SIR Model of Epidemics∗

Troy Tassier
September 15, 2005

1 Introduction
In the last class we discussed a model of disease spread where individuals were susceptible to
a disease, potentially contracted the disease, and then recover and are once again susceptible.
we called this model the SIS model. Today we study a model of different diseases. Those for
which the person contracting the disease becomes immune to future infections after recovery.
This model is called SIR for Susceptible - Infected - Recovered. As a variant on this title
SIR can also stand for Susceptible - Infected - Removed if we allow for people to die from
a disease and thus leave the population we are studying. Examples of diseases which fit the
first category include chicken pox and influenza. Others that fit the second category where
we may consider removals from a population would be HIV or the Bubonic Plague where
many or most individuals die from contracting the disease. In order to keep things simple
we will concentrate on the first version of the model so that we may keep the population
constant across time. This decision is entirely to keep things simple. And, you should be
able to directly extend the model to one of a fluctuating population (with births and deaths)
without much difficulty.

1.1 An Example
As an example consider the following data from an influenza epidemic in New York in the
winter of 1968-69. This strain of influenza was dubbed the “Hong Kong flu” due to its place
of discovery. Table 1 lists the number of “excess” pneumonia-influenza deaths that winter.
Of course not everyone dies from influenza thus the data do not count all instances of
the influenza. But we might view the number of excess deaths as an indication of how wide
spread the Hong Kong flu was in each week if we assume that the number of excess deaths
is proportional to the number of total incidents of the disease in our population of interest.
Our goal for today will be to attempt to devise a model that will describe this data.
∗
Data presentation on the Hong Kong Flu adapted from David Smith and Lang Moore, Duke University

1
 Table 1: Excess Deaths

Week Deaths
1 14
2 28
3 50
4 66
5 156
6 190
7 156
8 108
9 68
10 77
11 33
12 65
13 24

2 The SIR Model

In the introduction we mentioned that tonight we are interested in the spread of an infectious
disease where individuals may be susceptible to the disease, may be currently infected with
the disease, or may be recovered and immune from the disease. Thus we have three groups
or states in which we can place individuals. In addition we see that our data is a time series
where we have a number of infected individuals at each point in time. Similarly we also have
a number of susceptible and recovered individuals at each point in time. Let us begin with
some notation.
St = the number of susceptible individuals in the population at time t.
It = the number of infected individuals in the population at time t.
Rt = the number of recovered individuals in the population at time t.
N = the population size.
Correspondingly define the three groups as fractions of the total population N in lower
case.
st = St /N (the susceptible fraction of the population at time t.)
it = It /N (the infected fraction of the population at time t.)
rt = Rt /N (the recovered fraction of the population at time t.)
We will work with both of these sets of notation in the modelling process. Note that
each individual in the population is in one of the three groups. Thus St + It + Rt = N and
st + it + rt = 1.

2.1 Dynamics
If we think about the process of a disease that fits the SIR framework we have a flow of
individuals from the susceptible group to the infected group and then to the recovered or

2
removed group.
Susceptible → Infected → Recovered
An individual potentially moves from the susceptible to the infected group when she
comes in contact with an infected person. What qualifies as a contact depends on the
disease. For HIV a contact may be sexual contact or a blood transfusion. For influenza it
may be walking within a few feet of an infected person that has recently coughed. Suppose
that each infected person contacts γ individuals in each period of time on average. Now each
contact may not result in transmission of the disease. Perhaps only α percent of the contacts
result in transmission. Thus the potential number of transmissions may be at most γ ∗ α.
Let us define this value as β = γ ∗ α. β is the average number of transmissions possible from
a given infected person in each period.
Now, we must remember that there are three groups in the population. If we assume that
individuals are mixed randomly then each potential transmission may be from an infected
person to a susceptible person which results in a new infected person. Or a transmission may
occur from an infected person to another infected person which results in nothing happening
since the person is already infected. Or the potential transmission may occur from an
infected person to a recovered or immune person. In this case again nothing changes. Since
only st percent of the population is susceptible each infected person generates only βst new
infections each period. Each infected person recovers (or is removed/ dies) at some rate. Let
the fraction of the infected group that recovers be κ.
We are now ready to describe the SIR process. Given the current state of the population
in period t described by St , It and Rt we can write a series of difference equations that
describe the motion of the system. First let’s describe the susceptible population. We begin
period t with St individuals in the susceptible population. From this population we lose on
average βst It from the population. Thus in period t + 1 we have:

St+1 = St − βst It (1)

Through similar reasoning we see that:

Rt+1 = Rt + κIt (2)

and

It+1 = It + βst It − κIt = It (1 + βst − κ) (3)

Similarly we could write each of these in terms of the population fractions:

st+1 = st − βst it (4)

Through similar reasoning we see that:

rt+1 = rt + κit (5)

and

it+1 = it (1 + βst − κ) (6)

If we add up these equations we will find that st+1 + it+1 + rt+1 = st + it + rt = 1.

