This study aims to propose dual moderating roles of organizational unlearning in shaping the network drivers and the performance outcomes of interfirm knowledge transfer. First, the authors examine the moderating effect of unlearning on two key network drivers, structural network (measured as centrality) and relational network (measured as tie strength), on knowledge transfer. Second, the authors examine the moderating effect of unlearning on the relationship between knowledge transfer and firm performance.
To test the two proposed moderating models, the authors collected a unique data set comprising the entire sociometric network of 42 petroleum wholesalers in Tanzania and analyzed the survey data using moderated regression analyses.
The results support the authors’ hypotheses. Specifically, organizational unlearning weakens the effect of tie strength but strengthens the effect of centrality on knowledge transfer. Moreover, organizational unlearning amplifies the positive effect of knowledge transfer on firm performance.
By theorizing unlearning as a critical boundary condition, the authors show that it shapes when and how network drivers influence knowledge transfer and its performance consequences. In doing so, this study deepens our understanding of the network mechanisms underlying knowledge transfer and reveals the opposing moderating effects of unlearning on network drivers and knowledge transfer, which prior research has largely overlooked.
1. Introduction
Knowledge is a strategic asset vital for firm performance and survival (Grant, 1996; Grant and Baden-Fuller, 2004). Firms typically rely on their own employees to develop new knowledge internally. However, as the pace of competition fastens, firms increasingly leverage their connections within their business network to expand their sources of knowledge and to sustain mutual knowledge exchange (Cheng et al., 2026; Easterby-Smith et al., 2008; Phelps et al., 2012; Uzzi and Gillespie, 2002).
Interfirm knowledge transfer has been identified as an important component of organizational learning (Grant and Baden-Fuller, 2004; Ho and Wang, 2014), leading to innovation, economic growth, competitive advantage and ultimately superior performance (Farooq, 2023; Ferreras-Méndez et al., 2015; Ferrer-Serrano et al., 2021; Liu and Lui, 2020). In this context, social networks facilitate interfirm knowledge transfer (Li et al., 2014; Ngowi et al., 2022; Van Wijk et al., 2008). Recent research has explored network multiplexity as actors simultaneously reside in multiple networks (Ferraris et al., 2020; Gondal, 2022; Shipilov, 2012) and has advanced statistical methods to model complex network and centrality effects (Leung, 2023; Schoch, 2018).
Two dominant network views that offer distinct mechanisms to explain how social network affect knowledge transfer can be discerned: a structural view and a relational view (Ferrer-Serrano et al., 2021; Inkpen and Tsang, 2005; Moran, 2005). The structural view points out that a firm’s centrality within a network increases its opportunity to access knowledge across the network (Borgatti and Everett, 2020; Reck et al., 2021). In contrast, the relational view argues that tie strength between network members motivates the transfer of greater variety and depth of knowledge (Levin and Cross, 2004; Ozdemir et al., 2016; Tortoriello et al., 2012). In other words, network structure enables knowledge transfer by creating conduits for knowledge flows, whereas network ties enable knowledge transfer by motivating relational exchange through enhanced trust and reciprocity. The two mechanisms underlying knowledge transfer, one based on opportunity and the other on motivation, are distinct. However, previous research has not focused on differentiating the two underlying mechanisms. Therefore, the first objective of this study is to delineate the structural and relational arguments underpinning social networks and knowledge transfer.
The concept of organizational unlearning has been introduced to enrich the study of organizational learning and knowledge transfer (Huber, 1991; Tsang and Zahra, 2008; Maccioni et al., 2024; Xu et al., 2024). Organizational unlearning refers to the intentional disposal of obsolete knowledge, beliefs and routines of an organization (Hedberg, 1981). This concept stands as the counterpart of organizational learning, which involves the addition of new knowledge through searching, acquisition, internalization and assimilation (Tsang, 2008, 2017). A debate over whether learning and unlearning are two sides of the same coin (Howells and Scholderer, 2016) or independent but complementary concepts (Tsang, 2017) has persisted even up to date (Acharya and Mishra, 2022; Klammer et al., 2024).
Recent organizational-level empirical studies have gone beyond this debate to focus on organizational unlearning as an antecedent of knowledge acquisition (Wang et al., 2017; Xi et al., 2020), information search (Zhang et al., 2022) and innovation (Leal-Rodríguez et al., 2015; Lyu et al., 2020; Yeniaras et al., 2021; Zhang and Zhu, 2021; Zhang et al., 2024). As learning and unlearning capture interrelated processes of knowledge activities, they ultimately contribute to the absorptive capacity and performance of a firm (Oh and Kim, 2022; Yu et al., 2022). However, previous research has not explored the potential role of unlearning to shape the context of learning and knowledge transfer. Therefore, the second objective of this study is to revisit this fundamental learning-unlearning debate by investigating how organizational unlearning is related to interfirm knowledge transfer.
We integrate theories on interfirm networks, organizational unlearning and knowledge transfer (Hedberg, 1981; Huber, 1991; Inkpen and Tsang, 2005; Tsang, 2017) to propose a research model that posits organizational unlearning as a dual moderator of both the network drivers and performance outcomes of knowledge transfer. The dual moderating perspective highlights unlearning as a critical boundary for knowledge transfer.
Considerable effort was spent to collect a unique sample to test the research model. We surveyed 42 Tanzanian petroleum wholesalers, through which we obtained information covering the whole network of the petroleum wholesaler industry in Tanzania. Whole network data are difficult to collect, and the Tanzanian sample of this study supplements research on emerging markets that has often drawn their samples from Asia. The use of sociometric variable and archival performance record measured in sales volume of petroleum allows us to avoid subjective perception and social desirability in responses often found in surveys (Podsakoff et al., 2012).
This study contributes to the literature on knowledge transfer in two ways. First, we clarify the mechanisms through which structural and relational networks affect knowledge transfer. Although we do not directly observe and test the mechanisms, the proposed moderating effect of unlearning allows us to infer their operation. Second, we delineate the dual moderating roles of organizational unlearning on knowledge transfer. In doing so, we contribute to the enduring debate on organizational learning and unlearning. We posit that organizational unlearning acts as a boundary condition that constrains the network drivers and the performance outcomes of knowledge transfer. This suggests that unlearning could strengthen or mitigate favorable impact of structural and relational network conditions on knowledge transfer and firm performance. For firms to fully capitalize on knowledge transfer, they must also be willing and able to discard obsolete knowledge, thereby creating the space for new knowledge.
2. Theoretical development
Recent reviews on unlearning research have documented key definitions, antecedents and consequences of organizational unlearning at different analytical levels (Kim and Park, 2022; Klammer and Gueldenberg, 2019; Klammer et al., 2024; Kluge, 2023; Sharma and Lenka, 2021, 2022; Zhao et al., 2013). They also revealed a fundamental debate between organizational learning and unlearning that revolves around the controversy in the added value of unlearning on top of learning (Tsang and Zahra, 2008). On the one hand, Huber (1991: 104) has argued that unlearning can be subsumable under learning, and Howells and Scholderer (2016) has criticized the concept of unlearning as misguided. On the other hand, Tsang (2008) has examined how unlearning works in tandem with learning at every stage of the knowledge transfer process in international joint ventures. Tsang (2017) further suggested that unlearning and learning should be studied as two isolated yet complementary processes. Investigating the moderating effect of unlearning in interfirm knowledge transfer in this study addresses this fundamental debate.
