Networks for Innovation – But What Networks and What Innovation?



Innovation is a social and interactive process in which collaboration and exchange of knowledge and information play crucial roles. Two conflicting hypotheses have been raised in previous research: Burt's structural hole hypothesis and the density hypothesis. In brief, the former of these hypotheses builds upon arguments for open network structures in the acquisition of innovation; the latter one builds upon arguments for closed network structures for innovation. To shed some light on this state of confusion, this paper tests these two conflicting hypotheses on two separate measures of innovation in a service industry setting. One innovation measure is more incremental in nature and regards the implementation of employees' ideas. The other innovation measure is more radical in nature and regards new services. Findings suggest that social network measures are, indeed, powerful predictors of innovation and, further, that the impact of these are likely to be radically different depending upon the type and measure of innovation. Consequently, this paper recommends caution when studying the impact of social network measures upon innovation, and that more fine-grained measurements in particular are needed rather than focusing upon inter-relationships of an overly general and superficial nature.


Leonard and Sensiper (1998) conclude that innovation, to a large extent, is a social and communicative process. Hence, interaction between individuals constitutes an important aspect of innovation activities, potentially influencing the emergence of ideas as well as their further refinement and realization. Thus, it is hardly surprising that a considerable body of literature highlights the importance of networks for innovation, pointing to its importance for information flow, exchange of ideas, access to resources, etc. However, empirical studies show substantial inconsistencies in terms of the effects that different types of network structures are proposed to have upon innovation performance. A debate in this field addresses how innovation activities are influenced by network density. Where some research on social networks with an explicit focus on social capital has underlined the potential benefits of network openness (Burt, 2004), recent contributions have questioned this in the specific context of innovation and argue instead that a certain degree of density offers fruitful ground for knowledge creation and innovation. A shortcoming of several of the studies drawn upon in this debate is that they pay little attention to which type of innovation emanates from the various networks. In light of this, this paper aims to investigate whether different types of communication network structures serve separate ends in terms of incremental and radical innovation.

Networks and Innovation

The importance of networks for innovation has also been emphasized in a range of innovation management sub-fields. Allen and Cohen's (1969) research on the flow of new ideas in a R&D laboratory has paved the way for many followers. Although the term gatekeepers1 had been used in communication research since the 1940s, Allen and Cohen made the term known as being positively associated with innovation. Moreover, these studies reveal that the fundamental role played by informal communication flows in R&D settings point to the need to attend not only to formal organizational structures and roles, but also to more informal relationships between individuals and groups within and across firms.

Another stream of literature draws heavily upon cognitive aspects (including learning theory and knowledge management) pointing out informal network structures, such as communities of practice, as being fruitful arenas for knowledge sharing, learning and innovation (Brown & Duguid, 1991). These spontaneously emerging social networks allow for efficient and effective knowledge creation and knowledge transfer, as their members have a substantial shared knowledge base as well as high levels of reciprocity and trust (Wenger & Snyder, 2000). This leads to open communication and well-functioning feedback mechanisms.

We have also seen an increasing interest lately in the relationship between network structures and innovation, underpinned by the recent development of methods and tools for social network analysis. This renders it possible to map and measure network characteristics in a more explicit and detailed way than ever before. The general applicability of social network research is apparent in the broad variety of theoretical issues that are treated, including the effects social networks have upon the following: power (Kilduff & Krackhardt, 1994); individual performance (Sparrowe et al., 2001); individual creativity (Burt, 2004); and social support (Wellman & Wortley, 1990).

More recent empirical research furthers the interest in the way in which network structures impact innovation processes. Vibrant examples of this are an empirical study by Björk and Magnusson (2009), focusing specifically upon ideation performance, and Sammarra and Biggiero (2008), which emphasizes innovation processes' requirement to access and recombine diverse types of knowledge. Nevertheless, there is still only limited consensus on the specific inter-relationships between network structures and innovation.

Two Conflicting Hypotheses

While there is empirical evidence that networks are important for both individuals and organizations, there is still uncertainty regarding the effects of different network structures upon various types of organizational performance. Specifically, two competing views are still a matter of debate as to which structure is most advantageous (Walker, Kogut & Shan, 1997; Ahuja 2000). One view argues for more open networks where individuals have connections to others who, in turn, are less connected to each other; the other view argues for dense network structures where individuals are connected to others who, in turn, are connected to each other. Figure 1 illustrates networks with varying degrees of openness and density. The network openness declines from left to right whereas the network density increases in the same direction.

Figure 1.

