Evaluation of a Presidential Election Game

GUILLERMO OWEN
Rice University

1. Introduction Admittedly the electors are (constitutionally) free

There has been considerable discussion, over to vote for anyone, and in the last few elections
the past decade, of the merits of the presidential some have actually done so. Nevertheless it seems
election system in the United States. In particular, doubtful that this independence would be exer-
discussion has frequently centered on the exces- cised in a truly close election.
sive power which this system seems to give one Assumption (c), of course, does not always
group or another (the large states, the small states, hold. Nevertheless, referral to the House of
organized minorities within one or another of Representatives occurs so infrequently (this last
these kinds of states, etc.), though there is also happened after the 1824 election) that we con-
frequent disagreement about the identity of these sider its probability negligible. (It should be
favored groups. noted that in 1968, when this possibility seemed to
In this paper we carry out an evaluation of the be somewhat probable, Mr. Wallace seemed more
presidential election game. The Shapley value1 interested in making a deal within the Electoral
College than in letting the election be decided in
is computed approximately by the method of the House.)
multilinear extensions.2 A comparison with the
Banzhaf value3 will also be made. 2. Game-Theoretic Basis
The following simplifying assumptions will be
made: A game, as understood here, is merely a pair
(N; v) where N is a (finite) set of players, and v is
(a) The number of voters within a state is a function which assigns to each subset SCN, a
directly proportional to that state's popula- number v(S). The subsets are called coalitions;
tion, according to the last census. the elements / of N are players.
(b) The presidential electors for each state are The presidential election can be thought of as
pledged to the candidate who carries that a game (AT; «), where N is the set of all voters in
state, and have no choice but to abide by the country, and u assigns, to each coalition S
their pledge. of voters, the number 0 or 1, depending on
(c) The election will be decided by the electors whether S is "large" enough to win the election
rather than reaching the House of Repre- against its complement, N—S. Thus, «(5)= 1 if 5
sentatives. contains a majority of the voters within each of a
The reason for assumption (a) is that it allows set T of states, these states (in T) having at least
us to work directly with census figures rather than 270 electoral votes. Otherwise, «(5) = 0.
having to count voter rolls, and having to figure For an arbitrary game (N; v), the Shapley value
out the number of potential but unregistered is a vector d[v], whose components #,[i>] are given
voters. The assumption is reasonable, in the sense by
that we might expect voters to be approximately sl(n - s - 1)!
proportional to total population. There may, SAv = SaN
2-, ;
however, be certain states with large alien popu- 2.1 n\
lations for which this assumption is not very
accurate. Additionally, we will find that toward
the end of the decade, a fast-growing state (e.g.,
Florida or California) is underrepresented in the where s and n are the number of players in S and
Electoral College (i.e., by the 1980 election, N, respectively. This component &i[v] is known
Florida's population may be much larger than as the value to player i of the game v, and repre-
represented by the apportionment which was sents (in some sense) his power within that game.
based on 1970 census figures). It is this value which we shall study for the presi-
Assumption (b) also seems to be reasonable. dential election game.
'Lloyd S. Shapley, "A Value for n-Person Games," As can be seen, the formula (2.1) requires the
in Annals of Mathematics Studies No. 28 (Princeton: sum of a very large number of terms (i.e., one for
Princeton University Press 1953), 307-327.
2
each coalition which does not contain /) and is
Guillermo Owen, "Multilinear Extensions of therefore quite difficult to compute when N is
Games," Management Science, 18 (1972), 64-79. large. In particular, for a presidential election,
'John F. Banzhaf III, "One Man, 3.312 Votes: A
Mathematical Analysis of the Electoral College," Vil- where N is of the order of 10s, a direct evaluation
lanova Law Review, 13 (1968), 304-332. seems out of the question. Recourse is therefore
947
948 The American Political Science Review Vol. 69

