UMAP Module 305

Modules In :
Undergrad 岫 I The Growth of Partisan
Mathematics^ and Its
Application*' Support II: Model
Analytics
Carol Weitzel Kohfeld
Published InJ cooperation 細 h the
Society for Industrial and Applied
Mathematic^ the Mathematical
Association Qf America, the J National

Council of Tsacher* & the American

Mathematical： Association 折
TWoYear Conges； The Institute of
Management Sciences, and the
American Stattetlcal

Applications of First Order

Quadratic Difference Equations
to American Politics

COMAR Inc., Suite 210,57 Bedford Street,

Lexington, MA 02173 (617) 862-7878

Z
-----------------------------------

MODULES AND MONOGRAPHS IN UNDERGRADUATE

MATHEMATICS AND ITS APPUCATIONS (UMAP)
PROJECT

The goal 破 UMAP was to develop, through a

community of users and developers, a system of
instructional modules in undergraduate mathematics
and its applications to be used to siq^plement existing
courses and from which complete courses may
eventually be built.

The Project was guided by a National Advisory

Board of mathematidansr scientists, and educators.
UMAP was funded by a grant from the National
Science Foundation and is now supported by the
Consortium for Mathematics and Its Applications
(COMAP), Inc., a nonprofit corporation engaged in
research and development in mathematics education.

COMAP STAFF

Paul J. Campbell Editor

Solomon A. Garfunkel Executive
Director COMAP

This unit was presented in preliminary from at

the Shambaugh Conference for Mathematics in
Political Science Instruction held December; 1977 at
the University of Iowa. The Shambaugh Fund was
established in memory of F. Shambaugh who was the
first and for forty years
served as the chairman of the Department of Political
Science at the University of Iowa. The funds
bequeathed in his memory have permitted the
department to sponsor a series of lectures and
confeiences on research and instructional topics. The
Prefect would like to thank participants in the
Shambaugh Conference for their reviews and all
others who assisted in the production of this unit.

This material was prepared with the support of

National Science Foundation Grant No. SED76-19615
A02 and No. SPE 8304192. Recommendations
expressed are those of the author and do not
 necessarily reflect the views of the NSF or of
COMAE

© Copyright 1979 by COMAC Inc. All ri^its reserved.

Reprinted in 1992,1993.

THE GROWTH OF PARTISAN SUPPORT II:

MODEL ANALYTICS
by

Carol Weitzel Kohfeld

Department of Political Science
University of Missouri - St. Louis
St. Louis, Missouri 63121

TABLE OF CONTENTS

1. INTRODUCTION ............................................................................................................

2. THE MODEL.....................................................................................................................

3， SOLVING FOR EQUILIBRIA .......................................................................................

4. GLOBAL STABILITY ANALYSiS ................................................................................

5. A LOOK BEYOND...........................................................................................................

6. REFERENCES..................................................................................................................

7. ANSWERS TO EXERCISES...........................................................................................
 Sheet： UMAP
Unit 305

Title: THE GROWTH OF PARTISAN SUPPORT

II: MODEL ANALYTICS
Aujhor: Caro) Wei tzel Kohfeld
Department of Po1i t i ca1 Science of
Uni ver-i ty Missouri-St. Louis Missouri
St. louis > 63121
Review Stge/OwUu 111 1 /23/79

Clfi f APPL FIRST ORD QUAD OIFF

EQ/AMER POL

Sygg 竺 ted Support Haterials:

: See Section 6 of t”t・

Prerequisit， Skills:
). Be able to use first: order quadratic difference
equations.
2. FamiIiarity with Taylor series-
3- Fami)iari ty wi th part ial di ff&rent iation.

Output Skills:
17 te mathematical properties of first order
quadrat ic
d i fference equation a) equilibria, b) 1 oca I stab i1i
ty, and c) globa1 stabi i i ty.
2・ Use these analytlc results to better understand political
mobi1ization processes.

