Who tells whom?
Diffusion of innovations through social
networks of children.
The art of tracing the right child?
Diffusion of innovations through social
networks of children.
The art of tracing the right child?
Master Thesis Business Administration
Business Development
Rijks
universiteit
Groningen
Author:
Laurien
Kunst
Student number:
1272780
Organization: Stichting
Cinekid
Supervisor:
Dr.
Jan
Kratzer
Co-assessor:
Dr. Ir. Marjolein C. Achterkamp
Index
Introduction 3
Theoretical framework
Part one: Social networks and the diffusion of an innovation 5
Part two: The role of individuals in the diffusion process 9
Method 13
Results
Results part one 17
Results part two 19
Discussion of the results
Part one 21
Part two 23
Limitations and suggestions for further research 25
References 26
Introduction
There is a growing interest for the children’s consumer market in academically literature as well from businesses point of view. McNeal (1992) identified children as representing three markets in one: a primary market spending its own savings or allowances; a secondary market of ‘influencers’ on mainly parental spending; and a future market of potential adult consumers. Children have their own likes, dislikes, curiosities and needs that are not the same as their parents or teachers (Druin, 2002). Although this seems very obvious, designers, marketers and product-developers sometimes forget that young people are not ‘just short adults’ but an entirely different user population with their own culture, norms and complexities (Berman, 1977).
A lot of marketing research is done in the field of advertising to children and the impact children have on the consumer behavior of their parents (Procter and Richards, 2002). Also, much attention is paid to children’s use of mass media, as a consequence the role of personal communication between children and adults in their environment tends to be overlooked (Hansen and Hansen, 2005). It is widely accepted that word-of-mouth communication (WOM) plays an important role in shaping consumers’ attitudes and behaviors, and in the adoption of new products (Brown and Reingen, 1987). In 1955, Katz and Lazarsfeld found WOM seven times more effective than newspaper and magazine advertising, four times more effective than personal selling, and twice as effective as radio advertising in influencing customers to switch brands. In addition Day (1971) computed that WOM was nine time as effective as advertising at converting unfavorable or neutral predispositions into positive attitudes. Also several other studies suggest that WOM is the most important product success factor, since personal sources are viewed as most trustworthy (Day, 1971; Katz and Lazarsfeld, 1955; Kantona and Mueller, 1954). Furthermore, through multiple dyads and retransmission, one message can reach and potentially influence many receivers (Brown and Reingen, 1987).
Czepiel (1974) and Seth (1971) in addition claim that an active, functioning informal communications network is one of the most important factors that determines the successful diffusion of innovations. There are many studies that investigated the role of social networks in the diffusion of innovations (Rogers, 1962; Coleman et al., 1966; Granovetter, 1973; Burt, 1987). However, despite the amount of evidence about the important role of WOM in fostering the diffusion of innovations, the role of WOM among children has not yet been studied in detail (Procter and Richards, 2002).
Other studies on innovation diffusion highlight the existence of varying thresholds of consumers, which causes for the well-known S-shaped rate of adoption (Rogers, 1983; Granovetter, 1978). However, most theories are based upon research with adults. There are several studies that investigated the factors influencing the varying thresholds. The use of mass media is seen as one of these factors (Mahajan et al., 1990). Other researchers suggest that the involvement of customers into the innovation process (providing them with the experience of participating in the design of a new product) wins their loyalty and stimulates WOM, thereby speeding up the diffusion process (McKenna, 1995; Brown et al., 2005). Although detailed research is done about different roles children could play in the design of new technologies (Druin, 2002), so far little is known about the influence of customer-involvement (child-involvement) on the diffusion of innovations.
of this people vary in the positions and status they occupy in particular social networks, which can affect their impact on what spreads through their networks (Bandura, 2001).
Partly based on the different ways individuals are embedded in their social networks, Rogers (1962) identified some important characters in diffusion theory, such as opinion leaders, innovators, gatekeepers and change agents. Innovators are the first to adopt an innovation. Gatekeepers influence the ability of the individual to adapt the innovation and change agents is normally a person outside the group providing information and advise about the innovation in question (Rogers, 1962; Hansen and Hansen, 2005). Opinion leaders are the actors in the social environment from which others take advise and whom they tend to copy their behavior (Rogers, 1983). They are often found to have a closely-knit network and therefore are often identified as strong ties (Rogers and Cartano, 1962).
Other studies about diffusion of innovations pointed out the important role of the opinion leaders as well (Berelson and Steiner, 1964; Valente, 1996; Czepiel, 1974). These studies emphasize the central position of the opinion leader within his/her social network, which seem to foster the diffusion of innovations within their social network. Although this extensive research about opinion leaders in ‘adult literature’, the role of the opinion leader among children is often overlooked (Hansen and Hansen, 2005).
The research of Granovetter (1973) pointed out that whatever is to be diffused can reach a larger number of people, and traverse greater social distance, when passed through weak links. Burt (1987; 1992) tried to identify and characterize the weak links in networks. He argues that information access and control advantages are created when relations span the ‘structural holes’ between groups. Actors with a social network rich in structural holes monitor information more effectively, and they move information faster and to more people than they can accomplish through formal channels and rules (Staber, 2004). Hence, these weak ties are functioning as bridges between groups and so they tend to connect individuals with varying interests and different perspectives. Information flowing in weakly connected networks tends to be less redundant and thus of greater value to individuals seeking new information (Staber, 2004).
Regarding the information access, a weak tie has a greater chance of seeing good ideas and they know about more opportunities (Burt, 1987; Staber, 2004). The characteristics of a weak tie therefore seem to be highly equal to those of a lead-user, a concept developed by Von Hippel in 1976. Although the lead-user concept is studied extensively, the position of lead-users in their social networks has not been studied in detail (both in ‘adult’ and ‘child’ research). Since lead-user identification has proven to be a difficult task (Von Hippel, 1999), information about the network position of a lead-user could be of great value for the design of an instrument that makes it easier to identify a lead-user. In addition the network position of a lead-user could provide valuable information about the role of a lead-user in diffusing innovations.
Part one: Social networks and the diffusion of an innovation
Diffusion of an innovation
The diffusion of an innovation traditionally has been defined as the process by which that innovation is communicated through certain channels over time among the members of a social system (Rogers, 1983). As Rogers (1962) points out, “the diffusion process consists of (1) a new idea, (2) individual A who knows about the innovation, and (3) individual B who does not yet know about the innovation. The social relationship between A and B have a great deal to say about the conditions under which A will tell B about the innovation and the results of this telling”. Different studies indicate that an active, functioning informal communications network (word-of-mouth processes) plays an important role, if not the most important role, in the diffusion of an innovation (Czepiel, 1974; Sheth, 1971). How do social networks influence the diffusion of an innovation? Innovations need to be communicated between actors within a social system in order for them to start the diffusion process. Therefore communication is central in most theories about the diffusion of innovations (Larsen and Ballal, 2005).