3
3 Steady States
Unlike the SIS model studied in the last class there do not exist positive steady state levels
of infected individuals in the SIR model if there is a constant population. Let us see why
this is the case. First suppose that it is equal to 0. Then from Equation 6 you can see that
It+1 = 0. Thus if we start with no people infected we stay there. This was true in the SIS
model too. There has to be an initial infected person in order for there to be more infected
people. But in the SIS model there were other steady states as well. Let us see why those
do not exist here. Suppose that it > 0. I ask: will It+1 be greater than, less than, or equal
to It ? The answer is either greater than or less than but never equal to (except in a special
case I will mention in a moment.)
So, in most cases it will either be increasing or decreasing. How do we know which? We
see from Equation 6 that it+1 = it (1 + βst − κ). Let ρt = 1 + βst − κ. This is the epidemic
threshold for the SIR model with a constant population. If ρt is greater than 1 then we are
multiplying it by a number greater than 1 so it+1 > it . The number of infected individuals is
increasing. But if ρt is less than 1 we are multiplying it by a number less than 1 so it+1 < it .
The number of infected individuals is decreasing. Clearly any time ρt > 1 or ρt < 1 we are
not at a steady state.
Now let us consider the possibility of ρt = 1 and show why this can not persist. If ρt = 1
then either it = 0 or ρt+1 < ρt . The first case is not interesting. This just says that there are
no infected individuals so there is no possibility of anyone else being infected. The second
case is more tricky. Let us again look at the definition of ρt , ρt = 1 + βst − κ We see that
there are three constants in the equation: 1, β and κ as well as one state variable st . Let us
take another look at the equation for st given by Equation 4. We see that this equation is
decreasing whenever it > 0. Thus if it > 0 ρt is decreasing for any β > 0. Thus there can
never be a steady state where ρt = 1 and it > 0. Further if it is increasing it is doing so
at a decreasing rate. And if it is decreasing it is doing so at an increasing rate. Combining
these results indicates that the infected population will always disappear in the long run if
we have a constant population.
Now let us take a brief look at st and rt . First let us look at st . We know that st is
decreasing for it > 0. And we also know that it equals 0 in steady state. So, from this we
can deduce that st will also reach a steady state value since it is constant if it is equal to 0
(see Equation 4 to convince yourself.) Now, rt is increasing for it > 0 and at steady state
when it = 0. Like st rt will also reach a steady state value. What remains is to calculate
what those steady state values are and the dynamic path to steady state given a set of initial
conditions.
We could calculate this using analytical methods but we can most likely proceed quicker
and with more understanding if we use computational techniques. This also will allow us to
further expand our toolbox for understanding the rise and fall of epidemics.

4 Computational Analysis
To get a better feel for how the model works we will use an Excel spreadsheet to view the
dynamic path of the epidemic. (I will place a copy of the spreadsheet on the course web-page