Related Units:
Lxponential Mode 1s of Legislative Turnover (Unit 2d6)
The Dynamics of Poli tical Mobil!zatIon I (Unit 297)
The Dynamics of Political Mobilization 11 (Unit 298)
Public Support for Presidents I (Unit 299)
Public Support for Presidents 11 (Unit 300)
Laws that Fa i1 I (Unit 301)
Laws that Fail I I (Unit 302)
The Di ffus ion of Innovation in Fami)y Planning (Unit
303) Growth of Partisan Support I (Unit 304)
Discretionary Review by the Supreme Court 1 (Unit SOB)
Discretionary Review by the Supreme Court I I (Unit 307)
What Do We Mean By Po)icy? (Unit 320)
 THE GROWTH OF PARTISAN SUPPORT:
MODEL ANALYTICS

1. INTRODUCTION

In UMAP Unit 304, "The Growth of Partisan Support I： Model and Estimation," we
learned that a nonlinear, quadratic ,difference equation had useful applications for modeling
processes which can be explained by diffusion or contagion. In particular, we found the first order
quadratic applicable to the growth of the Democratic party during the Roosevelt revolution and to
the more recent phenomenon of Carter's growth of support during his campaign for the
presidential nomination. The simple first order linear difference equation model which assumes
indi- vidual level sources of change was extended to include the effects of political context. Political
context is defined as the effects of interaction between those who behave in a particular political
way and those who do not and is formalized with the classic formulation of an assumption of
random mixing in population diffusion or contagion models. Although the model is quadratic we
were able to find estimates for the parameters and to assess the relative contributions of
individual and contextual level effects on the mobilization process.

We turn now to the analytic properties of the model. We want to know what the
mathematical properties of the model are. In particular we want to investigate its 1imiting
behavior, conditions for convergence, and possible types of qualitative behavior the model can
produce. Is the S-curve the only qualitative behavior possible? Because the model is quadratic we
cannot determine an explicit solution but we can solve for equilibrium points and examine
stability properties using a Taylor series expansion. This provides information about the process

1
only in a neighborhood of the equilibrium point. Conditions for convergence (global stability) of
the quadratic form have been discussed in Chaundy and Phillips (1936) and we will use their
results as explicated by Sprague (1969) to further analyze the limiting and convergence behavior
o£ the model. Finally we will look beyond the behavior we obtain in our use of the model to see
what possible qualitative behaviors this relatively simple quadratic form can produce. It turns out
that very complex behaviors can be produced with this apparently simple quadratic recursive
form.

2. THE MODEL

In this model we assume that the rate of change specifies a simple model of diffusion.
Thus there is a loss term (£), a gain term (g) which can be interpreted as spread from a constant
source and operating on the out- population ,and an interaction term (s) which may have either a
positive or negative sign depending on whether the outcome of interaction is to swell or diminish
those behaving in the fashion characterized as Mt. The proportion of the population recruited by
means of the constant source effects are "removed" from the population of potential supporters
who are involved in the interaction with supporters. This argument is formalized as follows:

(1) = g(L - Mt) - fMt * sMt[(L - - g(L - M?).

In words, using the Carter example: the rate of change in mobilization of support for Carter is
proportional to potential supporters at rate g, is disproportional to present supporters at rate f,
and is affected by the interaction between Carter supporters and recruitable nonsupporters rate s.
L sets the limit of the potential supporters. reasons of descriptive adequacy we assume the
following restrictions: at
For

2
(2) 0 < f,g,L,Mt < 1 and -1 < s < 1.

These are reasonable restrictions. L and are propor- tions of the population and the restriction
keeps us from having substantively meaningless things such as negative populations or more than
100 percent of the population. Restricting f and g positive does not allow supporters to be gained
from supporters nor nonsupporters to be produced from nonsupporters. Finally letting s range
both positive and negative allows interaction to produce supporters or nonsupporters as a result
of contact between groups.

3. SOLVING FOR EQUILIBRIA

Even though it is relatively easy to obtain values for the parameters, an explicit solution
for the quadratic form of the first order difference equation is not avail- able. This means we have
no closed form for the nth term of the time series, however particular time paths can be
determined using the recursive form.

