refer to intra-company organizational processes that favor transformative and exploitative learning, allowing the internalization of external knowledge considered relevant and its application in the market.

Lewin, Massini, and Peeters (2011) share this “bipartite” view of ACAP. For the authors, the ACAP would be formed by a connected set of internal and external meta-routines expressed in observable practical routines. Internal meta-routines are associated with the process of intrafirm knowledge construction, refinements, and sharing, helping to build an intrafirm knowledge base; they are routines that help to (1) modify old routines, (2) emerge and select new intrafirm ideas, and (3) share existing knowledge and practices. The external meta-routines are more directly aimed at the search for external knowledge, being linked to the practical routines for recognizing the value of external knowledge and learning from external partners, such as the use of gatekeepers, participation in networks, and R&D partnerships. These routines are focused on identifying and acquiring external knowledge, aligning with the funnel concept of ACAP. According to the authors, the effectiveness of the ACAP depends on the connection between these two meta-routines, which would be accomplished by meta-routines focused on (1) creating stimuli to explore knowledge and opportunities (internal or external) and (2) transferring the external knowledge inside the company through specific processes, such as the sharing of external knowledge among gatekeepers and other employees. Without these processes, the effectiveness of external ACAP diminishes, making absorption highly reliant on the individual capabilities of gatekeepers (Lewin; Massini; Peeters, 2011). This point will be important in the empirical analyses, as well as Lane, Koka and Pathak’s (2006) interpretation of ACAP.

As highlighted, the ACAP is an organizational capability that also depends on processes and routines that disseminate knowledge internally and connect and drive the actions of individuals (Zahra; George, 2002). Some of these processes are highlighted below, focusing on those measured in the empirical analysis. An extensive review of them is made by Volberda, Foss and Lyles (2010).

First, Cohen and Levinthal (1990) already highlighted theoretically the importance of actions to disseminate intra-firm knowledge. These actions act as “social integration mechanisms”, facilitating the flow of intra-firm knowledge and the transformation and exploitation of external knowledge (Zahra; George, 2002). Examples of these actions include the promotion of informal interactions among employees, open communication of ideas, mutual support among departments, and the conducting of internal seminars (Schmidt, 2005).

Another important action in favor of ACAP is the participation of employees in innovative projects. Jansen, Van den Bosch, and Volberda (2005) interpret this participation as greater freedom in the agents’ decision-making, but it has a dubious effect. On one hand, it increases the ability and motivation of workers to acquire and assimilate external knowledge, but, on the other hand, it hinders consensus building, which would negatively affect the realized ACAP. However, Ebers and Maurer (2014) claim that this greater participation would have a positive effect on the realized ACAP by allowing greater collaboration and intra-company knowledge exchange.

Formalizations and routines also have dubious effects on ACAP. For Van den Bosch, Volberda and De Boer (1999), these formalizations allow the knowledge exchange to occur through formal means, such as information systems or procedures exposed in documents, manuals, etc. However, these have dubious effects on both potential and realized ACAP. For the potential ACAP, these formalizations could increase the efficiency in the acquisition of external knowledge (Vega-Jurado; Gutiérrez-Gracia; Fernández-De-Lucio, 2008) or restrict the integration among the members of the company, which disfavors the potential ACAP (Jansen; Van Den Bosch; Volberda, 2005). For the realized ACAP, formalizations help codify the best practices for the application of external knowledge, but, on the other hand, reduce the creativity and flexibility important for this dimension (Jansen; Van Den Bosch; Volberda, 2005; Vega-Jurado; Gutiérrez-Gracia; Fernández-De-Lucio, 2008).

The determinants discussed above involve more managerial and deliberate decisions. However, for Ebers and Maurer (2014), ACAP can also be favored by strong and trusting personal relationships. For the authors, this kind of internal relationship contributes to communication and knowledge exchange among employees, favoring the transformation and commercial exploitation of new knowledge. Strong and trusting relationships between the company’s employees and external partners, on the other hand, affect potential ACAP by favoring the identification and assimilation of external knowledge.