Network Structures

Research regarding specific network structures, such as Burt's structural hole hypothesis (2004), argues that being between others in a social system has distinct advantages because opinion and behaviour are more homogeneous within – rather than between – cliques, which makes an actor connected across cliques more knowledgeable of alternative ways of thinking and behaving. Burt claims that open networks– or bridging structural holes between cliques – is the mechanism by which brokerage becomes social capital. Burt (2004) exemplifies the value generation of betweenness, serving as a bridge, with four interrelated situations:

  • 1it makes people on both sides of a structural hole aware of interests and difficulties in the other clique;
  • 2it facilitates transfer of best practice by noting the need for a certain practice, as well as being able to translate this to the target clique;
  • 3it recognizes that the way other cliques think or behave may have implications for the value of operations in their own clique; and
  • 4it facilitates the synthesis of elements from both cliques' practices into new beliefs or behaviours.

Hence, an open network may create benefits in terms of non-redundant information and knowledge exchange. Previous research favours open networks on the basis of creating information diversity (see, e.g., Hargadon & Sutton, 1997; Burt, 2004). Moreover, the notion of open networks is a major aspect of the seminal work of Granovetter (1985), as well as his finding that weak ties are instrumental in finding new opportunities in the form of new jobs (Granovetter, 1973). Aligned with Burt's structural hole hypothesis, dense networks may create a larger amount of redundant contacts that do not contribute to additional information or resources.

The arguments for density, which have been put forward in close relation to innovation, oppose these arguments for structural holes. Obstfeld (2005) argues that open networks pose a fundamental problem for acting upon new ideas because the dispersed actors are more difficult to co-ordinate due to their opposing interests, unique perspectives and different languages. Conversely, dense networks have a structure conducive to collective action or co-ordination because they facilitate trust, shared interests, perspectives and language (Coleman, 1990; Nonaka, 1994; Wenger & Snyder, 2000). Clique density will bring benefits in terms of greater co-operation, mutual interdependence, greater information sharing, a stronger sense of accountability, greater agreement on roles and expectations, and less tendency to engage in social loafing. Dense ties enable information about any unco-operative behaviour of an actor to circulate more easily; they also facilitate sanctions while deterring unco-operative behaviour (Coleman, 1988; Walker, Kogut & Shan, 1997; Ahuja, 2000; Rowley, Behrens & Krackhardt, 2000).

Regarding innovation, a clear demarcation between these opposing hypotheses could be elusive. These separate network positions pose different opportunities for Obstfeld (2005); that is to say, they may both be applicable. On the one hand, open networks are better at creating opportunities in order to generate new ideas; on the other hand, dense networks facilitate the action to implement the ideas in a co-ordinated fashion. Although Obstfeld makes note of this possibility, he concludes that density is the network variable that has a positive impact upon innovation.

Others also raise this line of argument: that an advantageous network combines density and openness (see, e.g., Reagans & Zuckerman, 2001). For instance, cliques should have high density and also have external networks with structural holes (Reagans, Zuckerman & McEvily, 2004). Kijkuit and Van den Ende (2007) raise yet another relevant dimension; they argue that different types of network structures have advantages at different stages of the innovation process. They propose that open structures should be beneficial for idea generation; however, that the realization of these ideas is furthered by more closely-knit structures.

Diverging Empirical Results

The empirical support for these two hypotheses is as diverging as the antagonists' arguments. Burt (2004) found that managers who had networks that connected separate cliques had better compensation, more positive performance evaluations, were promoted more often, and also express ideas that were more likely to be accepted and evaluated as being valuable. Similarly, Kratzer, Leenders and van Engelen (2004) found that subgroup formation of communication has a negative impact upon team creativity. However, the results of empirical studies regarding Burt's arguments of the benefits of openness upon innovation are inconclusive; there is no evidence that created ideas lead to implementation efforts or implementation success (Obstfeld, 2005).

Obstfeld's own findings (2005) favoured density in an individual's social network, when this was shown to facilitate his or her involvement in innovation. Ahuja's (2000) predictions for the effect of structural holes on firm invention output (i.e., patents) support the density hypothesis; having many structural holes reduces innovation output.

While focusing exclusively upon ideation (i.e., the creative side of innovation), Björk et al. (2010) found that having a higher number of structural holes in an individual's ego network reduces the likelihood that employees will generate valuable ideas. Additionally, Reagans and McEvily (2003) found that both network density and openness improve the ease of knowledge transfer in an R&D organization.

The various manners in which innovations and networks are measured (Björk et al., 2010) are a tentative explanation as to why these radically different results regarding the relationship between network structures and innovation occur. Specifically, empirical studies in the area of innovation rarely make distinctions between different types of innovation; they simply measure different things. In particular, Burt's research is focused upon managers' performance (i.e., compensation, performance evaluations, promotions and expressed ideas), whereas Obstfeld focuses upon involvement in innovation. Ahuja studies patents, while Björk et al. study ideas. Furthermore, Reagans and McEvily's dependent variable is ease of knowledge transfer.