Tt is possible to solve the quadratic for its equilibrium points. Rearrange the model in
Equation (1) so it has the following form:

(3) = (sg - + (sL g - sgL)Mt + gL

If the process is at equilibrium then any successive values of Mt are of equal value, i.e. , = Mt. If
such a
value or values exist they are called stationary values. Thus we want to know what values for M
make the process stationary, that is, when is the change in the mobilization process from one time
period to the next zero, i.e., what values satisfy Mt + ^ = Mt? We can solve Equation (3) for these
points using the quadratic formula. Set AMt = 0 and substitute M* (a stationary value) in
Equation (3) for Mt. The result is

3
 0 = (sg - s)M*2 + (sL g - sgL)M* + gL.

Recall the quadratic formula:

(-b 士 - 4ac)‘
2a

where

sL - g

* gL.

Solutions for Equation (4) are represented by

_ (sL_f _g_sgL) ± - 4(S£_S)EL)

~2(sg-s)

We can apply this formula to the results we obtained in UMAP Unit 304 for Essex, Wayne and
Dupage Counties. A summary of the estimates for f, g, s, and L for the three counties are reported
in Table 1. These estimates will be

TABLE I
Summary of Parameter Estimates for Essex, Wayne, and Dupage Counties,
Democratic Presidential Voting, 1920-1972.

Parameters Least Squares

Coefficients
County
m
L f g 5 o HI?

Essex 1 .16 . 54 .16 .95 -.45

Wayne 1 .28 .15 .65 .15 b ■12 -.55

Dupage I .18 -.57 .18 .24 .47

used in the calculations which follow in the text and exercises. As an illustrative example consider
Essex County. Substituting into formula (7) we have

M* =(・°5J 土 /F 云注-4(-.45)(.16)
2(-.45)

(8)

4
which reduces to
M* = .05 ± /^905
^79
M* = .54, -.66 .

What does this tell us about the mobilization process?

Since substantively represents a proportion of the population which supports the Democratic
party, we are interested in values of M* which lie in the 0,1 interval. Values for M* greater than 1
suggest that the process is stationary when more than all the population are mobi1ized and
similarly values less than zero indicate a stationary process with negative populations. Both of
these cases are substantively meaningless for our model so we restrict our consideration to the
equilibria which lie in the 0,1 interval.

Exercise 】.Calculate M* for Wayne and Oupage Counties using the estimates for f, g,
s, and L. In UMAP Uni t 304 you calculated estimates for f, g, and s when L was set at
a lower value (.74). What value would M* have if you used these second sets of
estimates? Why?

In the case of Essex County we find M* = .54. This means that when the process
reaches .54, that is 54 percent of the population are mobilized in support of the Democratic party,
gains and losses balance and the process is stationary unless perturbed by some external force.
important to note that M* is a net state. continues at M* but the same, i.e. , the process the state--
a number--no 11 is

A logical question the process to ask next is what happens when The process level of

is bumped by external forces? process, mobilization stays the remains dynamic butInthe
the political
we are talking about short run political measure of longer changes.

forces such as political scandals and

political personalities. These short run forces would shock the system away from the

5
equilibrium. After these forces have shocked the political system will the process converge toward
the equilibrium point or will it diverge? We have solved the quadratic form to determine its
equilibria, but we have not yet provided a means to test for stability. Stability can be local or
global. Local stability means that within some specified neighborhood of the equilibrium point, if
is perturbed it v/ill converge toward the point. Whereas global stability implies that is stable no
matter what the perturbation. complete description of stability see Rosen, 1970; May, 1973.)

First we apply a technique to investigate local stability at the equilibrium points and then
the processwe will take advantage of some general results of Chaundy and Phillips
equilibrium the(1936), reworked by Sprague (1969), to determine global stability for the
process (For aspecific quadratic form. The technique used to investigate local stability has a
more wider and general application for more complex nonlinear models. An
analysis of small perturbations around the equilibrium point M* begins by writing the perturbed
mobilization level as
(10) Mt = M* + Xf

Here Xt measures a small disturbance to the equilibrium M* within some specified neighborhood.
An approximate difference equation for the perturbation measure is obtained by a Taylor
expansion of the equation for our original model (3) about the equilibrium point. The Taylor
series expansion provides a linearization of the model because the linear term in the expansion
dominates the series in a small neighborhood and terms of order 2 and higher can be neglected.
The expansion takes the following form

皿七=aXt or Xt+1 = (1 + a)Xt where a is the partial derivative of with respect to M evaluated
at the equilibrium point M* (obtained from Equation (3)).

