Vous êtes sur la page 1sur 10

Qual Quant (2012) 46:949958

DOI 10.1007/s11135-011-9454-9

Looking for voting paradoxes in Indian elections


Santanu Gupta

Published online: 27 February 2011


Springer Science+Business Media B.V. 2011

Abstract We try to find out a methodology for identifying Borda Paradox and Condorcet
Paradox in the absence of information on preference profile of citizens, and try to apply the
same to find out existence of such paradoxes in the 2004 Indian Parliamentary elections.
Keywords

Voting paradoxes Condorcet Borda Electoral alliances

1 Introduction
Although, voting paradoxes have been well well documented in the theoretical literature,
there are still attempts to identify the actual occurrences of these paradoxes. Klitgaard (2008)
has tried to look for voting paradoxes under proportional elections with data from the Danish elections. Coleman and Pountney (1978) look for the theoretical likelihood and electoral
occurrences of Borda Voting Paradox using sample surveys of preference ordering of citizens
from 1966 British Elections. However, since most electoral systems do not ask for preference ordering from citizens, deciphering the preference ordering and therefore identifying
the occurrences of the paradoxes becomes a difficult task. This paper tries to address this
specific issue of how much can be deciphered about voting paradoxes from election results of
the first past the post system of the kind that is prevalent in India. The next section discusses
some features of Indian elections which might affects the identification of voting paradoxes,
Section 3 discusses ways to decipher occurrences of a Borda Paradox in the Indian context, Section 4 discusses ways to identify the occurrence of a Condorcet Paradox. Section 5
concludes.

S. Gupta (B)
XLRI, School of Business and Human Resources, C. H. Area (East), Jamshedpur, Jharkhand, India
e-mail: santanu@xlri.ac.in

123

950

S. Gupta

2 Some features of Indian elections


With the spread of the private television news channels conducting exit poll before elections
has become very popular, but in most cases, the predictions of exit polls were far removed
from the actual results. However, none of these surveys have asked for the preference profile
of citizens with respect to different parties competing from their electoral districts. Absence
of such surveys, make it almost impossible, to come up with conclusive evidence of the
occurrence of voting paradoxes.
The other problem that makes it even more difficult is the fact, that the number of candidates contesting from any electoral district is very large. For example in the 2004 Indian
Parliamentary Election, of the 543 electoral districts, only three witnessed two candidate
contest, 15 witnessed three candidate contest and 31 witnessed four candidate contest. The
average number of candidates was 10 in the 2004 elections from any district. With such large
number of candidates contesting from any district, it will be difficult to conduct surveys
asking citizens to rank such large number of candidates, in situations where they may not
even have heard or known most of them. Most citizens would be knowing which party to
vote for, and at most they may be clear on which party not to vote for.
Major parties realize that there is a huge split of the votes due to so many candidates contesting the election, in order to consolidate their vote share, major parties in the recent years
have resorted to electoral alliances. As to how electoral alliances affected election results will
be discussed in the next subsection in detail. Parties may also not have had a pre-poll alliance,
however, after the elections they decide to have a post-poll alliance. A post poll alliance will
happen if neither of the parties have won enough number of seats to form a government of
their own, however their combined number of seats won gives them a majority in parliament.
If parties form a pre-poll alliance, there is the implicit agreement to form a post-poll alliance
in case their combined seats is good enough to form a majority in Parliament. We will discuss
later that we may be able to find out possibilities of a Borda Paradox if parties do not have a
pre-poll alliance but form a post-poll alliance.
2.1 Electoral alliances
Given the fact that the presence of a large number of parties may make it difficult for citizens to come up with proper ranking of candidates, the other issue which might make it
difficult for citizens to rank, even in the presence of small number of parties is parties target
certain population profiles and appeal to certain characteristics. For example, people have
income characteristics, may be rich (R) or poor (P). They may either belong to community
A, or community B. Table 1 gives us the spread of voters by both income and community
in three electoral districts. Let there be four parties competing in the electoral race, namely,
PR , PP , PA and PB , appealing to rich, poor, voters from community A and B respectively.
Voters would prefer a party with its own characteristics over one that is not, but are indifferent
between two parties appealing to its different characteristics. For example a rich voter from
Community A, prefers a party appealing to the rich or one that appeals to community A, over
parties that appeal to poor or to community B, but is indifferent between a party appealing
to the rich and another appealing to community A. In such situations, we assume that votes
of the rich voters from community A, would be equally split between, the party appealing
to the rich and the party appealing to community A, were both of them to stand for election
from the same electoral district. In a similar manner the voters from other income classes and
communities would be split equally likewise between two parties. If for a particular income,
community group, there is only one party contesting that either appeals to its income or the

