A MIP model for scheduling Indias

General elections and police movement

Bodhibrata Nag

OPSEARCH

ISSN 0030-3887

OPSEARCH
DOI 10.1007/s12597-013-0160-3

1 23
Your article is protected by copyright and all
rights are held exclusively by Operational
Research Society of India. This e-offprint is
for personal use only and shall not be self-
archived in electronic repositories. If you wish
to self-archive your article, please use the
accepted manuscript version for posting on
your own website. You may further deposit
the accepted manuscript version in any
repository, provided it is only made publicly
available 12 months after official publication
or later and provided acknowledgement is
given to the original source of publication
and a link is inserted to the published article
on Springer's website. The link must be
accompanied by the following text: "The final
publication is available at link.springer.com.

1 23
 Author's personal copy
OPSEARCH
DOI 10.1007/s12597-013-0160-3
A P P L I C AT I O N A RT I C L E

A MIP model for scheduling Indias General elections

and police movement

Bodhibrata Nag

Accepted: 26 August 2013

# Operational Research Society of India 2013

Abstract Conducting the General Elections for the 543 members of the Parliament
of India across the 833 thousand polling stations spread over the 35 States is a
mammoth exercise. Deployment of the Central Police Forces is essential to comple-
ment the role of the State police during the elections. However paucity of Central
Police Forces necessitates the conduct of elections over stages. This paper proposes
and demonstrates a MIP model to (a) schedule the elections with a minimum number
of stages (b) sourcing the appropriate number of security personnel from the most
convenient Central Police Forces bases (c)scheduling movement of security person-
nel between stages with the objective of minimizing men-miles.

Keywords OR in government . Scheduling . Assignment . Logistics . MIP model

1 Introduction

The Indian Republic comprises 28 States and 7 Union Territories. The Indian parliamen-
tary form of government is federal in structure with legislative powers distributed between
the Parliament of India and State Legislatures. The Parliament of India comprises two
legislative bodiesthe Upper House or the Rajya Sabha and the Lower House or the Lok
Sabha. The members of the Rajya Sabha are elected indirectly by legislators of States and
Union Territories comprising the Union of India. The 543 members of the Lok Sabha are
elected directly by universal adult franchise by the electorate of all the 28 States and 7
Union Territories through the General Elections. The term of office of each Lok Sabha is
5 years from the date of its first meeting, unless dissolved earlier.
The total membership of the Lok Sabha is distributed amongst the 35 States and
Union Territories (which will be referred to as States in the remainder of the paper) in
such a manner that the ratio of the population to number of seats allotted to any State is

Electronic supplementary material The online version of this article (doi:10.1007/s12597-013-0160-3)

contains supplementary material, which is available to authorized users.
B. Nag (*)
Indian Institute of Management Calcutta, Joka, Diamond Harbour Road, Kolkata, India
e-mail: bnag@iimcal.ac.in
 Author's personal copy
OPSEARCH

nearly the same. The geographical area of the State is then demarcated into a number of
territorial constituencies (with geographical boundaries), equal to the number of seats
allotted, such that population of all constituencies in that State is nearly the same. Each
constituency has a large number of polling stations distributed across the constituency
such that voters can reach the polling stations to cast their vote with minimum travel.
The General Elections of India are the worlds biggest election exercise. During
the 2009 General Elections, a 717 million strong electorate exercised their franchise
through 1.3 million Electronic Voting Machines deployed in 833 thousand polling
stations spread across the length and breadth of India to elect 543 Members of the
Lok Sabha from amongst 8 thousand candidates contesting the elections. The only
other comparable elections are the European Parliament elections with an electorate
of 500 million and the US Congress elections with electorate of 312 million.
Elections in the past have been marked by instances of voter intimidation through
violence or harassment in various forms, as well as clashes between political oppo-
nents [1]. These incidences have been largely arrested through deployment of addi-
tional police forces during the polling process in order to bring peace, restore
confidence in candidates and voters and thereby ensure fair and free elections.
The Constitution of India mandates that maintenance of law and order is the
responsibility of the States. Thus while all States maintain police forces totaling about
1.5 million, the average police-population ratio for all the States is only 133 police per
100,000 [2] in comparison with average international ratio of 342 [3]. The Central
Government therefore maintains Central Police Forces at various bases spread across the
country, to complement the State police, whenever and wherever required.
Since the State police are the arm of the State governments, allegations of partisan
conduct of police in enforcing law and order during the election campaign closing stages
and during the day of elections are likely. It has therefore become universal practice to
deploy Central Police Forces, in addition to State police at all polling stations during the
General Elections. However, the numbers of Central Police Forces that can be spared for
deployment during the elections are not enough for manning all the polling stations of the
543 constituencies. The General Elections are therefore spread over different days with each
day covering a few States only, such that the required number of Central Police Forces can
be deployed across all polling stations of all constituencies of those States. The days of
elections are spread a few days apart to allow re-deployment of Central Police Forces and
allow them to be familiar with their constituencies. However, elections for all constituencies
in a particular State are held on the same day. For example, the 2009 General Election was
conducted in five stages on 16 April, 23 April, 30 April, 7 May and 13 May.
The movement of Central Police Forces from their bases to the polling stations in the
different stages and their subsequent return to the bases is a gigantic exercise, requiring
coordination between different agencies such as Central Police Forces operations, Election
Commission and State Chief Electoral Officers, District Election Officers, Railways, airlines
and the Indian Air Force. In the 2009 General Election, 119 special trains, 65 sorties by Indian
Air Force transport aircraft, 600 sorties by Indian Air Force helicopters and Air India
chartered flights were used for the cross-country movement of Central Police Forces [4].
The exercise of conducting elections in all the States within the minimum possible
time, along with deployment of adequate Central Police Forces and minimum possible
police movement poses an interesting operations research problem. An instance of the
problem is visually depicted in Fig. 1, wherein elections have to be conducted in 12 States
 Author's personal copy
OPSEARCH