Obstfeld (2005) has also noted some of these empirically diverging results. Accordingly, his finding that density has a positive effect upon innovation may reflect the types of incremental innovations characteristic of the large-scale automotive design process that comprised the empirical setting for his study. As Obstfeld further notes, none of the innovations in his study qualify as a radical innovation; many constituted solutions to well-structured problems. This points to a central issue; it would be instrumental if studies of innovation included measures of both incremental and radical nature. Accounting for more than one type of innovation will allow for better comparisons of results between studies, giving more insight into the possible differences in factors that produce incremental and radical innovation, respectively. This will provide a more solid basis on which to compare innovations in one organizational setting against innovations in another. Radical innovations in one setting may potentially be incremental in another, which the use of multiple innovation measures will make possible to detect.

In explaining the empirical results of their study, Björk et al. (2010) argue along the same lines as Obstfeld: that innovation is measured in different ways and that no distinction is made between incremental and more radical innovations. They also raise the possibility that dense networks could be conducive to incremental innovations, whereas weaker ties are more likely to be beneficial for the generation of more radical innovation. Some theoretical arguments and empirical results support this hypothesis. Well-structured or systematic problem solving actually hampers innovative behaviour (Scott & Bruce, 1994). According to Scott and Bruce, the systematic problem solver will generate conventional solutions to problems, whereas the intuitive problem solver will generate novel problem solutions. This is based upon Jabri's (1991) theory of creative thinking (a development of Koestler's (1964) work) which regards problem solving as two separate and independent modes of thinking: associative and bisociative. The former thinking is systematic and is characterized by habit or following routines, disciplinary boundaries and the use of rationality and logic. The latter thinking is intuitive and is based upon concurrent overlapping domains of thought, a disregard of rules and disciplinary borders, as well as the propensity to exercise imagery and intuition.

These different problem-solving styles may depend upon the individual's social network. On the one hand, dense networks may be conducive for incremental innovations employing a structured and systematic problem-solving style. On the other hand, weaker ties (Granovetter, 1985) and open social networks may provide unexpected information and enable an intuitive problem-solving style, thus, generating radical innovation. Hence, two hypotheses are postulated:

Hypothesis 1: The higher the openness (or the lower the density in organizational members' ego networks), then the higher the organization's degree of radical innovation.

Hypothesis 2: The higher the density (or the lower the openness in organizational members' ego networks), then the higher the organization's degree of incremental innovation.

Furthermore, the inconsistencies in the empirical findings could be due to variations in how employee networks are measured. Network research is based upon various operationalizations, focusing upon a range of network contents (e.g., advice, work interaction or friendship networks), and does not frequently control for the effect of multiple network contents. Moreover, Reagans and McEvily (2003) claim that results are usually inferred regarding informal and formal aspects of networks and their relationship to organizational performance. Their findings suggest that an individual is more likely to exert greater effort to transfer knowledge to a close personal contact.

As mentioned above, Allen and Cohen (1969) reveal the fundamental role played by informal communication flows in R&D settings: that gatekeepers were critical for relaying information into the organization, and that communities of practice (Brown & Duguid, 1991) function as informally constituted networks of shared practice that benefit innovation. Group research often asserts that mutual trust arises from exchange reciprocity where group norms are established (e.g., Coleman, 1988, 1990). This line of research argues that group members are more willing to extend favours to one another since they know that their favours will be returned. This cohesion effect is the principal argument for the benefit of dense network structures. However, this does not account for the fact that, although they are related, group cohesion and mutual trust are, indeed, separate things. Group cohesion may arguably be the result of management directives rather than being the result of a self-regulating choice. Hence, it is reasonable to assert that informal relationships of whatever structure may be more prone to innovative exchanges than are formal relationships. Especially if the ideas and innovations are novel, they will entail the risk that others with whom one has formal relationships will perceive them as diverting attention away from the pressing work tasks at hand, thus triggering normative sanctions against it. This risk is smaller for informal relationships; however, formal work-related relationships, based upon the very nature of the interaction, will be more prone to be effective regarding routine work. Consequently, the following hypotheses are formulated:

Hypothesis 3: The effect in hypothesis 1 is larger for informal networks than it is for formal networks.

Hypothesis 4: The effect in hypothesis 2 is larger for formal networks than it is for informal networks.


Data and Sample

In order to provide an enhanced picture of this inquiry, 22 pharmacies of the government-owned National Corporation of Swedish Pharmacies (NCSP or Apoteket AB in Swedish) have been the subject of a cross-sectional study.