Cohesion Theory
The initial network approach to diffusion research was to count the number of times an individual was nominated as network partner. In turn this variable was correlated with innovativeness. Innovativeness is measured by an individual’s time-of-adoption of the innovation under study (Rogers, 1962; Coleman et al., 1966). These (cohesion) theories argue that informal communication networks provide a better map than formal communication networks for successful diffusion. The understanding of the network is built around the number of times an actor is nominated by other actors through survey or interview. These nominations determine the centrality of an actor in a social system. This specific kind of centrality refers to the number of ties a node has and is defined as degree centrality (Freeman, 1979). The more ties the higher the degree centrality, which cohesion theory argues has a direct impact upon their innovativeness (Coleman et al., 1966). Cohesion is seen as a strong ties theory since it is based around the centrality and closeness of actors and links this to their importance concerning innovation diffusion (Larsen and Ballal, 2005).
The focus on strong ties, was later changed in a focus on both strong and weak ties. This was caused by a study conducted by Granovetter (1973), which studied ‘the strength of weak ties’ (Granovetter, 1973; Liu and Duff, 1972). The strength of weak ties indicates that an innovation is diffused to a larger group of individuals and traverses a greater social distance when it is passed through weak ties rather than strong ones. In any kind of situation, an individual operates in a particular communication environment consisting of a number of friends and acquaintances with whom a topic is discussed most frequently. These friends are usually highly similar with the individual and with each other, and most of the individual’s friends are friends of each other, thus constituting an “interlocking network” (Rogers, 1976). Liu and Duff (1972) showed that an innovation spread most easily within interlocking cliques. The similarity of individuals stimulates and facilitates effective communication inside such a network, but it acts like a barrier preventing new ideas from entering the network.
facilitates the flow of that-which-diffuses, and is called betweenness centrality (Borgatti, 1995). Borgatti (1995) argues that betweenness centrality is one of the most important ways to assess an actor’s importance in the diffusion process. In the second part of this article I will discuss the varying roles actors can play in the diffusion process in more detail, therefore I will come back to both strong and weak ties in the second part of this article. The next two hypotheses (related to the adaptive behavior of a child) will be tested.
Hypothesis 1a: The higher a child’s degree centrality in his/her social network, the stronger a child’s adaptive behavior1.
Hypothesis 1b: The higher a child’s betweenness centrality in his/her social network, the stronger a child’s adaptive behavior.
Thresholds
Theories concerning thresholds in the diffusion of an innovation claims that an individual engages in a behavior based upon the proportion of individuals in the social system already engaged in the behavior (Granovetter, 1978). Threshold models argue that individuals have varying thresholds. Therefore individuals have varying times-of-adoption and thus thresholds are seen as the cause for the S-shaped rate of adoption (Granovetter, 1978). In line with the previous there is a widely accepted method to predict the pattern of diffusion of an innovation, proposed by Rogers (1983). When the cumulative adoption (which give rise to the familiar S-shaped curve) is plotted in terms of actual adoption per period of time over the life of the product, the result is a normal distribution. By using the parameters of such a distribution (the mean and standard deviation) Rogers (1983) developed a system to classify adaptors of an innovation. The first few people that adapt an innovation are considered the innovators (2,5%), followed by the early adopters (13,5%). Then the majority is divided into early and late (34% each) and the ‘laggards’ make up the remaining 16% (Rogers, 1983).
The categorization of adaptors based on innovativeness as measured by time-of-adoption (Rogers, 1983) can be used to create a network threshold distribution (Valente, 1996). Valente (1996) used the adopter categories (in fact Valente deleted the ‘innovators’ from Rogers model, the ‘innovators’ are included in the ‘early adaptors’) to create four personal network threshold categories. The ‘very low network threshold’ actors have thresholds one standard deviation lower than the average threshold. ‘Low and high network threshold’ actors have thresholds bounded by one standard deviation less than and greater than average. Finally, ‘very high network threshold’ actors have thresholds one standard deviation greater than average. The average threshold is defined as the mean threshold for the community (Valente 1996).
Thresholds have been postulated as one explanation for the success or failure of collective action and the diffusion of innovations (Valente, 1996). Threshold models can be used to predict the diffusion of an innovation, since the adoption of an actor can be seen as a function of the adoption behavior of the actor’s network (Valente, 1996). Knowing this, it seems very important to examine what exactly causes the varying thresholds among actors, because the faster the first threshold is reached (the early adaptors), the faster other actors start to adopt the innovation. Next the factors that appear to influence thresholds are pointed out, thereby examining what exactly influences the adaptive behavior of an actor.
Thresholds - External influence
The Bass model (1969) assumes that potential adaptors are influenced by two means of communication: mass media and word of mouth. The Bass model further assumes that the adaptors of an innovation comprise two
1 In this study ‘adaptive behavior’ is used to measure a child’s innovativeness, it is measured through the number of times
groups. One group is only influenced by the mass-media communication (external influence) and the other group is only influenced by the word-of-mouth communication (internal influence). Bass termed the first group ‘innovators’ and the second group ‘immitators’. Mahajan, Muller and Srivastava (1990) provide an approach that develops adopter categories using the same analytical logic as Rogers used and thereby show how the adopter categories based on the Bass model can be used to study differences among their profiles. Combining the Bass model with the network threshold categories of Valente (1996) the actors with very low and low thresholds (early adaptors and early majority) are influenced strictly by mass-media communication (external influence). The remaining actors, with high and very high thresholds will be influenced strictly through word of mouth (internal influence). Other studies confirm that external influence provide actors with earlier awareness of an innovation (Becker, 1970; Weimann, 1982) and freedom from system norms (Menzel, 1960) thereby enabling them to adopt an innovation earlier. In addition Valente (1996) found that early adoption is associated with high external influence. Based on the previous the following hypothesis will be tested.
Hypothesis 2: The higher the degree of external influence, the stronger a child’s adaptive behavior (the lower a child’s threshold).
Thresholds - Customer integration
Empirical research shows the high risk, which is associated with developing new products (Brockhoff, 1998). Accurate understanding of customers needs has been shown near essential to the development of commercially successful new products (Von Hippel, 1986). A way to reduce the market risk of innovations, and a way to overcome the difficulties with designing new products for children, is to integrate the customer into the innovation process (Cooper, 1979; Chesbrough, 2003; Druin, 1999). Empirical research additionally shows that customers are frequently the first to develop and use prototype versions of what later became commercially significant new products and processes (Von Hippel, 1976; Van der Werf, 1990).
These integrated customers thus can be categorized as actors with no threshold or a very low threshold, since they were the first to adopt the innovation. Although not frequently mentioned in diffusion literature, it seems presumable that customer involvement into the NPD process influences the threshold of an actor, in the sense that it reinforces an actor’s adaptive behavior. In addition McKenna (1995) claims that the integration of customers into the innovation process wins their loyalty and thereby speeding up the diffusion of an innovation. This argument is in line with some recent studies on intermediating antecedents of word-of-mouth intentions and behaviors. Brown et al. (2005) show that customer commitment/involvement exert significant influences on positive WOM intentions and behaviors.
