The Nature of Communication Networks Between Organizations Involved in the Diffusion of Technological Innovations

ABSTRACT - This paper examines the nature of the communication networks that exist between organizations involved in the adoption of technological innovations. We investigate the impact of differing network topologies and alternative models of social contagion on observed adoption patterns. We do this both by means of an empirical field study and a simulation model. The preliminary evidence given here supports the work of previous authors in suggesting that departures from commonly held assumptions about network structure may be the rule, rather than the exception. Our findings from the simulations suggest that such departures can have a substantial effect on the shape of the diffusion curve and supplier strategies to influence it.


David F. Midgley, Pamela D. Morrison, and John H. Roberts (1991) ,"The Nature of Communication Networks Between Organizations Involved in the Diffusion of Technological Innovations", in NA - Advances in Consumer Research Volume 18, eds. Rebecca H. Holman and Michael R. Solomon, Provo, UT : Association for Consumer Research, Pages: 635-643.

Advances in Consumer Research Volume 18, 1991      Pages 635-643


David F. Midgley, Australian Graduate School of Management

Pamela D. Morrison, Australian Graduate School of Management

John H. Roberts, Australian Graduate School of Management


This paper examines the nature of the communication networks that exist between organizations involved in the adoption of technological innovations. We investigate the impact of differing network topologies and alternative models of social contagion on observed adoption patterns. We do this both by means of an empirical field study and a simulation model. The preliminary evidence given here supports the work of previous authors in suggesting that departures from commonly held assumptions about network structure may be the rule, rather than the exception. Our findings from the simulations suggest that such departures can have a substantial effect on the shape of the diffusion curve and supplier strategies to influence it.


The networks of interpersonal communications that link organizations adopting technological innovations are of considerable economic and social importance. The operation of these networks can have a significant impact on observed adoption patterns - even to the extent of determining the success or failure of an innovation. This applies particularly to industrial markets where mass marketing techniques are less prevalent and often substantial technical information is necessary to evaluate the benefit of an innovation. Network communications leading to contagion effects in industrial marketing include requests for advice, site visits, and casual conversations. Despite the importance of such communications there have been few recent studies of industrial networks, and moreover the literature that does address this topic does not reflect current thinking on innovation.

The purpose of this paper is twofold. First, we contrast current thinking on the diffusion of innovations with a seminal paper on industrial networks. We use this contrast to suggest an agenda of three research questions regarding network process and structure. Second, we present preliminary findings from two pilot studies. The first is a field study of communication networks within one industry - designed to establish whether the phenomena we discuss do exist. The second study is a simulation model which allows us to assess the impact of these phenomena on adoption patterns.


The marketing discipline has long recognized the significant role of interpersonal communication in the diffusion of innovations. In their recent review of diffusion, Gatignon and Robertson (1985) advance interpersonal communication networks as one of the key elements of the diffusion process. Despite recognition of its importance little research effort has been devoted-to investigating interpersonal communication - particularly with respect to the structure of communication networks (one exception being Reingen et al 1984). Equally, most of this prior research is concerned with innovations in consumer products rather than industrial innovations. Indeed, since the pioneering work of Czepiel in the 1970's almost no effort appears to have been devoted to industrial networks.

Czepiel (1975) studied interpersonal communication networks relating to the adoption of continuous casting techniques within the US steel industry. Using sociometric techniques he found some support for his hypothesis that this industry formed an identifiable social community. Contrasting Czepiel's methods with the conceptual developments that have occurred since 1975 suggests at least three areas where there is potential for more sophisticated research approaches.

First, Czepiel's network maps were constructed on the basis of respondents nominating other organizations with which they had "regular advice/opinion relationships". Such a definition narrows our view with respect to the messages concerning a specific innovation, implying a rigid and deterministic view of the diffusion process which has been questioned in subsequent years (Gatignon and Robertson 1985). A potential adopter may only activate a small part of his regular (pre existing) network and may also perceive a need to create new links to other members of the population with whom he or she has no regular contacts. The possible linkages between two organizations in a network are detailed in Table 1.