(11)

6
 HAM.)
(12) a • -~g 时=2(sg-s)M* + sL - f - g - sgL .
It measures the mobilization growth rate in the immediate neighborhood of the equilibrium
point.

Equation (11) is a first order linear difference equation for which we have an explicit
solution. It has the form

(13) Xt = XQ(1 + a)t

where X。 is the initial small perturbation. The disturbance dies away if (1 + a) lies in the open
interval -1,1, diverges if (1 + a) > 1 or (1 + a] £ -1, and is constant if (1 + a) = 1. Thus the
neighborhood stability analysis of the equilibrium point M* shows the point to be stable i£ and
only or more simply - 2 < a < 0.

Now we apply these results to our example the mobilization of the Democratic party in
Essex County. First we substitute (12) into (11) to obtain

(14) AXt = (2(sg-s)M* + sL - f - g - sgL)Xt .

Disaggregating AXt gives

(15) Xt+1 = (1 + 2[sg-s)M* + sL - f - g - sgL)Xt where

(16) 1 + a = (1 + 2(sg-s)M* + sL - f - g - sgL). Evaluating the coefficient of Xt at M*

= .54 using the estimates for the parameters s, f, g, and L from Table 1 gives

(17) (1 + a) = (1 - .49 - .05) = ,46.

Since (1 + a) = .46 and is between 0 and 1 we know that the disturbance is monotonically
convergent and dies out.
This means that the equilibrium M* = .54 for Essex County is locally stable. The mobilization of
the Democratic party converges toward .54 of the population which is a

7
locally stable equilibrium. (This discussion is adapted from May
1973.)

Exercise 2. Do a local stabi1ity analysis of the equilibrium points M* calculated in

Exercise 1 for Wayne and Dupage Counties.

4. GLOBAL STABILITY ANALYSIS

In general there are no techniques for investigating global stability for nonlinear
models. We can investigate local stability by linearizing the model with Taylor series expansions
around the equilibrium points, but this only provides stability analyses in the small. There are,
however, some general results for a particular nonlinear form, the quadratic, reported by
Chaundy and Phillips (1936) and further explicated by Sprague (1969). Chaundy and Phillips
consider a difference equation of the following form:

2
(18) Mt+1 = AM； + BMt + C where A, B, and C are real numbers independent of t.
We can immediately see that our model is isomorphic to this form. Chaundy and Phillips do not
provide an explicit solution but conditions of convergence, divergence, and ultimate qualitative
behavior can be developed from their discussion. Only a few of the results are presented here,
the inquisitive reader will search out the original source for a complete explication of their
results.
First define a quantity K by

.1 ± A + 4[(B/2)2 - B/2 - AC]

(19)

where A, B, and C are from Equation (18). This produces 3 possibilities:

K's, or a pair of 2 real and unequal K* s, 2 real, equal complex K* s. Six cases
below. are considered

8
 Casg j, f f K given by (19) 1 R complex chon the process is divergent,
diverging to infinity.
Now choose that K which is >_ One of the K's should meet this
condition.
Co 登 U ・ If IAM0 ■* B/2 ( > K then the process Mt diverges
to infinity.
Case III- If IAMQ ♦ B/2| ・ K then the process Mt is
stationary. This does not mean the process will converge if
displaced.
Case IV, If |AM0 + B/2| < K and 1/2 < K < 3/2 then the
process Mt converges to a value
(20) M 倉■ tk- K - B/2)
The limit M* is dependent on A, B. and C since K depends on C. Convergence in 1/2
< K < 1. this case is monotonic if
Case V. If |AMQ process 、 * B/2| < K and 3/2 K < 2 then the
oscillates finitely.
Case VI. If |AM0 + B/2| < K
and K > 2 then the process 七 diverges to infinity with a certain exception, i-e,, if MQ
is chosen so that the expression AM。 + B/2 is an element of a set involving the
square roots of the expression ・ K then the process Mt oscillates finitely, (This
discussion is based on results from Spraguet 1969.)
We now apply these conditions for convergence to the data on Democratic
mobilization for Essex County. We have already determined that there is a locally
stable equilibrium at M* -・54. We now take advantage of the preceding results to
see if the locally stable equilibrium satisfies conditions for global stability. First the
real numbers A. B, and C are defined

