Lecture 57 - Coverage (Data Envelopment Analysis (DEA))

IIT Kanpur July 201
20 followers
Jun 11, 2024
[Music]
welcome back my dear friends and
students a very good morning good
afternoon good evening to all of you
whichever time you are listening and
hearing this
lecture the course title is multi
decision making which is another npt M
series and my good name is raghunandan s
Gupta from the ime department at I
khur and as you know this uh course is
spread over 12 weeks which is 60
lectures each uh week we have five
lectures each lecture being for half an
hour and we are in the last week which
is the 12th week in the last lecture
which we started for the 12th week we
started the discussion about the data
envelopment analysis and rather
than solve problems I did discuss in
details about the basic concepts of
Da so this is the 57th lecture and the
idea as you know for this whole course
is multicriteria decision making we have
covered multiobjective decision making
in details multi attri rute
Theory as I mentioned the rules then on
the multiattribute decision making we
have covered Electra Epsilon Electra top
say V core HP and we are now into da we
did spend for Electra Epsilon
Electra V core topis HP a lot of time
and the da part is the last part which
is left for the multi attribute or
different type of non-parametric
methods the coverage is continuation of
data envelopment
analysis so if you remember we were
discussing
about decreasing return to scale the
concept of input output as bundles along
the xaxis y axis or y- axis x-axis
depending on the way you want to
depict and the examples when it will be
increasing return to
scale decreasing return to scale or
constant return to
scale we will be discussing once more
and the point of of decreasing return to
scale I did mention and the same example
can be brought forward and discussed for
the increasing return to scale and the
constant return to scale also the
example was basically of production the
quantum and the revenues depending on
the input quantity of input and I
consider the quantity of input as the
number of
labors so here the increasing return to
scale the diagram which we have already
considered in the last class the last
slide in the last class was where input
was measured along
the xaxis output along the y- AIS here
we have just
reversed the diagram considering input
and output are being considered along
the Y AIS and x-axis respectively here
also the dmu which is inefficient can
reach efficiency which is in the
frontier either moving vertically up
going horizontally to the left or moving
in the radial Direction and do remember
the green one which is the radial
Direction in all the
diagrams touches or meets the curve in
such a way that at that point the angle
is
90° the angle I mean this
one even though in this diagram it may
not be apparent it is 90° but it should
be
90° and as I
mentioned that the DME inefficient can
become e efficient when it reaches the
frontier based on the concept of input
oriented model output oriented model and
this what we'll highlight in the further
slides and the discussion for this
lecture in the constant return to scale
as the diagram shows which is very
apparent again we are measuring input
along the x-axis output along the Y AIS
the dmu which is inefficient which I'm
just
marking
using the red color it can reach
efficiency either going vertically up
moving horizontally to the left or going
in the radial Direction radial W I'm
using in order to
signify the way it's moving to reach the
efficient
Frontier and do remember the angle which
I'm now drawing in the red color is 90°
even though it may have not been drawn
accordingly but that will be
90° again the concept of dmu1 efficient
in blue color it is output
oriented why output oriented again I'm
mentioning because the inputs are fixed
output is increasing
the dmu1 efficient input oriented which
is pink in color why it is input
oriented because here the output is
fixed input is
decreasing input won't increase it will
decrease remember that because you want
to basically increase the efficiency and
the variable to scale can be a
combination of the input
output uh idea or more uh to discuss in
details would be try to combine the
increasing return to to scale and
decreased return to scale the ideas of
increasing return to scale and
decreasing return to scale I've
discussed in quite details in the 56th
lecture and also in the 57th lecture
which is the one which is going on so I
won't go into much detailed discussion
about the graph here so input again here
even though input is along the y- axis
output is along the x-axis but if you
have the input along the x-axis and
output along the y axis you can
similarly draw conclusion with the
graphs where it is variable return to
scale considering a combination of
increasing and decreasing return to
scale now comes the
formulation and do
remember I'll very briefly tell about