Two general sources of data have been utilized in this study: self-report data from a questionnaire2 and NCSP's balanced scorecard data. There were 252 employees in the districts at the time of the study; all were asked to answer a cross-sectional questionnaire that was available on the internet from the first week of February to the first week of March 2004.3 The number of respondents for the total sample of 22 pharmacies was 239 (88 per cent), of which 93 per cent were women. The response rate for individual pharmacies varied. Eight pharmacies had a response rate of 100 per cent; the response rate of eleven pharmacies ranged between 82 per cent and 91 per cent. Two pharmacies had a response rate of 44 per cent and 50 per cent, respectively; due to this low response rate, these two pharmacies were omitted from the structural equation model analysis. The latter of these pharmacies had only four employees, explaining why two missing responses accounted for the low response rate. The former of these pharmacies had nine employees, two of whom were either on vacation or on leave with a child. The missing responses of the remaining three employees were explained in a follow-up meeting in the district: these employees were involved in the planning of reconstruction activities at the pharmacy. An additional pharmacy had a response rate of 67 per cent, which is rather low; however, this was retained for further analysis. After omitting the two pharmacies, the sample was 239 employees in 20 pharmacies and the response rate was 90 per cent. Nine failed responses were due to vacation, sick leave, being home with child, or because the employee was not considered to be part of the staff (e.g., being a replacement). Furthermore, 15 of the failed responses were for unknown reasons, although the follow-up pointed to these respondents having a particularly demanding schedule during the time the questionnaire was distributed.

Additionally, data from NCSP's balanced scorecard system has been collected to assess the detection of drug-related problems and to implement ideas. This data4 had been obtained for the first quarter of 2004.

Incremental Innovation

Incremental innovation was measured by the number of incremental ideas in the pharmacies. The emphasis of implemented ideas in NCSP is on every employee contributing incremental innovations rather than a few people contributing radical innovations.

The activities regarding the implementation of ideas at the pharmacies are described as continuous improvement activities based upon SIQ's (SIQ, the Swedish Institute for Quality) model for business development. These activities include education of all employees at NCSP in quality issues, stressing the importance of these activities at NCSP. In addition to the basic training, some employees receive extra education to be certified to supervise and support continuous improvement activities. The value that NCSP attributes to this is further evident when this work is described in NCSP's Annual Report for 2004 regarding process activities at NCSP: ‘In the activities of the local pharmacies, use is made of systematic, customer-oriented quality activities. There is a quality model in support of this, which is based upon the seven criteria of leadership, information and analysis, strategic planning, employee development, business processes, business performance, and customer satisfaction’ (p. 27).

Radical Innovation

The NCSP faces major challenges since the role of the pharmacy has been transformed from producing, controlling and distributing pharmaceuticals to one where the pharmaceutical industry produces medications and the pharmacy's role is reduced to that of a distribution channel (Claesson, 1989). A contemporary issue of professional transition is to transform the operations from providing the delivery of pharmaceuticals, to providing clinical pharmacy interventions or pharmacy care (Nimmo & Holland, 1999a, 1999b; Björkman, 2006). This is not exclusively a Swedish transition; it is a global one as well. This transition involves innovation of more radical nature. Moreover, this broadening of the pharmacist's role has been debated as giving the profession the chance to survive (Hepler & Strand, 1990).

One clinical pharmacy intervention is the detection of drug-related problems (DRP) among patients in community practice (i.e., the customers) and constitutes a radical innovative work practice for the pharmacies. Westerlund, Almarsdóttir, and Melander (1999) define a drug-related problem as a circumstance of drug therapy that may interfere with a desired therapeutic objective. Two examples of drug-related problems described by Westerlund, Almarsdóttir, and Melander (1999) are overuse of medication and side effects. The results of their study indicate the importance of education and training of pharmacy personnel in the detection of drug-related problems. DRP detection was measured by the number of detected problems per 1,000 handled prescriptions, and the data was collected from NCSP's balanced scorecard system. Data were obtained for the first quarter of 2004. This variable is measured by pharmacy.

Social Network

Social network interactions are measured by the respondents answering questions about their interaction with colleagues in the pharmacy. In each of the districts, the respondents received a list of all the names of their colleagues in their respective district. They were then asked to indicate with which of the colleagues on the list they have interactions. Interactions were assessed by asking the respondents to classify their interactions into one of the following categories: ‘regarding goals and strategies of work’, ‘regarding experiences of work’, ‘regarding work routines’, or ‘not regarding work’. This was achieved by asking the question: ‘Regarding [name], what is the nature of your interactions?’.

The responses from the first three questions were combined into one summary work network which was recorded as a tie in this network, if the respondent indicated that a tie existed between his or her colleagues for any of these relations. These were considered as work-related social interactions of a formal nature. The social interactions that did not concern work were considered to be of an informal nature.