Although some researchers pointed out the direct link between customer involvement and innovation diffusion, literature often focuses on the relationship between customer satisfaction and WOM. Customer involvement seems to foster product advantage in terms of quality, reliability and uniqueness, which in turn is positively correlated with product market performance and customer satisfaction (Li and Calantone, 1998). In addition many researchers have found that customer satisfaction/dissatisfaction with the quality of a new product plays a crucial role in the diffusion of innovations and in facilitating word-of-mouth communication (Ha, 2006). Anderson (1998) as well reports a positive relationship between customer satisfaction and word-of-mouth.
As presented in the Bass model the actors with higher thresholds (late majority and laggards) are presumed to be eventually persuaded to adopt the innovation through WOM. To consolidate the previous arguments, I will test the following hypotheses:
Hypothesis 3a: The higher the degree of involvement into the NPD process, the stronger a child’s adaptive behavior (thus the lower the average threshold of a child).
Hypothesis 3b: The higher the satisfaction with the quality of an innovation, the stronger the child’s adaptive behavior (thus the lower the average threshold of a child).
Part two: The role of individuals in the diffusion process
What exactly is the diffusion process and what factors are important determinants of the diffusion process? The first part of this article addresses to answer these questions. Social networks of actors seem to play a central role in the diffusion process. So far, we know more about the way social networks could influence the diffusion of an innovation. Yet, little attention is paid to the role and characteristics of the actors who are part of a specific social network. Therefore in this part I will present insights about the varying roles of actors in their social network and how these roles contribute to the diffusion of an innovation.
Scale free networks
For a very long time science treated all complex networks as being completely random (Erdös and Rényi, 1959). In random networks hubs will have approximately the same number of links (Barabasi and Bonabeau, 2003). Random graph models have been widely studied and are quite useful since many of their properties can be computed analytically. However in 1998 Barabasi and other researcher studied a major network, the World Wide Web, and found a totally different network than they expected to find. Instead of finding a random pattern of connections, they found that a few highly connected pages were essentially holding the World Wide Web (Barabasi and Bonabeau, 2003). They called this type of network ‘scale-free’. Random networks are characterized by normal distributions, instead a scale free network are characterized by the existence of a few hubs and therefore is described by a continuously decreasing function (Barabasi and Bonabeau, 2003).
After this first study many other researchers found evidence for the existence of scale free networks. Knowledge about scale free networks has implications for understanding the spread of computer viruses, diseases, but also for the diffusion of innovations. Because hubs are connected to many other nodes, they are likely to play a crucial role in the diffusion process (Barabasi and Bonabeau, 2003). One node has adopted an innovation and will ‘infect’ at least one hub. Once the hub has adopted an innovation it will influence other nodes, eventually compromising other hubs, which will then diffuse the innovation throughout the entire system. In order to speed up the diffusion of an innovation identifying ‘the hubs’ in a social network is essential. Who are the actors that can be identified as hubs? What are their characteristics and how can these actors foster the diffusion of an innovation?
Opinion leaders
As Barabasi and Bonabeau (2003) pointed out hubs are likely to play a crucial role in the diffusion of an innovation, because hubs take a central position in a specific network. Since this study examines social networks, the hubs can be identified as specific individuals (actors). In the study of Rogers (1983) important characters are identified. The most important character for the diffusion of innovations within their social networks is called the opinion leader. According to Berelson and Steiner (1964) opinion leaders are “trusted and informed people who exist in virtually all primary groups, who are ‘models’ for opinion within their group, who listen and read in the media, and who then pass on information and influence to their circle relatives, friends and acquaintances.”
Opinion leadership can be measured by the number of network nominations received (Rogers and Cartano, 1962). Opinion leadership is generally related to a high average of network nominations (Valente, 1996). Recall the definition of degree centrality from in first part of this article, hence a high degree centrality is often associated with opinion leadership, therefore opinion leaders are likely to play a crucial (central) role in the diffusion of an innovation.
than non-leaders. In addition Czepiel (1974) pointed out that opinion leadership with respect to the innovation is greatest among early adopters than among later adopters. Rogers (1962) argues as well, that early adaptors often are more integrated in a social system, like opinion leaders.
As pointed out in the first part of this article Coleman et al. (1966) as well found that network centrality was highly correlated to innovativeness. Coleman’s theory is classified as a strong ties theory. In other words opinion leaders are identified as strong ties, which means that they are important for the diffusion of an innovation, especially within their own social network. In order to reach a larger number of people and traverse greater social distance, weak ties will play a more crucial role than strong ties do (Granovetter, 1973). Before I will examine the role of weak ties, let me first set up the hypothesis to test the position of an opinion leader in his/her social network.
Hypothesis 4: The higher the degree centrality of a child in his/her social network, the higher the extent to which a child can be identified as an opinion leader.
Weak ties
A common way to measure the strength of ties is the number of sociometric choices an actor received from others in a specific study (Granovetter, 1973; Valente, 1996). The more choices (or nominations) an actor receives the more this individual is characterized as ‘central’. The actors with few choices, the weak ties, are characterized as ‘marginal’ (Granovetter, 1973). The ‘marginal’ actors thus have a low degree centrality. However, Granovetter (1973) claims that whatever is to be diffused can reach a larger number of people, and traverse greater social distance, when passed through weak ties rather than strong. Following this weak ties are likely to play a central role in diffusing an innovation. The question that remains is ‘what kind of central role’?
As I mentioned in the first part of this article, besides degree centrality there are other ways to highlight the differences between important and non-important actors in a social network. All such centrality measures attempt to describe and measure properties of ‘actor location’ in a social network (Wasserman and Faust, 1994).
Granovetter (1973) claims that actors with many weak ties are more important than others because removing those actors would do damage to transmission possibilities throughout the network. In fact, Granovetter (1973) highlights the importance of actors who act as ‘bridges’ between subsets of actors. Granovetter (1973) reasons that except under unlikely conditions, no strong tie is a bridge and thus weak ties can only be identified as bridges. A bridge is an actor that is critical to the connectedness of the network. If the bridge is removed the remaining network has two or more subsets of actors, between whom no communication can travel (Wasserman and Faust, 1994). This means that interactions between actors might be dependent on the other actors, especially the actors who lie on the paths between the two.
In addition, Burt (1987) argues that opinion and behavior is more homogeneous within than between groups, which creates holes in the information flow between groups: the structural holes. Actors whose network span the structural holes (bridges) have early access to diverse, often contradictory, information and interpretations, which gives them a competitive advantage in seeing good ideas and thus early access to innovations.
that betweenness centrality is one of the most important ways to assess an actor’s importance in the diffusion process.