9
 (21)
8 - s«L
C - gL.
Substituting the estimates for the paraneters for s r f, g, and L for Essex County
from Table 1 into the formulas in (21) we obtain the values A ■B ・.95. and C ・・16・
Using these values we calculate K as follows
金 /I ♦ • .47 厂厂
(22)
K ・.78. .23.
We choose K - .78 as the value for K and find that Case IV applies . Now examine
the value |AMQ + B/Z|< Substituting values for A and B we obtain |-•4SMQ +
.475|・ The condition for
(23) |-.4SM0 + ・47S| < process is convergence of the
・78.
Recall that convergence for is nonlinear difference equations
dependent upon initial conditions.
Thus the starting point of the Mobilization process is an important consideration
behavior. in the determination of long run limiting For what
(23) hold? M°・ If values of M° does the inequality in
it all values We begin by looking at the extreme values for holds
Both extreme
values 0 and 1 for for the extreme values then it holds for of MQ can
satisfy the range across the 0,1 interval.
inequality thus any permissible starting value satisfies the condition for
convergence. We also note that convergence of the process is monotonic since K lies
in the 1/2,1 interval, Finallyt M* calculated using Equation (20) equals .54 for Essex
County. This is the sane value obtained using the quadratic formula which is as it
should be-
3・ Use these results to perform a global stab! 1 ity analysis for Wayne and
Dupage Counties- Compare the M* you catcuUte with the M* you calculated using
the quadratic formal•.

In summary, then, we have investigated the mathematical properties of the

1
0
first order quadratic difference equation used to model mobilization processes
characterized by diffusion or contagion. Although explicit solutions for the
quadratic are not available t the quadratic can be solved for equilibrium points
using the quadratic formula. Local stability was investigated using a Taylor series
expansion around the equilibrium point- But this provides information about
stability only in the small, in specified neighborhoods of the equilibrium. In
general> for nonlinear models local stability can be investigated using this
technique. However, in the case of the quadratic, some general results are known
and conditions for convergence and divergence were reported and used to
investigate global stability for the mobilization process.
This really is not the end of the usefulness of this simple mathematical form
because it is capable of producing rather remarkable behaviors. In the next section
we briefly illustrate some of the more dramatic tine paths which are produced for
arbitrary assignments to the parameters. Although there is no substantive
interpretation for most of the behaviors they do suggest that some very complex
behaviors which appear to be random or stochastic may be generated by this
relatively simple quadratic form.

1
1
 5. A LOOK BEYOND

We want to examine some of the possible qualitative behaviors which can

be produced by the recursive foTm of the quadratic difference equation.
We have seen the
sim S-curve produced, but there are many other time
paths which can be produced which exhibit bounded behavior and which
are much more complex. To illustrate these behaviors we use the simple
logistic forw of the model (used in Harnon's module, UMAP Unit 303t
1978)* The model is fornalized as follows
(24) AMt • rMt(L - Mt)
whe
re
r • -the intrinsic growth rate of the process t
typically a species population
L • the natural limit of the growth process in the
population
M » the proportion of the population which
behaves in a specified Banner or species
number.
We want to know what interesting behaviors can be obtained by driving
this model with assignments of arbitrary growth
rate In particular what kind of behavior is produced
when we drive the model by assigning growth rates which exceed unity?
First put the model into the recursive form (25) Mt+1 - -rM： + (1 + rL)Mt .
Using this form particular time path^ can be generated by varying the
growth rate and the conditions. In
Figure 1 three time paths are exhibited: two are the familiar S-curve and
the third oscillates with period two. All three are generated by the
recursive form in (25). Specific parameters are presented in Table 2.