For each of these interaction contents, the level of openness and density in the employees' social network has been calculated using two measures: betweenness centrality (the percentage of the links between an employees' contacts that pass through the employee ‘in between’ to connect the others in the shortest geodesic path; Freeman, 1979) and density (the total number of ties divided by the number of possible ties). Density and openness are ideally considered to approximate two poles on a continuum – the more open the network, the less dense and vice-versa. However, the two operationalizations are different and they are not inversely related to each other in the case of informal networks (see Table 1). Hence, in this study the two measures are considered as two latent variables of both openness and density.

Table 1. Descriptive Statistics and Correlations
 Means.d.Informal OpennessFormal OpennessInformal DensityFormal DensityInformal SizeFormal SizeEfficiencyRadical InnovationIncremental Innovation
  1. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.

Informal Openness8.331.221.0−0.11−**−0.17*−0.11−0.070.25**
Formal Openness24.536.79 1.0−0.06−0.68**−0.28**0.72**−0.10−0.16*−0.12
Informal Density41.727.53  1.00.10−0.16*0.09−0.14*−0.15*0.12
Formal Density63.215.99   1.00.46**−0.42**0.060.14*0.26**
Informal Size5.64.43    1.0−0.18**−0.23**−0.130.47**
Formal Size20.98.99     1.0−0.25**−0.33**0.18*
Efficiency7.40.76      1.00.52**−0.24**
Radical Innovation2.82.69       1.0−0.06
Incremental4.75.32        1.0

Additionally, network size was calculated as the total number of each individual's direct links with other actors in the network: a measure also known as degree centrality (Freeman, 1979).

Consequently, six network variables were calculated: Formal Openness, Informal Openness, Formal Density, Informal Density, Formal Size, and Informal Size. All network variables have been calculated using UCINET VI (Borgatti, Everett & Freeman, 2002). The measure was calculated by using in-ties rather than out-ties to avoid the potential bias of self-reported centrality (see Burkhardt & Brass, 1990; Sparrowe et al., 2001). The correlations in Table 1 are all below 0.8 indicating absence of multicollinearity.

Control Variables

One control variable is based upon an individual characteristic (i.e., gender). The number of employees at each pharmacy is added as a control variable; this is in line with Tsai and Ghoshal (1998). One reason for this is that large organizations can potentially have more slack resources and, thus, may be able to develop more know-how or to innovate more. It is also more likely that a higher number of employees have the potential to generate and implement more ideas. Since slack resources are important for innovation, it is also important to control for alternative activities that will have an effect on the workload of the employees. Hence, the pharmacies' efficiency is calculated by dividing the number of over-the-counter transactions by the number of work hours that produced those transactions. This measure is also called the WPROD measure (Westlund & Löthgren, 2001). In addition, since the level of radical innovation may divert resources away from incremental innovation and vice versa, these were inserted as control variables in the respective model.

Results and Analysis

Model summaries for the regressions are shown in Tables 2 and 3, respectively; descriptive statistics are shown in Table 1. Tables 2 and 3 also present the results of the regression analysis testing the hypothesized relationships between the different constructs of individuals' ego networks and different types of innovation. To test the research hypotheses, the control variables and the social network measures were entered using a step-wise method. Model 1 presents the model with the control variables and Model 2 adds the variables gauging the various structural characteristics of the individuals' ego network. The overall model for Radical Innovation in Table 2 explains 35 per cent of the variance (R2). Of these, the social network measures contribute 8 per cent.

Table 2. Regression Results with Radical Innovation as the Dependent Variable
 Model 1Model 2
Std. BStd. ErrorStd. BStd. Error
  1. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.

(constant) 2.351 2.362
No. of Employees−0.235**0.018−0.1420.025
Incremental Innovation0.159*0.0360.237***0.037
Informal Openness  0.0130.007
Formal Openness  0.235*0.008
Informal Density  −0.129*0.006
Formal Density  0.232**0.017
Informal Size  −0.225*0.063
Formal Size  −0.337***0.032
adj R20.2860.346
change in R2 0.079
Table 3. Regression Results with Incremental Innovation as the Dependent Variable
 Model 1Model 2
Std. BStd. ErrorStd. BStd. Error
  1. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.