We argue that organizational unlearning, as a firm-specific characteristic, may mitigate interfirm knowledge transfer. Research has shown that firm-specific characteristic often plays a role in strengthening or weakening interfirm processes (Easterby-Smith et al., 2008). For example, a firm’s own R&D investment, theorized as absorptive capability, enhances its knowledge acquisition from other firms (Ferreras-Méndez et al., 2015). Similarly, goal clarity, a core feature of a team, shapes the flow of knowledge of the team with other teams (Lai et al., 2016). This points to a potential moderating relationship of unlearning on interfirm knowledge transfer, which has not been fully addressed in previous research.
Specifically, we argue that organizational unlearning plays dual moderating roles on interfirm knowledge transfer: it moderates both the network drivers and the performance consequences of knowledge transfer. While the first moderating role addresses the two distinct network mechanisms underlying knowledge transfer derived from extant network theory, the second addresses the theoretical debate on how learning and unlearning relate to firm performance. Accordingly, two independent regression analyses are conducted to test the research model. The first regressed knowledge transfer on network antecedents. The second regressed firm performance on knowledge transfer.
The first moderating role of unlearning facilitates the conversion of network drivers to knowledge transfer. Factors of knowledge transfer consist of four aspects: knowledge recipients, knowledge donor, the type of knowledge and interfirm dynamics (Easterby-Smith et al., 2008). As interfirm knowledge transfer is fundamentally a relational exchange process between two organizations, both structural and relational network drivers have been identified as part of the interfirm dynamics (Kilduff and Brass, 2010; Reagans and McEvily, 2003). Unlearning may moderate network drivers on knowledge transfer as unlearning empties obsolete knowledge and sets the stage.
The second moderating role of unlearning facilitates the conversion of knowledge transfer to performance outcomes. Knowledge transfer results in an increase in knowledge that leads to behavioral change (Argote and Ingram, 2000). As a change in behavior is difficult to measure, this is often proxied by increased creativity, innovation or subjective performance (e.g. Liu and Lui, 2020). The substantial variation in the performance benefits of knowledge transfer (Levine and Prietula, 2012) points to potential moderators. Unlearning provides space for new interpretative frameworks and responses (Maccioni and Ghiringhelli, 2024). This creates favorable conditions for positive knowledge transfer outcomes (Akgün et al., 2006).
To summarize, network theory explains how relational and structural networks affect knowledge transfer through different mechanisms and how knowledge transfer, in turn, affects performance outcomes. Unlearning theory suggests that unlearning is a process of intentional forgetting that provides a context in which an organization obtains and uses new knowledge from external parties. Integrating network theory and unlearning theory, we propose in this paper that unlearning may differently moderate the effects of relational and structural networks, as well as the performance outcomes of knowledge transfer.
3. Hypothesis development
3.1 Social networks and knowledge transfer
According to the structural view of social networks, interfirm networks provide the conduits for knowledge exchange, and network structure and position predominantly determines the amount of knowledge network members are able to access from the network (Kilduff and Brass, 2010; Phelps et al., 2012; Powell et al., 1996). It is argued that occupying a central network position is associated with preferential access to knowledge sources (Brunetta et al., 2015; Schillebeeckx et al., 2021; Szulanski, 1996), which enhances central firm’s knowledge transfer in all four stages of searching, acquisition, assimilation and integration within the network.
We argue that firms with high centrality have more opportunities for knowledge transfer than those with low centrality. Central firms have higher status; hence, other network members look up to them as exemplars and influencers in their networks (Borgatti and Everett, 2020; Phillips and Zuckerman, 2001). With high status, other network members will offer preferential attachment to the central firm (Bothner et al., 2011). In turn, many network members prefer to associate with and offer more knowledge to prominent members to boost their own reputations (Borgatti, 2005; Chandler et al., 2013). This tendency not only enables central firms to gain an eagle’s-eye view of the resources that other network members possess, thereby easing their knowledge search (Arranz et al., 2020), but also facilitates central firms’ knowledge acquisition and assimilation from other network members (Schillebeeckx et al., 2021). This process initiates a Matthew effect, in which the more central a firm is, the greater the reputational advantages the firm will accumulate (Bothner et al., 2011).
Moreover, central firms have more experience in processing knowledge. This experience reinforces their potential to create and identify opportunities for knowledge exchange within a network (Argote and Ingram, 2000; Inkpen and Tsang, 2005; Uzzi and Gillespie, 2002). By the same token, as opportunities increases, central firms are better able to exploit network resources by efficiently sorting out valuable resources and integrating the preferential knowledge they access from their networks into their internal memories (Caccamo et al., 2022; Reck et al., 2021; Zahra et al., 2020; Zahra and George, 2002).
Three meta-analyses (Li et al., 2014; Ngowi et al., 2022; Van Wijk et al., 2008) have provided consistent empirical evidence that the direct relationship between network centrality and knowledge transfer is primarily positive. Summarizing the arguments above, given that centrality increases the opportunity for a firm to access and handle knowledge, we hypothesize the following:
Firm centrality is positively related to knowledge transfer.
The other network approach, relational view, focuses on how the strength of the social ties with other network firms provides the motivation to transfer knowledge. Tie strength between partnering firms implies a history of repeated interactions and close relationships, which lead to greater trust and reciprocity (Reagans and McEvily, 2003; Tortoriello et al., 2012). Based on the relational view of social networks, strong ties motivate the desire to transfer knowledge (Levin and Cross, 2004; Uzzi and Lancaster, 2003). The core argument of the relational view is that access provided by a favorable network position is a necessary but not sufficient condition for knowledge transfer (Inkpen and Tsang, 2005; Moran, 2005). Strong ties motivate partners to exchange knowledge by encompassing trust and reciprocity norms that minimize opportunism (Cropanzano and Mitchell, 2005; Polidoro et al., 2011). Strong ties also provide safety and reciprocity expectations on knowledge transfer among network members. As a result, firms are more motivated to share valuable knowledge with close rather than arms-length partners (Uzzi and Lancaster, 2003; Zhong et al., 2017).
Additionally, firms with strong ties are more motivated to engage in frequent and meaningful communication, which particularly improves the later stages of knowledge transfer, namely assimilation and integration. Frequent communication also encourages familiarity in the form of shared norms, values, mental maps and common understanding, minimizing cognitive limitations to transfer knowledge (Lui et al., 2023; Rogan, 2014; Schillebeeckx et al., 2021). Consequently, firms with strong ties are more motivated to transfer knowledge beyond information available to network outsiders (Polidoro et al., 2011). To this end, we hypothesize the following:
Tie strength is positively related to knowledge transfer.
3.2 Moderating role of organizational unlearning on social networks and knowledge transfer
We propose that organizational unlearning moderates the two network mechanisms. Unlearning involves both cognitive and behavioral components, requiring a cognitive mindset to willingly give up old knowledge and behavioral practices to discard old routines (Klammer and Gueldenberg, 2019). When firms unlearn, they develop and refine activities in discarding obsolete ideas and routines while preparing for new ones (Starbuck, 2017). The unlearning process refers to who unlearns, how unlearning occurs and what is unlearned (Maccioni et al., 2024).