Fig. 1 Police deployment in different stages of elections

indicated by S1, S2,., S12. Central Police Forces are deployed for elections from the
three police bases P1, P2 and P3. The number of police forces available at base P1 are
more than or equal to the requirement of Central Police Forces required for conducting
elections in States S1 and S2 in stage 1, States S5 and S6 in stage 2 and States S8,S9 and
S10 in stage 3. Police forces first travel from base P1 to States S1 and S2 for stage 1
elections; secondly, police forces travel from State S1 to State S5 and State S2 to State S6
for stage 2 elections; thirdly, police forces travel from State S5 to State S8 and State S6 to
States S9, S10 for stage 3 elections; finally the police forces return from States S8, S9 and
S10 to base P1. Police forces can travel directly from the base to any State where elections
are being held in the stage 2 or stage 3; for example, police forces can travel directly from
base P2 to State S6 in stage 2 and State S8 in stage 3. Police forces can return back from
any stage to their respective bases, if there is no further requirement in the subsequent
stages; for example few police forces can return back to base P1 directly after completing
stage 1 elections in State S2 or stage 2 elections in State S5.
So far the entire process of determining the number of election stages, States included in
each stage, police personnel manning the polling booths in States where elections are being
held in each stage and the movement of police personnel is done manually. The manual
process of decision making involving movement of 300 thousand police personnel from
about 50 police bases to 833 thousand polling stations of 543 constituencies spread over 35
States, is likely to be sub-optimal resulting in waste of manpower and transport fuel and
resources. This paper proposes an operations research methodology to enable conduct of the
General Elections for all the 543 parliamentary constituencies with (a) minimum number of
stages, (b) with the available Central Police Forces and (c) with minimum police movement.
Thus we propose a methodology for: (i) the scheduling of elections for all 543 constituencies
such that elections for all constituencies of a State are held together on the same day, and the
elections are completed within the minimum possible time (ii) scheduling movement of
police from the bases to the polling stations and polling stations of a certain stage to polling
stations in the next stage, such that the requirement of police personnel at the polling stations
are met, appropriate number of police personnel from each base are deployed and the total
number of men-miles traversed by the police personnel are minimized.
The author has not come across any similar problem after extensive literature
survey. The only set of problems bearing similarity with the problem under discussion
 Author's personal copy
OPSEARCH

in this paper is the multi-dimensional assignment problem [5]. The multi-dimensional

assignment problem is an extension of the classical 2-dimensional or 2-index assign-
ment problem to tackle few problems in real life. An example of the classical 2-
dimensional assignment problem involves selecting p workers from a set of q
candidates to do p jobs at the least cost, where the cost of getting the job i done by
worker j is aij. Mathematically the 2-dimensional assignment problem can be stated as
an integer linear program [5], using an indicator binary decision variable xij:

Xp Xq
Minimize i1 j1 aij xij 1

subject to the constraints:

Xp
i1 xij 1; for all j 1; 2; :; q 2

X q
j1 xij 1; for all i 1; 2; :; p 3

The extension of the above 2-dimensional assignment problem to a 3-dimensional

assignment problems was introduced by Pierskalla [6]. An example of the 3-dimensional
assignment problem is the problem of scheduling one-to-one meetings between r vendors
with s customers over t time periods in a trade show(where r s t), such that: (a) each
customer meets one vendor and each vendor meets at most one customer during each time
period, and (b) the total cost of meetings is minimized, where aijk is the cost of
meetings (based on preferences of vendors and customers, with lower values
indicating greater desirability for meetings) of vendor i with customer j in time
period k given by the triplet (i,j,k) [7]. Mathematically the 3-dimensional assignment
problem can be similarly stated as an integer linear program, using an indicator binary
decision variable xijk:
Xr Xs X t
Minimize i1 j1 k1 xijk aijk 4

subject to the constraints:

Xr
i1 xijk 1; for all j 1; 2; :; s; k 1; 2; :; t 5

Xs
j1 xijk 1; for all i 1; 2; :; r; k 1; 2; :; t 6

Xt
k1 xijk 1; for all i 1; 2; :; r; j 1; 2; :; s 7
 Author's personal copy
OPSEARCH

Constraint (5) ensures that each customer meets exactly one vendor in each time period.
Constraint (6) allows each vendor to meet at most one customer in any time period.
Constraint (7) ensures that vendors and customers meet only once in all the time periods.
In contrast to 2-dimensional assignment problems, the 3-dimensional assignment problems
are demonstrated to be NP-hard requiring development of various heuristics for their
solution [8].
Other examples of multi-dimensional assignment problems discussed in literature are:
(a) Problem of assigning p jobs to q workers and r machines (p q r) so that all the jobs
are completed in the shortest possible time, where the time aijk required for worker j to
complete job i on machine k is given; a heuristic method is proposed for solving the
problem using a series of solutions to the 2-dimensional problem [5]. (b) Problem of
assigning students and teachers to classes and time slots [9]. (c) Problem of assigning
medical residents to rotations at a teaching hospital over a 12-month planning horizon, to
maximize the preferences of residents to specific rotations and the period of vacation; a
linear programming model is proposed with the objective function comprising of priority
weights associated with desirable assignments, followed by application of a rounding
heuristic to determine integer solutions [10] (d) Problem of scheduling of shifts for vehicle
drivers of transit systems such that transit system service requirements are met and all
drivers have an equitable share of undesirable or desirable routes; a two-phase method is
proposed, wherein feasible solutions are generated in the first phase followed by improve-
ment of the solution in the second phase [11] (e) Problem of simultaneously assigning
cadets at Japans National Defense Academy to science/engineering departments and to
branches of Japans armed forces while maximizing cadets preferences and considering
the constraints imposed by Academy class sizes and vacancies available in the armed
forces; greedy methods have been proposed for finding feasible solutions, followed by
local search heuristics to improve solutions [12].
The problem under discussion in this paper can be viewed as a multi-dimensional
assignment problem [5], wherein Central Police Force personnel of a particular base
(being the agents) are assigned to the tasks of manning a particular polling station of a
constituency of a particular State in a particular stage. Since the number of agents is
lower than number of tasks, there has to be a one-to-many assignment of agents to tasks.
This one-to-many assignment of agents is done by dividing the problem into stages and
having a one-to-one assignment of agents to certain tasks in each stage. For the first and
last stages, the cost of the assignment is the distance of the constituency from the police
base is considered. For intermediate stages, the cost of the assignment is the distance of
the constituency from a constituency of the earlier stage is considered.
The multi-dimensional assignment problem discussed in this paper is set apart from
the typical three-dimensional assignment problem by varying requirements of police
personnel in each State. The first implication of this difference is that there are various
feasible combinations of States where elections could be held together in a particular
stage. The second implication of this difference is that there are various feasible
combinations of stages. The third implication of this difference is that the cost of the
assignment cannot be obtained easily. Police personnel may travel from the police base
to constituencies directly or from earlier stages. Thus referring to Fig. 1, the cost of travel
to State S7 could be the combination of cost of travel from P2 to S3, P2 to S4, P2 to S7,
 Author's personal copy
OPSEARCH