123

Looking for voting paradoxes in Indian elections

951

Table 1 Spread of voters by characteristics in different districts


District 1

District 2
A

District 3
A

10

Table 2 Results in a four candidate election between PR , PP , PA and PB


District 1
Votes

District 2
PR

PP

PA

PB

Votes

District 3
PR

PP

PA

PB

Votes

PR

PP

PA

PB

R and A

R and A

R and A

R and B

R and B

R and B

P and B

P and B

P and B

P and A

P and A

P and A

Total

Total

Total

Outcome: PP wins

Outcome: PA wins

Outcome: PA wins

community characteristic, all votes from this voter profile go to that party. If there is no party
contesting that represents either the income or the community characteristic from an electoral
district, such voters would tend to abstain from the election. Table 2 gives the results in an
electoral contest if all the four parties were to contest from each of the three districts. In
District 1, the two votes of the voters who are rich and belong to community A, are divided
equally between party PR and PA , and likewise all the other votes are distributed. In such
a situation, we see that party PP is the plurality winner from district 1, and party PA is the
plurality winner from districts 2 and 3.
In such a scenario, parties PR and PB may decide to form an electoral alliance, which we
will say as a pre-poll alliance. According to the alliance agreement, party PR agrees not to
field a candidate in district 1, and party PB agrees not to field candidates in districts 2 and 3.
Under such an alliance, the electoral results are depicted in Table 3. After the formation of
the alliance, PP , PA , and PB are the competing parties in district 1, all the two votes of the
rich voters from community A, which were earlier split between party PR and PA now go to
party PA , however all the six votes of the rich voters from community B, which were earlier
split between party PR and PB , now go to party PB . Now party PB emerges as the plurality
winner from district 1, and party PR emerges as the plurality winner from districts 2 and 3.
Therefore, by having a pre-poll alliance, parties PR and PB are able to win all the seats and
form the government.
We will like to illustrate with the above example, the mechanism of electoral alliances,
and that parties entering into an alliance may be successful only if they choose their alliance
partners carefully keeping in mind the population profile in different districts. However, we
will also like to highlight, that the reason for the large number of candidates contesting Indian
elections, is due to the fact that each party is trying to locate a voter profile and trying to
appeal to a particular characteristic, or a set of characteristics. One of the mainstream national
party, the Indian National Congress, has had a presence in most parts of India, has largely
identified itself as one striving for the interests of the poor at large. The main opposition party,

123

952

S. Gupta

Table 3 Results with an electoral alliance between party PR and party PB


District 1
Votes

District 2
PP

PA

PB

Votes

District 3
PR

PP

PA

Votes

PR

PP

PA

R and A

R and A

R and A

R and B

R and B

R and B

P and B

P and B

P and B

P and A

P and A

P and A

Total
7
6
Outcome: PB wins

Total
8
7
Outcome: PR wins

Total
11
10
Outcome: PR wins

the Bharatiya Janata Party (BJP), initially drew its support base from small traders, then tried
to expand its base by trying to appeal to the religious sentiments of the majority community.
However, since the 1990s regional parties became strong, and the Indian National Congress
which had earlier drawn support from all sections of the poor found its support base being
taken away by regional parties which appealed to caste or other local characteristics of the
voters. However, the BJP realized, that merely by having appealing to the majority community will not enable it to form a government at the national level and therefore it went for
alliances with regional parties. Finally in the 2004 Parliamentary elections, even the Congress realised the loss of its support base amongst some sections and went in for electoral
alliances with regional parties, and 2004 elections saw two major alliances emerging, the
United Progressive Alliance (UPA) led by the Indian National Congress, and the other the
National Democratic Alliance led by the Bharatiya Janata Party.
We will like to emphasize that the notion of a Condorcet Winner in such a situation may
be inappropriate, a Condorcet Winner in this case even if it exists will not have the mandate
of the whole population. In such a situation, if we are looking at two candidate elections, it
may be so that a chunk of the voters will abstain if the parties participating do not reflect their
characteristics, in this case a Condorcet winner will necessarily have to be an alliance, which
beats any other alternative alliance only for situations when there is full voter participation.