As argued in H1, central firms benefit from their high status and knowledge processing experience to create opportunities for knowledge transfer. We posit that unlearning strengthens the opportunity benefit of centrality on knowledge transfer by reinforcing these effects. When central firms engage in unlearning, they shed obsolete knowledge and practices and adapt to changing network dynamics, thereby demonstrating their flexibility and responsiveness to market change (Zhao and Wang, 2020). This enhances their high status, further increasing opportunities for knowledge exchanges.
Moreover, unlearning helps firms use their knowledge processing experience more effectively. Unlearning reduces cognitive biases and increases ability to integrate new information, which is a key component of absorptive capacity (Lyu et al., 2020; Ortega-Gutiérrez et al., 2022). As central firms process more information than other network members, unlearning helps reduce knowledge waste (Wensley and Navarro, 2015) while boosting their absorptive capacity for more knowledge exchanges with prominent partners. This creates more opportunities for central firms for knowledge transfer. Therefore:
Organizational unlearning strengthens the positive relationship between firm centrality and knowledge transfer.
Conversely, we argue that organizational unlearning poses challenges that may weaken the motivational benefits of tie strength for knowledge transfer. As argued in H2, firms that form strong ties with network partners are more motivated to engage in knowledge transfer due to trusting expectations and effective communication. However, as firms unlearn, existing interaction routines and practices between the firms and their network partners are reshuffled and disrupted. Changes in interaction routines and practices plant seeds for distrust and suspicion of opportunism (Wang et al., 2010; Zhang et al., 2021). These relational hazards threaten reciprocity norms and stability of strong ties in relationships (Heidl et al., 2014; Yang et al., 2012). Unlearning therefore poses relational risks that could reduce the trust, safety and reciprocity norms arising from tie strength.
Further, communication with other network partners could break down when firms unlearn, weakening the motivation to transfer knowledge. Strong ties tend to provide a sense of security and familiarity, which ease frequent communications (Gu et al., 2008; Uzzi, 1997). Nonetheless, unlearning reduces shared understanding with partners and introduces cognitive barriers, which may diminish the sense of security and familiarity inherent in strong ties. This tendency weakens communication, especially when the unlearning is not synchronized between knowledge donors and recipients (Brooks et al., 2021). Hence, we hypothesize that the effect of tie strength on performance will be lessened when organizational unlearning is high rather than low:
Organizational unlearning weakens the positive relationship between tie strength and knowledge transfer.
3.3 Knowledge transfer and performance
Knowledge transfer refers to the process by which the behavior of an organizational actor is affected by the experience of another actor (Argote and Ingram, 2000; Inkpen and Tsang, 2005, 2016) and is vital for firms to sustain innovativeness and make evidence-based decisions (Ferrer-Serrano et al., 2021; Filieri and Alguezaui, 2014). As firms undertake knowledge transfer activities, they obtain information from other partners on customer needs, industry best practices and emerging market trends (Levine and Prietula, 2012; Liu and Lui, 2020). This increase in knowledge facilitates the introduction of new and valuable products and services that meet market needs (Ferreras-Méndez et al., 2015; Xie et al., 2016). Additionally, knowledge received from network partners provides more evidence for firms to consider when solving problems and improves decision quality (Argote and Ingram, 2000; Kachra and White, 2008). Better decisions and problem solutions, in turn, enhance their competitive advantage and performance (Bendig et al., 2018; Farooq, 2023). Thus, we hypothesize the following:
Knowledge transfer is positively related to firm performance.
3.4 Moderating role of organizational unlearning on knowledge transfer and performance
We argue that organizational unlearning amplifies the benefits of knowledge transfer on firm performance. When firms unlearn, they recognize and correct what is wrong with their current knowledge and behavior, allowing adaptation to a changing environment (Martignoni and Keil, 2021). As knowledge transfer provides firms with valuable resources that are building blocks for creativity and innovation, unlearning complements the process by eliminating obsolete beliefs that hinder organizational change (Leal-Rodríguez et al., 2015), challenging current knowledge transfer (Maccioni and Ghiringhelli, 2024) and increasing the value of new knowledge acquired (Akgün et al., 2006). Engaging in both learning and unlearning processes simultaneously positively influence innovation and performance (Cegarra-Navarro et al., 2013; Yeniaras et al., 2021; Zhang et al., 2022).
Furthermore, organizational unlearning helps firms reflect on their past failures and successes in light of new knowledge, thereby improving the quality of their future decisions (Martignoni and Keil, 2021). As a result, firms make better decisions when they simultaneously acquire new knowledge and unlearn their obsolete knowledge, which enhances performance (Yeniaras et al., 2021). Similarly, as interfirm knowledge transfer expands firms’ capabilities and competencies to cope with turbulence and uncertainty (Kraatz, 1998), unlearning fosters flexibility and adaptability that help effective application of firms’ diverse capabilities to build competitive advantage in dynamic markets (Lyu et al., 2020; Zhang and Zhu, 2021). Therefore, when firms engage in unlearning while participating in knowledge transfer, they enhance efficiency and productivity from new knowledge while reducing the costs and risks of poor decision-making by discarding obsolete knowledge and beliefs. Thus, we hypothesize the following:
Organizational unlearning strengthens the positive relationship between knowledge transfer and firm performance.
4. Methodology
4.1 Empirical setting and sample
The petroleum wholesale industry in Tanzania is chosen to verify the proposed model and hypotheses. Like other sub-Saharan African countries (Donbesuur et al., 2021), business relationships, mutual dependence and networking are paramount in business activities in Tanzania. Formal structures to collect and publicly disseminate important institutional and market information are underdeveloped in Tanzania (Tanzania National Bureau of Statistics, 2022), leaving interfirm networks as the major way firms access and exchange important information. This industry suits our study objectives because the industry comprises 44 active firms, making it feasible to obtain whole network information to measure centrality, which is crucial for this study. The firms regularly and frequently exchange knowledge among themselves about local and global supply conditions, market demand and government policy. The knowledge is vital for competitive advantage and survival in the turbulent petroleum industry, especially during the recent global energy crisis (Dodd, 2022). Finally, the petroleum industry is one of the largest revenue generators in the Tanzanian economy (Kamer, 2022), making results of this study relevant and vital for policymakers in Tanzania.
We obtained research ethics approval for this study and collected data from two sources:
face-to-face surveys filled in by the top managers of the Tanzanian petroleum wholesalers; and
Annual Report of the Energy and Water Utilities Regulatory Authority (EWURA), the main industry regulatory body of the petroleum industry in Tanzania.
The EWURA listed 60 firms as valid petroleum wholesaler licensees (Ewura, 2022). However, only 44 of them were in operation during the data collection period (September to December, 2022). After repeated emailing and calling on all 44 firms to solicit their participation, two firms opted out of participating in our research, leading to a response rate of 95.45%. Given the high response rate, we follow previous studies using a whole network design (e.g. Reck et al., 2021) to attest that we have captured a complete census of contractual partnerships among petroleum wholesalers in Tanzania.
We interviewed and surveyed top managers from the petroleum wholesaler firms using contact information available on the company websites. These top managers included founders, communication directors, alliance directors, chief executive officers and chief financial officers of the firms. They were selected because they were the most suitable informants for providing insights into steering and managing business relationships. Our survey data indicated that, on average, they had managed their respective companies for about seven years and had worked in the petroleum industry for about 12 years.