S3 to S7 and S4 to S7. Further police personnel have to return back to their respective
police bases, the cost of which is impossible to incorporate in the assignment problem.
Thus the problem under discussion in this paper is an original contribution to the field of
operations research.
The solution methodology elaborated in Section 2.2 has been inspired by the
minimum cost multi-commodity network flow problem [13, 14]. The minimum cost
multi-commodity network flow problem involves a directed network with nodes i=
1,2,.,n and directed arcs, where each arc has a capacity bij and is associated with a cost
cij per unit flow, where commodities k=1,2,q with flow rk are required to flow from a
source node sk to the sink node tk. The decision variable is the flow yijk of commodity k
over the directed arc (i,j). The objective function is to minimize the sum of product yijkcij
over all arcs and all commodities, subject to the following two constraints:
q
k1yijk bij ; for all arcs 8
8
  < rk for i sk
nj1 yijk yjik rk for i t k 9
:
0; otherwise
The constraint (8) pertains to the arc capacity while the constraint (9) pertains to the
flow conservation at each node. In the problem under discussion in this paper, we need
to ensure that police men belonging to a particular police base should start from that
police base and return to the same police base after traversing different constituencies of
different States where elections are held in different stages. Thus flow conservation
constraints similar to the second constraint are built into the model (constraint(19) of
model described in Section 2.2) to ensure that police men of a particular base are not
mixed up with police men of another base at any stage of their traversal from base to
base. The first constraint, may arise if there are situations where there is a restriction on
the total number of police forces that can be moved from one State to another State due
to situations such as weather or road conditions; this aspect has not been incorporated in
the model discussed in Section 2.2, but has been discussed in Section 4.
This paper proposes and demonstrates two MIP models to (a) schedule the
elections with a minimum number of stages (b) sourcing the appropriate number of
security personnel from the most convenient Central Police Forces bases (c) sched-
uling elections for all constituencies in each State in a particular stage such that the
total movement of security personnel(in men-miles) is minimized. The paper is
organized as follows: the methodology is described in Section 2, followed by
demonstration of methodology in Section 3 and conclusions in Section 4.

2 Methodology

2.1 Determination of number of election stages

The number of stages over which the elections will be held throughout the country is
dependent on the number of polling stations in the States, number of police personnel
 Author's personal copy
OPSEARCH

required at each polling station and the total number of Central Police Forces
havailable. The minimum number of stages is given by the iceiling function
Total requirement of Central Police Forces at all polling stations of all 543 constituencies
Total number of Central Police Forces available , provided that
the total requirement of police personnel required in all polling stations of any
State is less than the total number of Central Police Forces available.
For example, if the polling is being conducted in 4 States A,B,C and D which have
requirement of 2,3,4 and 5 thousand police personnel respectively and 10 thousand
Central Police Forces are available, then the election will have to conducted in 2 stages.
However if the 4 States A,B,C and D have requirement of 2,9,4 and 5 thousand police
personnel, then the election will have to be conducted in 3 stages since only 10 thousand
Central Police Forces are available and elections for all constituencies of a State have to be
held together in the same stage. Thus the elections for State B could be held in the first
stage (since 9 thousand are required against availability of 10 thousand police personnel);
elections for States C and D could be held in the second stage (since 9 thousand are
required against availability of 10 thousand police personnel); election for remaining State
A can be held in the third stage.
The exact number of stages can be determined using the following integer program model:
Indices used:

m index used for election stages

r index for States of India
Notation for data elements:

pr number of police personnel required for all polling stations in State r

Q total number of Central Police Forces available
M set of election stages. The number of election stages is chosen to be slightly more than
the minimum number of stages obtained by the ceiling function described above.
L very large number
Decision Variables:

xmr =1, if election for State r is held in stage m

=0, if election for State r is not held in stage m
ym =1, if election for any State is held in stage m
=0, if election for none of the States is held in stage m
The objective function is given by:
Minimize mM ym 10
Subject to the constraints:
mM xmr 1; for all rR 11

rR xmr pr Q; for all mM 12

rR xmr Lym ; for all mM 13
 Author's personal copy
OPSEARCH

Explanation of the objective function: The objective function (10) seeks to min-
imize the number of stages over which the elections are held.
Explanation of the constraints: Constraint (11) ensures that elections in any State
are held only once in any of the stages. Constraint (12) ensures that the police
personnel required for conducting elections in any stage is less than or equal to the
total number of Central Police Forces available. Constraint (13) ensures that ym=1, if
elections are held in any one or more States in stage m; else ym=0.