the formulation and move logically in
this direction now if you remember in
the output oriented model and the input
oriented model I always highlighted in
which direction it was moving it was
either moving vertically up or hor
left when it was moving vertically up
and if the input was being measured
along the xais then input was
fixed and if it was moving horizontally
to the left and again if it input was
being measured along the x-axis their
outputs were fixed so based on that
we'll try to analyze how the models can
be made and how the analysis can be
drawn in the output or oriented model
the model formulation like this I'll
first mention the model and as I'm going
along mentioning the model I'll tell you
the reason why it is being formulated
like
that now whenever uh you want to
understand a dmu which or a set of dmus
you will try to compare the
efficiency and if I use the word
efficiency the word which will
immediately come in your mind is that
how can we measure efficiency efficiency
can be measured
by taking the ratio of the total amount
of output divided by the total amount of
input so if it is amount of money which
you're investing and the output which
you're getting is Goods which you want
to sell obviously it will be the cost of
the goods cost means the selling price
of the goods which is selling the market
divided by the input cost of the goods
and that will give you the the rate of
the price increase which you have may
have basically
gained now when I consider the dmu it's
basically just a system a machine so in
that machine if I consider efficiency
again the efficiency would be considered
in a very simple way like a ratio of the
bundle of output divided by the bundle
of input divided that means input is in
the denominator and output is in the
numerator now let us see when it is an
output or model how we are trying to
basically analyze
now once again let us go back to the
diagram because again I'll come back to
the slide number
seven now if I consider the output
oriented model for the continuous return
to scale and Mark it once with the
different color say example green this
is the output oriented model as as it is
written in the suffix op output oriented
model so in the output ored model what
is fixed input is
fixed if I go back to this uh this
diagram which is increasing return to
scale but here input is along the y-
axis and output is along the xaxis the
output oriented model again I'm
highlighting is where the input is
basically fixed and output is changing
that means in the output oriented model
all your focus is on the output keeping
the input fixed and similarly for the
input oriented model it will be just the
reverse we'll revisit that idea later on
so in the output oriented model coming
back to the slide from where we left if
you see the ratio forget about the max
max Part Max is basically
maximization if you pay attention to the
ratio in the numerator what you have is
the summation of all the outputs which
are there and in the denominator you
have the summation of all the inputs
which is basically the the
efficiency so for each dmu you want to
maximize the efficiency that's why the
optimization model is basically
maximizing the ratio of the output
bundle divided by the input bundle now
in this case what would be the different
type of constraints and the constraints
are also very logical so when you are
trying to basically maximize the
efficiency of 1 dmu obviously you would
always assume and rightly so that the
efficiency of none of the other dmus
including
itself would ever exceed one which means
the ratio of the output the input for
all the other dmus including that one
has to be less than equal to one and
this is what is
given now this is the first constraint I
am not com to the fact what are the
suffix still now similarly you have the
K one for the K
dmu now here capital K is taken so you
are considering the efficiency of
the first K is equal to 1 then K is
equal 2 then it goes down to K isal to
capital K so what what is Cap small K is
equal to 1 is this
one
what is small k equal to capital K is
this one and in between you have small k
equal to 2 2 small K is equal to capital
K minus 1 which I have not
written and where are they being
highlighted based on the fact that small
K is equal to 1 till small K is equal to
capital K if you see the numerator and
the denominator in the such this
constraints you have VJ 1 into y J1
divided by U i1 into X i1 where v and y
are related to the output and U and X
are related to the input that's why it's
a ratio of the output bundle to the
input bundle and basically you sum them
up for all the amount of outputs to
inputs and what is the number of outputs
you have Capital any in number what is
the amount total number of inputs which
is capital M in in number and if you see
the summation in numerator is from Jal
to 1 to n and summation in the
denominator is is basically from IAL to