The entire roster of the 44 licensed petroleum wholesalers in Tanzania was presented to a key informant from each of the 42 participating firms. The top managers then identified the firm’s contractual cooperative agreements with other petroleum wholesalers over the last three years. The contractual cooperative agreements between these firms mainly included:
building joint petroleum depots;
co-marketing agreements;
joint supply agreement;
transportation fleets sharing; and
joint ventures in new markets.
It is common for petroleum wholesalers to enter into supply and marketing agreements between themselves because of demand and supply fluctuations in the petroleum industry.
The contractual cooperative agreements reported by the managers captured the interfirm networks among the petroleum wholesalers in Tanzania. We followed best practices in social network research by using the roster method, which is widely regarded as providing more reliable and accurate network data (Li et al., 2020; Provan et al., 2007). In the interviews, the top managers continued to answer a survey to rate their firm’s level of knowledge transfer, unlearning and interfirm tie strength. We used archival data to capture firm performance over past three years. This information is included in the EWURA’s annual reports available on their webpage (Ewura, 2023). Our research design used different data sources, minimizing common method bias risk (Podsakoff et al., 2012).
We sent an invitation email to all 44 firms to participate in a study of the Tanzanian interfirm network in September 2022. After the first email, we visited each firm thrice and sent two reminder emails. In the end, our sample consists of 42 firms, the majority of which were locally owned (72.7%), medium-sized with an average of 146 employees and has been established for 16.6 years (SD = 14.4). Twelve of these firms are foreign-owned and operate as multinational corporations (MNCs), while Tanzanian entrepreneurs own 32 firms that operate locally.
As we conducted face-to-face surveys with the respondents in this study, we were able to collect anecdotal examples of knowledge transfer and unlearning within the industry. Respondents indicated that important knowledge and information were regularly exchanged and acquired during meetings among key decision-makers of the wholesale firms. They shared daily updates with other firms on petroleum prices and political news that could affect the industry. They also met up every Friday to discuss broader industry issues such as environmental protection, petroleum shipments, global price fluctuation and strategic responses to these issues. These weekly formal meetings were often followed by informal networking sessions, where they exchanged business information about their companies, potential new markets and finance matters such as loans.
Respondents also shared with us examples of unlearning activities that they have undertaken. Unlearning involves discarding organizational possesses and attitudes through incremental or episodic changes (Tsang and Zahra, 2008: 1446). For instance, the wholesaler firms once relied on snitching tactics by monitoring and reporting competitor’s dishonest practices to regulators, in the hope that investigators would temporarily shut down their rivals and give them a competitive edge. However, these reports backfired, triggering greater government scrutiny and higher taxes for all wholesalers. As a result, the firms abandoned and unlearned such practices, shifting instead toward more cooperation. Respondents also suggested that barriers to unlearning were often rooted in organizational structure, rigid hierarchy and ingrained mindset. They noted that local firms led by entrepreneurs were generally more flexible and capable of unlearning than their multinational counterparts.
4.2 Measures
We adapted established survey scales to our context to increase reliability and validity of the measures. Except for the sociometric part of the survey, firm size, age, locality and firm performance, all other constructs are reflective, thus captured using a seven-point Likert scale ranging from 1 “strongly disagree” to 7 “strongly agree”. Questionnaire Items presents all scale items and their sources:
Items
Tie Strength (AVE = 0.73, CR = 0.84, α = 0.69)
Source: Hansen, 1999
How often does your firm communicate with your coopetitors? (0.969)
How closely does your firm work with your coopetitors? (0.725)
Knowledge transfer (AVE = 0.56, CR = 0.83, α = 0.73)
Knowledge search (AVE = 0.89, CR = 0.96, α = 0.94)
We frequently scan the environment for new knowledge (0.926)
We thoroughly observe market trends (0.950)
We observe in detail external sources of new knowledge (0.956)
Knowledge acquisition (AVE = 0.69, CR = 0.89, α = 0.84)
We have frequent interactions with other companies to acquire new knowledge (0.841).
Our employees regularly visit other companies (0.901).
We collect industry information through informal means (e.g. lunch with industry friends, talks with trade partners). (0.683).
Our employees regularly approach third parties such as accountants, consultants or tax consultants for new knowledge (0.888).
Knowledge assimilation (AVE = 0.96, CR = 0.98, α = 0.96)
We quickly understand new opportunities to serve our clients (0.984).
We quickly analyze and interpret changing market demands (0.979).
Knowledge integration (AVE = 0.57, CR = 0.84, α = 0.74)
We discuss issues until we arrive at a shared understanding (0.861).
Top management integrates information from different organizational areas (0.703).
We seek to achieve consensus by dialogue and reasoning (0.839).
We stress sharing and trying to understand management vision through communication with colleagues (0.570).
Organizational unlearning (AVE = 0.51, CR = 0.83, α = 0.75)
Sources: Lyu et al. (2020) and Zhang et al. (2022)
We are ready to acquire new technologies and knowledge from various channels (0.592).
We provide favorable context for changing obsolete beliefs (0.602).
We are ready to change the way our company operates (0.444).
We can establish new product processes based on real needs (0.908).
We are ready to abandon outdated beliefs and routines (0.891).
Factor loadings in bracket at the end of the survey item.
4.2.1 Knowledge transfer.
The first dependent variable was measured using 13 items (α = 0.82, Mean = 5.59, SD = 0.84). The scale accounts for the four stages of knowledge transfer proposed for this study. Specifically, knowledge search was measured by three items taken from Ferreras-Méndez et al. (2015)’s scale of knowledge recognition. Knowledge acquisition and assimilation were measured by four and two items respectively, each taken from Jansen et al. (2005). Finally, knowledge integration was measured by four items taken from Flores et al.’s (2012) scale of information integration.
4.2.2 Firm performance.
The other dependent variable was measured as a natural logarithm of a firm’s average sales volume (in liters of petroleum sold) in the past three years (2020–2022). We collected this information from the 2020–2022 annual reports reported by EWURA (Ewura, 2023). We chose this indicator because sales volume is an objective measure drawn from annual reports and is therefore free from personal bias associated with perceptual measures. Additionally, sales volume data is commonly used in management literature to measure financial performance (Inkinen, 2016). We used the natural log of sales volume data in the regression analyses to avoid skewness in data. The average annual sales volume was 99.067 million liters across three years (SD 148 million).
4.2.3 Firm centrality.
Eigenvector centrality captures firms’ prominence in a network by being connected to other firms who are well connected (Bonacich, 2007). This measure goes beyond just capturing the number of direct connections a firm has but weighs the value of these direct connections based on their number of connections (Schilling and Phelps, 2007). This measure of centrality is most appropriate for the petroleum wholesaling industry, where the product is globally standardized. Under such circumstances, reputational benefits make central firms more visible and desirable as partners, positioning them as informal leaders in their networks (Borgatti and Everett, 2020; Bothner et al., 2011). To compute this variable, we used the eigenvector routine as operationalized in UCINET 6, which sums the focal firm’s connections (captured by the contractual cooperative agreements reported in the survey) after weighing these connections by their own respective connections and then normalizes following standard practice (Borgatti, 2002).
