Unveiling the Microfoundations of Absorptive Capacity: A Study of Coleman’s Bathtub Model

Baseline Hypothesis

Even though this study focuses on multilevel effects, the organization-level relationship between integration mechanisms and absorptive capacity is initially considered as a baseline hypothesis, which is unpacked by further hypotheses concerning cross-level and microlevel effects. Regarding potential absorptive capacity, many firms have established formal integration mechanisms, such as liaison committees and interdepartmental task forces, to enhance lateral communication and reciprocal information processing, thereby overcoming differences and enabling a better understanding of novel knowledge from external sources (Gilbert, 2006; Jansen et al., 2005). In addition, Henderson (1994), for instance, showed how pharmaceutical firms used informal mechanisms, such as social networks, to explore new external technologies. By relying on the informal relationships among experts across different organizational units, these firms integrated a broad array of disciplines to make novel drug discoveries. While informal integration mechanisms maintain more flexibility in knowledge processes and are thus helpful in acquiring new knowledge (Burgers, Jansen, Van den Bosch, & Volberda, 2009), “formal mechanisms have the advantage of being more systematic” to ease the identification and interpretation of new trends (Zahra & George, 2002: 194).

Concerning realized absorptive capacity, firms use formal mechanisms, such as cross-functional teams, to integrate and combine diverse expertise coming from different functional areas, such as research and development (R&D) and marketing, and to foster the application of knowledge in new processes and products (Ordanini, Rubera, & Sala, 2008; Verona & Ravasi, 2003). Firms also rely on informal mechanisms to encourage trust and cooperation among different units, thus reducing conflicts regarding goals and interests and augmenting efficient knowledge exchange and implementation (Burgers et al., 2009; Jansen et al., 2005). Moreover, using informal means, such as personal and open communication, improves the richness of communication channels (Daft & Lengel, 1986). According to Hansen (1999), strong social relations within a firm are most beneficial when transferring and combining complex knowledge. Taken together, formal and informal integration mechanisms contribute to both components of absorptive capacity. Hence, I assume,

Role of Perspective Taking and Creative Behavior

In addition to their generally postulated impact on absorptive capacity, integration mechanisms may directly affect the conditions of individuals’ behavior, such as the cognitive process associated with perspective taking. Although perspective taking is often assumed to be a stable disposition, which an individual possesses either by nature or by development (e.g., Davis, 1983), it is also widely acknowledged that the process of taking another’s perspective depends on how a specific situation is cognitively assessed. Thus, to a great degree, the process is contextually malleable by, for example, organizational structures influencing the specific contexts and situations that employees face (Litchfield & Gentry, 2010; Parker, Atkins, & Axtell, 2008; Parker & Axtell, 2001).

Accordingly, formal integration mechanisms may determine the development of perspective taking among a firm’s knowledge workers, as these mechanisms expose employees to diverse perspectives and increase their perception of expertise in other functional units within the organization (Jansen et al., 2005). Through these mechanisms, employees develop an understanding of how their job is related to other functions or departments and how it corresponds to the organization as a whole. Such an integrated job understanding increases the likelihood of taking another’s perspective (Parker & Axtell, 2001). Perspective taking among people with different functional backgrounds is facilitated by providing them with formal forums where mutual learning can occur (Mohrman, Gibson, & Mohrman, 2001), as illustrated by one of my interviews ¹ :

What motivates our people here are our so-called employee information events. This is where R&D, construction, and sales employees regularly showcase their topics to each other to implement a broad awareness of new insights. (managing director, automation and robotics firm)

Formal mechanisms might be useful not only in understanding colleagues’ perspectives but also in developing insights into the perspectives of external people with whom colleagues interact. For instance, by taking the perspective of a salesperson, an engineer may also internalize the views of a customer to consider customer needs when designing a product (Dougherty, 1992). In addition, when a firm stresses informal mechanisms, such as open communication and frequent social interaction in its operations, employees build more interpersonal familiarity and personal affinity and are thus more likely to adopt another’s viewpoint (Parker & Axtell, 2001; Sethi & Nicholson, 2001). Such means can even influence employees’ attitudes toward external views, as explained by an interviewee:

We maintain a culture of open communication regardless of hierarchy and function. That’s how people become more open-minded, also regarding things from the outside. Employees shouldn’t feel they work in a high-security wing; if they do, they are more likely to believe they are an elite bunch and refuse external views. (managing director, automation and robotics firm)

In sum, formal and informal integration mechanisms are potential drivers to develop perspective taking among a firm’s knowledge workers. Thus, I suggest,

Cognitive processes are often the basis for tangible actions (Kaplan, 2011). Concerning perspective taking, studies into the construct’s behavioral outcomes showed that taking the viewpoint of another person fosters socially integrative behaviors (Galinsky & Moskowitz, 2000; Parker & Axtell, 2001). In this regard, perspective taking has been suggested to overcome interpretive barriers to successful knowledge integration and innovation caused by different thought worlds existing inside and outside the organization (Dougherty, 1992; Litchfield & Gentry, 2010). This implies that perspective taking among knowledge workers may stimulate their creativity (cf. Hoever et al., 2012).