frequently had to the multilinear extension (MLE) Electoral College after the 1950 and 1960 census
of the game. Essentially, the MLE of game (N; v) reapportionments.
is the function f(xi, • • • , xn) denned by As against this exact method, we may approxi-
mate the Shapley value by the method of MLE's.
J(XU • • • , Xn) In fact, the partial derivative f&cu • • • , xn) can
be expressed as
2-2 = E 11^11(1 - *MS).
scN >es as
3.2 /,(*i, • • • , « . ) = E[v(SV {i}) - v(S)},
It is shown in an earlier paper by the author4 that,
if / i s the MLE of v, then where 5 is a random coalition, such that player
j(jf*i) has probability Xj of belonging to S, and
2.3 di[v] = f fi(t, / , - • - , that these probabilities are independent. (Here, S
J
o is assumed not to contain Q Now, in 3.2, each of
the terms v(SKJ {«}) and e(S) equals either 0 or 1,
where ft is the i-th partial derivative i and P(5)<I>C?VJ{2"}), so that the variable
The advantage of using the MLE is that it can D ( J U { I ' } ) - P ( S ) can have only the two values 0
be given a probabilistic interpretation. Thus, or 1. Its expectation, then, is the probability that
f(xi, • • • , xn) is the mathematical expectation it equal 1, and this happens only if D(5) = 0 but
o(S(J{i})=l. By (3.1), this happens if
E[v(S)],
j < 269
where S is a random coalition, determined
probabilistically by saying that there is prob- and
ability Xi that player i will belong to 5, and that
the probabilities of different players belonging 270.
are independent. This interpretation will allow us
to give approximations based on the central limit
theorem. Thus,
fi(Xi, • • • , Xn)
3. The State Game
If we disregard the process by which the elec- 3.3
tors are themselves chosen, we can think of the = Prob J27O - Wi < £ u>j < 269>
electoral college as itself representing a game.
By assumption l(b) above, all the electors from a Let, now, y=X)s w i- Then Y is a random vari-
given state will vote for the same candidate, and able, equal to the sum of 50 variables Zj, where
so each state can be thought of as a single player.
We have, then, here, a weighted majority game
(M; c), where M is the set of all states (plus the
District of Columbia), and, for any T<ZM, \wj

0 if 269 As we have seen above, the variables Zj are

independent. Zj has the values 0 and wy with
3.1 v(T) = probabilities 1 —Xj and Xj, respectively. Thus Zj
1 if £ «V > 270
T has mean and variance
Wj being the number of electoral votes that state 3.4
j has.
For this 51-person game, a direct computation and Y, being the sum of these, has mean and
of the Shapley value (2.1) still seems out of the variance
question. Use can, however, be made of certain 3.5
properties of the game (e.g., that v has only the
values 0 or 1, and that v(T) depends only on
X r Wj) to obtain an exact result through a method 3.6
of generating functions. The exact details are
given in Shapley and Mann6 as calculated for the Since Y is the sum of many (50) independent
4 random variables, all with similar distributions,
Owen, "Multilinear Extensions."
5
Lloyd S. Shapley and Irwin Mann, "Values of Large it is possible to approximate Y by a normal ran-
Games, VI: Evaluating the Electoral College Exactly," dom variable having the mean (3.5) and variance
RM-3158-PR, The RAND Corporation, May 1962. (3.6). Since Y has only integer values, but the
1975 Evaluation of a Presidential Election Game 949
normal variable is continuous, a correction is game. This can best be described by saying that
made, setting 269.5 (rather than 270) as the neces-it is the composition v[uu «2, • • • , un] of the
sary number of votes. Then we have the approxi- game e, described in section 3, with the 51 games
mation Uj which are played within the individual states,
i.e., a game is played within each state to deter-
(xU • • • , Xn)
3.7 •
mine the state's electors, who then vote for the
^ Prob {269.5 - wt < Y' < 269.5}, presidential candidate according to the game v.
Each of the games UJ is very easy to handle—in
where Y' is a normal variable with the mean (3.5) fact, Uj is a simple majority game for my players—
and variance (3.6). nij being the number of voters in the state, which
Applying formula (2.3), now we notice that, if we assume proportional to the state's population.
Xi=Xi= • • • xn = t, then (3.5) and (3.6) take the Thus
form
3.8 0 if s <
and Uj(S) = •
nij
3.9 1 if j > —
2

