Lecture 56 - Coverage (Data Envelopment Analysis (DEA))

IIT Kanpur July 201
20 followers
Jun 11, 2024
[Music]
welcome back my dear friends and
participants uh for this multicriteria
decision making course under NP Muk
series and as you know uh this course is
spread over 12 weeks with 60
lectures and each week we have five
lectures each lecture being for half an
hour and after each week we have one set
of assignments of 10 questions and after
that total course is is over we'll have
a final
examination once again a very good
morning good afternoon good evening to
all the set of people who are watching
these
lectures my good name is raghunandan s
Gupta from the IM department at I
put and if you remember in the last few
weeks we have been going through the
non-parametric methods and we had
covered a lot and
as the methologies of trying to solve
the problem were almost similar so I
just highlighted the main context main
idea and the different methodologies we
have considered was the Electra method
epon Electra topsis
VOR and the last part we discuss the
HP so HP is basically the analytical
hierarchy
process one another method we'll discuss
in the first lecture of the 12th week
which is the 56th lecture is the idea of
data envelopment analysis and this is
very heavily used in
economics in econometrics
in production economics where you want
to basically rank different decision
making units or
organizations now all the other
methods of Electra Epsilon Electra
topsis F or
HP we have discussed is basically the it
it was leave aside ahp the other four
methods we discussed was basically the
closeness to the best solution furthest
it is from the worst
solution we use the concept of
concordance matric discordance Matrix in
some case we use the indifference
Concept in epsilon
elector we've used the concept of AI pis
NIS and then we found out the ranking
method for all the Alternatives based on
the
criterias first we compared them uh
taking one at a time and then
collectively in HP it was a little bit
different where obviously the comparison
was between Alternatives based on the
criteria so for the example where we
considered uh two students RAM and Sham
were selected for the MBA programs and
they are different set
of criteria to select the
Institute or else consider the example
where you are buying a
car so in all these methods including
ahp there was a combination of
quantitative as well as
qualitative different criteria in order
to judge how you will
rank da is also in the similar line but
it is more from the point of view of
operation
research and I'll come to that later on
because operation research and the
simple concept of linear programming and
all these things is considered to be a
part and parcel as a prerequisite for
this course so I'll just give you the
highlighting points of these
methodologies and then you can solve the
problem accordingly I'll only
concentrate on the formulation other
things are very
straightforward so this is the 56th
lecture out of the
60 uh set of lectures uh the idea is
multicriteria decision
making we have already covered
multiobjective decision making
multiattribute utility Theory now we are
considering multiattribute decision
making even though the methodology of Da
you may find out some mathematical
formulation but still I'm considering it
as an part of non-parametric methods and
why I'll just highlight few of the
points as we
proceed so the coverage is data
envelopment analysis and we'll consider
it accordingly now where is data
envelopment analysis used so for the
examples we considered if you remember I
was always trying to harp on three
examples trying to buy a car where there
can be different criteria some which are
quantitative some which are
qualitative which are some are objective
some are subjective trying to buy an
apartment in a particular City with
different sets
of criteria some may be price some may
be locality some may be closeness to the
school or how safe it
is and another case was trying to choose
the Institute where you want to go and
study in the similar line you can use da
but da considered that I want to analyze
and rank or I want to find out ranking
methodology or
ranking of different type of
Alternatives here the alternatives are
not mentioned or named as Alternatives
in D method they are known as the
decision making units or dmu as written
in the slide so what are the decision-
making units
consider that the same example we want
to choose the best college best
university so when you make a list of
the different typ of universities be it
engineering College list from where you
want to take or the management school or
social sciences you have different sets
of priorities set somebody wants to
basically go to the best college where
the quality of teaching is excellent
somebody wants to look at the job
opportunity somebody wants to have a
look at what are the internship
opportunities and so on and so
forth or else consider a totally a
different example
that you want to basically analyze how
are the different type
of companies under the same sector are
doing consider the car industry so there