4.2.4 Tie strength.
This independent variable, capturing the closeness and communication frequency of partnering firms, was measured using two items from Hansen (1999) (α = 0.69, Mean = 5.89, SD = 1.24):
“How often does your firm communicate with your coopetitors?”
“How closely does your firm work with your coopetitors?”
Respondents were asked to consider the average tie strength across all partnering firms when answering these questions. Accordingly, tie strength is measured as a scalar variable reflecting the average relational strength across all business partners in this survey. This scale showed a strong inter-item correlation (r = 0.53, p < 0.001) and adequate CR (0.86) and AVE (0.76) in the measurement model, providing satisfactory reliability for a two-item measure.
4.2.5 Organizational unlearning.
A five-item scale (α = 0.75, Mean = 5.76, SD = 0.80) was adapted from the scales of Lyu et al. (2020) and Zhang et al. (2022) to measure the extent to which a firm discard its obsolete knowledge, beliefs and routines. An item example is “We are ready to abandon outdated beliefs and routines.”
4.2.6 Control variables.
We controlled for the following variables: First, local firm. Locality of a firm might affect its network structure and knowledge-transferring tendencies (Hsiao et al., 2006). Local firms were identified by their registration with the EWURA as Tanzanian corporations – coded 1, otherwise foreign registered firms were coded 0. Second, firm age was measured using a single question: “How many years has this company been in operations?” The experience a firm has in operation might influence the number of partners they have and the strength of its relationships (Liu and Lui, 2020). Third, firm size was measured using a single question: “What is the number of full-time employees in this company?” We controlled for firm size because firm centrality and firm performance have previously been shown to relate to firm size (Donbesuur et al., 2021; Leal-Rodríguez et al., 2015; Wang et al., 2015). Fourth, brokerage. Brokerage means the firm is positioned between other unconnected firms (Burt and Soda, 2021). Brokerage is associated with knowledge transfer in interfirm networks and could be an alternative explanation of our results (Liu and Xi, 2023). We operationalized brokerage as an inverse of the constraint routine in UCINET 6.
Before running the regressions, we assessed the measurement model using PLS-SEM. As the Questionnaire Items shows, all average variance extracted (AVE) and composite reliability (CR) values meet conventional thresholds. However, factor loadings of five items (ranged from 0.44–0.68) are below the recommendation of 0.70. Hair et al. (2009) suggest that items with factor loading above 0.50 could be retained (cited in Cheung et al. (2024: 750)). Yu et al. (2022: 274) similarly suggest retaining items with factor loading between 0.40 and 0.70 if their removal does not improve AVE and CR values. Following their suggestions, we retained these items in our analyses. As a robustness test, we deleted the item with a factor loading below 0.55 to construct a four-item unlearning scale. We reran the regressions and obtained the same results and significance levels for all the hypotheses.
4.3 Analysis and results
Table 1 presents descriptive statistics and correlation matrix among the study variables. Information on the network ties was first entered as a binary adjacency matrix among the 42 firms in the sample into UCINET. There were no isolated nodes. All network variables calculated from UCINET were then transposed to firm-level variables for the regression analyses. Moderated regressions were conducted to test the hypothesized two-way interactions. We report HC2 heteroskedasticity-consistent standard errors, which correct for bias in small samples (Hayes and Cai, 2007). Results based on conventional standard errors were similar.
Descriptive statistics and correlation matrix of study variables
| Variables | Mean | SD | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
|---|---|---|---|---|---|---|---|---|---|---|
| 5.59 | 0.84 | ||||||||
| 0.13 | 0.07 | 0.17 | |||||||
| 5.89 | 1.24 | 0.41** | −0.23 | ||||||
| 5.76 | 0.80 | 0.66** | 0.24 | 0.15 | |||||
| 14.11 | 7.10 | 0.20 | 0.64** | −0.17 | 0.06 | ||||
| 0.71 | 0.45 | 0.15 | −0.05 | 0.40** | 0.11 | 0.06 | |||
| 146.57 | 178.64 | 0.19 | 0.45** | 0.09 | 0.21 | 0.34* | 0.27 | ||
| 16.36 | 14.56 | 0.16 | 0.42** | −0.28 | 0.07 | 0.47** | −0.09 | 0.45** | |
| 0.92 | 0.24 | 0.07 | 0.89*** | −0.08 | 0.21 | 0.59*** | −0.07 | 0.43** | 0.25 |
n = 42; *p < 0.05, **p < 0.01 and ***p < 0.001 (2-tail). All means reported before mean-centering
Table 2 presents the regressions on knowledge transfer and results for H1 to H4. We first entered all control variables (i.e. local firm, firm size, firm age and brokerage) in Model 1. Afterwards, we mean-centered the independent (i.e. centrality and tie strength) and the moderating (i.e. organizational unlearning) variables to test the direct relationships H1 and H2 in regression Model 2. Finally, we created and entered the interaction between centrality and unlearning (H3) in Model 3, followed by tie strength and unlearning interaction (H4) in Model 4 and both interactions in Model 5. All variance inflation factors (ranged from 1.20 to 8.17) were below the recommended threshold of 10 (Hair et al., 2009), suggesting that multicollinearity was not a concern.
Results for regressions on knowledge transfer
| Variables | Model 1 | Model 2 | Model 3 | Model 4 | Model 5 |
|---|---|---|---|---|---|
| Constant | 5.21*** (8.15) | 7.55*** (8.94) | 4.96*** (5.49) | 6.32*** (13.54) | 6.25*** (7.07) |
| Local firm | 0.24 (0.76) | −0.18 (−0.62) | −0.02 (−0.06) | −0.11 (−0.48) | 0.05 (0.25) |
| Firm size | 0.00 (0.45) | −0.00 (−0.51) | −0.00 (−1.46) | −0.00 (−0.13) | −0.00 (−0.97) |
| Firm age | 0.01(0.73) | 0.01(1.30) | 0.01 (1.66) | 0.01 (1.39) | 0.01† (1.76) |
| Brokerage | 0.01 (0.02) | −2.13* (−2.58) | −1.41† (−1.80) | −1.66* (−2.28) | −0.95(−1.12) |
| Direct effect | |||||
| (H1) Firm centrality | 7.54** (2.75) | 4.98† (1.78) | 5.80* (2.07) | 3.27(1.08) | |
| (H2) Tie strength | 0.35* (2.31) | 0.30* (2.35) | 0.14(1.53) | 0.17† (1.72) | |
| Organizational unlearning | 0.59*** (4.13) | 0.69*** (5.11) | 0.58*** (4.88) | 0.65*** (6.99) | |
| Moderating effect | |||||
| (H3) Firm centrality x | 3.87* | 3.84** | |||
| Organizational unlearning | (2.02) | (2.59) | |||
| (H4) Tie strength x | −0.31* | −0.26* | |||
| Organizational unlearning | (−2.14) | (−2.31) | |||
| F value | 0.59 | 9.42*** | 9.66*** | 10.75*** | 11.43*** |
| F change | 20.00*** | 4.50* | 7.48** | 6.94** | |
| Adjusted R2 | −0.04 | 0.59 | 0.63 | 0.65 | 0.70 |
| R2 change | 0.60 | 0.04 | 0.06 | 0.10 | |
n = 42. Unstandardized coefficients reported;t statistics in parentheses; Robust t statistics based on HC2 standard errors are reported for Model 3–5., † p < 0.10, *p < 0.05, **p < 0.01 and ***p < 0.001 (2-tailed)
H1 predicts that firm centrality is positively related to knowledge transfer. Model 2 in Table 2 shows that our hypothesis is supported such that the regression coefficient was positive and significant (b = 7.54, p = 0.02). In the same model, tie strength was also shown to positively affect knowledge transfer (b = 0.35, p = 0.001), supporting H2. In other words, when other factors in this model were held constant, two firms that differed by one unit in their centrality values were, on average, estimated to differ by 7.54 units in their knowledge transfer, while firms that differed by one unit in tie strength with their partners, were on average estimated to differ by 0.35 units in their knowledge transfer.