Innovators and lead-users
Several studies tried to identify the actors who lie on the paths between (subsets of) actors. Rogers (1962) argues that the very first adaptors (the so called innovators) of an innovation usually receive a small number of sociometric choices from other actors. The next group Rogers (1962) highlights, the early adaptors, often is a more integrated part of the social system. As I pointed out before, opinion leaders are often found to be early adopters (Czepiel, 1974). A major issue in the study of innovation has been whether opinion leaders are also innovators or whether different individuals play the two roles. Although overlap sometimes occurs, most studies found the roles were played by different individuals (Hansen and Hansen, 2005).
Empirical research shows that customers are frequently the first to develop and use prototype versions of new products and processes (Von Hippel, 1976; Van der Werf, 1990). Consequently these customers are the very first adaptors of an innovation. Innovation by customers tends to be concentrated among lead-users (Urban and Von Hippel, 1988; Morrison et al, 2000).
Lead-users have two specific characteristics. First lead-users experience needs for a given innovation earlier than the majority of the target market. Second lead-users expect attractive innovation-related benefits from a solution to their needs and therefore are motivated to innovate. These two characteristics make lead-users highly valuable for marketing research and therefore the most important people to integrate into the innovation process (Von Hippel, 1986). An important indicator of a lead-user is the dissatisfaction of the user with the current market offerings (Lühtje and Herstatt, 2004).
Lead-users are (frequently) the first ones to adapt an innovation (Von Hippel et al., 1999). Von Hippel (1986) reasons that the longer it takes for an innovation to diffuse the more lead-users will benefit from their information. He claims that users able to obtain the highest benefit from the solution to a new product need will be the ones who have devoted most resources to understand those needs. It follows that these lead-users should have the best real-world understanding of the new product need.
This indirectly implicates that lead-users will not share their understanding of the new product need with the other actors in their network, because this would devaluate the value of their information. Research into “informal information trading” has shown that users sometimes do trade and share their innovation-related information. However in marketplaces where users are direct rivals, hiding rather than sharing of innovations by user-innovators will be the norm (Morrison et al., 2000).
Lead-users and their position in a social network
Although the characteristics of a lead-user are investigated and exposed in detail, little attention is paid to the position of a lead-user in their social network. The identification of lead-users is found to be a difficult and time-consuming process (Von Hippel et al. 1999). The network position of a lead-user therefore could be of great value to design an instrument for the identification of lead-users. In addition information about the network position of a lead-user could have important implications for the role of a lead-user in the diffusion of an innovation.
select form and synthesize. Accordingly, a lead-user has early access to information since the lead-user is the first to innovate, and therefore it is presumable that a lead-user possesses a central position between groups. In this way the user is able to acquire the valuable information. This leads to the expectation that a lead-user will have a high betweenness centrality.
Another way to assess the extent to which an actor’s social network spans the structural holes is presented by Burt (1992). Burt’s measures of structural holes calculate the efficiency of an actor’s social network. The more efficient an actor’s network is, the more the actor’s network spans structural holes. Beside the expectation that a lead-user will have a high betweenness centrality it would be likely that the lead-user will have a highly efficient network. The previous leads to the next two hypotheses:
Hypothesis 5a: The higher the betweenness centrality of a child in his/her social network, the higher the extent to which a child can be identified as a lead-user.
Method
Study design, procedure and participants
In order to test the hypotheses different data were collected. To set limits for the research the decision was made to collect the data through examining school classes. School classes can be used easily to analyze social relationships between children (Defares et al., 1971). I used a pre-experimental research design (Baarda and De Goede, 2001); the experiment is conducted on three public primary schools in the Netherlands. This is done in combination with a survey, to be more specific with a (partly) self-administered questionnaire. To measure the differences in the adaptive behavior of children, on each school an innovation (called Kijkradio2) was
introduced. To examine the effect of different ‘treatments’ on the adaptive behavior of the children, the innovation was introduced in a different way on each school.
At the first school children were intensive involved into the new product development process of Kijkradio. The roles that children can play in the design of a new technology are studied in detail (Druin, 2002). Druin (2002) distinguishes four main roles: the child as user, tester, informant and design partner. The children at the first school played the role of both user and tester. The involvement of these children is done by taking into account research methods that are suitable for both the ‘user-role’ and the ‘tester-role’ (Druin, 2002). At the second school Kijkradio was introduced with a simulated mass media campaign, no further involvement of children occurred. Finally at the third school Kijkradio was introduced with no additional attention. About two weeks after the introduction of Kijkradio the children filled in a questionnaire.
Kijkradio is aimed at children with the age of 8 till 12, which are the children in the four highest classes of the primary school (‘groep 5’ till ‘groep 8’). The target population study therefore is ‘children in the Netherlands with the age from 8 till 12’. The operational population is ‘children from the four highest classes of public primary schools’. Piaget (1971) saw the age of 7 as a major cognitive turning point. Around this age children make the transition from preoperational to the concrete operational stage. This implies that children around this age become better at logical, systematic thought using multiple pieces of information. In addition language skills develop and children learn about classifications.
I decided to use a cluster sample, containing the children in ‘groep 5’ and ‘groep 7’. These groups contain children of all ages (8 till 12), as a result of recidivists. Moreover a whole group (or class) can be used easily to analyze social relationships between children (Defares et al., 1971). The final cluster sample contained three public primary schools, which means 3 classes ‘groep 5’ and 3 classes ‘groep 7’. The total sample contained 141 children.
Data collection
Collecting data from children is difficult because children’s interpretations of questions and definitions often are ambiguous. A child’s cognitive, communicative, and social skills are still developing as he/she grows older, and this affects a child’s ability to answer survey questions (Borgers et al., 2004). To collect data about children’s social networks I used sociometric method, called SAGS (Seracuse-Amsterdam-Groningen Sociometrische Schaal). This method uses a specific questionnaire (appendix 1). The questionnaire is designed as a complete list of actors (children in a specific class) and ratings are gathered from each actor about their ties to other actors. The ratings are made by choosing one of the five possible categories for the strength of each tie. SAGS
2Kijkradio is an interactive online tool, which enables children to make their own news program. Kijkradio is a product of a
has the advantage of being a reliable and valid instrument to examine the networks of (young) children (Defares et al., 1971). The SAGS questionnaire was filled in by 135 children.
As pointed out before two weeks after the first introduction of Kijkradio the children filled in a questionnaire. This questionnaire focused on the perception of the quality of Kijkradio, the extent to which children adapted the innovation and the different roles children could play in the diffusion process (appendix 2). The questionnaire is designed by taking into account the expert appraisal coding schedule for questionnaires for children and adolescents (Borgers et al, 2004). Eventually there were 129 children who filled in both the SAGS questionnaire and the second questionnaire properly.
Analysis and measures
Hypotheses 1a, 1b, 4, 5a and 5b contain specific measures of social networks. To provide insight about the way concepts are operationalized and data was analyzed, let me first explain more about social network analysis. The data that was gathered through the SAGS questionnaire resulted in a matrix containing valued and directed relations between the children of each class. There are different ways of collecting data on social networks, which results in different ways of representing and measuring social networks. A model that presents a social network with an undirected dichotomous relation is called a graph. So, a graph is a tie that is either present or absent between each pair of actors (Wasserman and Faust, 1994). The data gathered through SAGS thus contains more complicated (valued and directed) relations between children.