2.2 Determining the election schedule and police movements

The Central Police Forces personnel movement can be categorized in either of the
following: (a) movement from police bases directly to the constituencies before any stage
of elections, or (b) movement from constituencies to the police bases after any stage of
elections, or (c) movement from constituencies after any intermediate stage of elections to
constituencies before subsequent stages of elections. Thus if elections are being held in 3
stages, examples of P1 base police forces movement in reference to Fig. 1 are: (a) P1 to S1
or P1 to S5 or P1 to S8 (b) S1 back to P1 or S5 back to P1 or S8 back to P1 (c) S1 to S6 or
S1 to S5 or S1 to S8.
The following MIP model is proposed for scheduling the elections and police
movement. It is based on the multi-commodity network flow model, along with binary
variables to model the holding of elections in a particular State in a particular stage.
Indices used:

m Index used for election stages

r Index for States of India
j Index for constituencies of a State
b Index for police bases
Notation for data elements:

Sr Set of constituencies of State r

R Set of States of India
B Set of police bases
M Set of election stages determined by the integer program model described
in Section 2.1; if the number of stages is 3, M = {1,2,3}
F Set of pairs of election stages, where each pair is a feasible 2-tuple of election
stages between which police personnel movement is allowed; if the number
of stages is 3, F = {(1,2),(1,3),(2,3)} where the 2-tuple (1,2) indicates that
movement of police personnel is allowed from stage 1 to stage 2.
crj Number of polling stations in constituency j of State r
prj Number of police personnel required for each polling station in
constituency j of State r
cb Number of police personnel available at base b
dbjr Distance from base b to constituency j in State r
d j1 r1 j2 r2 Distance from constituency j1 in State r1 to constituency j2 in State r2
L Very large number
 Author's personal copy
OPSEARCH

Decision Variables:

xmrj =1, if election for constituency j in State r is held in stage m

=0, if election for constituency j in State r is not held in stage m
xbjrm Number of police moving from police base b to constituency j in State
r for elections in stage m
xjrmb Number of police moving from constituency j in State r after elections
in stage m to police base b
x j1 r1 m1 j2 r2 m2 b Number of base b police personnel moving from constituency j1 in
State r1 after elections in stage m1 , to constituency j2 in State r2 for
elections in subsequent stage m2; for example if m1=1, m2 could be 2
or 3 for a 3-staged election.
Thus the objective function is given by:
Minimize
  
bB;mM ;rR; jS r xbjrm xjrmb d bjr
 
bB;m1 ;m2 F;r1 ;r2 R; j1 S r ; j2 S r ;r1 r2 x j1 r1 m1 j2 r2 m2 b d j1 r1 j2 r2
1 2

bB;m1 ;m2 F;r1 ;r2 R; j 14

1 S r1 ; j2 S r2 ;r 1 r 2 x j1 r1 m1 j2 r2 m2 b L
Subject to the constraints:
mM xmrj 1; for all rR; jS r 15

xmr j1 xmr j2 ; for all mM ; rR; j1 ; j2 S r 16

bB xbjrm bB;m1 M ;m1 ;m F;r1 R;r1 r; j1 S r1 x j1 r1 m1 jrmb prj crj xmrj ; for all mM ; rR; jS r
17
mM ;rR; jS r xbjrm cb ; for all bB 18

xbjrm m1 M ;m1 ;m F;r1 R;r1 r; j1 S r x j1 r1 m1 jrmb 19

1
xjrmb m1 M ;m;m1 F;r1 R;r1 r; j1 S r1 xjrm j1 r1 m1 b ; for all bB; mM ; rR; jS r
xbjrm Lxmrj ; for all bB; mM ; rR; jS r 20

xjrmb Lxmrj ; for all bB; mM ; rR; jS r 21

x j1 r1 m1 jrmb Lxmrj ; for all bB; m1 M ; m1 ; m F; r1 R; r1 r; j1 S r1 ; mM ; rR; jS r 22

Explanation of the objective function(14): The first term of the objective function
is the number of men-miles travelled by police personnel leaving the police bases
before election stages and entering the police bases after completion of election
stages. The second term is the men-miles travelled by police personnel travelling
 Author's personal copy
OPSEARCH

between constituencies after completion of an election stage and before commence-