4
for you to use in a HW assignment at the end of class.) People usually recover from influenza
in about 3 three days. Note that what we are really interested in here is the length of time
that it takes to become unable to pass on the disease. So, even if one still does not feel well
after three days it may be that they are past the time that they are infective. With this
in mind a good estimate for κ is 1/3. β is probably a little tougher to estimate. So, let us
begin by looking at a range of parameter values for β and see what looks probable. What
value of β creates a time series that looks like the data? Answering this question also will
give us a chance to see how changing the value of β affects the course of the epidemic.
Let us start with κ = 1/3 and β = .1 and I0 = 1. What happens? Essentially nothing.
Similarly if we increase β to 0.2 and then to 0.3 nothing happens. And this makes sense if
we look at the analytics of the model. Recall that It will only grow if 1 + βst − κ > 1. So,
since st is approximately 1.0 with only 1 infected person, ρt is approximately 1 + β − κ. Thus
if β < κ then an epidemic will not occur. But let us look at what happens when we increase
β to 0.4. Now we get a small spike in infections around period 200. And about 32% of the
people get the disease. Now note that while 32% of the people get the disease there are never
more than about 1.5% that are infected at any one time. If we increase β further, to 0.5, we
start to get something that resembles the data with which we began this lecture. Here we
get a peak in infections around period 100 with about 6.5% of the population infected and
just under 60% of the population has been infected by the end of the epidemic.
If we keep increasing β we find that the height of the spike in infections increases and a
larger percentage of the population becomes infected over the life of the disease. For instance
if β = 1 then about 96% of the population gets infected at some point.
Now let us return to parameters that look fairly similar to the Hong Kong flu outbreak:
κ = 1/3 and β = 0.6. And let us use our spreadsheet to view the effect of the parameter κ
which is the period of infectiousness. First let us decrease κ to 0.2. This means that 20% of
the infected population recovers in each period. Or, it take 1/0.2=5 periods for an infected
person to recover. When this occurs we get a bigger spike in infections at the height of
the epidemic and a much larger percent of the population is infected over the course of the
epidemic (95% compared to 75%.) If we lower κ further to 0.1 we see and even bigger spike
and a larger percent of the population (essentially everyone) gets infected. In addition the
height of the epidemic lasts longer. By comparison if κ is smaller (people recover from the
epidemic faster) the epidemic is smaller. For instance if people recover in two periods instead
of 3 (κ = 1/2 instead of κ = 1/3) only 32% of the population becomes infected instead of
75%. Thus a shorter period of infectiousness implies that an epidemic is of smaller magnitude
and lasts for a shorter period of time. And as we saw above if we increased κ even further
so that κ = β the epidemic never occurs.
Recall that there are two ways that people leave the infected pool. One is that the
individual recovers. The other interpretation of the model is that people move to the R pool
because they are removed, or in other words, they die. One interpretation of our results above
then is that epidemics that slowly kill people can be far more dangerous to the population
as a whole than epidemics that kill people quickly. Thus if we compare some of the more
sinister diseases of the past, epidemics such as the black plague (which usually resulted in
death after less than one week) are potentially far less dangerous than something like HIV/
AIDS which can take years. Of course this is tempered by our current medical knowledge,
treatment procedures, and understanding of the transmission of diseases. Yet, this is a result

5
that many find surprising. If a disease removes its carriers quickly the disease is not likely
to have a long life itself.

5 Herd Immunity
In the last section we saw one surprising fact, if people are going to die from an infectious
disease it is better that they die fast for the purpose of ending the epidemic. In this section
we will see another potentially surprising fact: You do not need to immunize everyone in the
population in order to prevent an epidemic.
To see this first recall that it is decreasing in the amount of susceptible individuals in
the population. More exactly we saw above that if 1 + βst − κ is less than 1 the number of
infected individuals will decrease. The social contact structure and the rate of removal, β
and κ, are parameters of the social structure of the population and the disease of interest.
There is nothing that we can do about κ since it is a function of biology and physiology.
(Well, doctors and medical researchers can work on this problem but for the most part we
cannot.) We do have some control over the β since it is a function of the population. We
can make sick children and workers stay home when they are ill so that they do not infect
others in the population. Recall the days of school closings if too many students are sick.
(You probably recall those days fondly if you weren’t one of those infected.) We also can
quarantine people for certain diseases. This would probably happen if a small pox outbreak
occurred. In the next section of the course we will spend a great deal of time discussing
specifically the contact structure of social populations and the potential effects of public
policies designed to deal with epidemics.
But for now we will do something more simple. Let us assume that the good doctors
and medical researchers of the world have done their job and found a vaccine for our disease
of interest. And that if we immunize a person they are forever immune from catching the
disease and thus from passing it on to someone else. Now, from a modelling perspective this
essentially means that when a person is immunized they shift directly from the susceptible
pool to the recovered pool without going through the infected phase. Instead of the following
picture:
Susceptible → Infected → Recovered
we have:
Susceptible → Recovered for those that are immunized.
Immunizing people from a disease is costly. There are the direct costs of producing and
administering the dosage as well as indirect costs of providing information to the public.
Thus we would like to be able to provide safety from disease at the lowest possible cost.
What we will see in a moment is that we only have to immunize a fraction (many times a
small fraction) of the population in order to succeed in our task. Specifically we need to
move enough people from the susceptible pool to the recovered (or in this case immune)
pool such that the disease disappears of its own accord. In terms of our model we need
to move enough people such that: 1 + βs0 − κ < 1. In other words we need to lower the
epidemic threshold below one. Or to write it more directly the fraction of people that needs
to be immunized is such that s0 < κ/β. Thus you need to immunize 1 − κ/β percent of the
susceptible pool. We should see that this makes intuitive sense. If κ is small that means