To find out which network measures could be calculated, I analyzed whether the collected data could be presented in a matrix containing undirected relations, thus as a symmetric matrix. This was done for each class, using UCINET VI (Borgatti, Everett, and Freeman, 2002). First each data matrix was transposed. The transpose of a matrix is constructed by interchanging the rows and columns of the original matrix (Wasserman and Faust, 1994). To measure the value of reciprocity between each pair of classmates, the correlation between the original and the transposed matrix was computed. A Pearson’s correlation coefficient higher than 0 implies a positive correlation between the two matrices. A (strong) positive correlation is seen as enough evidence to symmetrize the matrix.
The Pearson correlation coefficient for each transposed matrix was on average higher than 0.5, which implies a strong positive relation. This correlation was significant at a 0.01 significance level. To make sure that a symmetrized matrix would not influence the values of the network measures, both in- and outdegrees were calculated. The differences in the in- and out degrees of each actor were very small. In combination with the high correlations, this was enough evidence to symmetrize the matrices. The matrices were symmetized using the minimum symmetrizing method, because friendship can be defined as a feeling that needs to be mutual. If it is not (completely) mutual the lowest score will represent the value of friendship between two actors.
To compute specific network measures, such as betweenness centrality, data was dichotomized. In the SAGS questionnaire children valued their friendship relations from 1 to 5, where 1 refers to ‘no friendship at all’ and 5 refers to ‘best friends’. When a child valued the friendship with 3, this means the child is not really a friend, but is closer related to the child than children who are valued with 1 or 2. Therefore the data was recoded as follows: the values 1,2 and 3 were recoded into 0, which means there is no friendship between the children. Values 4 and 5 were recoded into 1, which implies there is a (strong) friendship between the children. After the preparation of the data, the network measures could be calculated.
Degree centrality
Freeman, 2002). The valued and undirected relationships between the children of each class were taken as input for this calculation.
Betweenness Centrality
Betweenness centrality refers to the probability that a ‘communication’ from actor j to actor k takes a particular route. Hereby it is assumed that lines have equal weight and communications will travel along the shortest routes, and therefore it is assumed that such a communication follows one of the geodesics (Wasserman and Faust, 1994). If Bjk is the proportion of all geodesics linking actor j and actor k, which pass through actor i., the betweenness of actor i is the sum of all Bjk where i, j and k are distinct (Borgatti, Everett and Freeman, 2002). This measure proposed by Freeman (1979) and is calculated with UCINET VI (Borgatti, Everett and Freeman, 2002).
Network efficiency
Network efficiency is the effective size of a network divided by the number of alters in an ego’s network (Burt, 1992). Burt's measure of the effective size of ego's network is the number of alters minus the average degree of alters within the ego network (not counting the ties to ego). This measure proposed by Burt (1992) is calculated with UCINET VI (Borgatti, Everett and Freeman, 2002).
Adaptive behavior
The adaptive behavior of a child is measured through the number of times a child has used Kijkradio. The use of each child is recorded through the number of times a child has signed in on the website.
Degree of external influence
The degree of external influence is measured as follows. As previously mentioned I conducted a pre-experiment on three public primary schools, in which I introduced Kijkradio on three different ways. On the second school Kijkradio was introduced with a simulated media campaign, which contained promotional posters, brochures and a video demonstration. The degree of external influence on the second school (42 of the 129 children in the sample) therefore is presumed ‘high’. The remaining children (the control group) are presumed to have a low degree of external influence. This variable is included as dummy, with low degree of external influence = 0 and high degree of external influence = 1.
Degree of involvement into the NPD process
The degree of involvement into the NPD process is measured the same way as the degree of external influence. Only the children on the first school (48 of the 129 children in the sample) were involved into the NPD process. As mentioned before, this is done by using children in the role of ‘user’ and in the role of ‘tester’. The degree of involvement of each child on the first school therefore is presumed as ‘high’. The remaining children (the control group) are presumed to have a low degree of involvement. This variable again is included as dummy, with low degree of involvement = 0 and high degree of involvement = 1.
Satisfaction with the quality
The ‘satisfaction with the quality (of Kijkradio)’ measure consists of four indicators, which were measured on a 1-to-5 Likert-type scale. These four indicators were based on a current way to describe the characteristics of software/computer games (Howland, 1998). Based on their internal consistency (Cronbach’s alpha = 0,653) the
four indicators were combined into a single measure of ‘satisfaction with the quality (of Kijkradio)’.
3 Nunnally (1976) has indicated 0,7 to be an acceptable reliability coefficient, but lower thresholds often are used in
The extent to which a child can be identified as an opinion leader
The opinion leadership measure consists of three indicators, also measured on a 1-to-5 Likert-type scale. A self-designating technique that demonstrated to provide a valid measure for opinion leadership is by asking how frequently one is contacted for advise and how often one gives advise (Corey, 1971). In addition to these two indicators I asked if a child in comparison to his/her friends knew significant earlier that there was a new toy/ game. A Cronbach’s alpha of 0.797 pointed out that the internal consistency of the three indicators was very high. Therefore the indicators were combined into a single measure of opinion leadership.
The extent to which a child can be identified as lead-user
The measure for a lead-user consists of six indicators, measured again on a 1-to-5 Likert-type scale. The indicators refer to the two characteristics of a lead-user Von Hippel (1986) pointed out. In addition the indicator of Lühtje and Herstatt (2004) was included, which refers to the dissatisfaction of a user with current market offerings. The value of Cronbach’s alpha was 0.714, which point out the internal consistency was high enough to combine the indicators into a single measure of a lead-user.
Control variables
There are many other factors that have been shown or may be shown to influence social networks of actors and the adaptive behavior of an actor. While it is not possible to include all other variables in this study, I chose to include two variables that have been suggested to affect the social networks of actors and the adaptive behavior of actors.
Gender
Boys and girls develop in the same way and have the same fundamental needs. In contrast to this the way they express and satisfy their needs and feelings is different (Del Vecchio, 2002). Therefore differences in the adaptive behavior of children are likely to occur. Kalmijn (2003) in addition reports that gender influences social networks, since women are likely to have more frequent contacts with friends than men do. This variable is included as dummy, where male = 1 and female = 2.
Age
Children of different ages have diverse likes and dislikes, as a child grows older the thoughts, expectations and feelings of a child change (Craig and Baucum, 2003). This does not only influences their adaptive behavior of an innovation, it also influences their friendships (and thus social networks) with other children (Craig and Baucum, 2003). Additionally research shows that social networks are not stable over time. Stages in the life course will influence the social networks of actors (Kalmijn, 2003).
Table 1 shows the descriptive statistics about the variables included in this study.