Specifically, perspective taking addresses the two conditions that creativity must fulfill per its definition: the creation of novel ideas and the creation of useful ideas (Oldham & Cummings, 1996). With regard to the novelty of ideas, considering the perspectives of others stimulates people in the production of new ideas, as they are more able to combine, build on, and experiment with different viewpoints (Perry-Smith & Shalley, 2003). It also enhances their divergent thinking abilities (Ford, 1996): Seeing problems from others’ perspectives enables people to ask new questions, apply unusual interpretations, identify nonobvious linkages, and create many alternative solutions to open problems. This way of taking different perspectives is also exemplified in the interviews:

Our knowledge workers often look far into other fields that might have analogies with our business. When you develop endoscopic devices, you can look at how, for example, users of microtechnology see and do things, or you can look at how watchmakers work . . . to get ideas out. (chief executive officer, medical technology firm)

Concerning the usefulness of ideas, taking other persons’ perspectives may enable knowledge workers to transform novel ideas into ideas that are useful (Grant & Berry, 2011). After having generated several novel ideas, employees must select those that are most valuable and practical to others (Woodman, Sawyer, & Griffin, 1993). By taking numerous and different views into account, employees develop a detailed understanding of the ideas and whether different stakeholders and colleagues would regard the ideas as useful (Amabile, 1996). An interviewee highlights how important this is:

Some of our user groups, for example, the dentists and [ear, nose, and throat] physicians, are very reluctant to adopt innovations. Therefore, we need to better understand how these users tick to figure out whether they would see value in a novelty at all. (chief executive officer, medical technology firm)

Thus, perspective taking may serve as a filter for determining the utility of an idea (Boland & Tenkasi, 1995; Grant & Berry, 2011). Moreover, perspective taking also facilitates a more constructive appraisal of others’ ideas, thus fostering a reciprocal elaboration of one another’s ideas to attain the highest possible usefulness (Hoever et al., 2012). Taken together, I expect,

Relying on the two conditions that creative behavior must fulfill per its definition, knowledge workers’ ability to create novel ideas may be linked to potential absorptive capacity, while their ability to create useful ideas may be related to realized absorptive capacity. As creativity involves the ability to think divergently (Ford, 1996), it contributes to potential absorptive capacity when employees make new associations and connect seemingly different external or internal information and elements that were once isolated (Amabile, 1996). By thinking outside the box, employees come up with multiple, entirely new ideas that may represent new solutions for problems or constitute potential business opportunities for their organization (Gaglio & Katz, 2001). The creative process is further reinforced because employees seek additional information to increase their understanding of the new ideas generated (Tang, Kacmar, & Busenitz, 2012). By ascribing meaning to new associations and relating them to previously held knowledge (Baron, 2006), employees improve the comprehension of new ideas for their organization, which lies at the core of a firm’s assimilation capacity. One of the interviewees echoes this approach of generating and understanding new ideas:

With our R&D colleagues, we often ask, for example, “Could a certain substance be applied to entirely different medical indications than it is designed for?” If you ask such trigger questions, you come up with very interesting thoughts that wouldn’t have sprung to your mind spontaneously. Then we gather background information to get a better feeling of whether a new idea represents a real opportunity for the firm. (head of R&D, biopharmaceutical firm)

Regarding realized absorptive capacity, the emphasis is on the usefulness of new ideas for the organization and its stakeholders. While divergent thinking is crucial for the creation of a large number of novel ideas, it is not the key when it comes to the practicability of an idea (Woodman et al., 1993). Here, the ability to think convergently by relying on facts takes center stage in evaluating which new idea is the most valuable and should be implemented (Basadur, Graen, & Green, 1982; Reiter-Palmon & Illies, 2004). Organizational transformation of knowledge can be achieved through employees’ mutual creative act of reflective reframing. By questioning one another’s original ideas and shifting the perception to new aspects of the problems to be solved, employees give adjusted or new meaning to original ideas, thus making them more appropriate for subsequent implementation (Hargadon & Bechky, 2006), as reflected in one of my interviews:

When our developers have a new idea for a software application, they need to anticipate: Can a certain customer problem be solved exactly like this, or is this only the tip of the iceberg? At this point, it is important to explore all potential possibilities and check what is already known. . . . Often we need to refine an initial solution to avoid making a value proposition that the firm later cannot fulfill in an economically reasonable way. (chief executive officer, software firm)

For successful knowledge exploitation, employees’ creative problem solving is necessary to consider possible obstacles to implementing ideas and matching the requirements for their application in new products or processes (Reiter-Palmon & Illies, 2004). Moreover, creative employees want their ideas to pay off eventually and, thus, also extensively engage in the realization of their ideas, for example, by overcoming resistance to them (Sternberg, 2006). To sum up this reasoning, I propose,