The righthand side of (3.7) can now be evalu- where 5 is the number of players in 5.
ated (for given values of t) by means of a table of It is, of course trivial—by symmetry—to com-
the normal distribution; it is then necessary to pute the Shapley value for each of the games «,-.
integrate this (numerically), to obtain the power Unfortunately, composition of games does not
of state i. generally give rise to a composition of values, and
In Table 1, values are given both by the exact so it is not possible to obtain exact values for the
method and by the MLE approximation, for the composed game.
1970 apportionment. The approximation, while Turning to the MLE's, we find, however, that
not exact, is really not bad. (Maximum error 0.367 MLE's do compose when the games compose.
per cent.) Thus, if the game v has MLE /, while the games
« ] , • • • , M61 have MLE's g,, • • • , g a respec-
4. The Composed Game tively, then the compound game v[u\, • • • , uw]
We proceed now to the presidential election has the MLE h, where

Table 1

Number MLE Approximation

Electoral of Exact Value Rescaled Index Approximation Ratio
Votes States 538 ^» di'

45 .08830938 47.510446 .08852076 1.00239

41 .07972734 42.893309 .07975829 1.00039
27 .05096311 27.418153 .05113219 1.00332
26 .04897682 26.349529 .04915304 1.00360
25 .04699893 25.285424 .04715678 1.00336
21 .03916926 21.073062 .03927665 1.00274
17 2 .03146563 16.928509 .03157240 1.00339
14 1 .02576694 13.862614 .02586151 1.00367
13 2 .02388196 12.848494 .02396389 1.00343
12 3 .02200413 11.838222 .02207416 1.00318
11 1 .02013337 10.831753 .02019398 1.00301
10 4 .01826959 9.829039 .01833326 1.00349
9 4 .01641273 8.830049 .01647300 1.00367
8 2 .01456270 7.834733 .01461150 1.00335
7 4 .01271944 6.843059 .01276382 1.00349
6 4 .01088284 5.854968 .01092260 1.00365
5 1 .00905301 4.870519 .00907755 1.00271
4 9 .00722957 3.889509 .00724257 1.00180
3 7 .00541245 2.911898 .00543059 1.00335

Total 51 0.99999997 537.999803 1.00296983

950 The American Political Science Review Vol. 69
Table 2

Value
Electoral Total Per Inh. Banzhaf
Population Votes Value *i/m, Ratio Ratio
m< (X10-9) U./Pd.c. P>
1960
Alabama 3,266,740 10 .013971 4.2767 1.615 1.632
Alaska 226,167 3 .0011068 4.8937 1.845 1.838
Arizona 1,302,161 5 .0044025 3.3809 1.275 1.281
Arkansas 1,786,272 6 .0062101 3.4766 1.313 1.315
California 15,717,204 40 .130848 8.3251 3.143 3.162
Colorado 1,753,947 6 .0061309 3.4955 1.321 1.327
Connecticut 2,535,234 8 .0098302 3.8774 1.464 1.477
Delaware 446,292 3 .0015535 3.4809 1.313 1.308
Dist. of Columbia 763,956 3 .0020251 2.6508 1.000 1.000
Florida 4,951,560 14 .024200 4.8873 1.845 1.870
Georgia 3,943,116 12 .018404 4.6674 1.762 1.789
Hawaii 632,772 4 .0024692 3.9022 .472 1.468
Idaho 667,191 4 .0025366 3.8019 .434 1.429
Illinois 10,081,158 26 .065190 6.4665 2.430 2.491
Indiana 4,662,498 13 .021761 4.6672 .762 1.786
Iowa 2,757,537 9 .011564 4.1936 1.581 1.596
Kansas 2,178,611 7 .0079966 3.6705 .385 .392
Kentucky 3,038,156 9 .012143 3.9968 .509 .521
Louisiana 3,257,022 10 .013967 4.2882 .619 .635
Maine 969,265 4 .0030605 3.1575 .192 .186
Maryland 3,100,689 10 .013609 4.3890 1.657 .675
Massachusetts 5,148,578 14 .024652 4.7881 1.808 .834
Michigan 7,823,194 21 .046039 5.8849 2.223 2.262
Minnesota 3,413,864 10 .014300 4.1888 1.581 1.597
Mississippi 2,178,141 7 .0079966 3.6712 1.385 1.392
Missouri 4,319,813 12 .019273 4.4615 1.683 1.710
Montana 674,767 4 .0025455 3.7724 1.423 1.421
Nebraska 1,411,330 5 .0045350 3.2133 1.211 1.231
Nevada 285,278 3 .0012467 4.3701 1.649 1.636
New Hampshire 606,921 4 .0024238 3.9936 1.506 1.499
New Jersey 6,066,782 17 .032595 5.3727 2.026 2.063
New Mexico 951,023 4 .0030342 3.1905 1.204 1.197
New York 16,782,304 43 .146181 8.7104 3.287 3.312
North Carolina 4,556,155 13 .021538 4.7272 1.781 1.807
North Dakota 632,446 4 .0024692 3.9024 1.472 1.468
Ohio 9,706,397 26 .064013 6.5949 2.487 2.539
Oklahoma 2,328,284 8 .0094188 4.0454 1.528 1.541
Oregon 1,768,687 6 .0061779 3.4929 1.317 1.321
Pennsylvania 11,319,366 29 .078088 6.8986 2.604 2.638
Rhode Island 859,488 4 .0028828 3.3541 1.264 1.259
South Carolina 2,382,594 8 .0095422 4.0050 1.509 1.524
South Dakota 680,514 4 .0025588 3.7601 1.419 1.415
Tennessee 3,567,089 11 .016093 4.5115 1.702 1.721
Texas 9,579,677 25 .061003 6.3680 2.404 2.452
Utah 890,627 4 .0029358 3.2963 1.245 1.237
Vermont 389,881 3 .0014597 3.7440 1.415 1.400
Virginia 3,966,949 12 .018454 4.6519 1.755 1.784
Washington 2,853,214 9 .011754 4.1196 1.555 1.569
West Virginia 1,860,421 7 .0073532 3.9524 1.491 1.506
Wisconsin 3,951,777 12 .018415 4.6599 1.758 1.788
Wyoming 330,066 3 .0013411 4.0631 1.532 1.521
1975 Evaluation of a Presidential Election Game 951
Table 2. (Continued)

