Let's talk about best practices in social influence study with a network approach. A lot of what I'm talking about comes from the Gate Robin's book, Doing Social Network Research, as well as my own experience. As usual, we start with the data. In another book study, be clear whether your data are directed or undirected, and make sure you're using the methods of network analysis consistently. Because most of the methods are coded in R, you have to make sure that explicitly you specify your network as either directed or undirected, and check whether the default command is for directed or undirected network. Remember, libraries differ quite a bit. When you talk about your study, report the number of actors in the whole network with a multiple network studies, such as network of a number of school classrooms. You put the mean and standard deviations across networks of the number of nodes. With an egonet study, even a qualitative study, report the mean and standard deviation of the number of alters across egos. When you're doing descriptive analysis, report descriptive statistics of major attribute variables in ways consistent with the standards applied in your own research area. Often that would include means and standard deviations for continuous attributes or frequencies under different categories for categorical or binary attributes. For egonet studies, where you have collected alter data, report means and standard deviations across egos of important alter attributes. When talking about network density, always report network density. An alternative, of course, is to report the average degree. Perhaps use average degree rather than density when you study focuses primarily on the number of ties for each actor, then density may not be as important. With multiple networks, report the mean and standard deviation of density, and probably the range as well, such as the highest and lowest density. You can also do so with a graph. For egonet studies with alter-alter ties, report the density among alters the same way. With a multiplex network study, report the densities for each relational tie, and perhaps record the amount of overlap between different types of a tie. Report the degree distribution in more detail, the types of network configurations present in the network, and provide a network visualization. Remember, always graph the data. With dyad and triad census, we have quite a bit of flexibility because the full triad census not really used that much anymore as a method of analysis. But the triad census contains important ideas that they've picked up in more sophisticated models, particular statistical models. For example, networks can be thought of as being built up by small local structures and then triad census is important. Different types of configurations indicate different structural processes that could have generated the network, such as reciprocity and closure. We might want to look at those triads as well. Then there are different types of closure indicating local hierarchies in the network. When it comes to other network descriptives, report degree distributions. A scale-free degree distribution is evidence that the preferential attachment type process could be operating in the network, and of course, the node importance centrality, can go without it. The degree distributions the feature of the network as a whole but it is a simple step to ask which nodes have high degrees and thereby identify the most central nodes by their degree. When it comes to other centrality, remember the degree centrality of a node is the activity or the popularity of the node. There is a distinction between incoming ties for the in-degree and outgoing ties for the out-degree. Betweenness centrality measures the importance of a node connecting the network through short paths. Closeness centrality is the sum of all geodesic distances from the node to all others. But remember, it only exists for the network without disconnected components. Of course, you can calculate for individual components within disconnected network. Eigenvector centrality for node indicates how central other nodes network partners. There's one more centrality which is called Beta centrality. We'll come until we've talked about it though I'm sure you have looked at up in our center package. It's a measure of the total number of ways of getting from node i to all other nodes. You can use cuts centrality, you can use page link or prestige. No matter what you use, explain to your readers, just like I did, what each centrality means. Don't assume they know it already. Then there are structural holes. Structural holes present conceptual similar indices. The most prominent is the network constraint, which measures how much the egonet is constrained by egos ties, concentrate on the densely connected set of alters. Then there is effective size. A measure of how much redundancy in the egonet in the sense that ties are connected to alters and then there are the alternatives. For instance, the density of an alter-alter ties in the egonet, is an indication of the presence of structural holes. You have all ways of getting from node i to all other nodes. If it is important to talk about connectivity then talk about connectivity and cohesive subsets. Cohesive subset of nodes is an induced subgraph with high density of connectivity. The basic idea is that of a clique. A clique is a complete induced subgraph, a subset of nodes with all possible ties present. We'll talk about them in community detection but in social influence models, that's also an important concept. This idea is important in general in graph and network theory because as cliques often overlap, analysis using this approach is not always easy. But [inaudible] and it's actually quite time-consuming. The use of density as a defining feature for cohesion transcripts into notions of connectivity. Connectivity including important concept of geodesic, often comes into play in the analysis of network. Trade-offs between density in the form of a network closure and connectivity in the form of short average geodesics are important in determining whether network is a small world. The last set of parameters from the structural variables is the network closure. Network closure is such an important and prevalent effects menu human social networks, that this is important to consider whether it is present in your own data. Closure is sometimes referred to as network clustering. In a sense that the network nodes cluster into triangles. For non directed graph, there are a couple of clustering coefficients that are often reported. The global clustering coefficient and the local clustering coefficient. Of course, the other structural parameters that we're going to talk about. We don't have time for all of them, but I hope you get the idea. What we have left is the attribute evaluation. There are different approaches to the use of attributes. We can use them as dependent variables or criteria. What we do is we take the attributes and we try to use network indices as predictors of the attribute change. Of course, attributes can also act as independent variables or predictors, and then we can just use other attribute variables as the criterion variable, or we might in fact use some of the network indices as the dependent or the criterion variable. Now that we have talked about theory, let's turn to our case study and see how we can build a social influence model using network approach.