Mediation Hypothesis

In addition to the single effects hypothesized so far, potential indirect relationships among the variables proposed must be taken into account to provide further arguments for the overall theoretical model of this study. For instance, integration mechanisms may only affect creative behavior through perspective taking. Even if employees are given the opportunity to share knowledge by establishing integration mechanisms within the firm, idea creation is not automatically ensured (Hoever et al., 2012). Clashes among perspectives may impede knowledge sharing and thus limit creative outcomes (Dougherty, 1992), as one of the interviewees explains:

We cannot just put our electronics developers in the field of optics—these areas are different worlds. But if the electronics and optics developers don’t understand each other’s needs, they won’t be able to pass knowledge relevant for the respective others. (chief executive officer, medical technology firm)

Perspective taking can avoid these problems by reducing the psychological isolation of people with different views and knowledge (Litchfield & Gentry, 2010) and by affecting how information to be exchanged is framed (Boland & Tenkasi, 1995).

Furthermore, perspective taking may only affect absorptive capacity through creative behavior. Linking cognitive microfoundations such as perspective taking to tangible behaviors is essential, because without such relationship it remains unclear how and why cognitive processes lead to the formation of a firm capability (Litchfield & Gentry, 2010). It is through the employees’ concrete behavior of producing new and valuable ideas that employees’ cognitive conditions determine a firm’s ability to absorb and leverage new knowledge (Lane et al., 2006). In sum, to the extent that a firm establishes integration mechanisms to coordinate its activities across and within its different organizational units, knowledge workers’ perception both of others’ knowledge and of different perspectives is encouraged. Assuming that knowledge workers also engage in perspective taking, their creative behavior will be leveraged as a consequence, and in turn, the firm’s absorptive capacity will be enhanced (Cohen & Levinthal, 1990; Lane et al., 2006). Hence, I conclude,

Sample and Data Collection

As part of an exploratory prestudy, I conducted interviews with chief executive officers and senior innovation managers in 12 German high-tech firms. These interviews increased my understanding of which employees are involved in activities related to knowledge absorption and leveraging and how they are influenced by the organizational context. To test the study’s hypotheses, a survey among firms from the German medical technology industry was set up. This industry setting was chosen because of its short product lifecycles and high innovation rate. Moreover, the heterogeneity in technologies and product categories across firms in this sector allows for capturing sufficient variance in their absorptive capacity. To account for the multilevel design of my analysis and to limit common method variance in examining top-down and bottom-up relationships (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003), I collected data from multiple informants at two levels of analysis. For this purpose, I administered two questionnaires: one with a focus on organization-level constructs and the other with a focus on individual-level constructs.

For the organizational level, I first approached one key informant who had a detailed understanding of the firm’s organizational structures and mechanisms as well as its knowledge-related capabilities. The exploratory interviews revealed that this was a member of the top management or another senior employee with a long firm tenure. For the individual level, I applied a procedure similar to that of Smith et al. (2005) and asked the first informant to identify two or three core knowledge workers who are critical to the firm’s innovation activities and work closely with other knowledge workers. To address the multilevel structure of my examination, I employed multilevel structural equation modeling (MSEM), for which collecting fewer microlevel entities for the benefit of collecting more macrolevel entities is recommended (Preacher, Zyphur, & Zhang, 2010). Therefore, I concentrated on increasing the number of firms instead of the number of knowledge workers per firm to ensure satisfactory performance of the estimation methods.

The sampling frame consisted of 407 medical technology firms; it was derived from the Creditreform database and checked against the member lists of the main industry associations for medical technology in Germany. ² Of these firms, 152 participated in the survey, yielding a response rate of 37%. Specifically, I obtained 148 questionnaires pertaining to the organizational level and 267 pertaining to the individual level. Four individual-level respondents did not directly qualify as core knowledge workers. To be consistent with the study’s research design, I did not consider them for further analyses. I received the minimum required number of respondents for each level for 106 firms. Thus, regarding the organizational level, the final sample consisted of 106 key informants, mainly top and senior managers with an average firm tenure of 11.8 years. For the individual level, the final sample was composed of 236 core knowledge workers, corresponding to 82 firms with two core knowledge workers and 24 firms with three. Most core knowledge workers (i.e., 72%) had an R&D function, while the remaining held positions in marketing (14%), product management (4%), or another functional area (10%). Knowledge workers had, on average, 9.4 years of firm experience.

The primary survey data were supplemented with secondary data for firm size, firm age, and industry segments collected from company databases of the Bureau van Dijk and Hoppenstedt and other publicly available sources. The firms of the final sample had a median age of 34.5 years since founding (mean = 55, SD = 46.7) and median size of 216 employees (mean = 1261.9, SD = 5,249), with 70% being medium sized, ranging from 50 to 500 employees. Thus, they were large enough to have established more formal organizational mechanisms. However, they were also small enough to ensure that the responding core knowledge workers had reliable insights into the characteristics of their colleagues since the group of these employees was quite small versus the total number of employees. Tests for nonresponse and late-response bias showed no significant results.