Finally, the development of capabilities cannot be separated from the company’s structure (Nelson, 1991). Van den Bosch, Volberda and De Boer (1999) analyze three types of structural forms (functional, divisional, or matrix), identifying that more hierarchical forms - such as the functional - tend to contribute to the efficiency of the external knowledge absorption, but not to its scope or flexibility. On the other hand, less hierarchical forms - such as the divisional - favor the flexibility of this ACAP, but limit it as a whole due to the low integration between the departments that characterize them. Thus, the ACAP tends to be favored by an intermediate degree of hierarchy, in which there are gatekeeper positions and stronger connections between R&D and manufacturing departments, but with some departmental autonomy around diverse projects. This favors the efficiency, scope, and also flexibility of ACAP (Van Den Bosch; Volberda; De Boer, 1999; Duchek, 2015).

The next sections empirically investigate the effect of the mentioned organizational determinants on the ACAP of firms that interacted with universities. These firms tend to have distinctive capabilities compared to those that do not do this, because, on the one hand, this interaction is a way to develop ACAP by the improve of the knowledge base (Bishop; D’este; Neely, 2011) and, on the other hand, it requires peculiar prior ACAP, built by distinct organizational processes and innovative efforts (Lane; Lubatkin, 1998; Murove; Prodan, 2009; Negri, 2006; Schmidt, 2005). However, in a Brazilian context, the development of ACAP can be limited because the firms’ short-term strategies that use the universities not only as complementary but also as substitutes for their internal R&D activities (Rapini et al., 2009; Teixeira; Rapini; Caliari, 2020).

The present study uses a database developed between 2015-2016 specifically to measure the ACAP dimensions and their organizational determinants, focusing on firms that had interacted with universities in Brazil, according to the 2010 census of Research Groups Directory of the Brazilian National Council for Scientific and Technological Development (DGP/CNPq, in Portuguese). It is also collected information about the innovative activities of firms and their interaction with universities. As discussed before, these firms have different ACAP, but it is not possible to say a bigger or better ACAP them others, necessarily. A diversity of ACAP will be identified during section 4.

A detailed description of this base is provided by Teixeira (2020). Here, we focus on presenting its basic characteristics and the analyzed variables. Tables 1, 2, and 3 show the variables and references used for their construction. The questionnaire was applied in Portuguese, but the questions are translated here.

Table 1
Questions about the dimensions of ACAP

Table 2
Questions to measure the organizational determinants of ACAP (a)

Table 3
Questions to measure the organizational determinants of ACAP (b)

As shown in these tables, the analyzed variables were measured on a Likert scale with 4 levels of importance, plus a fifth level called “unable to give an opinion” (or to respond). It is argued that responses to this item indicate (i) a lack of understanding of the respondent about the question or (ii) a lack of knowledge about a particular action.

Regardless of the interpretation, it is assumed that if the respondent chose this option in all items referring to at least one “construct” (i.e., a dimension or a determinant of the ACAP), the answer does not portray the construct in question and, therefore, such a company is removed from the sample. On the other hand, if at least one item of the construct is answered in another way, then this company is kept in the sample, understanding that it provides some minimal information for it. With this strategy, it was possible to maintain a greater number of firms, but, a different treatment was necessary via the Categorical Principal Components Analysis (CatPCA), discussed in section 3.2.

It is important to highlight that specific ACAP surveys are important to capture ACAP multidimensionality and its dependence on organizational factors that go beyond R&D efforts (Ebers; Maurer, 2014; Jansen; Bosch, Van Den; Volberda, 2005). However, there is no consensus on how to measure absorptive capacity, given that different studies focus only on the ACAP dimensions and do not measure their organizational determinants in the same questionnaire (Camisón; Forés, 2010; Flatten et al., 2011; Jiménez-Barrionuevo; García-Morales; Molina, 2011). This makes it difficult to separate the antecedents of the ACAP and the “core” of its dimensions - a criticism similar to the use of R&D as a proxy for the ACAP (Flatten et al., 2011) -, in some cases requiring the use of the original concept to measure a given dimension.

Regarding the general features of the database, it is worth noting that the collection process resulted in 57 responses¹. After the process of cleaning, it has resulted in 39 valid responses², whose characteristics are described in Table 4. The firms are of different sizes, sectors, and regions. There is a predominance of national firms that interacted with universities in the three years before the survey (until 2015 or 2016, depending on when the company answered it)³. Regarding the respondents, more than half of them have been in the company for at least 6 years (66.7%), have a post-graduation, specialization, or MBA (61.5%), and are linked to management or R&D activities and innovation (53.8%). Such characteristics suggest, in general, a respondent with the ability to understand the questionnaire and with reasonable knowledge about the organizational characteristics and processes of the company.