For the interaction effects, H3 is supported, as shown in Model 3 of Table 2, such that organizational unlearning strengthened the effect of centrality on knowledge transfer (b = 3.87, p = 0.04). R2 of the model increased from 0.59–0.63 (F change = 4.50, p < 0.04). Specifically, at low levels of unlearning, centrality has no significant effect on knowledge transfer, but as unlearning increased by one unit, the effect of centrality on knowledge transfer, on average, increased by 3.87 units. Figure 1 presents the moderating effect graphically.
The plot presents knowledge transfer against centrality for plus 1 standard deviation and minus 1 standard deviation unlearning. At plus 1 standard deviation, knowledge transfer increases steadily as centrality moves from low to high. At minus 1 standard deviation, knowledge transfer decreases steadily as centrality moves from low to high. The bands surrounding both lines widen towards the ends of the centrality range.Two-way interaction between centrality and organizational unlearning on knowledge transfer
The plot presents knowledge transfer against centrality for plus 1 standard deviation and minus 1 standard deviation unlearning. At plus 1 standard deviation, knowledge transfer increases steadily as centrality moves from low to high. At minus 1 standard deviation, knowledge transfer decreases steadily as centrality moves from low to high. The bands surrounding both lines widen towards the ends of the centrality range.Two-way interaction between centrality and organizational unlearning on knowledge transfer
We also found support for H4, whereby organizational unlearning weakened the positive effect of tie strength on knowledge transfer (b = −0.31, p = 0.01). R2 of the model increased from 0.59–0.65 (F change = 7.48, p = 0.01). Specifically, at low levels of unlearning, one unit increase in tie strength was related to a 0.43 unit increase in knowledge transfer, but as unlearning increased by one unit, the effect of tie strength on knowledge transfer decreased by 0.26 units. Figure 2 presents the moderating effect graphically. Model 5 shows the full model that still supports our predictions when both interaction terms are entered together.
The plot presents knowledge transfer against tie strength for plus 1 S D and minus 1 S D unlearning. At plus 1 S D, knowledge transfer remains near 6 across low to high tie strength. At minus 1 S D, knowledge transfer increases steadily from about 3.3 at low tie strength to about 5.7 at high tie strength. The surrounding bands are wider at low tie strength and narrower towards high tie strength.Two-way interaction between tie strength and organizational unlearning on knowledge transfer
The plot presents knowledge transfer against tie strength for plus 1 S D and minus 1 S D unlearning. At plus 1 S D, knowledge transfer remains near 6 across low to high tie strength. At minus 1 S D, knowledge transfer increases steadily from about 3.3 at low tie strength to about 5.7 at high tie strength. The surrounding bands are wider at low tie strength and narrower towards high tie strength.Two-way interaction between tie strength and organizational unlearning on knowledge transfer
Table 3 presents the regressions on firm performance for H5 and H6. Model 2 shows the direct effect of knowledge transfer on firm performance (H5), and Model 3 shows the interaction effect of knowledge transfer and organizational unlearning on firm performance (H6). As shown in Model 3, our results provide support for H5 (b = 3.53, p = 0.04) and H6 (b = 1.91, p = 0.05). Figure 3 presents graphically how organizational unlearning strengthened the positive effect of knowledge transfer on firm performance.
Results for regressions on firm performance
| Variables | Model 1 | Model 2 | Model 3 |
|---|---|---|---|
| Constant | −5.47 (−1.38) | −6.15 (−1.55) | −5.54 (−1.63) |
| Local firm | 2.68 (1.34) | 2.46(1.25) | 2.17(1.11) |
| Firm size | −0.01(−0.87) | −0.01(−0.83) | −0.01(−0.88) |
| Firm age | 0.20** (3.02) | 0.18**(2.72) | 0.16*(2.52) |
| Brokerage | 16.51***(4.23) | 17.72***(4.54) | 16.64***(4.75) |
| Direct effect | |||
| (H5) Knowledge transfer | 2.31†(1.75) | 3.53* (2.06) | |
| Organizational unlearning | −2.30(−1.64) | −2.23†(−1.80) | |
| Moderating effect | |||
| (H6) Knowledge transfer x | 1.91* | ||
| Organizational unlearning | (2.00) | ||
| F value | 8.97*** | 6.80*** | 6.63*** |
| F change | 1.74 | 4.03* | |
| Adjusted R2 | 0.44 | 0.46 | 0.49 |
| R2 change | 0.05 | 0.04 | |
n = 42. Unstandardized coefficients reported; t statistics in parentheses; Robust t statistics based on HC2 standard errors are reported for Model 3; F change and R2 change of Model are reported. † p < 0.10, *p < 0.05, **p < 0.01 and ***p < 0.001 (2-tailed)
The plot presents firm performance against knowledge transfer for plus 1 S D and minus 1 S D unlearning. At plus 1 S D, firm performance increases steadily from about 7 at low knowledge transfer to about 14 at high knowledge transfer. At minus 1 S D, firm performance increases more gradually from about 13 to about 16. The surrounding bands are wider at low knowledge transfer and narrower towards high knowledge transfer.Two-way interaction between knowledge transfer and organizational unlearning on firm performance
The plot presents firm performance against knowledge transfer for plus 1 S D and minus 1 S D unlearning. At plus 1 S D, firm performance increases steadily from about 7 at low knowledge transfer to about 14 at high knowledge transfer. At minus 1 S D, firm performance increases more gradually from about 13 to about 16. The surrounding bands are wider at low knowledge transfer and narrower towards high knowledge transfer.Two-way interaction between knowledge transfer and organizational unlearning on firm performance
As a robustness test, we reran the regressions replacing eigenvector centrality with three other common centrality measures (i.e. degree, closeness and betweenness). All hypotheses were supported except those involving betweenness. Since different centrality measures capture distinct aspects of centrality (Ngowi et al., 2022; Pappas and Wooldridge, 2007), the largely consistent results across alternative centrality measures provide support for our arguments. Furthermore, to ensure that our results were not biased by a potential network autocorrelation in the residuals, we performed a QAP-style node-label permutation test with 5,000 repetitions, where we permuted the network variables (centrality and brokerage). This process confirmed that the coefficients for centrality and brokerage remain statistically significant and the impact of the (unpermuted) unlearning variable remain stable, showing that our primary results are not driven by underlying network structure, further enhancing the credibility and robustness of our results. Results from the UNICET settings, SPSS, STATA and R code, VIF tests, the Fornell–Larcker Criterion discriminant validity test, and all robustness tests are available upon request.