Measures

To measure the different organization- and individual-level constructs, I adapted multi-item scales from the extant literature (see appendix) that were based on a 7-point Likert format in which 1 = strongly disagree and 7 = strongly agree. The constructs’ reliability was examined by following the procedures of Geldhof, Preacher, and Zyphur (2014) for calculating the Cronbach’s alpha consistency coefficient (α) and the composite reliability coefficient (ω) based on multilevel confirmatory factor analysis (CFA) via the software Mplus 7 (L. K. Muthén & Muthén, 1998–2015).

Specifically, as the individual-level constructs are based on several individual ratings of core knowledge workers nested within firms and thus have variance between and within firms, the reliability coefficients for these constructs were computed at both the between- and within-firm levels. For the organization-level constructs, however, the reliability coefficients were computed only at the between-firm level because these constructs are based on one rating per firm and thus have only between-firm variance. For α and ω, values ≥0.70 suggest good reliability (Hair, Black, Babin, & Anderson, 2010), which I found for all constructs, as reported in this section. CFAs conducted to evaluate construct validity are presented in the subsequent section.

Confirmatory Factor Analyses

To assess construct distinctiveness, I performed a multilevel CFA in Mplus with all the main measures. Specifically, I loaded all individual-level items on their respective constructs at the within- and between-firm levels and all organization-level items on their respective constructs at the between-firm level, resulting in a large measurement model with 10 factors. I used bayesian estimation, which incorporates prior distributions of the estimated parameters and which thus enabled me to fit this more complex model that would yield convergence problems and improper estimates via conventional maximum likelihood (ML) estimation given my limited sample size (Johnson, Van De Schoot, Delmar, & Crano, 2015). Moreover, with a Bayes estimator, I was able to apply B. Muthén and Asparouhov’s (2012) bayesian structural equation modeling procedure to detect potential model misspecifications by freeing all residual covariances that would, however, lead to a nonidentified model within an ML framework.

In particular, I followed Asparouhov, Muthén, and Morin’s (2015) stepwise approach. First, I estimated the hypothesized CFA model with uninformative priors for the factor loadings but without any residual covariances. This model was rejected on the basis of assessing model fit with posterior predictive checking (Zyphur & Oswald, 2015): The posterior predictive p value (PPP) was <.05 (i.e., PPP = .01), and the corresponding 95% confidence interval [9.55, 200.10] of the difference between the observed and replicated χ² values had a positive lower limit, with both indicating poor model fit. ³ Next, I added informative inverse Wishart priors for all possible residual covariances and conducted a sensitivity analysis to examine the impact of varying these priors with different degrees of freedom on convergence speed and model fit (Asparouhov et al., 2015). Finally, on the basis of the model with the fastest convergence and excellent model fit, I identified two large isolated residual correlations that likely occurred due to parallel or similar item wording: The one between two indicators pertaining to creative behavior at the within-firm level and the other between two indicators pertaining to transformation. ⁴

Consequently, only these two residual correlations were included in the original model with uninformative priors to improve model fit and the estimates’ accuracy (cf. Asparouhov et al., 2015). Appendix A shows the results of this modified CFA model, which had an acceptable fit (PPP = .10; 95% confidence interval [–31.03, 147.87]). Regarding convergent validity, the standardized factor loadings were all significant and above the minimum value of 0.50, and the factors’ average variances extracted all exceeded the recommended cutoff of 0.50 (Hair et al., 2010). Regarding discriminant validity, as illustrated in Appendix B, the factors’ correlations are not too high (i.e., well below 0.90; Kline, 2011). Moreover, each factor’s average variance extracted was larger than the squared value of the correlations of this factor to other factors (Fornell & Larcker, 1981).

In addition, I conducted smaller CFAs with a robust ML estimator to further establish the validity of the constructs’ factor structure in separate models. First, a multilevel CFA model with two correlated factors at the between and within levels reflecting the items of perspective taking and creative behavior, respectively, demonstrated a much better fit (χ²[39] = 66.48, comparative fit index [CFI] = 0.98, Tucker-Lewis index [TLI] = 0.98, root mean square error of approximation [RMSEA] = 0.06, standardized root mean square residual–between [SRMR_between] = 0.04, SRMR_within = 0.03) than a model in which all these items were loaded on a common factor at each level (χ²[41] = 355.10, CFI = 0.82, TLI = 0.75, RMSEA = 0.18, SRMR_between = 0.13, SRMR_within = 0.17).