Value
Electoral Total Per Inh. Banzhaf
Population Votes Value Ratio Ratio
mi Wi
(xio-») Di/Cd.0. ft
1970
Alabama 3,444,165 9 .012003 3.4849 1.406 .421
Alaska 302,173 3 .001186 3.9253 1.584 1.582
Arizona 1,772,482 6 .005740 3.2385 1.307 .314
Arkansas 1,923,295 6 .005977 3.1079 1.254 1.261
California 19,953,134 45 .156584 7.8476 3.166 5.177
Colorado 2,207,259 7 .007474 3.3861 1.366 .376
Connecticut 3,032,217 9 .011271 3.7171 1.500 .515
Delaware 548,104 3 .001597 2.9131 1.175 .175
Dist. of Columbia 756,510 3 .001875 2.4783 1.000 1.000
Florida 6,789,443 17 .032070 4.7326 1.906 .941
Georgia 4,589,575 12 .018515 4.0341 1.628 .651
Hawaii 769,913 4 .002521 3.2804 1.324 .325
Idaho 713,008 4 .002428 3.4073 1.375 .376
Illinois 11,113,976 26 .063718 5.7332 2.313 '.361
Indiana 5,193,669 13 .021349 4.1106 1.659 .684
Iowa 2,825,041 8 .009662 3.4202 1.380 .392
Kansas 2,249,071 7 .007544 3.3542 1.353 .363
Kentucky 3,219,311 9 .011609 3.6062 1.455 .470
Louisiana 3,643,180 10 .013730 3.7686 1.521 .538
Maine 993,663 4 .002863 2.8860 1.164 .166
Maryland 3,922,399 10 .014238 3.6300 1.464 1.482
Massachusetts 5,689,170 14 .024084 4.2333 1.708 1.736
Michigan 8,875,083 21 .045549 5.1322 2.071 2.113
Minnesota 3,805,069 10 .014027 3.6864 1.487 1.505
Mississippi 2,216,912 7 .007490 3.3787 1.363 1.373
Missouri 4,677,399 12 .018688 3.9954 1.612 1.635
Montana 694,409 4 .002397 3.4516 1.393 1.394
Nebraska 1,483,791 5 .004375 2.9488 1.190 1.194
Nevada 488,738 3 .001508 3.0854 1.245 1.244
New Hampshire 737,681 4 .002470 3.3486 1.351 1.353
New Jersey 7,168,164 17 .032935 4.5946 1.854 1.889
New Mexico 1,016,000 4 .002897 2.8516 1.151 1.153
New York 18,190,740 41 .134168 7.3756 2.976 3.004
North Carolina 5,082,059 13 .021122 4.1563 1.677 1.702
North Dakota 617,761 3 .001695 2.7435 1.107 1.107
Ohio 10,652,017 25 .059855 5.6191 2.267 2.314
Oklahoma 2,559,253 7 .008042 3.1423 1.268 1.278
Oregon 2,091,385 5 .006231 2.9793 1.202 1.209
Pennsylvania 11,793,909 27 .068299 5.7910 2.337 2.385
Rhode Island 949,723 4 .002802 2.9498 1.190 1.192
South Carolina 2,590,516 8 .009257 3.5734 1.442 1.454
South Dakota 666,257 4 .002347 3.5260 1.423 1.424
Tennessee 3,924,164 10 .014242 3.6292 1.464 1.482
Texas 11,196,730 26 .063950 5.7115 2.305 2.352
Utah 1,059,273 4 .002958 2.7924 1.127 1.129
Vermont 444,732 3 .001438 3.2362 1.306 1.305
Virginia 4,648,494 12 .018631 4.0080 1.617 1.640
Washington 3,409,169 9 .011942 3.5030 1.413 1.428
West Virginia 1,744,237 6 .005695 3.2648 1.317 1.324
Wisconsin 4,417,933 11 .016631 3.7645 1.519 1.539
Wyoming 332,416 3 .001244 3.7423 1.510 1.509
952 The American Political Science Review Vol. 69