4.4 Post-hoc analyses
We conducted several post-hoc analyses to further explore the regression results from the main analysis. First, we tested for a potential two-way interaction between tie strength and centrality on knowledge transfer. This is because structural and relational networks could interactively affect knowledge transfer, resulting in Simmelian ties, where both centrality and tie strength are strong (Krackhardt, 1998; Tortoriello and Krackhardt, 2010). However, results (not shown here but available on request) indicated that the interaction was insignificant (b = −0.14, p = 0.44) for this sample.
Second, we conducted separate analyses for the four stages of knowledge transfer to explore the potentially different effects of social networks on each stage. While knowledge transfer is treated as an overall learning process in the hypotheses, we followed Tsang (2008), who qualitatively examined how unlearning affects different stages of knowledge transfer in joint ventures. Four generic stages of interfirm knowledge transfer can be identified (Filieri and Alguezaui, 2014; Flores et al., 2012; Jansen et al., 2005; Zahra and George, 2002). The first stage, knowledge search, refers to seeking and identifying valuable knowledge outside firms’ boundaries. The second stage, knowledge acquisition, refers to accessing or getting hold of valuable knowledge. The third stage, knowledge assimilation, refers to processing and making sense of the externally acquired knowledge. The fourth and final stage, knowledge integration, refers to combining new knowledge with existing knowledge for utilization. These four stages unfold gradually as opportunities and motivations for the knowledge transfer process emerge.
As a corollary of our hypotheses, centrality and tie strength are expected to show varying levels of significance across the different stages. Table 4 presents the regression results on each stage of knowledge transfer. Overall, results are consistent with our theoretical argument on the different mechanisms of centrality and tie strength. Results show that centrality has a positive and direct effect on knowledge search (b = 14.65, p = 0.02) and assimilation (b = 11.69, p = 0.04), but no significant effect on knowledge acquisition and integration. In contrast, tie strength has a positive and direct effect on knowledge acquisition (b = 0.45, p = 0.01), assimilation (b = 0.38, p = 0.01) and integration (b = 0.34, p = 0.001), but no significant effect on knowledge search.
Results for regressions on individual stages of knowledge transfer
| Knowledge search | Knowledge acquisition | Knowledge assimilation | Knowledge integration | |||||
|---|---|---|---|---|---|---|---|---|
| Variables | Model 1 | Model 2 | Model 3 | Model 4 | Model 5 | Model 6 | Model 7 | Model 8 |
| Constant | 9.45*** (5.66) | 8.13*** (5.72) | 6.14 (3.81) | 4.32† (1.74) | 8.90*** (5.82) | 7.11*** (4.78) | 6.95*** (7.67) | 6.33*** (6.32) |
| Local firm | 0.19 (0.44) | 0.44 (1.64) | 0.21 (0.52) | 0.56 (1.14) | −0.35 (−0.91) | −0.06 (−0.17) | −0.69** (−2.99) | −0.58* (−2.41) |
| Firm size | −0.00 (−1.28) | −0.00 (−1.35) | 0.00 (0.09) | −0.00 (−0.41) | −0.00 (0.53) | −0.00 (−0.39) | 0.00 (0.39) | 0.00 (0.41) |
| Firm age | 0.01 (0.49) | 0.01 (1.07) | 0.02 (1.26) | 0.03 (1.43) | 0.01 (0.36) | 0.01 (0.68) | 0.12 (1.54) | 0.01† (1.70) |
| Brokerage | −3.95* (−2.39) | −2.80 (−1.92) | −1.68 (−1.05) | −0.07 (−0.03) | −2.97† (−1.96) | −1.30 (−0.91) | −1.02* (−1.14) | −0.44 (−0.46) |
| Direct effect | ||||||||
| Firm centrality | 14.65* (2.50) | 10.51† (1.84) | 4.77 (0.84) | −1.03 (−0.15) | 11.69* (2.18) | 5.59 (1.01) | 3.15 (0.99) | 1.03 (0.30) |
| Tie strength | 0.17 (1.02) | 0.04 (0.25) | 0.45** (2.85) | 0.25 (1.11) | 0.38* (2.55) | 0.01 (0.10) | 0.34*** (3.82) | 0.22* (2.03) |
| Organizational unlearning | 0.56* (2.56) | 0.68* (3.29) | 0.47* (2.24) | 0.62*** (3.31) | 0.57** (2.81) | 0.55*** (3.18) | 0.74*** (6.23) | 0.74*** (5.78) |
| Moderating effect | ||||||||
| Firm centrality x | 5.17† | 6.69* | 2.84 | 1.14 | ||||
| Organizational unlearning | (1.83) | (2.41) | (0.86) | (0.57) | ||||
| Tie strength x | −0.11 | −0.21 | −0.63*** | −0.20† | ||||
| Organizational unlearning | (−0.69) | (−1.03) | (−2.65) | (−1.88) | ||||
| F value | 2.61 | 2.28 | 3.27** | 3.30** | 3.08* | 5.44*** | 10.28*** | 8.88*** |
| F change | 1.08 | 2.43† | 8.78*** | 1.93 | ||||
| AdjustedR2 | 0.21 | 0.22 | 0.28 | 0.34 | 0.26 | 0.49 | 0.61 | 0.63 |
| R2 change | 0.04 | 0.08 | 0.22 | 0.03 | ||||
n = 42. Unstandardized coefficients reported; t statistics in parentheses; Robust t statistics based on HC2 standard errors reported for Model, 2, 4, 6, 8; † p < 0.10, *p < 0.05, **p < 0.01 and ***p < 0.001 (2-tailed)
We also explored the moderating role of organizational unlearning at each stage of knowledge transfer. We predicted that organizational unlearning would moderate the effects of structural and relational network on knowledge transfer in opposite ways (H3 and H4). As Table 4 shows, organizational unlearning significantly strengthened the effect of centrality on the earlier stages of knowledge transfer – knowledge search (b = 5.17, p = 0.07) and knowledge acquisition (b = 6.69, p = 0.02) – only. Moreover, organizational unlearning significantly weakened the effects of tie strength on the later stages of knowledge transfer of knowledge assimilation (b = −0.63, p = 0.0003) and knowledge integration (b = −0.20, p = 0.07) – only.
Integrating the main analysis and the post hoc tests, results show that interfirm networks have asymmetric effects on different stages of knowledge transfer. Specifically, the effect of centrality was more prominent in the early stages of knowledge search and acquisition; the effect of tie strength was more prominent in later stages of knowledge assimilation and integration. This illustrates further the opposite effects of structural and relational networks on different stages of the knowledge transfer process.
5. Discussion and conclusion
This study examines the dual moderating roles of organizational unlearning for interfirm knowledge transfer. We first examine its moderating role on the relationship between social network and knowledge transfer. We found that organizational unlearning moderates the two network effects in opposite directions: whereas it strengthens the positive effect of firm centrality on knowledge transfer, it weakens the effect of tie strength on knowledge transfer. This finding provides initial evidence of the proposed causal mechanisms between network and knowledge transfer. When central firms are simultaneously able to unlearn, unlearning enhances the opportunities provided by centrality to search for and acquire knowledge. In contrast, higher levels of organizational unlearning weaken the motivation of firms with strong ties to exploit network resources.