Second, as all items of the integration mechanisms scales jointly demonstrated a high degree of internal consistency (α = 0.88), I ran a CFA model with one second-order factor that consisted of two first-order factors corresponding to formal and informal integration. ⁵ While this model fit the data well (χ²[19] = 28.85, CFI = 0.97, TLI = 0.96, RMSEA = 0.07, SRMR = 0.04), a model with integration mechanisms as a first-order factor with all items loaded to only this factor indicated a poor fit (χ²[20] = 130.08, CFI = 0.68, TLI = 0.55, RMSEA = 0.23, SRMR = 0.12). Third, to reflect the multidimensionality of absorptive capacity and the complementarity of its dimensions (Zahra & George, 2002), I performed a CFA model with absorptive capacity as a second-order factor containing four first-order factors pertaining to the four dimensions. ⁶ This model had a good fit (χ²[72] = 87.69, CFI = 0.97, TLI = 0.97, RMSEA = 0.05, SRMR = 0.06) and fit the data much better than a model with absorptive capacity as a first-order factor with all items treated as separate indicators (χ²[76] = 215.14, CFI = 0.77, TLI = 0.72, RMSEA = 0.13, SRMR = 0.09). As a result, I treated absorptive capacity and integration mechanisms as second-order constructs in the main analyses but additionally considered their single dimensions individually in robustness analyses.

Analytical Procedures

To test my hypotheses, I employed MSEM using Mplus 7 (L. K. Muthén & Muthén, 1998–2015) and followed Preacher and colleagues’ (2010) recommendations for modeling multilevel mediation. The intraclass correlation coefficients (ICCs) of the individual-level measures further indicated the adequacy of this approach (i.e., all ICCs >0.05). MSEM partitions the variance of variables measured at the individual level into a within-level element (within-firm variance) and a between-level element (between-firm variance). Specifically, such variables can be modeled with intercepts, which are permitted to differ across firms. These intercepts are defined as latent variables at the between level with individual respondents of each firm acting as indicators (cf. Nohe, Michaelis, Menges, Zhang, & Sonntag, 2013). Thus, relationships between individual-level variables can be specified as between- and within-level effects independently and simultaneously in one model, thereby avoiding the conflation of between and within parts that usually occurs in traditional multilevel linear modeling (Preacher et al., 2010).

For my analysis, these methodological features are particularly suitable because they allow for the specification of a bottom-up effect by relating the latent between-level component of individual-level variables to absorptive capacity treated as an upper-level outcome. Moreover, the mediation effect—also the indirect effect—can be computed more precisely as a pure between-level effect because two variables of my overall mediation model (i.e., integration mechanisms and absorptive capacity) vary only at the between level (i.e., between firms; Nohe et al., 2013). However, because of the multidimensionality of the organization-level constructs, the specification of several pathways at two levels, and the inclusion of many control variables, the number of parameters to be estimated versus the number of observations is quite high in my analysis (Kline, 2011). Thus, to minimize the number of free parameters, I computed the constructs’ arithmetic means to be used in multilevel path analysis as a special form of MSEM with only a structural model but no measurement model (cf. Den Hartog, Verburg, & Croon, 2013; Sun, Zhang, Qi, & Chen, 2012).

I applied bayesian estimation because it offers several advantages over conventional ML estimation that are suitable for the analysis of my multilevel mediation model (for an overview of advantages of bayesian over frequentist analysis, see Kruschke, Aguinis, & Joo, 2012). First, bayesian analysis relies on a prior distribution for each estimated parameter that is combined with the observed data to generate a posterior distribution. Thus, parameters are not treated as fixed values but as random variables, which offers a conceptually more natural analysis of multilevel mediation models (Yuan & MacKinnon, 2009). Second, bayesian statistics are not reliant on the restrictive normality assumption and large-sample approximation of conventional ML analysis (B. Muthén & Asparouhov, 2012). Thus, with bayesian analysis, the estimation of the indirect effect, which is not normally distributed, is more appropriate, and the resulting inference is exact even for limited sample sizes (Yuan & MacKinnon, 2009). Third, through the generation of a posterior distribution, an automatic bootstrapping of the indirect effect is obtained, which avoids the application of (computationally more expensive) parametric bootstrapping or Monte Carlo procedures that would be required under an ML approach (Yuan & MacKinnon, 2009). ⁷

I employed uninformative priors to ensure that all parameter values were equally probable a priori. Thereby, I mimicked frequentist estimation and applied a (null) hypothesis testing view but avoided the limitations of conventional ML estimation (Zyphur & Oswald, 2015). As introduced for CFA, I evaluated model fit based on the PPP value and associated 95% confidence interval. I reported the median of the posterior distribution as a parameter estimate with the respective posterior standard deviation (SD). As a test of significance, I indicated the one-tailed bayesian p value referring to the proportion of the posterior distribution below zero if the estimate is positive or above zero if the estimate is negative. I also assessed whether the 90% or 95% credibility interval (CI) of the posterior distribution of a parameter comprises zero (Zyphur & Oswald, 2015).