4.1 h =fo (gh It is now possible to approximate/i(*i, • • • ,Xn)

6
as in section 3 above; we apply the formulae (3.5),
is the composite function. (See Owen for details (3.6) and (3.7), as described there. Note that,
of this.) although (4.7) is formally an improper integral,
Now, the MLE / was discussed in section 3 it is in practice possible to neglect all values of T
above. To construct the MLE gy, we note that outside some small interval so that only a finite
g]{xi, • • • , xmj) is the probability that the coali- integration is necessary. Once 'J';, the value to all
tion 5 will have more than rrij/2 members, given voters in the i-th state, has been computed, this
that the several players have probability can be divided by TM< to obtain the value to each
xi, x2, • • • , xmj of belonging, and that these are individual voter from that state (clearly all voters
independent. We are interested, however, only in in the same state have equal powers).
values of gy along the "diagonal" xi = A certain difficulty arises here; namely, that we
• • • xmj=t, and there we find that do not know /n», the number of voters. We know
the population, and have an assumption that w,- is
4.2 gj(t, proportional to it, but we lack a constant of pro-
, 0 = Prob <Z > — portionality. Since, however, we are not so much
interested in the actual power of each voter, but
where Z is a binomial random variable with only in the disparity of power between voters
parameters my and /. This can be approximated from different states, the problem is not a serious
(with truly negligible error) by a normal variable one. We shall use the population figures for me,
with mean tm, and variance /(I — t)mj. Thus, the eventual power obtained per individual can
then be divided by the power per individual in the
>0 most disadvantaged state (really, the District of
4.3 Columbia) to obtain power ratios.
Table 2 shows the results of these computations,
together with a further approximation obtained
by dividing S ,• (as given in Table 1) by y/mT. (This
where <£ is the cumulative normal distribution is, in effect, the Banzhaf ratio.)
function. As in any approximative method, one should
Now, as is shown in Owen,7 the value to all worry somewhat about the errors introduced by
voters in the i-th state together is given by the approximation. Briefly speaking, there are
'x dxt three sources of error here: (a) the error intro-
fi(xu Xt, • • • , xn) dt, duced by the approximation in formula (4.3); (b)
/ o dt the error introduced by the approximation in
formula (3.7); (c) the error arising from numerical
where the integration is taken over the path integration.
Of these three, (a) can safely be neglected; m,-
*v(0 = &•(', • • •. 0 is clearly large enough so that the normal ap-
4.5 proximation to the binomial will be extremely
accurate. Similarly, (c) can be made very small if
the steps of integration are small enough, as
indeed they are here. The main source of error,
For the sake of simplicity, it is easier to change then, is (b). It is difficult to estimate this error a
the parameter from / to T, where priori; we might, however, expect that the error
< _ i here will be approximately the same as that ob-
I 2
4.6 tained for the state game (section 3). Comparison
- t) of the exact figures with the MLE approxima-
tions, as seen in Table 1, suggests that the error
Then T varies from — « to + « as f goes from 0 here is not large either. Thus we feel these results
to 1. Thus we will also have are reasonably accurate.