1 to n and here the second suffix after
V after y after in after U and after X
all are one that means for small K is
equal to 1 similarly if I go to the last
constraint again the summation in the
numerator and the denominator are
respectively from G is equal to 1 to n
in the numerator and I is equal to 1 to
m in the
denominator and in the bundle of outputs
you have the variables and the scaling
Factor the variables is y scaling factor
is V in the numerator and in the
denominator the variable is X and the
scaling factor is U and the suffix if
you see is J
capital K then also J capital K and
similarly in the denominator I capital k
i capital K so these are all the
constraints so there are a capital K
number of constraints and what do we
have in the objective function objective
function is basically the ratio of the
output to the input which basically
gives you the efficiency of the kth dmu
now in that case it means the optimation
problem are capital K number each for
one of the
dmus so as it is mentioned
I is = to 1 to m j is equal to 1 to n k
is equal to 1 to capital
K now immediately the problem
formulation would make you pause and
think so I've been mentioning is an easy
optimation problem but this is as you
see in the problem formulation this is
not linear which is true so I want to
basically bring linearity into the
optimization problem and similarly
ensure linearity in the constraint also
so can it be done the answer is yes it
can be done how it can be done if you
remember I did
mention that
the efficiency of any dmu including the
K one small K one can never exceed One
Max it can be one
which means if you pay attention to the
objective function maximization can
occur by ensuring that I keep the value
in the denominator which is input fixed
and basically keep increasing the
numerator with the maximum possible
value and this is what we are going to
do so if you see the output oriented
model for a case where your
outputs for dmu as K is equal to 1 2
3 which I'm marking in blue so capital K
is
3 J which is the number of uh outputs is
four which I'm again marking in Blue
double tick and the input which is
denoted by the nomenclature I is I is
going to 1 to5 which I'm marking in
triple tick so if I consider the pro
problem formulation for the first D dmu
and here I'm marking the suffix in red
color in order to differentiate that you
want to maximize the bundle of output to
input for the first dmu based on the on
the idea that the efficiency for the
first dmu which I will now use a
different color first dmu is less than
equal to 1 Max can be
one similarly the idea of the second dmu
also being less than equal to one is
which I'm marking in blue one blue that
means in the constraint and the idea the
constraint for the third dmu is also
less than equal to 1 is marked in Violet
similarly the idea can be replicated for
dmu2 dmu3 and in that case the problem
formulation would only remain the same
only the the
suffix of the coloring scheme which I
put as one would change corresponding to
two for the second dmu to three for the
third
dmu and this is what is happening if I
consider the output oriented model you
have again same thing IAL
to5 Jal 1 to 4 K is equal 1: 3 the
optimization model for the second dmu
just notice the suffix two which is
marked in red which is for the second
dmu similarly the first tick mark which
is for the constraint is for the first
dmu the second tick mark which I'm using
the green color is specifically for the
second
dmu and the third tick mark which I'm
using the blue one is for the third dmu
similarly for the third dmu problem
formulation you have again the same idea
I = 1: 5 J is = 1: 4 K is = 1 to 3 the
objective function is for the Third dmu
and the constraints are exactly the same
only pay attention that the number three
which is k equal to 3 is being
highlighted to make the things much more
apparent now when I come to the output
oriented optimization model now all
these
three optimization model for DM 1 2 3
were nonlinear now I want to make them
linear and if you remember what I did
mention few minutes back I would try to
ensure the denominator is kept fixed and
we try to maximize the numerator that
means output and that's what I meant
that the input would be kept fixed
output would be increased to the maximum
level so if I remove the denominator but
removing it doesn't mean that I'm
totally removing it from the system as
such it is not being removed it is being
transferred here I'll use a color red
the the denominator was
there and for the bundle of inputs as it
is an input uh fixed model it is being
transferred
here see here the summation of the
bundle of inputs is being fixed as one
such that the maximation problem is now
simple linear problem maximization on
the bundle of outputs and similarly the
constraints have also been very nicely
changed rather being nonlinear they have
been convert into linear part how if you
notice one of the