Second, and in common with much of the marketing and diffusion literature, Czepiel assumes that interpersonal influence is transmitted directly by communication between adopters and potential adopters. While this is a common sense assumption, it has subsequently been brought into question by the debate over the "cohesion" and "structural equivalence" models in the sociological literature. The cohesion model makes the traditional assumption that influence is passed directly from adopters to potential adopters during discussions on the merits of the innovation. On the other hand the structural equivalence model stresses competition between people of similar status and roles within a social structure. That is, the more attractive the adoption of an innovation makes one person to the rest of their social contacts, the more likely is a person of similar status also to adopt. Awareness is spread via third parties or by observation. In his re-analysis of the classic Coleman, Katz and Menzel (1957) study Burt (1987) found more support for the structural equivalence model than for the cohesion model.





In many ways the structural equivalence model seems more relevant to a competitive industrial setting than the more commonly assumed cohesion model. In such a setting there may be strong norms restricting direct communication, particularly if the innovation provides a significant competitive advantage to the adopter. Czepiel took the social cohesion view of diffusion and therefore concentrated only on communication links between companies within the industry. While this view may have been legitimate for an innovation such as continuous casting, many innovations have wider relevance and cross industry boundaries; for example, personal computers, facsimile machines, and cellular telephones. For these innovations messages may not be confined to sources originating within the industry and may well come from a variety of sources outside the industry. Equally the more competitive an industry the more likely are members of it to turn to outside sources. Examples of structural equivalence communications in the industrial setting would include information from (and to) suppliers, customers, consultants or industry groups. These communications might be formal advice, casual conversations or direct observations. Such indirect communications would be emphasized under the structural equivalence view of the process rather than under the social cohesion perspective. Table 2 illustrates the possible extremes of structural equivalence/social cohesion t models and regular network / innovation specific network structures. A continuum of possibilities along either axis is possible.

The nature of these links is important both from the perspective of modelers who wish to write equations describing the dynamics of diffusion, and from the perspective of marketing strategists who wish to target promotional campaigns in an effective manner. Most existing diffusion models assume direct contact between "innovators" and "imitators" and homogeneous mixing of the population. This corresponds to a special case of the General Social Cohesion cell in Table 2. This model is also the one assumed by Czepiel (1975). For marketing strategists structural equivalence implies more subtle social cues to adoption behavior and therefore more sophisticated promotional campaigns.

Essentially all the research questions raised so far are also bound up with the topology of the communication network for a specific innovation. This being the case it is instructive both to outline Czepiel's findings with respect to topologies and to make some suggestions about future research. Czepiel (1975, p15) identified two cliques which had significantly different firm characteristics but were connected through a small number of links or "bridges". The implication of findings such as these is that we cannot automatically assume perfect mixing of the population in a given diffusion situation. A network with two cliques linked through a small number of organizations may well display different patterns of diffusion to one in which all members are linked to all other members, as is assumed by most quantitative diffusion modelers. In summary we have raised three research questions which we consider important to the way in which network structure could impact upon the industrial innovation diffusion process, namely:

Do messages about the innovation flow primarily through the regular communication channels or through networks specific to individual innovations?

Is cohesion (direct communication with adopters) or structural equivalence (influence inferred via indirect relationships to others) a better explanation of observed adoption patterns?

What impact do different network topologies have on the rate and patterns of diffusion?

Given the scarcity of marketing network studies this is a difficult research agenda. Here we report the findings from two pilot studies which point to the extent of problems of the dominant perfect mixing paradigm and suggest ways in which they can be addressed. With these two studies we essentially first ask whether the conditions of social cohesion, regular network structures and absence of cliques assumed by most diffusion modelers do actually pertain in a real setting. We then assess the implications of departures from these three assumptions on a typical model of industrial diffusion.


To examine the prevalence of different types of network structure and process we selected the Australian life insurance industry. With 32 major companies this industry is large enough to provide for interesting network topologies and the companies are easily identifiable for research purposes. The innovation we chose to investigate was facsimile machines; a technology which has primarily been adopted in the last 3 years. The respondent was the senior manager in charge of communications and the questionnaire was administered by mail with a telephone follow-up. Respondents were first asked to name all those organizations with which they had regular discussions regarding telecommunications technology, and to indicate the organization type of each contact. Each respondent was also asked a similar question but in the context of communications specially concerning facsimile.


Survey Response

17 out of 32 major life insurance companies responded to our survey. As this sample represents over 80% of industry turnover and the vast majority of industry potential for facsimile machines we feel that it is adequate for our purposes here.

