Learn how to model social and economic networks and their impact on human behavior. How do networks form, why do they exhibit certain patterns, and how does their structure impact diffusion, learning, and other behaviors? We will bring together models and techniques from economics, sociology, math, physics, statistics and computer science to answer these questions. The course begins with some empirical background on social and economic networks, and an overview of concepts used to describe and measure networks. Next, we will cover a set of models of how networks form, including random network models as well as strategic formation models, and some hybrids. We will then discuss a series of models of how networks impact behavior, including contagion, diffusion, learning, and peer influences. You can find a more detailed syllabus here: http://web.stanford.edu/~jacksonm/Networks-Online-Syllabus.pdf You can find a short introductory videao here: http://web.stanford.edu/~jacksonm/Intro_Networks.mp4
5.6: Solving the SIS Model
Loading...
From the course by Stanford University
Social and Economic Networks: Models and Analysis
208 ratings
Stanford University
Social and Economic Networks: Models and Analysis
208 ratings
From the lesson
Diffusion on Networks
Empirical Background, The Bass Model, Random Network Models of Contagion, The SIS model, Fitting a Simulated Model to Data.
Meet the Instructors
Matthew O. JacksonProfessor
Economics
Hi folks. So, let's talk a little bit more about getting solutions in the SIS model. So, as we had talked about, in terms of finding the steady state, we have an equation which relates the fraction of people that you're going to randomly meet who are infected. To an expression which involves the degree distribution and parts of that degree distribution, as well as parameters indicating the relative frequency of infection compared to recovery. So what we need to do, is see what this H looks like. And then figure out what the steady state solution looks like. So again, as we pointed out, H is going to be a function where, if we're finding we want to find thetas that equal H of theta then basically if we've got a situation where H prime at 0 is less than 1 and it's a concave function. There's going to be no steady state and otherwise, we'll find a positive steady state. So let's take a look at this function in a little more detail. So we want to take derivatives of this with respect to theta to figure out what its shape looks like. And if we take the first derivative of this with respect to theta that's quite easy. And in fact we'll find that this expression is greater than 0 and therefore we we end up with a, an increasing function. So you can take the derivative here. And verify that this is the expression you get, and indeed, what we end up with is a positive function. and then if we take the second derivative of this function then we get a, an expression, which is less then 0, so here I'm not going to explicitly solve for the derivative. You can go through and, and take those derivatives.
it's relatively straight forward, these are just polynomial functions. Take the derivatives, the first derivative is positive, second derivative is negative. and so here, we've got a function which is going to be increasing and strictly concave, and that's going to tell us that the picture that I was showing you before is an accurate picture. And then the question is, where is this going to hit?
and have an intersection. So are we in a situation where if H prime at 0 is bigger than 1, then initially, it's going above the 45 degree line again, it's given as strictly concave. Either we're going to have a steady state all the way up at 1, so we'll end up hitting up here, or it'll cross somewhere and have a, a non-zero steady state. Or if H prime 0 is less than 1, then basically, there's no possibility of sustaining a, an infection, the only theta solution's going to be 0. And that's going to be in a situation basically where the degree distribution is putting weight on very low degrees, and the lambda's fairly low, so you don't have much infection that's going to be a situation where H prime has a fairly low derivative. So if we look at H prime at 0, we can plug in 0 for theta and then see what this thing looks like. And if we calculate H prime at 0, it looks out to be lambda e of d squared over e of d, so it's looking at the relative expectation of the square of the degree compared to the expectation of the degree, and weighting that by lambda, where we recall lambda's looking at the relative infection rate compared to the recovery rate. So we have those expressions and so the theorem then we get that the conditions for a steady state process of of the SIS model to have a non-zero steady state. Is going to be if and only if, this land is large enough, and what it needs to be larger than is e of d compared to e of d squared. Okay, so you need the infection recovery rate to be high enough relative to the average degree divided by a second moment of, roughly think of variance there, so large enough lambda is, is going to give us a steady state that's non-zero. Now the interesting thing here is increasing the variance, if you do a mean-preserving spread. So you increase the variance. That makes it easier to satisfy this equation, right? So in a situation where we keep E-d constant and increase E of d squared, then we're going to be in a situation where basically, what we're doing is spreading out the degrees. But that makes higher degree nodes, which are going to have. serve as hubs and be conduits for infection and that aids in, in the spread of, of the infection. And allows us to have a steady state that's non-zero.
So, if we think about this condition, then we can plug in what we know for various different models. And for regular network, what does this turn out to be? Well in a regular network, everybody has the same degree. So the expected degree is just whatever that degree is. the expected degree squared is just the expected degree of squared. So, in this case for regular network, E of d squared, is just equal to E of d. Everybody has the same squared. So, in that case then you just need lambda to be bigger than 1 over the expectation. So, in that case, the larger the expected degree, the easier it is to satisfy this, so that, that makes sense, and, and also the larger lambda obviously the easier it is to sustain a positive steady state. For an Erdős–Rényi random network if you work through what E of d is and E of d squared are for, a Posone random network, then you end up with a situation where the e of d squared is just equal to e of d times 1 plus e of d, so in that, in that model, then we end up with in this case, lambda with 1 over 1 plus the expanded degree. If you work in a power-law network. so if we have, say for instance we work with one where the density function looks like c times d to the minus gamma. Then what we end up with is if you do a calculation say, integrate that and, and look for the variance e of d squared actually becomes infinite. And if that becomes infinite, then this whole expression becomes 0, and so we end up with lambda greater than 0. And so, it's you, you basically always have a non-zero steady state. And what's happening here is, in a power-law network, at least in the, in sort of the limit, if you have a very large network, you're going to have very, very large degree nodes. They're going to interact and, and always become infected, and carry the infection through the society. So, in that setting you end up putting weight on the tails. And sufficient weight on the tails that you always sustained an, an infection. Now, if you, you, you , if you do a power-law where you actually truncate the distribution and have some maximum degree, then you won't quite find this. But you'll find that that that expression converges to 0 as you let that, whatever the maximum degree you allow in the society, to go to infinity. So, so in the limit you always have a non-zero steady state in that model. And so, basically what we find is the, the presence of hub kinds of nodes helps substantially in sustaining non-zero steady states. So the idea here, is these high degree nodes are more prone to infection. They serve as conduits. Higher variance allows more such nodes, and that enables infection, and we see that directly in the theorem. So this is one of the kinds of insights that comes out of the SIS model. Which is a useful insight and, and made explicit in this particular model. And it also then allows us to compare degree distributions, showing that if you have the same mean but you're increasing the e of d squared, then it's easier to satisfy these conditions. Okay, so, so that shows some insights that we get out of the SIS model. next will take a little more, a close look. So what this did, is allow us to know when it is that we get a non-zero steady-state. We can also ask questions about how large that steady state is and, and whether we can do comparative statics in that. And that's not going to be exactly the same kind of answer as just when there exists one, which was just looking at that derivative at 0. And more generally we can, we can go through and try and solve this model, and say something about, what's the average infection rate in the society?
Coursera provides universal access to the world’s best education, partnering with top universities and organizations to offer courses online.
© 2017 Coursera Inc. All rights reserved.
