Thread by @KevinZollman

This week in my Network Epistemology class, we looked at the analogy between the spread of ideas and the spread of disease. (I planned to do this topic before the pandemic, but it seems especially relevant now.)

The underlying idea is that many process involve diffusion of ... something ... through a population. That "something" might be a disease, it might be an idea, it might be a mutation, or it might be a technology. Perhaps the underlying models might be the same.

In epidemiology, the basic model is the SIR model (S = susceptible, I = infectious, R = resistant or recovered). This model is called a "binning" or "compartmental" model because we assume that the population is compartmentalized into three different groups S, I, and R.

This model uses differential equations to represent the amount of the population who currently reside in each compartment. And those number influence the rate of flow from one compartment to the other. The number of infectious determine how fast people become infected.

The basic model in the innovation diffusion literature is almost exactly the same, except there are only two compartments "non-adopters" of the new technology and "adopters" of the new technology.

And, the underlying model of transmission is the same. With a disease, the more infected people there are, the more people will catch the disease. With technology, the more adopters there are, the more likely people are to adopt the new technology.

But of course, that analogy breaks down in various ways. We focused on exploring that.

We started by reading this review that focused on agent based modeling approaches to innovation diffusion that led away from the basic disease model.

link.springer.com/article/10.100…

This paper showed how some conclusions of the basic disease model change when you start modeling individuals as heterogeneous, embedded in a network, and subject to different types of social influence (like negative word of mouth).

We spent a good part of the class discussing the methodological differences between "population level" modeling (like the SIR model) and "agent based" modeling. In the later individuals are explicitly represented -- for better or worse.

We also talked about reasons to prefer one model over another including various types of fit to data and reduction to micro-processes verses focusing on macro phenomena.

It turns out the literature on innovation diffusion is a great case study for this.

The push toward agent based models is also occurring in epidemiology, and we talked about the motivation for this as well. This is one field where there are several modeling paradigms that coexist in an ecology of models and they are often used in concert.

We also read this paper by Grim, Singer, et al. They argue that the analogy between the diffusion of disease, genetic material, and ideas has been pushed too far, and that these processes lead to fundamentally different conclusions.

journals.uchicago.edu/doi/abs/10.108…

They argue this by giving three examples, one model for each process, and showing how they lead to different judgments about which social networks result in faster rates of diffusion.

In class, we focused on how convincing the argument was by asking this question: was each of their models clearly a model of only the target phenomena? That is, was their model of genetics clearly only about genetics. Or could it also be a model of ideas?

We agreed that the different processes were indeed different, but some were skeptical that they had shown that the individual targets were actually different from one another.

More broadly we discussed the nature of analogy in science. What does it mean to say these diffusion processes are similar? What kinds of differences would lead one to reject the conclusion? Do they have to be identical or just similar in some respects?