Test of the Hypotheses

I specified one two-level mediation model in which all direct and indirect pathways were estimated simultaneously (cf. Den Hartog et al., 2013). To account for the direct pathway, absorptive capacity was regressed on integration mechanisms. Concerning indirect pathways, perspective taking was regressed on integration mechanisms, creative behavior on perspective taking, and absorptive capacity on creative behavior. All these regressions were specified at the between level and included all control variables. In addition, the regression of creative behavior on perspective taking was modeled at the within level and included only the individual-level controls. Unlike ML estimation, individual-level variables, which are used as predictors only, are not automatically decomposed into their within- and between-variance components with bayesian estimation in Mplus. Thus, I followed Zhang, Zyphur, and Preacher’s (2009) recommendation and entered the individual-level controls as group-mean centered variables at the within level to retain only their within-variance part. ⁸ At the between level, however, I considered between-firm variance by entering the cluster means of these variables.

Figure 2 illustrates the (standardized) results of the hypothesized two-level path model based on 10,000 iterations. With a PPP value of .32 and a 95% confidence interval of the χ² difference ranging between −29.09 and 47.83, this model fits the data well. The variance inflation factors for the predictor variables range from 1.07 to 3.92 and are well below the cutoff value of 10 (Kline, 2011). Regarding the effects of integration mechanisms, Hypotheses 1 and 2 are supported: Integration mechanisms are positively and significantly related to absorptive capacity (0.37, SD = 0.12, p = .012, 95% CI [0.08, 0.55]) and perspective taking (0.27, SD = 0.12, p = .015, 95% CI [0.02, 0.52]). Hypothesis 3 also is strongly supported because the relationship between perspective taking and creative behavior is positive and significant at the between level (0.74, SD = 0.22, p = .000, 95% CI [0.34, 1.19]) and the within level (0.49, SD = 0.06, p = .000, 95% CI [0.36, 0.60]). The findings also confirm Hypothesis 4, with creative behavior positively and significantly associated with absorptive capacity (0.74, SD = 0.27, p = .000, 95% CI [0.34, 1.40]).

OPEN IN VIEWER

Figure 2

Estimated Two-Level Path Model

With respect to Hypothesis 5, assuming a sequential mediation between integration mechanisms and absorptive capacity through perspective taking and creative behavior, I find the unstandardized parameter estimate of the indirect effect, computed as the product term of the respective between-level coefficients, to be positive (0.14, SD = 0.11, p = .016). The corresponding 95% CI [0.01, 0.45] with equal tail percentages—the default for bayesian estimation in Mplus—does not include zero. As an additional check, I computed CIs of the indirect effect’s posterior distribution that indicate the highest posterior density region (Gelman, Carlin, Stern, & Rubin, 2004). Here, only the 90% CI [0.01, 0.32] does not include zero. However, for testing mediation hypotheses, it is often justified to use 90% CIs to refer to one-tailed significance tests at alpha = 0.05 because they reflect directional hypotheses (Preacher et al., 2010; Sun et al., 2012). Thus, in sum, I find evidence supporting Hypothesis 5. To examine whether full or partial mediation is present, I specified an alternative model without the direct path between integration mechanisms and absorptive capacity. The model fit did not improve but rather very slightly decreased (PPP = .31, 95% confidence interval of the χ² difference [–24.89, 48.16]), and because the direct path between integration mechanisms and absorptive capacity in the original model is significant, a partial mediation can be concluded.

Robustness Checks

Several additional analyses were conducted to establish the robustness of the results. First, I tested two alternative path models by specifying the relationship between integration mechanisms and absorptive capacity as being mediated by either only perspective taking (Alternative Model A) or only creative behavior (Alternative Model B). As shown in Table 2, when compared with the hypothesized model, both alternative models are worse in terms of model fit, as indicated by relatively lower PPP values, and the 95% CIs of the alternative indirect effects did include zero. ⁹ These results suggest that alternative pathways are less probable and provide additional evidence supporting Hypothesis 5.

OPEN IN VIEWER

Table 2

Indirect Effects and Model Fit

Two-level path models	Indirect effect				Model fit
	Estimate	Posterior SD	One-tailed p	95% credibility interval	PPP	95% confidence interval
Hypothesized model
Integration mechanisms → perspective taking → creative behavior → absorptive capacity	0.14	0.11	.016	[0.01, 0.45]	.32	[−29.09, 47.83]
Alternative Model A
Integration mechanisms → perspective taking → absorptive capacity	0.19	0.28	.037	[−0.02, 0.94]	.27	[−20.70, 39.27]
Alternative Model B
Integration mechanisms → creative behavior → absorptive capacity	0.18	0.17	.064	[−0.07, 0.60]	.27	[−20.83, 38.80]

Note: Estimates and credibility intervals for the indirect effects are based on the unstandardized path coefficients of the respective mediation chain. Only the indirect effect pertaining to the hypothesized model is significant at the 5% level because, in contrast to the alternative models, the corresponding 95% credibility interval does not include zero. PPP = posterior predictive p value.