3 Borda Paradox
A weak Borda Paradox is said to occur when a majority of the voters prefer a losing candidate to the plurality winner and a strong Borda Paradox can be said to occur when a majority
of the voters prefer any of the losing candidates to the plurality winner. Attempts to trace
a Borda Paradox were done by Coleman and Pountney (1978). They used an election survey data after the 1966 general elections in UK, to find out the proportion of voters with
different preference profiles and only 15 of the 261 constituencies with three candidate contests displayed the Borda effect. Nurmi and Suojanen (2004) use a similar methodology to
look for a Borda Paradox in the 2001 British elections, and they found such a possibility in
eight such constituencies. Given the absence of survey data giving preference profiles, in the
Indian elections, we resort to alternative information available in such a situation that helps
us decipher a Borda Paradox.
In India, in order to avoid situations like the Borda Paradox, parties normally resort to
what is called electoral alliances and it boils down to having seat adjustments in electoral
districts. The official reason as given by the parties entering into an alliance is that they

123

Looking for voting paradoxes in Indian elections

953

have similar agenda, then there is a possibility of votes being split up between these parties,
and a third party with lower votes than the combined vote of these parties win. However, as
explained in the last section, electoral alliances help in the consolidation of votes of voters
with different characteristics. Seat adjustments would imply that such parties divide the total
electoral districts amongst themselves, and only one of these two parties fields a candidate in
any district. Parties then urge the voters, in electoral districts they are not contesting to vote
for the party they have alliance with. If alliance partners are chosen carefully, both parties
have a chance to win a larger number of seats and are in a better position to stake claims to
government formation.
In the 2004 election both the national parties, the Congress and BJP went for electoral
alliances with different parties in the states. However, in the state of Uttar Pradesh, which
has 80 seats of the total 543 electoral seats, talks of electoral alliance failed between both the
national parties and the two regional parties, Bahujan Samaj Party (BSP) and the Samajwadi
Party (SP). So in most of the electoral districts, all the four parties contested. However, after
the elections, the Samajwadi Party joined the ruling coalition led by the Congress. Electoral
analysts had then commented that had these two parties gone for a pre-poll alliance, it would
have been better for both these parties. To analyze this logic, one needs to look at electoral
districts where both the Congress and the Samajwadi party lost, but their combined votes were
more than that of the plurality winner, and if their combined votes is more than 50% there is
a strong possibility of a Weak Borda Paradox. There are 20 such electoral districts where the
combined vote share of Congress and SP is more than the vote share of the Plurality Winner,
but the largest combined vote share is as high as 49%. It might be interesting to rank these
20 districts in descending order of the possibility of Borda Paradox. We will say the higher
is the combined vote in these 20 districts, the higher is the possibility of a Borda Paradox,
and the higher is the difference between the combined vote share of these two parties and
the vote share of the plurality winner, the higher the chances of a Borda Paradox.
Let x1 = combined vote share of SP and INC in these 20 districts and x 2 = difference in the
combined SP, INC vote share and the vote share of the plurality winner. Let us construct an
index


x1 min(x1 )
1
(x2 ) min(x2 )
I =
+
2 max(x1 ) min(x1 )
max(x2 ) min(x2 )
where max(xi ) i {1, 2} is the maximum value of xi in these 20 districts, and min(xi ) is
the minimum value of xi in these 20 districts. The rankings of these 20 districts and the value
of the index is reported in Table 4. We see that the maximum possibility of a Borda Paradox
exists in Azamgarh while the least possibility is in Jalaun.