Structure of the Intra-Industry Network

In Figure 1 we present a network diagram of all the regular communication links between the 17 life insurance companies in our sample (including links to companies not included in the survey). In the diagram the companies are arrayed around the circumference of the circle and the nominated communication links are represented by the connecting lines. As can be seen in Figure 1, there are three companies (nodes) who had more extensive links to other organizations (A, B and C). However, the general pattern is for most life insurance organizations to have relatively few links to other members of the industry. Indeed while each company could have links to all others (the 32 top companies were listed, together with an "other" category), the average in this industry is 5.2 (or 3.1 excluding the three major nodes), representing a 16% connected network.

Also of note in Figure 1 is that the three major nodes are linked to each other, and that most other organizations are linked to at least two of these nodes. In contrast to Czepiel's findings for the US steel industry there is little evidence of a clique structure in the Australian life insurance industry. Rather we have a pattern of strong opinion leadership. So, like Czepiel, we can not support the assumption of a completely connected network.

General Versus Innovation Specific Networks.

In Table 3 we present an analysis of the communication links that assisted the diffusion of facsimile machines. As can be seen in the table both regular and new links were activated during the diffusion of facsimile machines. Across the sample about 60% of the pre-existing or regular links were activated for this innovation (17.4% out of 29.2%). However, new links equivalent in number to about 40% of the regular links were also established (6.9% versus 17.4%). Overall we can see that, at least for this industry and innovation, communications were transmitted through both regular and innovation specific links.





In addition to their general communication links with other insurance companies, respondents were also asked to list links with other organizations (including outside companies, consultants, suppliers, overseas companies, and government and professional bodies). Preliminary qualitative research showed that customers (insurance brokers) were not a source of information about telecommunications innovations, nor did they arise during the survey. In the general communication networks there were 90 links between respondents and other insurance companies, while there were 56 with organizations external to the industry. That is, approximately one third of the companies' links are of a structural equivalence type and two thirds are of a social cohesion type.

However, those statistics relating to regular networks understate the role of outside companies in the communication process with respect to the facsimile innovation. Many of the new links formed are external to the industry, while of the preexisting, intraindustry links only a minority are activated. Table 4 presents a breakdown of network links used by the 17 respondents, both internal and external to the life insurance industry. (Suppliers have been omitted from the table as outside organizations because all respondents reported seeking advice from them. To include them would further reinforce the role of external organizations.) It may be seen that network links external to the industry are employed by 10 of the 17 respondents, while internal links are used by only 4. Only one respondent relied solely on word of mouth from within the industry. However, we do see both forms of communications in evidence. It is therefore possible that both the structural equivalence and social cohesion processes were operative during the diffusion of this innovation.





In Figures 2 and 3 we display the details of these external links in the form of inter-industry network diagrams (links between insurance companies not being shown). In the diagrams we can see the heavy reliance on suppliers, removed from Table 4. Presumably suppliers are a primary but biased source of product information while the other companies are less biased but not as informed.


To examine the effect of departures from perfect mixing on the diffusion curve we developed a model of industrial diffusion which we could adjust to reflect these changes. With a product category such as facsimile machines we have a diffusion process within the organization. That is, the organization initially adopts a few machines and progressively buys more until it reaches some saturation level. This intra-organizational diffusion process is occurring at one level. At another level there is a related inter-organizational diffusion process whereby the adoption of facsimile technology by one company leads to its adoption by others. We can use simulation to study the effects of different network structures on both diffusion processes.



The model we use for intra-organizational diffusion is the same as that applied by Bass (1969) although our interpretation is somewhat different. We regard the individual organization i as the population throughout which facsimiles are diffusing. The change in penetration during time t, Sit, is given by

Sit = (pit + qi * Yi,t-1/mi)*(mi - Yi,t-1)     for t > ti0

     = 0                                                     for t < ti0


Yit = level of cumulative sales to organization i

Pit = coefficient of external influence

qi = coefficient of internal influence

mi = organizational potential for innovation

ti0 = Time of initial adoption

The coefficient of external influence, pit, will depend on the organization's environment and its connectedness to that environment, as well as organizational traits unrelated to its current level of adoption. Under the cohesion model it will vary directly with the penetration of the innovation in other organizations, while under the structural equivalence model it will depend on communications with bodies related to the industry who in turn will draw their information from early adopters in this and other industries. Under both scenarios, as the external environment adopts the innovation the pressure on organization i will increase, so pit will be monotonic increasing in the penetration levels of the various other companies within the market. At time t=ti0 (when Yit=0) Piti0*mi gives the level of trial at the time of adoption. The degree of lead, ti0, enters the market penetration equation not only in determining how soon an organization becomes an adopter and thus starts purchasing, but also in generating inter-organizational influences through the network.