4.7 ¥<= I /,(*!, • • -,xn)-^dT, 5. Conclusions

Generally speaking, four factors seem to con-
tribute to the disparity of power in presidential
where the path of integration is elections. Two of these, of course, have always
been known. The others have been more or less
4.8 xj(T) = $(Ty/mj)- suspected, but it is only with game-theoretic
analysis that they can be measured.
' Owen, "Multilinear Extensions."
7
Owen, "Multilinear Extensions." a. The first factor is what might be called the
1975 Evaluation of a Presidential Election Game 953
integrity factor, which is best explained by minor, a necessary evil to contend with, while (b)
saying that the number of electors from each was part of a compromise which made the smaller
state must be an integer. The obvious conse- states more willing to join the union. The other
quence is that there may be states with two factors, however, are probably quite un-
sizable differences of population but the expected. Factor (c) was certainly not expected by
same number of electors, e.g., Alaska with the Constitution's framers, since the present
302,173 inhabitants and North Dakota with method of pledged electors was not then foreseen;
617,761. Both have one U.S. Representa- in any case it is minor, as Table 1 shows. The
tive, and hence both have 3 electors. This strongest factor, from Table 2, is clearly factor
factor clearly benefits the voters from states (d). Somehow, people have been aware of this
whose population is low among those with factor's existence, as evidenced by the traditional
equal numbers of U.S. Representatives. rule which says that the more populous states are
b. The second factor might be called the Senate usually "swing" states. It is only with game-
factor, and is explained by saying that the theoretic analysis, however, that the extent of this
number of electors is equal to a state's total factor has been measured, first by Banzhaf,8 and
congressional representation. Now the now in this paper.
state's delegation to the House is (more or In closing, it might be worthwhile to comment
less) proportional to population, but all on a recent controversy over allocation of re-
states have two senators. Thus, this factor sources in a presidential election.
tends to favor states with low population. In one paper9, Brams and Davis recommend
c. The third factor, which was first noticed by a "3/2 power rule" for allocation, and suggest
Shapley and Mann (see note 5), might be that this is empirically tested by strategies ac-
called the unit rule factor. Essentially, we see tually used in the last four presidential elections.
in Table 1 that, if the values to the states in In another paper10, Colantoni, Levesque and Or-
the state game are multiplied by 538, the deshook disagree.
power for the larger states is slightly more It should be pointed out that our model is quite
than their electoral vote; that for the smaller different, treating the election as an n-person game
states is slightly less than their electoral vote. among the voters, while the authors in both the
This factor favors the large states. above papers treat an election as a two-person
d. The fourth factor can best be described as game between the candidates. Moreover, it is
the composition factor. This can best be seen dangerous to try to find empirical validation of a
mathematically from formula (4.7), where non-obvious and complicated mathematical model.
the factor dxi/dT appears. From (4.8), we Nonetheless, it is interesting to note that the
will have Shapley value as here obtained is very similar to
the allocation recommended by Brams and Davis.
8
Banzhaf, "One Man, 3.312 Votes."
5.1 "Steven J. Brams and Morton D. Davis, "The 3/2's
IT Rule in Presidential Campaigning," American Political
Science Review, 68 (1974), 113-134.
which, for small values of T, is roughly pro- 10
Claude S. Colantoni, Terrence J. Levesque, and
portional to yfmj. (For larger values of T, of Peter C. Ordeshook, "Campaign Resource Allocations
course, the exponential term decreases Under the Electoral College," American Political Sci-
ence Review, 69 (1975), 141-154; followed by a
rapidly as irij increases.) Since most of the "Comment" from Brams and Davis, 155-156, and a
integral (4.7) will be concentrated near "Rejoinder" by Colantoni et al, 157-161.
T=0, it follows that there will be an addi-
tional factor, not quite equal to \/Tn~j, Additional Bibliography
favoring the larger states. John F. Banzhaf III, "Weighted Voting Doesn't
Work: A Mathematical Analysis," Rutgers Law Review,
19 (1965), 317-343.
Of these four factors, it is clear that (a) and William F. Lucas, "Measuring Power in Weighted
(b) were expected by the original framers of the Voting Systems," unpublished manuscript, Cornell Uni-
Constitution. Factor (a) was probably considered versity, Center for Applied Mathematics, October 1973.