4 Condorcet Paradox
Given a first past the post electoral system, a Condorcet Paradox is possible in an election
involving three or more than three candidates and the winner receives less than 50% of the
votes in which case there exists no Strong Condorcet winner who can defeat every other candidate in a two candidate election involving him and any one else. With an election involving
three candidates, A, B, and C, let A receives the largest votes and hence is the winner, B
receives the second largest votes and C the least number of votes. Let A , B and C be the
share of total votes obtained by A, B and C respectively. If there exists no strong Condorcet
Winner, it is the case that A < 0.5. With such an electoral outcome, Table 5 lists the eight
preferences profile that are theoretically possible, provided all citizens who voted for A, or B

123

954
Table 4 Ranking of districts in
Uttar Pradesh with possible
Borda Paradox

S. Gupta
No.

Electoral district

Value of I

Azamgarh

0.9768

Maharajganj

0.8127

Domariaganj

0.8100

Unnao

0.8085

Mirzapur

0.7426

Basta

0.6947

Sultanpur

0.5969

Sitapur

0.5916

Misrikh

0.5221

10

Fatehpur

0.5134

11

Aonla

0.3747

12

Bareilly

0.3326

13

Chandauli

0.3123

14

Robertsganj

0.2961

15

Padrauna

0.2766

16

Pilibhit

0.1707

17

Faizabad

0.1464

18

Bilhaur

0.1445

19

Bara Banki

0.1391

20

Jalaun

0.0050

or C as their first choice have the same preference profile. Given the set of profiles generated
in Table 5, we see the outcome in these profiles in Table 6. Profile 3 and Profile 7, generate A
as the Condorcet Winner, Profile 2 and Profile 4 generate B as the Condorcet Winner, Profile
5 and Profile 6 generate C as the Condorcet Winner and Profile 1 and Profile 8 generate a
Condorcet Paradox.
Therefore, even in the absence of a Strong Condorcet Winner in a three candidate election, there is only a 0.25 probability that a Condorcet Paradox exists. A Condorcet Paradox
may also be said to be a situation with outcomes with no clear mandate (OWNCM), that
is there is no candidate who can defeat anyone everyone else in a two candidate election.
However, in the absence of information on whom the second and third votes went to, it is not
possible to comment on the likelihood of a Condorcet Paradox looking only at the first vote
shares of different parties, we will look for outcomes with no clear mandate (OWNCM).
We will say that OWNCM most likely exists for a vote share profile that is most distant from
profiles where either, A, B or C are strong Condorcet Winners. If there happens to be a strong
Condorcet Winner, the candidate will naturally be a plurality winner with more than 50% of
the votes. We are also aware that in a Saari Vote Share Triangle in a representation of vote
shares with three candidates, a strong Condorcet Winner is concentrated at one of the edges
of the triangle, and the further the profile is from any of the edges, we will say are profiles
with OWNCM.
In Fig. 1, we have the representation of a Saari Vote Share Triangle where any point
inside the equilateral triangle represents the vote share of each of the three candidates A, B
and C (See Nurmi 1999, pp. 3133, for more details). The triangle is a plane (simplex) in a
three-dimensional space that intersects all three axes at points (1, 0, 0), (0, 1, 0) and (0, 0, 1).
So, the coordinates of any point within the triangle sum up to unity. From any point inside an

123

Looking for voting paradoxes in Indian elections

955

Table 5 Possible preference profiles in the first past the post electoral system with three candidates
Profile 1

Profile 2

Profile 3

Profile 4

Profile 5

Profile 6

Profile 7

Profile 8

Table 6 Results in different


preference profiles in the first
past the post electoral system
with three candidates with no
strong Condorcet Winner