are many different car Industries
companies in India it can be Hyundai it
can be Honda it can be Volkswagen it can
be Tata it can be Mahindra so you want
to basic analyze how does these set of
companies compare amongst themselves
considering there are different sets of
inputs and sets of outputs now the word
of sets of inputs and outputs you are
hearing for the first time in context of
Da and obviously we have not encountered
these two words in other non-parametric
methods so we consider a dmu in a very
simple sense is a blackbox or a unit
where different inputs are utilized in
in different quantums in different
units and the units of input can be
electricity can be Manpower can be water
can be raw materials anything and
outputs can be different type of outputs
considering the case of car so it can be
what type of car the companies are
making is it a passenger car of certain
Cc or is it a heavy vehicle or SUV and
so on and so
forth now if I consider the case
of the choosing the educational
institute consider for our example
education institutes are all the iits so
in the I obviously the input would be
the students input would be the
infrastructure input would be the
faculty members input would maybe the
location and what would be the output
output can be the the type of jobs with
the students get type of research output
on the front of of paper books different
type of projects which are which are um
sold in the
market so based on the set of inputs and
set of outputs you want to basically
rank this decision making units so
basically this is the background based
on which we are going to study the da
method so let me continue the discussion
further on by reading the slides
consider you have a system or soall
decision making unit or units which
takes some inputs in different Quantum
different units process them and
produces some
outputs now when I'm using the word
processing processing can be different
it can be a chemical process it can be a
mechanical process for an for an
organization which is uh producing some
goods and for um the educational
institute you are basically processing
the students to be well acquainted with
the job market well acquainted to for
the academic
world now in another important point is
if you remember in I did mention and I
did mention it few times when we were
solving the problem that if you remember
the concept of discordance set or Matrix
if you remember the concept of how far
it is from the most positive Point how
far it is or how close it is from the
most negative point so those theoretical
points so we are considering the
negative
value of for taking a decision for
choosing an
alternative so in this case when you're
considering a dmu the question may crop
up is that can we have some outputs
which are negative in nature yes we can
have say for example when producing um
some steel making some steel or making
some cards the pollution can be
considered as an output which has
negative impact so in the same way as we
are analyzed you can consider the
negative out um discordance concept or
the how close it is to the negative
theoretical point we can also analyze
the concept of negative outputs
considering the um da methodology
examples of systems or dmus can be a
machine a factory an Hospital an airport
and a
university one example of the University
I have just mentioned and
factories for the example consider that
I'm considering all the different type
of uh small car manufacturing companies
which I mentioned tatas Mahindra Hundai
Honda then Volkswagen and all this list
invariably there may be many such
similar systems of dmus which are taking
these
inputs in in some
combinations or sets of inputs in some
combinations or in different quantums
processes them and produces the outputs
also in different
quantums
example consider two different the old
iits which was um built in India just
after Independence and consider the new
I maybe the amount of land the old IES
have with respect to the new itties may
be different maybe the total
infrastructure may be different so we'll
consider the outputs and
inputs here its inputs and outputs would
also change accordingly so is basically
what the situation
is another point is that when you're
considering
uh the factory example it may be
possible a certain CC of car is not
being manufactured by tataa Motors so in
that case the output for for that
company or for that dmu which is tataa
Motors will be consider nail but still
we'll try to basically fathom it how it
can be considered in trying to model the
da uh optimization model I'm using the
word optimization for the first time and
as I did mention it is a different
version of optimization and you will see
it's a very simple in nature in such
cases we are required to find out what
is the rate of processing of these
inputs or outputs or what is the
efficiency of the system so that we can
compare them amongst thems and also find
out to what degree they are deficient
when compared to so-called ideal system
or a full efficient system or an
efficient dmu
maybe we are comparing two companies and