1
2
Flgug 1. Time paths generated by the simple legist Fc form: ■「七（!》%）.
（Parameters exhibited in Table 2,）

TABLE 2
Parameters for Time Paths Exhlbi ted in Figures 1 and 2
Generated by the Logistic Form ■ rMt（L ■ %）

Figure Tine Path Parameters Initial Condition

r L %
1 I .0 . .1
! 11 ・ . J
1 111 5 9・ J
2 1 6 ・ •）
5
This appears orderly but note the departure from smooth growth
for the process when the growth parameter r is set greater than 1・ But
even more is possible. In Figure 2 we exhibit a process generated by the
same recurs I vo form M (25) wl th r * 6, t -nnd M(1 ■・1・ Increasing r
makes the process wore reactive and produces

1
3
wild oscillation. But notice even this wild oscillation
is bounded behavior. Much still has to be learned about
the possible behaviors produced by this simple deterministic rule.
Looking at the time path in Figure 2 it is
hard to imagine experiencing this process as deterministic» (See Li and
Yorke, 1975； Mayf 1973; May, 1974; May, 1975;
Mayt 1976; and May and Oster, 1976.) 1.00

fl

Figure 2・ Time path generated by simple logistic form:七■啪七(1>■网

卩・ (r - 6, L -・5， »0 ■・1)

1
4
 6- REFERENCES

Chaundy, T.W- and Phil1ips> Eric. 1936, ”The Convergence of

Sequences Defined by Quadratic Recurrence Formulae,"
Quarterly Journal of MatheMatics, Oxford Series, pp・ 74-80.
Harmont Kathryn Newcomer. 1978. "The Diffusion of Innovation in
Family Planning," UMAP Unit 303, EDC/UMAP, Newton,
Massachusetts.
Li, T.Y. and Yorke, J-A. 1975. "Period Three Implies
Chaos 妙竺切四血 them tics Monthly 82:98$・992,
May, R.M. 1973. Stability “nd Cgplexiey in Model
Ecosystems- Princeton: Princeton University Press.
1974. ''Biological Populations with Nonover-
lapping Stable Cycles,
Generations: Stable Points, and Chaos
Scitgflge 186:645・647 .
197$ ・"Biological Populations Obeying Difference
Equations: Stable Points, Stable Cycles, and Chaos r“ JQutgal
of Theoretid 49 :511- S24.
May. R.M., ed. 1976. Ecology： Principles
项d ・ Philadelphia: B • Saunders t
pp- 11-1S.
May, R.M・ and Oster, G・F. 1976. ''Bifurcations and
Dynamic Complexity in Simple Ecological Models ,T,
MericmNSurUlsl: 110:573-599.
Rosen, Robert. 1970- DyrumlcU Theory in Biology,
y。】.I・ Stabili 卄 Theory ・nd Its ADDllcation, New York:
Wiley Interscience.
Sprague, John. 1969. "A Nonlinear Difference Equation."
Mimeographed, Washington University, St. Louis.
. 1973. ''Three Applications of Contextual Theses :
Cross Section, Across Time, and Across Parameters-"
Mimeographed, Washington Universityf St. Louis-

1
5
 7. ANSWERS TO EXERCISES

Wf Cgynty
由.・.12 £ 而 2)2 + 虹.55)(.15) 册.-i(.55)一
M* - .64, -.*>23.
Du*e CpMnty
・76 *.76)2 •蛆丄 7)(.曲
M* - ~rnrr
M*・.29, 1.33・
You get the same values for H* with the other estimates
for the parameters.

(I + a) - (1 - 2(.55)(・64) ♦ -12)-・姐 6.
•A76 V I- Disturbance is monotonically convergent and
dies away. Th* process Ts toca1!y atabte-
DupaQ County
(I ♦ •) ■ (! ♦ 2(.柘)(.29)・・76)-・员・
⑸ < 1. Disturbance is monotonically convergent and
dies away. Th* process is locally stable.
W“ne County
K -・79, .22 H* - I -專學..64
|--55MO + .56] < .79.
This Enequaf i ty it satisfied for «I1 permltslble values of therefore
tha process Vs globally stable. K I )es In 1/2,1 Interval t ther«fcr« the
prccus U gngTgM.

1
6
 =.74, .26 M* = I 1 .12 = .29
~^7
|.47% + .I2| < .74.

Thi s inequali ty r$ sat isfied for al 1 permissible values of MQ therefore the

process is globally stable. K Iies in 1/2,1 interval, therefore the process is monotonically
convergent-

17