Chapter 3 – Literature Review
3.5 Diffusion Networks and Mathematical modelling
3.5.1 Diffusion Networks
3.5.1.3 Communication network analysis
The analysis of communication networks is complex. Although networks have a certain degree of structure or stability, they consist of numerous interconnected individuals who are linked by patterned flows of information. Rogers (2003) identifies the complexity of analysis with the example of 100 members of a system that can mean 4950 possible links. 200 members would be 19,900 links….
Rogers’ work on communication network analysis was clearly influenced by Thomas Valente and as such this research will now be discussed in more detail. Valente’s work on diffusion focuses on a more quantitative and predictive approach than that of Rogers.
While Rogers sought generalisations of individual’s behaviour in order to develop his diffusion model, Valente seeks to understand the relationships between individuals and utilise these relationships to estimate the speed in which innovations may diffuse among a social system.
Valente (1995 p.xi) states that,
The diffusion of innovation occurs among individuals in a social system, and the pattern of communication among these individuals is a social network. The network of communication determines how quickly innovations diffuse and the timing for each individual’s adoption.
Two key aspects of Valente’s work are that of thresholds and critical mass. Valente sees thresholds as an individual characteristic with critical mass models focusing on the entire social system. The differentiation between these will be discussed shortly.
Valente (1995 p.2) is very enthused by network analysis and the use within a diffusion context as diffusion allows a real world application to compare network models. On the subject of network models, he states that,
Network analysis is a technique used to analyse the pattern of interpersonal communication in a social system by determining who talks to whom. Network analysis can be used to understand the flow of personal influence by enabling researchers to define who influences whom in a social system.
Valente (1995 p.4) states that very few diffusion studies investigate social networks and how the flow of communication influences diffusion. In fact, when selecting previous studies for his work Valente found that he was limited by the quality and extent of the previous research.
The history of network models of diffusion can be traced from Coleman et al (1966) and Menzel and Katz (1955). Also Rogers (1962) discussion on opinion leadership through to Granovetter’s (1973, 1982) strength of weak ties, through to Rogers and Kincaid’s (1981) communication networks and finally Burt’s (1987) structural equivalence model.
Valente (1995 p.5) argues that,
Network models allow the specification of whom and to what degree individuals monitor others in the social system, based on the systems social structure. Thus, network models capture the structure of communication and incorporate this structure into predications of individual behaviour.
As was identified by Rogers, it is the uncertainty of an innovation that leads an individual to find out more to minimise their risk and increase the adoption potential.
This means that individuals are more likely to rely on the behaviour of immediate peers rather than mass media or a perception of what the social norm is.
Data Sets Analysed
Before identifying Valente’s concepts, it is important to identify the data sets with which he worked. In order to fulfil Valente’s requirements for network analysis he required both time of adoption data and also network data. This is extremely difficult data to obtain and as such is quite rare. At the time of writing Valente could only identify five studies, which had previously collected the relevant data, and of those five only three had data that existed in the public domain. The three studies used were:
• Medical innovation (Coleman et al. 1966) o 4 communities.
o 125 respondents.
o 18 months.
• Korean family planning (Rogers and Kincaid 1981) o 25 communities.
o 1047 respondents.
o 11 years.
• Brazilian farmers (Rogers et al. 1970) o 11 communities.
o 692 respondents.
o 20 years.
Rather unsurprisingly Rogers, who has already been identified as having a long history in the diffusion research domain, compiled two of these studies.
As one of the most famous studies, some further consideration will now be given to the medical study to provide a clear context and setting for the work of Valente. The
Coleman et al (1966) study is used as an example in numerous diffusion publications as it contained elements of good practice and was one of the first to consider network analysis. In addition, the time of adoption was accurately recorded as the prescription records were used rather than relying on a respondent’s recall, which has already been identified as an important methodological enemy (Rogers 2003). However, the data can still be seen as skewed as the prescription records were only analysed on three
consecutive days each month, therefore any other prescriptions would have been missed thereby altering the actual date of adoption. In addition to this Valente (1995) suggests that the prescriptions were in fact only the first trial and that this may not constitute adoption in the Rogers sense of the word where it implies continued use.
The social network data was obtained by asking doctors to name three doctors with whom they most frequently sought for discussion, friendship and advice. The friendship element of this study formed an important aspect as it showed that even though the
doctors had extensive networks those who were friends adopted within a short time of each other.
The Coleman et al (1966) study was therefore not without its limitations, and the other two studies were also flawed as they relied upon recall for time of adoption and in addition they were carried out in rural communities in developing countries and as such have limited parallels to modern developed countries. However, as Valente (1995) identified, they were the only three studies that combined the necessary data with which to undertake a network analysis.
3.5.1.4 Contagion
Rogers (1983) refers to contagion as the diffusion effect. This is where individuals are exposed to innovations through their peer network and this exposure has a cumulatively increasing influence on adoption. This is essentially referring to peer pressure.
Valente (1995 p.12) states “contagion refers to how individuals monitor others and imitate their behaviour to adopt or not adopt innovations.”
Once again the main diffusion concepts are of interest as it is these communication channels and influence that affect the rate of diffusion. Valente (1995) argues that contagion is modelling the behaviour of those in direct contact, which is similar to the principles of cohesion. Burt (1987) however argues that contagion is more likely to occur through structural equivalence. Valente counters this by suggesting that structural equivalence is the imitation of others in a similar position within the social system, but not necessarily others with whom the adopter communicates. This is an important distinction as network models are focused on communication, and not on people’s perceptions and therefore structural equivalence does not have a role to play in network analysis.