Table 1.. Descriptive statistics about variables in this study
Variables Minimum Maximum Mean Std. Deviation N
1 Adaptive behavior 0 68 7,80 8,23 129
2. Degree centrality 18 67 41,06 10,63 129
3. Betweenness centrality 0 98,6 14,95 23,77 129
4. Degree of external influence 0 1 0,67 0,47 129
5. Degree of involvement into NPD 0 1 0,63 0,49 129
6. Satisfaction with quality 2,5 5,0 4,27 0,56 129
Results
Results part one
Hypotheses 1a and 1b are tested by conducting a regression analyses for adaptive behavior. Hypotheses2, 3a and 3b are also tested by conducting a regression analysis for adaptive behavior, but are presented separately from the first two hypotheses since they are related to innovation diffusion in a quite different way. Before I present the results of these regression analyses, let me first show the bivariate correlations for each of the variables of the hypotheses of part one. Table 2 shows the correlation coefficients between variables 1, 2, 3, 4, 5, 6, 10 and 11.
Table 2. Bivariate correlations for variables of part one
Variables 1 2 3 4 5 6 10 11
1 Adaptive behavior -
2. Degree centrality 0,383** -
3. Betweenness centrality 0,342** 0,434** -
4. Degree of external influence 0,077 0,013 -0,015 -
5. Degree of involvement into NPD -0,066 0,309** -0,092 -0,535** -
6. Satisfaction with quality 0,102 -0,146 -0,030 0,190** -0,233** -
Control variables
10. Gender -0,136 -0,045 -0,090 0,126 0,001 0,028 -
11. Age 0,271** 0,425** 0,142 -0,190 0,157 -0,113 -0,186* -
* Significance at 0,05 ** Significance at 0,01
As the correlations coefficients indicate, there is a significant positive relationship between the adaptive behavior of a child and the degree centrality (0,383), the betweenness centrality (0,342) and the age of a child (0,271). Another result is that degree centrality positively relates in a statistically significant manner to betweenness centrality (0,434) and to age (0,425).
The degree of involvement into the NPD process relates in a significant positive way to degree centrality (0,309). Since the network data was gathered before I conducted the experiment (the involvement of the children into the NPD process), this positive correlation does not implicate that higher involvement of children causes a higher degree centrality. This positive correlation just implicates that the children on the first school have a significant higher degree centrality than the children on the remaining two schools. The negative relation between degree of external influence and degree of involvement (-0,535) is a logic consequence of the pre-experimental design, in which a high degree of external influence was only present on the second school and a high degree of involvement was only present on the first school.
Then there is a significant negative relation between the degree of involvement into the NPD process and the satisfaction with Kijkradio (-0,233). This indicates the children that were involved into the NPD process, were less satisfied with the quality of Kijkradio than the remaining children in the experiment. Finally the analysis shows a negative correlation between age and gender (-0,186), which implies that the sample contains more boys when the children have a higher age.
explains 14,0 percent of the variance in adaptive behavior. When betweenness centrality is included in the regression analysis (model 2) the explained variance rises to 17,2 percent. Looking at the coefficients of both centrality variables, it follows that degree centrality and betweenness centrality relate to adaptive behavior in a significant positive way.
To test hypotheses 2, 3a and 3b another regression analysis for adaptive behavior was conducted. The regression analysis started with the same base model, presented in table 3. As pointed out before the base model explains 6,6 percent of the variance in adaptive behavior. However, the stepwise regression analysis showed that the explained variance in adaptive behavior did not improved when the independent variables degree of external influence, degree of involvement and satisfaction were included in the model. This suggests that none of these variables relate in a significant way to adaptive behavior. The correlation analysis in table 2 supports this idea, since the table shows no significant correlations between the three independent variables and adaptive behavior.
However, to draw proper conclusions about the hypotheses, I conducted an Independent Samples T Test for hypothesis 2 and 3a. The results of these analyses are presented in table 4. The table shows that both the degree of external influence and the degree of involvement did not cause for significant differences in the average adaptive behavior. These results confirm the previous findings. Hypothesis 3b consists of the independent variable satisfaction. The satisfaction-measure was combined out of 4 indicators that were measured on a 1-to-5 Likert scale (see Method). Therefore a One-way-ANOVA analysis was conducted. This way I examined whether the adaptive behavior is significant different for the varying values of satisfaction. The degrees of freedom were 128; the value of the test statistic F is 0,799, with a P-value 0,618. This result implies that the varying values of satisfaction do not cause for significant differences in the average adaptive behavior. Again this result supports the findings from the regression analysis and the correlation analysis.
The last hypothesis in first part of this article (hypothesis 3c) is tested by conducting an Independent Samples T Test. Hence, this test analyses whether the degree of involvement into the NPD process causes for significant differences in the average satisfaction with the innovation. The results in table 5 show that the mean satisfaction of the children with a high degree of involvement was significant lower than the mean satisfaction of children with a low degree of involvement. This result corresponds with the result of the correlation analysis, which showed a significant negative correlation between degree of involvement and satisfaction (-0,233).
Table 3. Regression analysis for adaptive behavior (hypotheses 1a and 1b)
Variables Base model Model 1 Model 2
Constant -12,905 (6,558) -4,373 (2,691) -2,505 (2,751) Gender -1,449 (1,431) -1,644 (1,365) -1,353 (1,345) Age 2,078** (0,654) 0,858 (0,701) 0,973 (0,689) Degree centrality 0,297** (0,063) 0,224** (0,069) Betweenness centrality 0,075* (0,031) Adjusted R 0,066 0,140 0,172 R 0,074** 0,147** 0,038* * Significance at 0,05 ** Significance at 0,01
Table 4. Independent Samples T test for hypotheses 2 and 3a
Variables (adaptive behavior) Mean (sd) t df Sig.
Degree of external influence: High (N=42)
Low (N=87) 8,71 (10,63) 7,37 (6,81) -0,87 127 0,386
Degree of involvement: High (N= 48)
Table 5. Independent Samples T test for hypothesis 3c
Variables (satisfaction with Kijkradio) Mean (sd) t df Sig.
Degree of invovlement: High (N=48)
Low (N=81) 4,10 (0,53) 4,37 (0,55) 2,69 127 0,008**
** Significance at 0,01
Results part two
Part two starts with the research of Barabasi, who claims that networks are scale free. A scale-free network is characterized by a continuously decreasing function (Barabasi and Bonabeau, 2003). Figure 1 presents the distribution of the social network of the children in this study. This distribution can be described as normal, which characterizes a random network, and supports the earliest findings on social network distributions (Erdös and Rényi, 1959). This implies that there are just a few children who have a small or large number of friends; the majority of the children have a number of friends that is somewhere in between. Although Barabasi and other researcher found evidence for the existence of scale-free networks, the social networks of children (in this study) do not seem to be scale-free.
Table 6 presents the correlations between variables 2, 3, 7, 8, 9, 10 and 11.