ment of subsequent election stage. The third term penalizes instances of police
personnel staying back in a constituency after completion of the election stage.
Explanation of the constraints: Constraint (15) ensures that elections for any
constituency in any State are held in a particular stage. Constraint (16) ensures that
elections for all constituencies in any State are held in the same stage. Constraint (17)
ensures that the required number of police personnel is available at all polling stations
of all constituencies in the States where elections are being conducted for a particular
stage. Constraint (18) ensures that the total number of police personnel travelling to
constituencies is less than or equal to the available number of police personnel at that
particular base. Constraint (19) ensures that the number of police personnel of a
particular police base entering a particular constituency before an election stage
equals the number leaving that constituency after the election stage. Constraint (20)
ensures that police movement from police base for a particular election stage can only
occur to those constituencies where elections are being held in that particular stage.
Constraint (21) ensures that police movement back to the police base after completion
of a particular election stage, can only occur from those constituencies where
elections are being held in that particular stage. Constraint (22) ensures that police
movement from constituencies where elections have been held in earlier stages can
only occur to those constituencies where elections are being held in the subsequent
stage.
If the objective is to conduct the elections in the minimum number of stages as well as
to minimize the police travel movement, the number of stages m is chosen as exactly
equal to that obtained by the integer program model described in Section 2.1. However,
if the objective is to conduct the elections in a manner to minimize the police travel
movement only, the number of stages m may be chosen higher than that obtained by the
integer program model described in Section 2.1; the model ensures that a minimal subset
of the stages are actually utilized for conducting elections in all the States contained in R.
Since the number of stages chosen affects the model size and complexity, it is advisable
to choose the number of stages as slightly higher (say, about twice higher) than that
obtained by the integer program model described in Section 2.1.
There could be problem instances wherein a feasible solution with 4 stages could
have lower police movement (in terms of men-miles) than a feasible solution with 3
stages. If the decision maker is interested in carrying out the elections in the minimum
possible number of stages, a sub-optimal solution in terms of police movement
might be chosen. It is up to the decision maker to choose the appropriate m depending
on whether the decision makers objective is (i) to minimize police movement alone
or (ii) minimize both the number of stages and the police movement.

3 Demonstration of methodology

The number of binary variables required for the MIP model proposed above is given
by the product of the total number of constituencies (543) and total number of
election stages. Thus if elections are held in 3 stages, the number of binary variables
will be 1629.
 Author's personal copy
OPSEARCH

The Central Police Forces personnel movement is generally by air, except the last
mile movement by road/rail to and from the constituencies. The average distance
between constituencies, including air travel between nearest airports and last mile
road/rail movement, is 735 miles. The average last mile distance of constituencies to
the nearest airport is only 34 miles. Since the last mile distance is quite small
compared to the total inter-constituency distance, we can ignore the last mile
distance and reduce the problem to minimizing the men-miles travelled between
airports only. The airport servicing the constituencies of a particular State can thus
replace those constituencies; the airport will thus be servicing the total number of
polling stations of those constituencies. For example, in Table 1 the Agra airport
services 43 constituencies and the Gorakhpur airport services 37 constituencies of the
state of Uttar Pradesh; thus Uttar Pradesh can be considered as having two constit-
uencies with the number of polling stations indicated in the table. Thus the number
of binary variables reduces to 117 only, since the number of constituencies have
reduced from 543 to 39. This method of problem reduction has been used for
demonstrating the methodology.
The following data is used for demonstration of the proposed method:
a. Number of police personnel to be assigned to each polling station has been
assumed to be the same (4 numbers) for all polling stations.
b. Five Central Police Forces bases are assumed to be located next to nearest
airports at Mumbai, Kolkata, Delhi, Shillong, and Hyderabad. Each base is
assumed to have three hundred thousand policemen available for deployment.
c. Inter-airport distances calculated using geodetic coordinates of the airports. The
inter-airport distance (in miles) is calculated by the Haversine formula
3963.1cos1((sin11sin21)+(cos11cos21cos(1222))), where 11, 21 are the
latitudes of the airports (in radians) and 12, 22 are the longitudes of the airports
(in radians) [15].
Since the (a) available Central Police Forces is 1.5 million, (b) the total number of
polling stations is 833,701 and (c) 4 police personnel are assigned to a polling station,
the minimum number of election stages is 3 as given by the ceiling function in
Section 2.1. The exact number of stages also works out to 3 using the integer program
model described in Section 2.1.
The MIP model described in 2.2 gives a solution of 1.005 billion men-miles. The
model takes about 1.22 min for processing and solution using CPLEX 12.1.0 on a
1.3 GHz computer. The optimal schedule obtained is:the first stage elections are held
in the 13 States (194 constituencies, 317,319 polling stations) of Chandigarh,
Chattisgarh, Jharkhand, Karnataka, Lakshwadeep, Maharashtra, Meghalaya,
Mizoram, NCT-Delhi, Puducherry, Punjab, Rajasthan, West Bengal; followed by
second stage elections being held in the 15 States (162 constituencies, 230,670
polling stations) of Arunachal Pradesh, Bihar, Dadra & Nagar Haveli, Daman &
Diu, Goa, Haryana, Jammu & Kashmir, Madhya Pradesh, Manipur, Nagaland,
Orissa, Sikkim, Tamil Nadu, Tripura and Uttarakhand; followed by third stage elections
being held in the 7 States (187 constituencies, 285,712 polling stations) of Andaman &
Nicobar Islands, Andhra Pradesh, Assam, Gujarat, Himachal Pradesh, Kerala and Uttar
Pradesh. The optimal movement of Shillong base Central Police Force personnel is
 Author's personal copy
OPSEARCH