Second, the findings of three separate multilevel regression models with conventional ML estimation and reflecting only each direct relationship with all control variables further supported Hypotheses 1 through 4. Third, I reran the bayesian analysis with weakly informative priors to ensure that my results were not affected by the choice of the uninformative priors. ¹⁰ Fourth, I reestimated the two-level path model and the three separate multilevel models with balanced clusters of only two knowledge workers per firm to rule out any potential bias caused by unbalanced data (Preacher et al., 2010). Fifth, to obtain top and senior managers’ assessment of their firms’ knowledge workers, the questionnaire for the first informants also asked them to rate those employees’ perspective taking and creative behavior. ¹¹ The results of ordinary least squares regressions with data from all 148 first informants were consistent with those obtained from multilevel regressions.

Sixth, I found that empirically distinguishing among the single dimensions of absorptive capacity as well as combining the corresponding dimensions to form potential and realized absorptive capacity, respectively, yielded results similar to those of the main analysis in terms of direct and indirect relationships—though in some cases at only the 90% confidence level. However, when considering only the indirect effect of one subdimension of integration mechanisms while controlling for the other, I found that the respective coefficient lost significance. When both subdimensions were inserted as separate predictors of absorptive capacity and perspective taking in single multilevel regressions, the size of the individual effects was smaller (and in some cases significant at lower levels or not significant at all) than when they were treated as a combined construct of integration mechanisms.

Seventh, as opposed to relationships across levels, within levels the chance for common method bias might be higher because the main variables were taken from the same informants. However, when I loaded all items from the same data source on a single factor, the resulting models demonstrated poor fit. Finally, my findings are potentially vulnerable to endogeneity due to the cross-sectional study design. Although I did not find suitable instruments in the context of my analysis to rule out this issue, I sought to strengthen the inferences by including appropriate controls to alleviate worries regarding omitted variable bias (Antonakis, Bendahan, Jacquart, & Lalive, 2010). For instance, while I attempted to account for unobserved influences specific to a firm’s particular segment by incorporating medical technology segment dummies, I controlled for the influence of characteristics specific to the responding knowledge workers on their perception of themselves and their colleagues.

Implications for Research

My findings have noteworthy implications for research into the antecedents of absorptive capacity, the microfoundations of organizational capabilities, and the role of perspective taking for knowledge integration. With regard to absorptive capacity research, this study addresses three essential gaps concerning the antecedents of absorptive capacity spotted in a previous literature analysis (Volberda et al., 2010). First, I explain what and how individual-level factors influence organization-level absorptive capacity. I reveal that perspective taking is an important cognitive, situationally motivated prerequisite for identifying and processing diverse internal and external knowledge. Influenced by this cognitive process, subsequent creative behavior is identified as another critical individual antecedent because it helps a firm in creating value from newly acquired knowledge. Thereby, my findings revitalize the conceptual roots of Cohen and Levinthal’s (1990) foundational work that introduced absorptive capacity as an organizational counterpart to the psychology-based creativity and cognition constructs.

Second, I address the neglect of studying the relative contributions of formal and informal structures to absorptive capacity in prior work (Volberda et al., 2010). Since I find formal and informal integration to be highly correlated and the combined measurement of integration mechanisms directly and indirectly (via the micro level) related to absorptive capacity, it seems that the two types of integration are complements rather than substitutes in influencing absorptive capacity (cf. Gulati & Puranam, 2009). This implication is further corroborated by the post hoc robustness analyses, which reveal that the two types’ combined effects exhibit larger sizes and lower significance levels than their separate effects. Third, I enhance the understanding of the interplay between organizational and individual antecedents (Lane et al., 2006; Volberda et al., 2010). My findings suggest that a firm’s absorptive capacity is not just the sum of its employees’ cognitions and behaviors; it is also contingent on the organizational mechanisms by which individual contributions are integrated to form a collective outcome (cf. Gupta, Tesluk, & Taylor, 2007). Accordingly, these findings highlight the need to conduct more multilevel studies on absorptive capacity because an isolated analysis of only one level may lead to erroneous results (cf. Dansereau, Yammarino, & Kohles, 1999).

With regard to microfoundations research, the findings indicate that differences in the characteristics of knowledge workers among firms explain differences in firm capabilities. Thus, the study confirms the theoretical consideration of influential microfoundations research suggesting that individuals cannot be assumed to be homogeneous across organizations (Felin & Foss, 2005; Felin & Hesterly, 2007), and it emphasizes the critical role that key employees play in the formation of an organizational capability (Gavetti, 2005). By using data collected at two levels of analysis, following the referent-shift model in capturing the attributes of a specific group of individuals, and applying a recently developed method (MSEM) allowing for modeling bottom-up effects, the study tackles the challenges of empirically accommodating Coleman’s (1990) bathtub framework.

Specifically, I provide empirical corroboration to explain relationships between organizational antecedents and capabilities in terms of a sequential mediation of individual conditions and individual actions when theorizing about microfoundations (Abell et al., 2008). The comparison of the results of different models indicates much stronger empirical evidence for a four-path model according to Coleman’s logic than alternative three-path models, including either only individual condition or only individual action. The empirical approach undertaken in this study meets recent desires to bridge micro and macro levels in management research (Aguinis, Boyd, Pierce, & Short, 2011; Molloy, Ployhart, & Wright, 2011), and it may be valuable for future quantitative studies in further advancing the microfoundations movement (Coff & Kryscynski, 2011; Felin et al., 2015).