Table 6. Bivariate correlations for variables of part two
Variables 2 3 7 8 9 10 11 2. Degree centrality - 3. Betweenness centrality 0,434** - 7. Network efficiency -0,082 0,267** - 8. Opinion leadership 0,543** 0,381** 0,083 - 9. Lead-usership 0,282** 0,693** 0,257** 0,400** - Control variables 10. Gender -0,045 -0,090 0,014 0,076 -0,121 - 11. Age 0,425** 0,142 -0,059 0,207* 0,028 -0,186* - ** Significance at 0,01 * Significance at 0,05
The correlation coefficients indicate there is a significant positive relation between degree centrality and betweenness centrality (0,434). Another positive relation with degree centrality is found for opinion leadership (0,543), lead-usership (0,282) and the age of a child (0,425). Betweenness centrality is positively correlated to network efficiency (0,267), to opinion leadership (0,381) and to lead-usership (0,693). Then the analysis shows a significant positive relation between network efficiency and lead-usership (0,257). Another result is that opinion leadership is positively related to lead-usership (0,400) and to the age of the children (0,207). This
implies that children of higher age are more often identified as an opinion leader. Like the analysis of the first part, this second correlation analysis also shows the negative correlation between age and gender (-0,186). To test hypotheses 4, 5a and 5b two separate regression analyses are conducted. Multicollinearity does not appear to have affected the precision of the regression estimates; this was examined by checking the variable inflation factor. In each regression analysis all the network variables (variable 2, 3, and 7) were stepwise included, to analyse their influence on opinion leadership and lead-usership. Table 7 shows the regression analysis for opinion leadership; table 8 shows the regression analysis for lead-usership.
The analysis in table 7 shows that the base model age has a positive statistically significant relation to opinion leadership. Age explains 3,5 percent of the variance in opinion leadership, which is small. However this result is in line with the results presented in the correlation analysis, which shows just a small positive correlation between opinion leadership and age (0,207). The regression analysis in table 7 shows that the independent variable degree centrality causes for the greatest amount of variance in the dependent variable opinion leadership, since the explained variance increased to 28 percent. Model 3 in table 7 shows the inclusion of betweenness centrality into the regression analysis also improves the model’s fit with a slight 2,7 percent. Furthermore the inclusion of network efficiency (model 1) did not influenced opinion leadership in a significant way. Again this is in line with the results of the correlation analysis presented in table 6, which presents an almost non-existent relation between opinion leadership and network efficiency (0,083).
Table 8 shows that betweenness centrality explains 47,6 percent of the variance in lead-usership. Although the inclusion of network efficiency and degree centrality in model 1 and model 2 at first seem to influence lead-usership in a significant way, the stepwise regression analysis excluded these variables in the third model. Also, the control variables have no significant impact on lead-usership (see base model). Hence, the extent to which a child can be identified as lead-user is best explained by betweenness centrality, as model 3 in table 8 indicates.
Table 7. Regression analysis for opinion leadership (hypothesis 4)
Variables Base model Model 1 Model 2 Model 3
Constant 0,235 (0,851) -0,906 (1,341) 0,637* (0,279) 0,817** (0,286) Gender 0,222 (0,164) 0,222 (0,164) 0,185 (0,142) 0,211 (0,141) Age 0,199* (0,076) 0,204* (0,077) 0,000 (0,073) 0,008 (0,072) Network Efficiency 1,918 (1,741) 2,551 (1,503) 1,540 (1,581) Degree centrality 0,048** (0,007) 0,041** (0,008) Betweenness centrality 0,008* (0,003) Adjusted R 0,042 0,043 0,289 0,302 R 0,057* 0,011 0,285** 0,027* * Significance at 0,05 ** Significance at 0,01
Table 8. Regression analysis for lead-usership (hypothesis 5a and 5b)
Variables Base model Model 1 Model 2 Model 3
Discussion of the results
Part one
The way the results of the first part are presented, one could say there has been made a distinction between the hypotheses in part one. Hypotheses 1a and 1b deal explicitly with the role of social networks on the adaptive behavior of a child, these hypotheses are based on cohesion theories which emphasize the role of weak (Granovetter, 1973; Burt, 1987) and strong ties (Coleman et al., 1966) in the diffusion of innovations. The remaining hypotheses of part one (2, 3a, 3b and 3c) relate to innovation diffusion (adaptive behavior) in a different way. The variables used in these latter hypotheses namely are presumed to impact the thresholds of children, and therefore impact their adaptive behavior. Hence, these hypotheses were based on threshold theories that point out that innovation diffusion depends on the varying thresholds of actors (Rogers, 1983). The results presented in table 3 confirm hypotheses 1a and 1b. The base model showed that age was significant related to adaptive behavior in a positive way. However with stepwise regression degree centrality and betweenness centrality were brought in, which significantly improved the model’s fit (model 1 and model 2). This result implies that the central position of a child impacts the adaptive behavior of a child, and this central position is better at predicting the adaptive behavior of a child than is the age of a child. Table 3 shows that model 2, including degree centrality, with 14 percent causes for the greatest explained variance in adaptive behavior. Betweenness centrality causes for the additional, but much smaller 3,2 percent of explained variance in adaptive behavior. It follows, that these findings support the cohesion theories. Especially those who emphasize the important role of strong ties in the diffusion of innovations, since the results show the higher the degree centrality of children (a way of measuring strong ties) the higher their adaptive behavior. The weaker relation between betweenness centrality and adaptive behavior suggests that weak ties play a less significant role in the diffusion of innovations among children than strong ties do.
Becker (1970) found that when an innovation is relatively safe and uncontroversial, central figures (strong ties) are the first to adopt an innovation, otherwise the weak ties (high betweenness centrality) would lead in its adoption. Since Kijkradio is a free online application, the innovation is very safe and uncontroversial. This could explain why degree centrality seems to impact the adaptive behavior of a child in a stronger way than betweenness centrality does.
In addition to previous results, the correlation analysis presented in table 2 shows that degree centrality is statistically significant related to the age of a child in a rather positive way (0,271). This implicates that as children grow older their social networks tend become more concentrated, in the sense that older children seem to have more ties than younger children have. This result corresponds with the social cognitive development theory and theories about friendships development of children, which point out that as children grow older they feel the need to have more steady friendships and they want belong to a group (Craig and Baucum, 2003). The practical implication of this result is as follows: to foster adaptive behavior among children the focus should mainly be on strong ties among children. Since the identification of strong ties can be time consuming, the argument that older children are more often identified as strong ties (higher degree centrality) might help with the identification of strong ties. Nevertheless, although the relation between age and degree centrality is significant, the correlation coefficient is small. So the implications should be interpreted with caution.
Test for hypotheses 2 and 3a (table 4) and the One-way-ANOVA analysis for hypothesis 3b show that there is no statistical evidence to prove that degree of external influence, degree of involvement and satisfaction with Kijkradio lead to significant stronger adaptive behavior. The results thus disconfirm hypothesis 2, hypothesis 3a and hypothesis 3b.