constraints which was I'll write it
down in black color so it was bundle
of I'm not using the suffix anymore
because that would be apparent for each
and every problem formulation which I do
it it is
VY and summation of
ux that was less than equal to one
remember so in that case what we do is
that it is summation of
Vy summation of ux is taken on to the
right hand side again brought to the
left hand side it becomes minus
summation of ux is less than equal to Z
so here you see the nonlinear
constraints each of them can be
converted into linear constraints and
that what we have done if I consider the
first constraint the first constraint
which I highlight using the green color
is
this this is related to the first dmu
similarly there would be for our second
dmu third dmu for dmu so this one is for
k equal to small K is = to 1 and the one
which I'm now marking in
Violet is for small K is equal to
capital
K and the one which I marked in red is
the extra constraint which will being
brought from the objective function onto
the constraints in order to ensure that
it's an output oriented
model so now you have a very simple
India
programming maximization needs to be
solved for each dmu and the constraints
have just increased by one for each of
the separate optimization problem but
remember that considering there were
capital K number of dmus initially all
the optimization problem had capital K
number of constraints the extra one
which is capital k + 1th one would keep
changing depending depending on what dmu
or which number of dmu optimation
problem you
solving so if it is the first one the
first K number capital K number of
constants are same the capital k + 1 one
would be the denominator part
corresponding to the first dmu which is
being transformed and being made into a
linear
programming for the second optimization
problem which is related to
dmu2 the constraint capital K would all
remain the same
the capital k + 1 one would basically be
the denominator which is coming from the
optimation problem related to the dmu
number two and we'll continue basically
changing the k + one constraint
accordingly based on the fact that what
is the dmu we are concentrating on in
order to solve the optimization
problem so once we do that without going
to the details because I have already
explained and the slides would be there
with you
so you'll ma want to
maximize a linear programming and
remember the same problem which we had
considered I is equal to 1: 5 Jal 1 to 4
K isal 1:
3 and here the corresponding parts are
related to the first second and
third similarly when I have for for the
second
dmu it is for IAL 1 125 J is = 1 to 4 K
= 1 to 3 this is the maximization
corresponding to
dmu2 as you can find out from the suffix
which is red two given Red two means the
dm2 and the constraints are as
usual as was the case for
the first optimization problem first
linear optimization problem but only the
highlighting factor is the second
dmu respective
the denominator which is being
transformed to the object uh constraint
part finally for the third dmu again I
is equal to 1: 5 J is equal to 1 to 4 K
is equal to 1 to 3 you maximize
the output bundle for the third dmu and
the constraints are based on the first
second and the third part has been
highlighted using the red three so it
would basically highight give you the
idea that dmu is being
optimized now coming to the input
oriented model the idea on the input
oriented model if you have understood
the idea for the output oriented model
is just the
reverse now in the in output oriented
model the idea was keep the out input
fixed concentrate only on the output
increasing it obviously you would not
decrease the output you will increase
the output in the input oriented model
the idea is keep the out output fixed
and keep decreasing the
input now this is what we are doing so
if you want to basically decrease the
input what will happen is that rather
than maximization you will want to
basically minimize minimize the inverse
of the efficiency now what is what was
the efficiency in the first case when it
was an output oriented model there you
found out the ratio of the bundle of
outputs divided the bundle of inputs now
in this case rather than maximizing you
will basically try to minimize the
inverse that
means
minimize the ratio of the bundle of
inputs to bundle of
outputs in the similar way in the
maximization problem what we did was
basically maximize the numerator keep
the numerator fixed at
one in this case input oriented model
you will minimize the numerator keeping
the denominator fixed because this is
now an input oriented model you want to
decrease the input and that's what the
formulation is you want to basically
minimize as I said and here the ratio is
basically bundle of inputs because you
see U and X in the numerator v and y in
the denominator the summation of U and X
is basically from I equal 1 to M
summation of v and y is basically from J