Table 1 Number of constituencies and polling stations in each State

Sl. State-Serving Airport(s) No of No of Sl. State-Serving Airport(s) No of No of

Members polling Members polling
of the stations of the stations
Lok Lok
Sabha Sabha

1 Andhra Pradesh- 22 35702 2 Arunachal Pradesh-Zero 2 2057

Hyderabad
Andhra Pradesh- 20 31058
Rajahmundry
3 Assam-Jorhat 14 18828 4 Bihar-Patna 40 57020
5 Goa-Dabolim Goa 2 1339 6 Gujarat-Rajkot 8 12912
Gujarat-Vadodara 18 29656
7 Haryana-Hissar 10 12894 8 Himachal Pradesh-Kulu 4 7253
9 Jammu & 6 9129 10 Karnataka-Hubli 28 46576
Kashmir-Srinagar
11 Kerala-Kozhikode 20 20510 12 Madhya Pradesh-Bhopal 29 47812
13 Maharashtra-Akola 14 25541 14 Manipur-Imphal 2 2193
Maharashtra-Mumbai 34 57057
15 Meghalaya-Shillong 2 2117 16 Mizoram-Aizawl 1 1028
17 Nagaland-Dimapur 1 1692 18 Orissa-Jeypore 21 31617
19 Punjab-Ludhiana 13 18846 20 Rajasthan-Jaipur 25 42699
21 Sikkim-Darjeeling 1 493 22 Tamil Nadu-Madurai 39 52158
23 Tripura-Agartala 2 3008 24 Uttar Pradesh-Agra 43 69654
Uttar Pradesh-Gorakhpur 37 59792
25 West Bengal-Malda 42 66109 26 Chattisgarh-Raipur 11 20984
27 Jharkhand-Ranchi 14 23696 28 Uttarakhand-Dehradun 5 9003
29 Andaman & Nicobar 1 347 30 Chandigarh- Chandigarh 1 422
Islands
31 Dadra & Nagar 1 161 32 Daman & Diu-Daman 1 94
Haveli-Daman
33 NCT of Delhi-Delhi 7 11348 34 Lakshadweep-Agatti 1 40
35 Puducherry-Pondicherry 1 856

given in Fig. 2 as an illustration. The optimal movement for all personnel, States and
stages is given in Electronic supplementary material ESM1.

4 Conclusion

The method described in Section 2 and demonstrated in Section 3, enables (a) scheduling
of elections in all States within the minimum number of stages (b) sequencing the stages,
such that the movement of Central Police Forces (measured in men-miles) is minimized
and (c) sourcing the appropriate number personnel from the most convenient bases. This
 Author's personal copy
OPSEARCH

Inter-Airport Distances (in miles)

Shillong-Aizawl Shillong-Malda Shillong-Dimapur

137 238 116

Shillong-Jorhat Aizawl-Imphal Malda-Zero

165 108 394

Malda-Imphal Malda-Darjeeling Malda-Agartala

365 139 215

Malda-Jorhat Dimapur-Jorhat Zero-Jorhat

396 66 62

Darjeeling-Jorhat Agartala-Jorhat Imphal-Jorhat

368 273 136

Fig. 2 Movement of Shillong based Central Police Force personnel

paper has two original contributions: the first being the description of a problem which
has not been discussed so far in operations research literature, and secondly proposing a
solution methodology utilizing a combination of set covering and multi-commodity
network flow techniques. This method can also be utilized for scheduling and planning
of any nation-wide event requiring scarce resources; for example, a road show for
marketing of a new product which involves demonstration of the product working to
prospective clients and only a few prototypes of the product are available for the road
show.
The model described in Section 2 can be used for various extensions of the problem
such as :(a) Differing requirements of police personnel for different States/constituencies
depending on threat perceptions. This will be taken care of suitable data for prj (b)
Imposing conditions on police personnel that they cannot be deployed in certain States;
for example Shillong based police personnel cannot be deployed in States of Meghalaya
and Assam. This can be modeled by additional constraints: xbjrm=0,for r=1,2,b=1; xjrmb=0,
 Author's personal copy
OPSEARCH

for r=1,2,b=1; x j1 r1 m1 jrmb 0; for r 1; 2; b 1 , where r=1,2 pertain to States of