6
that it takes longer to recover from infection and an infective person has more time to infect
people. Thus as κ decreases 1 − κ/β increases; you need to inoculate a larger fraction of the
population. And as β increases, each infected person contacts more people in a given period,
1 − κ/β increases. Thus again you need to inoculate a larger fraction of the population.
Let’s return to our computational model of the epidemic with base parameters of β = 6/10
and κ = 1/3. We start the model with I0 = 1; there is one infected individual. With these
parameters we end up with about 75% of the population becoming infected. Now let’s see
what happens when we inoculate some fraction of the population. Start with 50,000. From
looking at the graph not much changes. But we do drop to about 74% of the population
becoming infected. Go to 500,000 inoculated. You should notice three things: First, the
number of people who become infected drops to about 66% of the population. Second, the
number of infected individuals at the peak of the epidemic drops. And third, the peak
of the epidemic occurs later in the model. Go to 1,000,000 people inoculated. Again we
see similar results. Go to 2,000,000. Here we see that the peak level of infections has
dropped dramatically as to be almost imperceptible and only about 35% of the people
become infected. When we go to 3,000,000 the peak of the epidemic is almost imperceptible
and only about 12% of the population becomes infected. And at 4,000,000 the epidemic
never occurs.
Let us see how these computational results compare to our analytical results above.
We calculated that the number of people that you need to inoculate in order to stop an
1/3
epidemic is κ/β. In our example this is 6/10 or 5/9 of the population. And we saw that the
disease disappeared when we inoculated just over 1/2 half of the population (4 million of 7.9
million.) So, in this exercise we have seen how one can use both analytical techniques and
computational techniques to help gain a better understanding of a problem. The analytical
techniques do a nice job of helping us to find the herd immunity threshold but do a poorer
job of intuitively describing the dynamic story. The computational model does a great job
of helping us easily see the dynamic story but we have to find the herd immunity threshold
by trial and error.

6 Home Work: Non-Constant Populations

Before we close we should briefly consider our assumption of a constant population. In most
societies (at least all that I know of!) individuals enter and leave the population either
through immigration or even more simply by birth and death. We have not included any
of these ideas in our model so far for the sake of simplicity. However, these can easily be
incorporated into our equations above.
For homework let us suppose that each period n individuals are born into the population
and that m individuals die each period of random causes. Further let us assume that all newly
born individuals are born into the susceptible population and that individuals are equally
likely to die in any of the population groups. Again let us make a simplifying assumption.
Let us assume that n = m so that the population size is constant.
The difference equations for St , It , Rt , st , it , and rt with the birth and death rates included
are:

7
 Our equation for St+1 is now:

St+1 = St − bst It + n − mst (7)

Our equation for Rt+1 becomes:

Rt+1 = Rt + kIt − mit (8)

and our equation for It+1 becomes:

It+1 = It + bst It − kIt − mit = (1 + bst − k)It − mit (9)

Or in terms of the proportion equations:
n − mst
st+1 = st − bst it + (10)
N
Our equation for Rt+1 becomes:
mrt
rt+1 = rt + kit − (11)
N
and our equation for It+1 becomes:
mit mit
it+1 = it + bst it − kit − = (1 + bst − k)it − (12)
N N
Note that these equations are identical to those without birth and death (and those in
the class spreadsheet) except for the last term in each case. Thus when you change the
spreadsheet in question 2 below you only need to add this one term in each equation.
What you should be able to see with a little work is that if there are enough births and
deaths in the population there can be steady state levels of the infected population greater
than 0. We will use an Excel spreadsheet to see this in the homework.
1) Recreate our spreadsheet from class with parameters β = 3, κ = .2, N = 7, 900, 000
and I0 = 1. Use Excel to graph 300 periods of each population group. Print the graph to
hand in next class. (Be sure that your spreadsheet matches the spreadsheet from class before
you move on!)
2) Add a birth rate and a death rate to your spreadsheet equations where there is a
constant number of people born each period and a constant number of people that die each
period. Each new person born enters the susceptible pool. Deaths occur with equal likelihood
from each group. Hand in the answers to the following five questions. Print the graph for
each of these to hand in with your answers. (Print them small so that we don’t waste too
much paper.) You can write your answer on the back of your graph or on a separate sheet
of paper.
a) Let n = m = 1, 000. What happens?
b) Let n = m = 2, 500. What happens?
c) Let n = m = 7, 000. What happens?
d) Let n = m = 10, 000. What happens?
e) Let n = m = 100, 000. What happens?

8
 Answer these in an intuitive fashion in two to three sentences. I do not need to know
exactly what the equilibrium level of, for example, s is. I want you to tell me something like
“s is higher than it was in the base case.” Or, “s is increasing at the end of period 300.” Tell
me about the interesting phenomena that you observe as you change parameters. If nothing
substantial changes tell me “nothing substantial changed”!
3) In question 2e) above you should have found a steady state level of infected individuals.
Do you expect the disease to affect the old, the young, or all age groups evenly? Explain in
two or three sentences. (Think about Chicken Pox and Measles!)