Profile

A and B

B and C

A and C

Implication

A > B

B > C

C > A

Condorcet cycles

B > A

B > C

C > A

B > C >A

A > B

B > C

A > C

A > B >C

B > A

B > C

A > C

B > A >C

A > B

C > B

C > A

C > A >B

B > A

C > B

C > A

C > B >A

A > B

C > B

A > C

A > C >B

B > A

C > B

A > C

Condorcet cycles

equilateral triangle, the sum of the length of the perpendiculars drawn to the three sides of
a triangle are equal. In Fig. 1, D, E and F are the midpoints of AB, BC and AC respectively,
therefore all points on the line DF represent vote share profiles where A gets exactly 50% of
the vote. Therefore, the region ADF represent vote share profiles with a Strong Condorcet
Winner, which is candidate A in this case. The region AFD represent the region where A is
a Strong Condorcet Winner, and the region AHD represents the region where A receives the
largest vote share, B the second largest and C the least. Therefore, the region GHD represents
the region, where, where A gets the largest number of votes, B second and C the least number
of votes, but A gets less than 50% of the votes so there exists no Strong Condorcet Winner. We
know in this region there are equal possibilities of A, B or C being the Condorcet Winner, and
of a Condorcet Paradox, from Table 6, we therefore divide this region into four equal regions,
GIKD, DKJH, HJIG and IJK. The first three regions, GIKD, DKJH and IJK represent regions
where candidates, A, B, and C have performed their best, respectively, relative to that in a
plurality outcome with no Strong Condorcet Winner. The triangle IJK, represents the region
where neither party B nor C can be said to be giving a tough fight to the plurality winner A,
neither can A be said to be the majority mandate, we will then say that profiles in this region
represent OWNCM. It is important to note that in a plurality outcome when we question, the
chances of the winner being a Condorcet Winner, we need to ask as to who could have been
the Condorcet winner, and in the absence of information on preference orderings, we need
to give weightage to the first votes also due to the fact, that voters may care only for their
most preferred candidate.
Now that one has identified the region where OWNCM can be said to take place, one can
now calculate the limits within which the vote shares of candidates should be, for one to
say that the profile represents an OWNCM (see Appendix 1 for the complete proof). Table 7
reports the critical values within which As vote share should lie if A is the plurality winner,

123

956

S. Gupta

Fig. 1 Representation of Vote Shares of candidates A, B, and C in a Saari Triangle


Table 7 Vote shares of
candidates in situations of
OWNCM

As vote share lower limit

0.3889

As vote share upper limit

0.4722

Cs vote share lower limit

0.0972

Cs vote share upper limit

0.2639

Table 8 Vote shares of candidates with most and least votes in the five electoral districts with no Strong
Condorcet Winner with Three Candidate Elections: Indian Parliamentary Elections 2004
No.

Electoral district

State

Share first

1
2
3
4
5

Share third

Anakapalli

Andhra Pradesh

0.4928

0.0341

Bijapur

Karnataka

0.4367

0.1741

Nowrangpur

Orissa

0.4611

0.1097

Sambalpur

Orissa

0.4818

0.0546

Daman and Diu

Daman and Diu

0.4951

0.0207

Note: The underlined electoral districts are situations with OWNCM

and the critical values within which Cs vote share should lie, if C is the candidate getting least
votes in a three candidate election for a situation of OWNCM. In order to be in the triangle
IJK, As vote share limit must be between 0.3889 and 0.4722, and Cs vote share, must be
between 0.0972 and 0.2639. In Table 8, we report the vote shares in the five electoral districts,
from the 2004, Indian Parliamentary Elections with three candidate elections with no Strong
Condorcet Winner, and only two electoral districts, namely Bijapur and Nowrangpur can be
identified as OWNCM.
It will be interesting to identify situations with OWNCM in elections with four or more
candidates, a similar methodology could be used, first in determining the list of possible preference profiles, then identifying the various Condorcet Winners and their frequency and then

123

Looking for voting paradoxes in Indian elections

957

the frequency with which cycles occur. Once this has been done, the critical values of OWNCM can be worked out depending on the probability of such an outcome. Since geometric
method cannot be used to compute the critical values, this remains a challenge and scope for
future work. This exercise will be all the more relevant if since in such situations it is most
unlikely that citizens will have a clear idea of their preference profile, they will however,
know for sure their most preferred candidate, and at most their least preferred candidate.