The coefficient of external influence may be modeled as a function of the cumulative penetration of other organizations within the industry. Thus we may write:

pit = p'it +p"it *S(dij * eij * Yj,t-1)


p'it = coefficient of influence external to industry

p"it = coefficient of industry influence

dij = 1 if i and j are linked in the network, 0 otherwise

eij = effectiveness of the link between i and j

p'it will change with communication levels about the innovation from outside the industry. In general it will initially increase as consultants and suppliers, for example, move to exploit the growing industry market. p"it, the effect of intra-industry word of mouth may also be dynamic. It is interesting to examine the implications of regular versus innovation-specific links and social cohesion versus structural equivalence on the above model of intra- and inter-organizational diffusion. This we do by changing the assumptions and model structure to match each hypothesis. For example, if communication links are enduring across innovations then we would expect exchanges of information to occur soon after organizations gain experience. If they are innovation-specific then we would expect to see a lag in the establishment of the link. That is, if organizations i and j have no regular communications it is reasonable to surmise that it will take i longer to realize that he has something to learn from j. Under the model of social cohesion we would expect influence external to the industry (p'it) to be small relative to influence internal to the industry (p"it). Under the model of structural equivalence we would expect (p'it) to be large relative to (p"it).




In this section we look at standard network topologies and process assumptions to see if the simulation model can shed insight into the effect of differing innovation networks on the diffusion of innovation. We start with 30 firms in a hypothetical industry. We assume them to be of equal size and distributed equally amongst three cliques. In each simulation we assume that there is one innovator in this market. This firm has a 10% penetration of the innovation at the beginning of the simulation but in every other respect all firms are equal This limited scenario allows us to contrast the results of different network structures on aggregate diffusion patterns. A more complete set of network structures would allow a correspondingly richer set of diffusion patterns.

The Envelope of Diffusion Feasibility

In the first scenario we assume no communication between organizations. For all but the seeded firm (firm 1 in clique 1) penetration is slow and in fact would not occur if we did not include a small time trend in the coefficient of innovation external to the industry (p'it), representing increasing pressure from suppliers and other related organizations. The aggregate result of this "no links present" scenario is shown in Figure 4. This can be contrasted with the second scenario where all members of all cliques communicate with each other (a complete network, shown as "all links present"). Because the lower curve represents the minimum rate of diffusion (associated with no communication) and the upper curve represents the maximum rate (with total communication), any intermediate (incomplete) network structure will lead to a diffusion pattern falling within the area bounded by these two curves. Thus we call the area the envelope of diffusion feasibility. Also included in Figure 4 (as "links within cliques only") is the diffusion curve associated with a network in which there is intra-clique communication, but no inter-clique communication. Although we do not show this here a link between, say, a member of clique 1 and clique 2 greatly accelerates the penetration rate of clique 2, leaving clique 3 being slow to start. Multiple links between cliques obviously speed the innovation process further and can change the trajectory of penetration at the aggregate level. A more detailed description of these simulations and an introduction to the concept of network efficiency may be found in Midgley, Morrison and Roberts (1990).




To examine the effect of establishing new links for a specific innovation relative to using preexisting ones we can introduce a lag into the communication system, representing the time that it takes to establish a link specifically for that innovation. If we look at the resultant diffusion pattern relative to the previous simulations we see that the lag in hearing about the lead of another firm and the delay in establishing links with it, slow down the diffusion process considerably, both in the penetration of clique 1 and then later in the penetration of cliques 2 and 3. This is shown in Figure Sa.

Social Cohesion Versus Structural Equivalence

To compare the social cohesion models we have examined so far to a structural equivalence model we assume there are no intra- or inter-clique links. Rather there are links to a new group of outside organizations (consultants, suppliers, or other groups external to the industry) and that these feed information back into the industry. We also assume that links from industry organizations to these third parties are less effective than intraindustry links, since they describe an environment in which organizations try to protect knowledge about the competitive advantage of an innovation. Links from third parties to other firms within the industry are assumed to be highly efficient because third party organizations have a vested interest in promoting the information they have managed to acquire, either because it will aid in their sales (suppliers) or because that is their function (consultants). If we contrast this network to one with a series of weak inter-organizational links (a complete network) we can see that for networks that are of the same average effectiveness the social cohesion network is more effective early in the diffusion process, while the structural equivalence model gives stronger communication support after the point of inflection. These results are shown in Figure 5b.