(constant) 4.541 4.488
No. of Employees0.705**0.0290.413***0.045
Radical Innovation0.148*0.1320.216***0.131
Informal Openness  −0.0060.012
Formal Openness  −0.284**0.015
Informal Density  0.159**0.011
Formal Density  −0.152(t = 1.63)0.032
Informal Size  0.365***0.116
Formal Size  0.325***0.061
adj R20.3340.405
change in R2 0.086

Hypothesis 1 proposes a positive relationship between the openness in individuals' ego networks and radical innovation in the organization; this is tested in Table 2. Formal Openness has a positive influence upon Radical Innovation (β = 0.235; p < 0.05), providing support for hypothesis 1. Further strengthening the value of openness, the coefficients of the control variable Informal Size (β = −0.225; p < 0.05), Formal Size (β = −0.337; p < 0.001) and the variable Informal Density (β = −0.129; p < 0.05) are negative and significant. However, Formal Density is also positive and significant (β = 0.232; p < 0.01) suggesting partial support for hypothesis 1. In general, when employees have large numbers of social interaction partners and when these interaction partners are themselves connected with each other, this results in less radical innovation in the pharmacies. However, this is a truth with modification, specifically regarding Formal Density.

The variable for Informal Openness is not significant and Informal Density has a negative influence, whereas Formal Density has a positive influence. Hence, hypothesis 3 is rejected. Informal and formal network structures are of more equal importance.

The control variables show that the number of employees in the pharmacy has a negative impact upon radical innovation: an effect that is attributable to social networks, whereas the involvement in incremental innovation and pharmacy efficiency do not hinder the creation of radical innovation. Incremental innovation (β = 0.237; p < 0.001) and efficiency (β = 0.342; p < 0.001) actually have a positive significant impact upon radical innovation, even when controlling for social networks. The overall model for incremental innovation in Table 3 explains 41 per cent of the variance (R2). The social network measures contribute 9 per cent of these in Model 2.

Hypothesis 2, as tested in Table 3, predicts a positive impact of network density upon incremental innovation; this is also supported. The control variables Informal Size (β = 0.365; p < 0.001) and Formal Size (β = 0.325; p < 0.001), as well as the variable Informal Density (β = 0.159; p < 0.01) are all positive and significant. Providing more support for this hypothesis, the variable gauging formal openness is negative and significant (β = −0.284; p < 0.001) and Informal Openness is not significant. In general, when employees have large numbers of social interaction partners and when these interaction partners themselves are connected with each other, this results in more incremental innovation in the pharmacies. The only indication of an exception to this is that the variable gauging Formal Density is negative, although not significantly.

In Model 2, the control variables show that the number of employees in the pharmacy has a positive impact upon incremental innovation; this is an effect that is attributable, to a certain extent, to employees' social networks. However, employees' involvement in pharmacy efficiency does not hinder the creation of incremental innovation. Radical innovation is positive and significant when controlling for the social network (β = 0.216; p < 0.001). Hence, employees' engagement in radical innovation has no negative impact upon their level of incremental innovation.

There is no support for hypothesis 4 in the models in Table 3. Rather, the analysis suggests that formal and informal interaction content is of a more equal importance.


The empirical findings support that social network measures are, indeed, powerful predictors of innovation and, further, that the impact of these are likely to be radically different depending upon the type and measure of innovation.

In line with the arguments and hypotheses presented in this paper, the structure-innovation relationship is dependent upon whether the innovation is of radical or incremental nature. Björk et al. (2010) and Obstfeld (2005) also highlight this when explaining their respective empirical findings. Accordingly, openness is hypothesized to facilitate more radical innovation, based upon arguments of its propensity to lead to new information as well as conditions for creating and recognizing novel ideas. Denisty is hypothesized to facilitate incremental innovation as this structure is argued to further instrumental co-ordination and emotional aspects such as trust. Both of these hypotheses are supported in the present study.

Interestingly, hypothesis 1 was only partially supported when formal dense networks also had a positive impact upon radical innovation. This provides some support for Obstfeld's (2005) argument that dense networks have a structure conducive for collective action. More precisely, as Reagans and Zuckerman (2001) and Reagans, Zuckerman and McEvily (2004) note, an advantageous network combines openness and density. As the results in the present study suggest, this applies to radical innovation, yet not to incremental innovation. This finding is in line with Kijkuit and Van den Ende's (2007) argument that open structures should be beneficial for idea generation, while the realization of ideas is facilitated by dense structures. At the very least, the results of the present study suggest that open and dense networks work in parallel regarding radical innovation. Future research will benefit from studying the potential interaction between open and dense networks. Central issues are whether both open and dense network structures are needed to produce radical innovation, whether they work independently, or whether certain contingencies determine if the formal or latter assertion is applicable.