is equal to 1 to n and small K is
basically the k number of dmu now the
constraints were in the initial case of
the output oriented model were
efficiency cannot be more than one they
were basically less than equal to one
now if you are taking the inverse so
obviously the inverse of the efficiency
would be greater than equal to one so if
you see it here the bundle of inputs
divided by bundle of outputs for each
and every dmu has been written such that
they are all greater than one initially
it was what if I write it in the output
oriented model it was I'm not using the
suffix I'm only using V and
and
Y and U and
X so for the output oriented model they
were all less than equal to 1 now as I
find out the
inverse so obviously it has to be less
than greater than equal to one because
just do a multiplication can find it out
so also we would now have similar
problem K number capital K number of Sol
dmu problem solution minimization type
all the constraints are basically of
greater than equal to 1 and here also I
equal to 1 to m j is = to 1 to n k is =
to 1 to
K and here again for the same problem
formulation where I is equal to 1: 5 J
is = 1: 4 K is = to 1 to 3 the problem
formulations are given only pay
attention when you get the slides is for
which dmu the problem formulation is
being done here it is basically being
done for DM to one that's why the suffix
one has been numbered accordingly to
highlight the importance of which dmu
optimization is being done it is being
highlighted by the red color and here
also remember the constraints are all
greater than equal to one similarly in
slide number 17 which you see now it is
basically minimization or the bundle of
inputs divided by the bundle of outputs
because obviously it has to be
minimization and it is for the dmu
number two the constraints are given as
usual and they are same
for DM 1 2 3 being greater than equal to
1 for each of the case and here input
numbers is five output number is four
number of dmus is three finally for the
third dmu input oriented model you have
input uh numbers are five output numbers
Four dmu 3 you try to minimize the
objective function which is the ratio of
input bundle by the output bundle for
the third dmu and the constants are
exactly the same now all this all these
three problem formulations for dmu 1 2
three are nonlinear in nature we want to
convert into linear in nature what we do
is that fix fix the denominator now the
denominator is not related to input it
is basically related to the output
because it's an input oriented model we
want to minimize minimize the input
keeping the output at the fixed level so
output is fixed so you want to minimize
and the output factor which was there in
the denominator is now no more in the
objective function it has been brought
into the constant as you can see bundle
of outputs is equal to one so we ensured
for the K dmu which you are trying to
optimize it's equal to one and it's a
minimization problem not the
maximization as in the output oriented
model and here if you remember in the
output oriented model all the constants
were of what type it was basically
summation of
Vy summation of u x were less than equal
to 1 now they are the reverse so
obviously in this case they are now I'll
use a different
color they are summation of ux which is
the bundle of input divid by VY which is
bundle of outputs is greater than equal
to 1 so when I basically convert into
linear format they are all greater than
equal to zero so now you have all linear
constraints and 1 equal to 1 the the one
which is coming from the denominator of
the minimization objective function and
the objective function is is basically
simple a minimization problem so when I
utilize that for DME 1 2 and 3 the
problem formulations are are like this I
would not repeat it because once you go
to the slides it will become evident
only remember I 125 J 124 dmu 123 and
here the highlighted color is basically
related to the dmu so this is for first
dmu this is for second
dmu and this is for the third dmu so
with this I will end the end the the
second lecture for the last week which
is related to the DEA and before I end
the lecture
you all of you may be thinking that I've
been talking about the formulation what
is the solution solution we have for if
you remember when I started the da class
I did mention that ideas on optimization
of simple optimization problem of linear
programming which are a prerequisite
would all have been solved by you so
just take one of the problem solution or
Simplex method and try to utilize that
to get the variables and solve the
problems accordingly obviously there's a
huge branch of study in da but
considering the course is basically
multicriteria decision making covering
all the things may not be possible but I
thought I'll just give you the ideas of
Da in brief accordingly thank you very
much and have a nice day
[Music]
[Music]
[Music]