The results which we have presented have implications for researchers studying the diffusion process and for managers attempting to direct it. For researchers, the simulations we present above indicate that network structure can have a significant impact on the diffusion process. Of particular relevance here are our findings on the impact of (i) lags in establishing network links, and (ii) asymmetrical effectiveness in network links to third parties. Lags may not only slow down the diffusion curve but also alter the rate at which the innovation is adopted by various cliques. Asymmetrically effective network links can lead to a slower take-off but faster finish to the diffusion process. Such asymmetries would be expected under the structural equivalence model rather than the social cohesion model.

The preliminary evidence from our empirical pilot study suggests that such innovation specific and structural equivalence effects may well arise. We found clear evidence of organizations establishing new (innovation specific) links as well as organizations activating pre-existing links. A substantial proportion of these links were to organizations outside the life insurance industry.

For managers, networks with cliques or highly connected opinion leaders call for directed targeting strategies to ensure that there are not pockets of laggards due to communications isolation and communication occurs as rapidly as possible. The value of seeding individual organizations can be estimated using this modeling approach (see Midgley, Morrison and Roberts 1990 for details).



In conclusion, we have suggested that the effects discussed in this paper can have an important impact on the diffusion process. By our pilot study (as well as drawing on the work of others) we have provided evidence that these effects exist in some situations. In the future we would hope that diffusion researchers will at least test for the presence or absence of clique structures, innovation specific links and influence via structural equivalence before assuming perfect mixing. Alternatively, they should test the sensitivity of their models to such phenomena.


Bass, Frank M. (1969), "A New Product Growth Model for Consumer Durables," Management Science, 15 (5), 215-227.

Burt, Ronald S. (1987), "Social Contagion and Innovation: Cohesion versus Structural Equivalence," American Journal of Sociology, 92 (May), 1287- 1335.

Coleman, James S., Elihu Katz and Herbert Menzel (1957), "The Diffusion of an Innovation Among Physicians," Sociometry, 20 (December), 253-270.

Czepiel, John A. (1975), "Patterns of Interorganizational Communications and the Diffusion of a Major Technological Innovation in a Competitive Industrial Community," Academy of Management Journal, 18, l(March), 6-24.

Gatignon, Hubert and Thomas S. Robertson (1985), "A Propositional Inventory for New Diffusion Research," Journal of Consumer Research, 11 (March), 849-867.

Midgley, David F., Pamela D. Morrison and John H. Roberts (1990), "The Effect of Network Structure in Industrial Diffusion Processes," Working Paper Number 90-019, Australian Graduate School of Management, University of New South Wales, Kensington, NSW

Reingen, Peter H., Brian L. Foster, Jacqueline Johnson Brown and Stephen B. Seidman (1984), "Brand Congruence in Interpersonal Relations: A Social Network Analysis," Journal of Consumer Research. 11 (December). 771-783.



David F. Midgley, Australian Graduate School of Management
Pamela D. Morrison, Australian Graduate School of Management
John H. Roberts, Australian Graduate School of Management


NA - Advances in Consumer Research Volume 18 | 1991

Share Proceeding

Featured papers

See More


Symbolic sustainable attributes improve attitude toward low-quality products: A warm-glow feelings account

Valéry Bezençon, University of Neuchâtel
Florent Girardin, University of Neuchâtel
Renaud Lunardo, Kedge Business School

Read More


Toward Optimal Symbolic Recovery: Why and When “Thank you” is Better than “Sorry” in Addressing Service Delays

Yanfen You, New Mexico State University, USA
Xiaojing Yang, University of Wisconsin - Milwaukee, USA
Lili Wang, Zhejiang University
Xiaoyan Deng, Ohio State University, USA

Read More


M5. The More Expensive a Gift Is, the More It Is Appreciated? The Effect of Gift Price on Recipients’ Appreciation

Jooyoung Park, Peking University
MENGSHU CHEN, Tencent Holdings Limited

Read More

Engage with Us

Becoming an Association for Consumer Research member is simple. Membership in ACR is relatively inexpensive, but brings significant benefits to its members.