How to Model Real World Networks
When trying to simulate a network effect on some population it is almost always infeasible to collect the data necessary to construct a network graph for that group. Instead, scientists use network models which generate graphs for a given population based on factors such as the population size or location. The hope is that these models are accurate depictions of the actual network topology of the population. In the article I read, Rafal Kasprzyk examined the best ways to model real world networks and how the choice of model can have major impacts on network simulations especially with regards to network diffusion
As it turns out, most real world networks (social networks, epidemiological networks or otherwise) have a very high clustering coefficient and a somewhat low diameter. In other words, populations are densely connected to each other. Several models for generating networks with these properties have been proposed over the years. According to Rafal Kasprzyk, regular graphs (graphs where all vertices have the same number of neighbors) and random graphs (graphs where edges are chosen randomly under some probability) are poor models for real world networks. Instead, it is better to construct graphs which grow by adding nodes sequentially and adding edges to new nodes according to the connectivity of surrounding nodes. These types of graphs are called scale free graphs, and they provide a good approximation for real world networks.
The choice of the network model is extremely important. Kasprzyk’s study found that even in a simple model of disease diffusion, the network model meant the difference between disease eradication and epidemic. In particular, realistic scale free graphs are more likely to cause widespread epidemics than random or grid based graphs. When dealing with large scale networks it becomes increasingly important to choose the right network model for the task as the network structure will always have large implications for how these situations play out in real life.