With regard to research on the role of perspective taking for knowledge integration, this study complements prior work proposing that individual perspective taking can be scaled to firm capability, helping to combine apparently incongruous information (Litchfield & Gentry, 2010). As such, perspective taking should foster knowledge integration processes related to transformation as one particular dimension of absorptive capacity. I greatly extend and refine this view in two ways. On one hand, my empirical findings suggest that perspective taking is a microfoundation of an organizational capability related to knowledge integration rather than an organizational capability in itself. I find that perspective taking needs to be expressed in tangible actions in the form of generating creative ideas before it can contribute to knowledge integration at the organizational level.

On the other, I suggest that in addition to transformation, perspective taking indirectly affects the other three dimensions of absorptive capacity through creative behavior. Perspective taking may be important for the acquisition and assimilation of new knowledge because it stimulates people to explore more broadly and give meaning to new associations. Perspective taking may also be crucial for the exploitation of knowledge because it is directed to attune to the needs of others and thus helps in implementing useful ideas. Moreover, this study underscores that integration mechanisms are important determinants of perspective taking, implying that perspective taking is a motivated cognition and to some extent malleable (cf. Boland & Tenkasi, 1995; Litchfield & Gentry, 2010; Parker & Axtell, 2001).

Implications for Practice

From a managerial perspective, my findings imply that firms may not directly act on organizational capabilities such as absorptive capacity (cf. Foss, 2011). Rather, they may form capabilities more indirectly by establishing certain organizational mechanisms that affect the conditions of employees proven to favor a certain behavior or by hiring key employees with the required traits. For building and maintaining absorptive capacity, firms need knowledge workers who are highly capable of taking different perspectives and exhibit a high degree of creative behavior. This need is even more apparent in a digital era in which industry boundaries are blurry and new knowledge continuously originates from multiple sources, as seen, for example, in development of the self-driving electric car that involves technologies from several industries (e.g., car manufacturing, electronics, information technology).

In such environments, firms can hardly sustain high levels of specialized knowledge stocks in all potentially relevant fields. Rather, firms require employees who can empathize with new viewpoints and different knowledge perspectives to flexibly integrate novel insights from various stakeholders. As routine work is increasingly being taken over by machines and algorithms, a firm may defend its competitive advantage based only on its knowledge workers’ creativity in combining and leveraging new knowledge—a characteristic that (still) differentiates human beings from machines and computers. In addition to establishing integration mechanisms, firms can employ human-centered innovation methods, such as design thinking, to nurture perspective taking and creative behavior. Design thinking embraces tools promoting empathy, ideation, and prototyping, thereby reducing cognitive biases among knowledge workers and helping them find more novel and valuable solutions (Liedtka, 2015).

Limitations and Outlook

This study has several limitations that may open avenues for future research. First, the cross-sectional setting does not allow for making causal assertions. Although the directions of the hypothesized relationships are theoretically well underpinned and based on a thorough application of Coleman’s model, alternative interpretations of the results may exist. For instance, a firm with a strong capacity to absorb external knowledge might provide its knowledge workers with the necessary stimuli to come up with ideas explicitly valuable to external partners (e.g., customers). Future research should econometrically prove the causal directions of the relationships and (completely) rule out potential endogeneity issues by employing appropriate instrument variables, a longitudinal design, and/or (quasi)experimentation. Second, the findings of the two-level path model show that perspective taking and creative behavior only partially mediate the relationship between integration mechanisms and absorptive capacity. This might indicate that perspective taking cannot be perfectly managed by organizational determinants but is to some degree a stable disposition of an individual. Moreover, it is possible that the organization-level relationship is additionally mediated by other microlevel variables not covered in this study.

Third, the findings of this study represent the situation of the German medical technology industry. The generalizability of the results to other populations might be queried. Finally, the sample for the micro level may not be representative and may suffer from selection bias, as the core knowledge workers were selected by the first informant and not randomly drawn from a larger population (Felin et al., 2015). Future studies can adopt a more costly approach by sampling all knowledge workers per firm and adopting self-referential measures. Of course, the identification of these employees from outside the firm is a major challenge, which may further justify the sampling procedure undertaken in this study. However, future studies could, for example, identify star scientists through publication and citation databases (Rothaermel & Hess, 2007) or use register data.

To gain a deeper understanding of how firm-level absorptive capacity emerges from individuals’ actions (i.e., bottom-up relationships), future research can explore how these associations are moderated by organizational factors such as organization design and reward systems. Econometrically, such models can be specified by using the possibilities of MSEM. In a similar vein, further opportunities exist to examine antecedents at the project, business unit, and interorganizational levels and how they interact in influencing absorptive capacity. This approach could also be adapted to examine the microfoundations of other organizational capabilities, for example, related to product development, alliances, and acquisitions.