Table 4
Description of the basic characteristics of the firms analyzed (Total: 39 firms)

Therefore, some characteristics of the database demand alternative methods of analysis: (1) a small number of valid responses; (2) responses such as “unable to express an opinion” are classified as missing; (3) non-linear Likert-type responses; (4) possible non-linear relationships between variables. To meet these specificities, the Categorical Principal Component Analysis (CatPCA) techniques will be combined with fuzzy-set Qualitative Comparative Analysis (fsQCA), where the first builds “super conditions” (Berg-Schlosser; De Meur, 2009) to be used in the second to explain the desired outcomes (i.e. the high ACAP). These techniques are detailed below.

The CatPCA is analogous to the linear Principal Component Analysis (PCA) but with a non-linear feature. The CatPCA reduces many variables into a smaller set of uncorrelated principal components, minimizing the loss of information. The method consists of considering each observed value of the ordinal categorical variable as a “category”. Each category is converted into a numerical value through a non-linear function (via the optimal scaling quantification method), transforming the categorical variables into quantifiable variables (Linting et al., 2007). In the present study, there are four “valid” categories: 1, 2, 3, and 4. This method quantifies each category, allowing, for example, that the difference between category 1 (strongly disagree) and category 2 (partially disagree) is less than the difference between this second category and category 3 (partially agree). As a result, the ordinal character of Likert-scale responses is considered in the analysis more appropriately than in traditional techniques such as linear Principal Components and Factor Analysis (Linting et al., 2007). Besides accounting for the ordinal nature of the variables, this method has other advantages, such as: (a) suitability for datasets with few observations or more variables than observations; (b) no requirement for multivariate normality; and (c) internal handling of "unable to give an opinion" responses, ensuring more comprehensive data use (Linting et al., 2007; Meulman; Heiser, 2011; Meulman; Kooij; Van Der; Heiser, 2004).

Specifically, about point c, the CatPCA allows treating responses of the type “not able to give an opinion” in two ways: passive and active. In both cases, this response is treated as missing, however, its contribution is different for the quantification of the categories and the construction of the principal components. In passive form, for example, if a company answers a question - for example, question “Ac_search_const”as “not able to give an opinion “, but answers all the others questions as 1, 2, 3, or 4, that company is not considered for the transformation of the variable “Ac_search_const”, but it is considered for all other variables. Besides, the object scores (i.e. the value of indicators) do not take into account the variable “Ac_search_const” to calculate it. The active form treats missing values as an additional category and includes them in all stages of the quantification process (IBM, 2011; Meulman; Kooij; Van Der; Heiser, 2004). In the present study, the passive form was chosen⁴.

However, this way of dealing with missing values has consequences. First, the object scores do not necessarily have a weighted average of zero. Second, the CatPCA does not maximize the P values of the correlation matrix between the transformed variables, but of a similar matrix, which differs especially by incorporating a diagonal matrix Mj that considers two possible values: (i) equal to zero if the object (i.e. the company) is missing in variable j and the passive strategy was adopted; (ii) equal to the weight of the variable, otherwise. As a result, component loadings do not necessarily vary between -1 and 1 and cannot be directly interpreted as the correlation between the transformed variables and the principal component in question. However, a high component loading, in the mathematical module, continues to represent a high importance of a certain variable in the composition of a given component. The direction of this relationship also continues to be represented by the sign of this component loading (IBM, 2011; Meulman; Kooij, Van Der; Heiser, 2004). Therefore, in the analysis of the results of the CatPCA (section 4.1) the term “importance of a given variable” will be used instead of the term “correlation” to interpret the values of the component loadings. A mathematical detail of this method can be found in IBM (2011).

The indicators built via CatpCA are used as conditions or outcomes in QCA. The last technique is relevant to this study because (i) it identifies different behavioral patterns leading to the same result and (ii) is suited for studies with few cases, typically fewer than 100, but with many explanatory variables (Berg-Schlosser et al., 2009). Both are characteristics of the present study.