We then examine its moderating role on the relationship between knowledge transfer and firm performance. We find that organizational unlearning enhances the effect of knowledge transfer on firm performance. Organizational unlearning helps effective knowledge application by organizations to avoid repeating detrimental mistakes, enabling better decision-making and flexible strategies that are relevant to the changing environment (Martignoni and Keil, 2021; Ortega-Gutiérrez et al., 2022; Zhao and Wang, 2020).
5.1 Theoretical implications
An understanding of the boundary role of unlearning in knowledge transfer contributes to the literature in two ways. First, by identifying the boundary condition of the social network-knowledge transfer relationship, this paper extends research on the effect of social networks on knowledge transfer. Specifically, we delineate two distinctive mechanisms through which centrality and tie strength facilitate knowledge transfer. While centrality provides opportunities to identify and access network resources, tie strength provides the motivation and ability to realize these opportunities. The opposite moderating effects of unlearning highlight the unique contributions of structural and relational networks to knowledge transfer.
Our results support prior research suggesting that the structural network dimension mainly increases the potential to access more resources while the relational dimension enhances the reciprocity and trust necessary to mobilize network resources (Moran, 2005; Ozdemir et al., 2016; Reinholt et al., 2011; Yu et al., 2022). However, these mechanisms have rarely been tested directly. Different from previous studies, we indirectly validate these mechanisms by demonstrating that unlearning exerts opposite moderating effects on structural and relational networks. Hence, we advise social network researchers to consider both aspects of interfirm networks for a deeper understanding of their impact on knowledge transfer.
Second, by identifying the boundary condition of the knowledge transfer-performance relationship, we weigh in on the debate concerning the distinctiveness of organizational unlearning from learning. Previous research has cast doubt on the conceptual validity of organizational unlearning (Howells and Scholderer, 2016; Klein, 1989). In contrast to prior studies, this paper demonstrates that these two constructs are distinct by showing that firms achieve superior performance when they cultivate a context of unlearning that facilitates learning. Intra-firm and inter-firm knowledge activities combine to contribute to firm-level outcomes (Oh and Kim, 2022). By conceptualizing unlearning as an intrafirm characteristic, we argue that unlearning provides the boundary for interfirm learning to take place. This moderating effect of unlearning on knowledge transfer complements prior studies that focus primarily on the direct and positive outcomes of unlearning (Martignoni and Keil, 2021).
5.2 Managerial implications
The practical value of this study lies in advising business managers regarding suitable conditions to engage in organizational unlearning, which not only supports knowledge transfer but also complements the positive effect of knowledge transfer on firm performance. If organizations fail to unlearn, they cannot respond to a changing external environment and may fail (Starbuck, 2017). Organizations could approach unlearning as an internal process to kick things out while knowledge transfer is an interfirm process to bring something in (Tsang, 2017).
Initially, we inform managers how structural and relational networks affect knowledge transfer differently. We recommend firms to practice unlearning when they occupy central network positions. Since centrality involves more frequent knowledge exchanges with more partners, unlearning helps sort and discard obsolete knowledge, creating space for more valuable knowledge acquisition.
Conversely, we warn managers to approach unlearning with caution when exchanging knowledge with their firms’ strong ties. This is because unlearning may be accompanied by knowledge loss, which may disrupt the familiarity, communication flow and shared understanding necessary to transfer more knowledge with network partners (Klammer and Gueldenberg, 2019). Alternatively, managers could invest in having trained administrators who would prevent accidental unlearning associated with vital knowledge loss while planning intentional unlearning that is less disruptive to interfirm relationships (Wensley and Navarro, 2015). We encourage firms to expand their access to resources through occupying central network positions and foster the motivation to share resources in their close relationships while recognizing the role of unlearning in enhancing knowledge transfer within their networks and financial performance.
Finally, we offer some recommendations for the Tanzanian petroleum wholesalers studied here. While the African market environment has become increasingly professionalized, social networks remain salient (Ado et al., 2017; Nachum et al., 2023). In general, we recommend the petroleum wholesalers to be agile and ready to remove obsolete practices as unlearning indirectly enhances sales. However, those wholesalers who have strong relational ties need to manage the unlearning process carefully. This is because engaging in unlearning for these wholesalers may disrupt their network. To address this issue, their managers need to proactively network with other firms during industry meetings and gatherings. In doing so, they could regain the social capital lost from undertaking unlearning activities.
5.3 Limitations and future research agenda
The limitations of this study point to promising research opportunities in the future. First, this study only focused on network centrality and has not considered other structural aspects of the network. The structural network theory has been extended to include structural hole theory (Burt, 2018) and embeddedness theory (Uzzi, 1997). Future research could explore further how unlearning interacting with these structural aspects to influence knowledge transfer.
Second, our data analysis is restricted by our sample, which is a cross-sectional survey based on one single network type (i.e. coopetition network) of one single industry (i.e. petroleum industry) in one single country (i.e. Tanzania). Although this research design allows us to control for many potentially unobservable factors, thus increasing our internal validity, the generalizability of our data could be limited. While we expect the findings to hold true in other national contexts, it is noteworthy that some relationships could be stronger or weaker in other national contexts due to cultural or institutional factors. Additionally, a small sample size of 42 firms, while essential for sociometrical data collection, limits the statistical power to conduct more sophisticated statistical analyses.
Third, cross-sectional survey data limits the examination of causal relationships. The possibility of reverse causation where knowledge transfer or firm performance impact interfirm networks cannot be ruled out. We therefore invite other researchers to accumulate more research to compare the findings of this study across industries, network types and cultural contexts. We also encourage longitudinal studies to capture the dynamic nature of interfirm networks and experiments that could be used to make causal claims.
Finally, the moderating effect of unlearning rests on two network mechanisms that we have not directly tested. Research in future could verify the hypothesized mechanisms by measuring the mediating variables and formulating hypotheses for each stage of knowledge transfer.
5.4 Conclusions
The key idea of this paper is that unlearning provides an important boundary condition for both network drivers and performance consequences of interfirm knowledge transfer. Firms having many prominent ties and strong relationships within their networks are involved in more knowledge transfer. Importantly, firms with high centrality enhance knowledge transfer when they are able to discard their obsolete knowledge and beliefs, while firms with strong ties hinder their knowledge transfer by doing the same. Additionally, the interplay between knowledge transfer and organizational unlearning improves firm performance. Hence, this study deepens the understanding of how and when organizational unlearning provides the boundary condition for knowledge transfer within interfirm networks.
Carolyn Yesse Ngowi is based at School of Management and Governance, UNSW Business School, Sydney, Australia.
Steven Lui is based at School of Management and Governance, UNSW Business School, Sydney, Australia.
Salih Zeki Ozdemir is based at School of Management and Governance, UNSW Business School, Sydney, Australia.
The authors would like to thank Editor-in-Chief Professor Manlio Del Giudice, Associate Editor Insaf Khelladi, and two anonymous reviewers for their constructive comments and suggestions. Authors would also like to express their sincere gratitude to the Tanzanian petroleum companies for their participation in this research. Authors are grateful for the support provided by the Energy and Water Utilities Regulatory Authority, the Tanzania Association of Oil Marketing Companies, and the United African University of Tanzania.