The rejection of hypothesis 2 implies that the use of mass media, such as promotional posters and video, does not influence the adaptive behavior of a child in a significant way and hence does not support the Bass model and other theories that claim adaptive behavior is influenced by the degree of external influence (Becker, 1970; Weimann, 1982). The disconfirmation hypothesis 3a implicates that the degree of involvement is not statistically significant related to the adaptive behavior of a child. The involved children were expected to feel more empowered, therefore more committed to Kijkradio, which should result in stronger adaptive behavior. Instead the correlation analysis in table 2 and the regression analysis show a (very small) negative relation between the involvement of children into the NPD process and the adaptive behavior (-0,066). Then again this negative relation is not statistically significant, and thus the results imply there is no relation between the two variables at all.
The disconfirmation of hypothesis 3b is interesting, especially when one takes into account the results from the Independent Samples T Test for hypothesis 3c (table 5). Let me explain this in more detail. The argument that is tested through hypothesis 3b, is that a higher degree of satisfaction results in a stronger adaptive behavior. Hence a positive relation is suggested. The result of the One-way-ANOVA analysis shows there is no significant relation between satisfaction and adaptive behavior (neither a positive relation, nor a negative relation). Nevertheless, table 5 shows that the degree of involvement relates in a significant negative(!) way to satisfaction. This result consequently disconfirms hypothesis 3b, which expects the complete opposite, which is a positive relation between degree of involvement and satisfaction. The empirical findings thus show that children that were involved into the NPD process, were significant less satisfied with the quality of the innovation than the remaining children in the experiment. Conversely, since there is no significant relation between satisfaction and adaptive behavior (hypothesis 3b), the lower satisfaction of the children with a high degree of involvement does not seem to influence their adaptive behavior.
An explanation for the previous findings could be that the involved children looked at Kijkradio through the eyes of a ‘user’ and a ‘tester’. The children were asked to be critical in their feedback, so that Kijkradio could be improved. Since these children paid more attention to the strengths and weaknesses of Kijkradio, this could cause for a more critical review of the quality of Kijkradio. Also the lower satisfaction of the involved children may well be due to the fact that the suggested changes of Kijkradio made by these children, were never carried through. In other words, the children never saw the result of their contribution to the development of Kijkradio. Druin (2002) pointed out that although children feel more empowered in their role as ‘user’ and ‘tester’, if they feel that their suggested changes are not taken seriously, this results in frustration and disappointment. This may have caused the lower satisfaction with the quality of Kijkradio, though not influenced the adaptive behavior.
Part two
Figure 1 shows that the social networks of the children in this study are random, which implies that each child will have approximately the same number of friends. This is in contradiction with the argument of Barabasi and Bonabeau (2003) and supports the argument that social networks are (approximately) random (Erdös and Rényi, 1959). However both strong and weak ties are present in a random network as well. Providing insights about the characteristics of these ties, will explain more about the role children can play in their social network and how this contributes to the diffusion of an innovation.
The results in table 7 shows that degree centrality (model 2) and betweenness centrality (model 3) are significant related to the extent in which a child can be identified as an opinion leader in a positive way. These two variables thus significantly improve the model’s fit. However, degree centrality is causing for the greatest amount of variance in opinion leadership (explained variance changes with 28,5 percent). As a result hypothesis 4 is confirmed. This implies that children with a higher degree centrality are more often identified as an opinion leader. Children with a high betweenness centrality are more often identified as opinion leaders as well, however looking at the coefficients (table 7, model 3) one can see that betweenness centrality has not a major impact. Like the results in the first part already indicated, degree centrality was positive related to the adaptive behavior of a child (confirmation hypothesis 1a). Since degree centrality is positively related to opinion leadership as well, opinion leaders are likely to have stronger adaptive behavior than children who are not identified as opinion leaders. This suggestion is confirmed when the bivariate correlations between opinion leadership and adaptive behavior are calculated.
As with the results of part one, the findings in part two support the cohesion theories. More specifically the findings support the strong ties theories who claim that opinion leadership is positively related to high degree centrality (Valente, 1996) and to innovativeness (Schiffmann and Gaccione, 1974; Czepiel, 1974). In addition, table 7 shows that age (initially) influences opinion leadership in a significant positive way. Also the bivariate correlations in table 6 show, age and opinion leadership have a significant positive relation (0,207). Nevertheless, this relation is not very strong. All in all this result could implicate that the older a child gets, the more often he/she is identified as an opinion leader. However, since the relation is not strong this implication should be interpreted with caution.
Table 8 shows the results of the regression analysis for lead-usership. As model 3 shows, betweenness centrality causes for 47,6 percent of the variance in lead-usership. The coefficients of both network efficiency and degree centrality are no longer significant in model 3. Accordingly, the result form the regression analysis implies that a child with a high betweenness centrality will be identified as a lead-user more often than children with a low betweenness centrality. As a result of this hypothesis 5a is confirmed. In contrast hypothesis 5b is disconfirmed. Although betweenness centrality and network efficiency are both measures that try to identify the actor’s between groups (who span structural holes), the small (significant) correlation between these two variables (0,276) show that both variables vary in their way of calculating this specific position of an actor. Regarding the very large influence of betweenness centrality, in this study betweenness centrality seems to be a better way to measure whether an actor takes in a central position between groups.
the social network, which seems to give them early access to information and provides them with different types of information.
The confirmation of hypotheses 4 and 5a does need to be interpreted carefully. This is because the confirmations of these hypotheses suggest that totally different actors play the roles of lead-user and opinion leader, since opinion leadership is significant related to degree centrality and lead-usership is related to betweenness centrality. Looking at the results of the correlation analysis in table 6, one must conclude that it occurs that actors play both roles at the same time, that is the one on opinion leader and the one of lead-user. In more detail, table 6 presents a considerable correlation between degree centrality and betweenness centrality (0,434), which implicates it is not unlikely that a child with a high degree centrality also functions as a bridge between groups (a high betweenness centrality). The same applies for opinion leadership and lead-usership, table 6 shows that these two variables are significant correlated in a rather positive way (0,400). So it is important to mention that children in this study sometimes play two roles, that of an opinion leader and that of a lead-user. Nevertheless, the confirmation of hypotheses 4 and 5a implicates that even if the children have both a high degree centrality and a high betweenness centrality, the extent to which a child is identified as opinion leader or as lead-user is primarily caused by respectively degree centrality and betweenness centrality.
The considerable correlation between opinion leadership and lead-usership is interesting. Hansen and Hansen (2005) argue that overlap in the roles sometimes occurs, however most studies found that they were played by totally different actors. As Morrison et al. (2000) pointed out, lead-users tend to hide their information about innovations since they want to obtain the highest benefit from their information (Von Hippel, 1986). Opinion leaders on the contrary do share their information about innovations, since they provide friends of innovation related advise. The considerable overlap between opinion leaders and lead-users in this study implies that child lead-users sometimes do share their information with other users, instead of hiding their information. An explanation for this could be that children around the age of ten are just starting to understand simple marketing processes and bargaining principles (Gunter and Furnham, 1998). Hence children are less concerned with making profit out of their information, consequently sharing information with other children occurs more frequently.