Both hypotheses 3 and 4 where rejected, suggesting that both formal and informal interactions are of importance for incremental and radical innovation. However, these results should be interpreted with some caution since the presence of formality and informality in the interactions may be too complex for the present method. In this study, formal interactions are operationalized as interactions regarding work and informal interactions regard non-work issues. Respondents may potentially have difficulty separating these from each other post hoc. These types of interactions may alternate from one to the other during the same conversation, making it problematic for the respondents to ascertain whether or not the conversation was job-related. Moreover, the present method relies upon the relationships between two employees being predominately formal or informal. In reality, these may be combined where relationships consist of a balance of informal and informal interactions.

As this research underlines, different innovations depend upon different contingencies, in terms of social networks. Hence, it is important to continue to make distinctions between different types or phases of innovation. Two categories are offered in the pursuit of making further distinctions: idea generation and idea implementation.

Regarding idea generation, Burt5 himself proposes further refinement. Consequently, social networks may contribute to the following aspects pertaining to the generation of new ideas:

  • 1the awareness of interests and difficulties in separate cliques;
  • 2the transfer of best practice as well as being able to translate this to the target clique; and
  • 3the synthesis of elements from two cliques' practices into new beliefs or behaviours.

However, as Obstfeld (2005) points out, good ideas are not enough; hence, these aspects should be assessed together with aspects pertaining to the implementation of ideas. For this aim, Obstfeld argues that mobilized action depends upon pre-aligned or normatively constrained interests and perspectives, as well as readily available language and trust. What is needed in future research is to consider whether or not various innovations, to varying degrees, depend upon these aspects. Some innovations may be more dependent upon awareness, whereas other innovations may be more dependent upon transfer or synthesis. The relationship between these aspects and social networks will further increase our understanding of innovation in organizations.

Further research interest in this matter is manifest if trust and co-ordination are posed as necessary elements for radical innovation as well, even though they may be more of a challenge to obtain in combination with open communication network structures. Hence, moderating or mediating effects of trust and co-ordination in the relationship between network structure and innovation is also of interest for empirical research.

Additionally, realizing ideas may also depend, to varying degrees, upon the extent of collective action required for adoption and implementation. Consequently, individually oriented innovations may be favoured by openness, whereas collectively oriented innovations may be favoured by closeness. Some innovations may be more in the hands of individual employees to adopt without the involvement of colleagues; other innovations may be very dependent upon collective action. This also relates to the possibility of opportunism that has been raised as a potential difference between open and dense networks, and argued to be more likely in open network structures (Ahuja, 2000). Opportunism may be seen as the breaking of the social norms resulting in, or being the result of, less trusting relationships. This opportunism is regarded as hindering the implementation of innovation; however, opportunism may also be a central characteristic in innovative work if the innovation is of a more individual orientation. For these innovations, it may be beneficial with opportunism and the possibility for employees to diverge from prevalent constraining social norms.

The present findings further raise serious questions regarding the strategic orientation for innovation in organizations. One such question is whether the organizational design fits with the organization's environment. This issue has been the basis for arguments regarding the favours of fit between organizational structure and environmental stability (e.g., Burns & Stalker, 1961). Contingency research originally lacked clarity regarding what constitutes a fit or match between task dimensions and organizational structure (Tushman & Nadler, 1978). Aiming to make such a fit explicit, Tushman and Nadler introduced information processing ideas where fit is a match between information processing requirements facing the organization and information processing capacity of the organization's structure. Social networks are likely to be vital for this information processing. However, the question of how social networks work to produce these effects has received less attention. The present findings suggest that networks do, indeed, create contingencies for different types of innovation; it is probable that this is the case for other kinds of organizational outcomes as well.

Another question of strategic orientation is whether the organization should pursue a single objective in terms of innovation type or whether it should venture into the acquisition of multiple objectives. Given that the social networks in the present study are separate ends on a single continuum, this shows that organizations cannot simultaneously maximize both incremental and radical innovation; they must choose between them or try to optimize. In both of these cases, social network has proven to be a fruitful avenue for further research.

The empirical setting was highly appropriate for organizational studies. Since the separate pharmacies were comparatively homogeneous, they were well suited for research on the influence of the constructs in question when the work tasks, environment, directions from the head office, formal education, occupation, and demographics were generally similar between the pharmacies. The difference between pharmacies was mainly in terms of size. However, the fact that the research is based upon a fairly homogeneous group of respondents is also a limitation. Although homogeneous samples have several advantages (e.g., allowing meaningful and consistent performance metrics), they do not allow investigators to ascertain the generalizability of the findings. Thus, it is important for future research to build upon this research by investigating across different professions and settings. The pharmacies in the present study have close similarities with other knowledge-intensive service organizations, especially those involved in retail service. General health service providers as well as bank offices may be analogue settings where similar results to the present study would be expected.