5 Conclusion
This paper tried to discuss unique features of Indian elections that may help us decipher the
existence of voting paradoxes in Indian elections. We also try to highlight the fact that that
in situations with large number of candidates, identifying traditional voting paradoxes may
not be so much relevant, especially since voters may not know the candidates at all therefore
may not be able to come up with a proper ranking. Even when the number of candidates are
small, the conventional voting paradoxes may not be relevant since there may not be full
participation in situations with only a small number of candidates. Therefore future work
needs to redefine the voting paradoxes in the context of electoral contests with large number
of candidates and also identify a suitable methodology of identifying them.
Acknowledgments The author wish to thank European Research and Educational Collaboration in Asia
(EURECA) which sponsored his visit to the Public Choice Research Centre (PCRC), in the University of
Turku, Finland in March to May 2009, during which time this paper was written. The author wish to thank
Hannu Nurmi in particular for valuable feedback on this work and introducing him to the area of Voting
Paradoxes.

Appendix 1
ABC is an equilateral triangle of length

2. Therefore AB = AC = BC =

AD = D B = AF = FC = C E = E B =

AE = B F = C D = 23
DF =
DG =

1
2 BC
1
2EB

HD =

1
3C D

=
=

= 1
 6

2
2
G H = H D DG = 16
IK =

1
2
1

2 2
1
3

3
2

1
8

1
24

2.

1
2

2 6

1
1

2 DG = 4 2
1
1
I J = 2GH =
4 6
1
J K = 21 H D =
2 6
1
1
Area of G D H = 21 DG G H = 21

= 1
2 2
2 6
16 3
1
1
Area of I J K = 21 I K I J = 21

= 1
4 2
4 6
64 3
Area of G I K D = D K J H = H J I G = 1
64 3

1
Area of G I K D = I N 21 (G D + I K ) =
64 3
1
1
3
GD + I K =
+
=
2 2
4 2
4 2

123

958

S. Gupta

IN =

1
64 3

LG = I N =
LM = I J =

4 2
3

1
12 6

1
12 6
1

4 6

M H = G H LG L M =

H E = 13 AE = 13 23 = 1
MH + HE =

6 6

LM + MH + HE =

2 6

6
7
1 =
6
6 6
1
7

+
4 6
6 6

1
12 6

4 6

6 6

17

12 6

To find out critical values


of As an Cs vote share in order to be OWNCM, they have to

be normalized since AE =

3
2.

For OWNCM, As vote share lower limit


7
18

= 0.3889

For OWNCM, As vote share upper limit

2
3 (L M

= 0.4722
Area of H J I G = M J 21 (G H + I J ) =
17
36

GH + I J =
MJ = 2

2 6

1
64 3

1
3

=
4 6
4 6
4 6
1
3 = 12 2

J H2 = M J2 + M H2 =

7
J H = 16 24

1
12 2

2

6 6

2

H P2 = J H 2 J P2 =
HP =


1
6

7
24

2
3

6 6

2
17

3 12 6

=
=

1
62

1
8

1
6

1
62

7
24

1
64 3

JK + HD =
JP = 2

+ M H + H F) =

1
64 3

Area of D K J H = J P 21 (J K + H D) =
1
3

+ 1 =
2 6 6
2 6
1
2 3 6 = 1
64 3
24 2

(M H + H E) =

2
3

1
24 2

2

1
246

7
243

7
6

1
8

25
6242

5
24 6

PD = HD HP =
QD = PD PQ =

5
1
6
24 6
19
1

24 6
2 6

Therefore, Cs vote share lower limit =



19
share upper limit = 23 P D = 243
=

=


19

24 6
7
24 6

2
3 QD

19
72

7
72

= 0.0972 and Cs vote

= 0.2639

References
Coleman, A.M., Pountney, I: Bordas voting paradox: theoretical likelihood and electoral occurrences. Behav.
Sci. 23, 1520 (1978)
Klitgaard, P.K.: Voting paradoxes under proportional representation: evidence from eight Danish elections. Scand. Political Stud. 31(3), 242266 (2008)
Nurmi, H.: Voting Paradoxes and How to Deal with Them. Springer, Berlin (1999)
Nurmi, H., Suojanen, M.: Assessing contestability of electoral outcomes: an illustration of Saaris geometry
of elections in the light of 2001 British Parliamentary elections. Qual. Quant. 38, 719733 (2004)

123

Vous aimerez peut-être aussi