Meghalaya and Assam and b=1 pertains to Shillong police base. (c) Imposing condition
for simultaneous elections in certain States in a particular stage; for example elections in
States of Puducherry, Kerala and Tamil Nadu must be held simultaneously. This can be
modeled by additional constraints: xmr1 j xmr2 j ; xmr1 j xmr3 j ; where r1, r2 and r3
pertain to the States of Puducherry, Kerala and Tamil Nadu. (d) Imposing capacity
constraints on arcs connecting particular States; for example capacity constraint could
be imposed for the route connecting the States of Himachal Pradesh and Jammu-Kashmir
due to snow fall on the roads. This can be modeled by an additional constraint:
m1 M ;m1 ;m F; j1 S r ; jS r x j1 r1 m1 jrmb D , where where r1 and r pertain to the States of
1
Himachal Pradesh and Jammu-Kashmir and D is the upper bound of police movement
between the two States.
The paper discussed above has the objective of minimizing men-miles. However
in practice, police personnel will be transported in vehicles of different types having
varying capacities. These vehicles are either owned or outsourced by the police bases.
Each type of vehicle will have a fixed and variable cost (associated with distance
covered) associated with it. Further research could be carried out to obtain a schedule
of vehicle deployment so that police personnel are transported according to require-
ments of different constituencies at the least cost.
Each State may have a preference ranking for police personnel of a particular base,
depending on the familiarity of the police personnel with the terrain, customs and
language. Further research can be carried out on determining the deployment of
police personnel of different police bases to maximize the preferences as well as
minimize the total cost of police deployment.

Acknowledgement The author is most grateful to the referees who suggested changes that have
improved the paper.

References

1. Scharff M. Policing Election Day: Vulnerability Mapping in India 20062009. Innovations for successful
societies. 2011. Available at: http://www.princeton.edu/successfulsocieties/content/data/policy_note/PN_
id173/Policy_Note_ID173.txt. Accessed April 6, 2012.
2. National Crime Records Bureau. Crime in India 2009 Statistics. New Delhi 2010.
3. Stefan Harrendorf MHM). International statistics on crime and justice. Helsinki 2010.
4. Election Commission of India. Lok sabha election 2009-reinforcing indian democracy. New Delhi:
Election Commission of India; 2009.
5. Gilbert, K.C., Hofstra, R.B.: Multidimensional assignment problems. Decision Sciences 19(2), 306
321 (1988)
6. Pierskalla, W.P.: The multidimensional assignment problem. Operations Research 16, 422431 (1968)
7. Gilbert KC, Hofstra RB, Bisgrove R. An algorithm for a class of three-dimensional assignment
problems arising in scheduling operations. Inst Ind Eng Trans. 1987:2933.
8. Burkard, R.E.: Selected topics on assignment problems. Discrete Applied Mathematics. 123, 257302
(2002)
9. Pentico, D.W.: Assignment problems: a golden anniversary survey. European Journal of Operational
Research. 176, 774793 (2007)
10. Franz, L.S., Miller, J.L.: Scheduling medical residents to rotations: solving the large scale multiperiod
staff assignment problem. Operations Research 41(2), 269279 (1993)
 Author's personal copy
OPSEARCH

11. Carraresi, P., Gallo, G.: A multi-level bottleneck assignment approach to the bus drivers rostering
problem. European Journal of Operational Research 16(2), 163173 (1984)
12. Yamada, T., Nasu, Y.: Heuristic and exact algorithms for the simultaneous assignment problem.
European Journal of Operational Research 123(3), 531542 (2000)
13. Tomlin, J.A.: Minimum cost network flows. Operations Research 14(1), 4551 (1966)
14. Kennington, J.L.: A survey of linear cost multicommodity network flows. Operations Research 26(2),
209236 (1978)
15. Longitude and Latitude Calculator. Available at: http://www.ig.utexas.edu/outreach/googleearth/
latlong.html. Accessed April 30, 2013.