Practical implications of this research relate to the facilitation of innovation in general, both incremental and radical, in organizations. Social networks are proven to be important and highlight the necessity to manage networks in order to facilitate innovation. In this task, social network analysis is a valuable tool. Those managers that aim to stimulate incremental innovation are advised to pay close attention to the presence and development of dense networks; managers interested in facilitating radical innovation are advised to also monitor and develop open networks. As Kratzer, Leenders and van Engelen (2004) suggest, management should prevent sub-groups from forming by changing work-related relationships of the team members. According to their study, creating inter-dependencies between members of different backgrounds or functionalities prevents the formation of sub-groups.

A prominent challenge in this work is that informal networks are of significance. These types of networks are more difficult to monitor and to manipulate since managers lack direct influence over them. This necessitates high trust relationships between managers and workers.

Future research is recommended to direct attention to the longitudinal perspective of innovation processes, e.g., to activities pertaining to awareness, transfer and synthesis. These stages of the innovation process are likely to require different network structures. Assuming that different stages, in fact, are facilitated by different network structures, it would be interesting to disclose the exact mechanism at play. For instance, Tertius Iungens is a suggested mechanism whereby open networks transform into dense networks; however, other kinds of mechanisms are of importance if they result in network transformations. Longitudinal research designs will enable the capture of these mechanisms and, if the design is given enough granularity, they will also be able to disclose the relevance of one-on-one communication in comparison to group communication, as well as the relevance of the medium of communication.

Furthermore, the use of longitudinal research approaches will reveal the temporal aspects of network dynamics. Hypothetically, open and dense network structures could exist simultaneously and work in combination with each other. Apart from the network structure, the importance of network content may adhere to similar arguments where the influence of formal and informal interactions may vary with time or exist simultaneously. Apart from arguments pertaining to various degree of trust and compliance to norms in various types of network contents, research will also benefit from relating interaction content to the type of knowledge acquired, e.g., complex or simple, in these interactions. Tentatively, and in line with media richness theory (Daft and Lengel, 1986), certain interactions are better in capturing knowledge, hypothetically due to their ability to reduce uncertainty or equivocality.


The major contribution of this study is that it provides further support for employees' social networks being central to innovation and, additionally, that the structure of these networks will contribute to radically different ends in terms of innovation type: radical or incremental. The results provide support that both of the two opposing views of openness or closeness in employees' network structure are beneficial for innovation in organizations, with the crucial amendment that this is a question of the type of innovation. Dense networks are better at supporting incremental innovation, whereas open networks are better at facilitating radical innovation.

Since the type of innovation is suggested to be crucial for explaining diverging results in past research, the present findings highlight the importance of considering the innovation measures used. Researchers interested in the impact of social networks are advised to take note of this. This is especially the case for innovation, where empirical support has now been given; however, it may very well be the case for different types of other performance outcomes as well (e.g., efficiency and customer satisfaction).

The present findings suggest that networks do, indeed, create contingencies for different types of innovation; this is something that is ostensibly the case for other kinds of organizational outcomes as well. Moreover, the results show that organizations cannot simultaneously maximize both incremental and radical innovation; they must choose between them or try to optimize. Social network analysis has proven to be important to use in practice as a tool to choose and optimize between performance outcomes. As an instrument for the attainment of knowledge, social network analysis has proven to be valid for research.


  • 1

    For instance, the term was used by Kurt Lewin when describing the internal decision-making process within the journalistic profession, entailing decisions to relay or withhold information from the public.

  • 2

    The Social Network and Process Innovation items used in this study were chosen from a larger sample in a questionnaire that was meant to gauge a broader set of aspects pertaining to collective learning and organization (Backström, 2004; Hemphälä, 2005).

  • 3

    The four districts had approximately two weeks each to respond to the questionnaire. The questionnaire was available for each respective district in sequence in order to facilitate the administration of the survey.

  • 4

    One new pharmacy was set up shortly before the study and, at the time of the study, had not been connected to the balanced scorecard system. Thus, outcome measures are lacking for this particular pharmacy and it was omitted from further analysis.

  • 5

    Here displayed in edited form.

Jens Hemphälä ( is a researcher at the School of Industrial Technology and Management at KTH in Stockholm, department of Integrated Product Development. He holds an MSc in Engineering Design and a PhD in industrial and organizational psychology. His areas of research are social network analysis, business management, and innovation, focusing on organizational obtainment of multiple goals.

Mats Magnusson ( is Professor of Product Innovation Engineering at KTH Royal Institute of Technology in Stockholm and Director of the Institute for Management of Innovation and Technology in Sweden. He has been Visiting Professor at Luiss Guido Carli University in Rome, the University of Bologna and Aalborg University, and is chairman of the Continuous Innovation Network. His main research interests concern continuous innovation, management of ideas and knowledge, innovation networks and dynamic capabilities.