Regarding point (i), it is not necessary to recognize a single specific causal model that best fits the data, allowing the identification of different combinations and configurations that lead to the same result, the so-called conjunctural causality (Berg-Schlosser et al., 2009). Regarding this, consider that variables A, B, C, D, and E may be possible conditions to generate the same product Y (e.g. high ACAP level). Then, the application of the QCA may conclude that the configuration of factors AB or ACD generates Y (AB or ACD → Y). Or, still, the lower value of a factor may be important to achieve the same result: aC → Y, where [a] is the designation for lower values of factor [A]. In theoretical terms, this is relevant because it allows different firms to combine differently the organizational processes analyzed and achieve similar levels of absorptive capacity. Furthermore, this is in line with the vision of ACAP as an organizational capability (Cohen; Levinthal, 1990), dependent on intra-firm routines (Lewin; Massini; Peeters, 2011) that can be used differently between firms (Nelson, 1991; Nelson; Winter, 2005). This heterogeneity is difficult to capture in econometric models or linear analyses (Berg-Schlosser et al., 2009), being less present in the literature.

In this work, the fuzzy-set QCA method (fsQCA) is used, which allows the score of association to a group within the range between 0 and 1. Values greater than 0.5 indicate that the company belongs more to the group of firms with high values for a certain condition; values less than 0.5, are the opposite. In this method, the rule used to assess the relationship between the outcome and the variables analyzed is the inclusion rate, also called the “consistency score” (Longest; Vaisey, 2008). This is given by:

Where X is the configuration of the predictor (i.e., AB), Y is the product (outcome), x_i defines each association in configuration X, and y_i denotes each association in the Y configuration.

Considering the conditional probability, the closer the unit is to the value of I_XY, the greater the consistency of the data when stating that X is a subseries that defines Y (X → Y). Several methods can be used to decide which configuration of X predictors is sufficient to explain Y. An important step is to define which configurations will compose the truth table. A first cut is to keep only configurations with I_XY> 0.800. In addition, configurations considered to be “contradictory” are also excluded from this table, that is, they explain both outcome Y and 1-Y. This elimination is based on the principle of asymmetric causality (Berg-Schlosser et al., 2009) and is performed by comparing the consistency of configuration X against outcome Y vis-a-vis the consistency of X with outcome 1 - Y. When this difference is not statically significant, this condition is considered contradictory, being disregarded in the assembly of the truth table (Longest; Vaisey, 2008).

With the truth table in place, an internal minimization procedure is implemented to identify sufficient minimum settings to explain outcome Y. These settings are evaluated in terms of their consistency (given by I_XY) and their coverage (C_XY), which is given by:

Although similar to I_XY, it provides different information: it is seen as the portion of the sum of outcome Y covered by configuration X. This can be interpreted as analogous to the “portion of the variance explained” by this configuration (Ragin, 2006) or as the portion of the outcome that can be understood from the final configuration found (Longest; Vaisey, 2008). Ragin (2006) considers a C_XY above 0.2 to be adequate. Finally, the model allows the best configuration (or fit) to be presented for each observation (that is, a firm), considering its variables in comparison to the result Y and the other groups analyzed. All fsQCA procedures were implemented using the "fuzzy" program by Longest and Vaisey (2008) for Stata® software.

The indicators from the previous sections will be employed in the analysis of the fsQCA. The role and summary of each indicator are presented in Table 12.

Table 12
Summary of the variables for fsQCA

* These codes refer to indicators of the third column transformed into fuzzy sets.

Three separate analyses will be studied, seeking to identify, for each type of ACAP, the sufficient minimum configurations among the organizational determinants. For this, the first step was to transform these indicators into fuzzy sets through standardization so that they vary between 0 and 1, preserving the position of firms in the distribution of a given variable. These transformed variables are represented by a single uppercase letter (see the second column of Table 12). In the analysis, uppercase letters denote variables with a score of association above 0.5, while lowercase letters indicate scores below 0.5. To exemplify this transformation, one can look at the indicator ACAP 1 (item X in the fsQCA): when the company has a score of association above 0.5 for ACAP 1, it will be interpreted that this company has a high ACAP 1 (represented by ‘X’); when this value is less than 0.5, it is said to have a low ACAP 1 (represented by ‘x’).

The considered “conditions” allow for 16 logically possible configurations that would compose the truth table⁸. However, not all of these logically possible configurations should be considered in the analyses, especially in relatively small databases (LONGEST; VAISEY, 2008; RAGIN, 2009). Three criteria were adopted for this exclusion: (1) the configuration is logically possible, but it was not observed in the database; (2) it is a contradictory configuration⁹; (3) the configuration did not show a consistency score statistically higher than 0.8. In all cases, the significance level of 10% was considered.

Table 13 depicts the sufficient minimum configurations for each ACAP and the accuracy of the final adjustment. The three minimum configurations for each ACAP presented high final consistency (above 0.8 each) and a satisfactory coverage rate (Ragin, 2006). Therefore, these configurations can be considered as sufficient conditions and will be interpreted as “minimum organizational configurations” to obtain a certain high ACAP.

Table 13
Minimum organizational configurations sufficient to obtain high levels of each ACAP

These minimal organizational configurations allow for three following logical relationships:

(1) If the company combines strong and reliable internal and external relationships (conditions “C” and “E”), regardless of the organizational management processes (indicated by the absence of conditions A and B in the minimum configuration), then the company has a high ACAP 1 (focused on exploratory learning).

(2) If the firm combines highly trustworthy internal (C) and external (E) relationships with organizational processes that favor and qualify the participation of employees in the execution of strategic innovative projects (A), regardless of the degree of internal hierarchy (indicated by the absence of condition “B”), so the company has high ACAP 2 (focused on the adaptation of technologies);

(3) If the firm combines low external trustworthy relations (e) with strong internal confidence (C), high internal hierarchy (B) and also high employees’ participation in innovative strategic projects (A), then the company has high ACAP 3 (focused on exploitative learning and results of greater knowledge appropriability).

These logical relationships imply theoretical discussions.

First, the minimum configurations for a high ACAP 2 are a subset of the minimum configurations for a high ACAP 1. This has important theoretical implications for the interpretation of the knowledge absorption process when analyzed together. To build an ACAP aimed at expanding the knowledge base without a defined application (i.e. ACAP 1), the combination of strong external and internal personal relationships is sufficient. The first allows employees linked to the innovative process to act as gatekeepers, accessing external knowledge more easily and bringing it to the company. However, for the assimilation in the internal knowledge base, the external knowledge needs to be disseminated and shared internally in the company (Cohen; Levinthal, 1990). In the present study, this is done through strong internal personal relationships (Ebers; Maurer, 2014) and does not necessarily depend on more managerial organizational processes. These processes, in turn, are necessary when the company seeks to build capabilities to adapt existing technologies, that is, to have a high ACAP 2. For this, the company must combine strong internal and external relationships with organizational processes that favor, qualify, and direct the participation of the employees in strategic innovative projects (i.e., the “presence of “A” in the final solution, together with “C” and “E”). These actions, when combined, can be seen as intra-firm “pipelines” (Lane; Koka; Pathak, 2006) that allow to connect workers internally, disseminate external knowledge internally and direct the application of external knowledge (Ebers; Maurer, 2014; Vega-Jurado; Gutiérrez-Gracia; Fernández-De-Lucio, 2008).

In other words, the results suggest that, when the firm seeks to access new knowledge without a defined application, only personal relationships can be sufficient; but, when seeking to exploit this knowledge to adapt technologies, managerial organizational processes are necessary for employee participation. These processes can be seen as a way of connecting individual actions and stimulating the exploitation of external knowledge, making ACAP less dependent on individual characteristics (Lewin; Massini; Peeters, 2011). It is worth mentioning that ACAP 2, in addition to focusing on the adaptation of technologies, is also dependent on individual skills, but it demands internal processes that induce the application of them.

The second interesting result refers to ACAP 3, which focused on the application of external knowledge in short-term, specific, and more direct results. To have a high ACAP 3, the company needs to balance the low confidence in external relations (represented by “e”) with the combination of a high internal hierarchy (“B”) with managerial organizational processes (“A”) and internal relations (“C”) that connect the members of the company, favor their participation in the innovative process and the consequent dissemination of knowledge within this more hierarchical structure. Van den Bosch, Volberda and De Boer (1999) already suggested that a more hierarchical organizational structure led to greater efficiency in absorbing external knowledge, but a smaller scope of it. This is in line with this ACAP 3 focused especially on more specific results, such as “working more effectively” and “gaining market share”, unlike ACAP 1 (focused, for example, on responding more quickly to changes in the market and having flexibility in the use of knowledge in innovations). However, given that ACAP 3 is also associated with the modification of the internal knowledge base¹⁰, the high hierarchy is not enough for this. Processes related to the coordination capacity among members are necessary to counterbalance the loss of scope and flexibility brought about by this higher hierarchy (Van Den Bosch; Volberda; De Boer, 1999). In short, the high hierarchy allows for some efficiency in the application of external knowledge in more specific results, but the other organizational processes for the participation of the employees in innovation and the strong internal relations favor the scope and flexibility necessary for the ACAP 3.