the first company is operating at 100%
efficiency so obviously you'll try to
find out that what is the efficiency of
Company 2 and how it can be improved now
here is one important point to mention
and we will revisit that later
on
consider uh I'll give you two examples
consider that you as a teacher you you
are trying to basically give scores for
one examination which you have conducted
and the grading system can be based on
two factors one is relative grading one
is absolute
grading in the case of relative grading
and absolute grading what you do is that
in the absolute grading you consider the
maximum marks as it is and then
basically give the grades or the marks
of the other St system exactly as the
the student he or she would get in the
relative concept the the marking of the
highest student is is normalized to the
maximum scale and corresponding to that
scaling system the other students marks
are also
analyzed now the question is that why
did I mention this example consider you
are trying to compare all the Steel
Industries or steel companies in
India so consider the companies uh steel
authority of India sale it can can be
other private companies which are GSW
steel and so on and so forth which are
operating in India and when you compare
them it may come out say for example
that out of the 10 different steel
companies which you are considering jsw
steel and steel authority of India have
an efficiency of
one and correspondingly tataa steel has
an efficiency of say for example
.95 now the question is does that mean
that uh the first two companies which I
mentioned GSW steel and steel and sale
steel authority of India are operating
at their highest level of efficiency
answer may not be uh emphatic yes it may
not be possible why because you're
comparing the companies of the sample
set which is in India so if I basically
increase the sample set of trying to
compare like say for example bring other
companies in into the picture nepon
steel posco and other companies in
Germany it may be possible sale and GSW
steel then their efficiency may be less
than one that means when the comparison
basically is much more based on the
population then actually we will be able
to get the overall efficiency of those
set of companies so this would also be
remembered as we proceed with this da
formulation now what are the
nomenclature so I'll basically mention
the nomenclature they have similar set
of
analysis or the ideas which have
already mentioned so here m is the
number of inputs if you remember there m
was number of
Alternatives n is the number of outputs
there n was the number of criteria here
are the number of of Alternatives which
is the dmus we are considering as k and
I Mark with a different color k as in
blue so we want to find out based on the
set of inputs m in number for each
dmu outputs being n in number for each
dmu we are going to find out the
efficiency of the K number of
dmus and what are the variables
variables are if you read it is x i k is
the ath input variable for the k dmu y
JK is the
J output variable for the k
dmu u ik is the input weight for the k i
input weight for the K dmu and finally
vjk is the jth output weight for the K
dmu and what is I J K let me verify once
again small I changes from 1 to M which
I'm am highlighting with blue color
here small J changes from J equal 1 to n
which I'm highlighting in green here
till capital
n and
finally uh the value of K which is for
the dmu changes from K is equal to 1 to
capital K so you have basically i j k i
changes from 1 to m j changes from 1 to
n and K changes from 1 to capital
K now before we consider the simple
model we are going to consider the very
simple model there can be three
different types of Productions and
returns concept now this basically comes
from the concept of how the economy is
doing so we'll consider that is in an
increasing return to scale Market or
economy one is a decreasing return to
scale and one is a constant return to
scale Market increasing return to scale
means the more you produce more the
profit is more the output is more the
revenues are so per unit increase would
basically mean your to total profit
function is on an upward
Trend decreasing return to scale means
more you produce the overall output if
it increases it gives a negative return
in the sense it is decreasing revenues
are decreasing and constant means the
more you produce the rate of change of
the revenue remains constant now these
three graphs of increasing decreasing
and constant we have already seen but
not in this context we have already seen
in the concept of the risk if you
consider the concept of risk there was
marginal utility was increasing at an
increasing rate were increasing at a
decreasing rate and was increasing as a
constant rate so the graph which we had
in the con context of risk was very
simply given like this if you have the
wealth W along the
x-axis utility I'm only utilizing the
symbol U which is along the Y
AIS the graphs would be the blue one
would be where the rate of increase is
constant the green one would
be let me where the rate of increase is
increasing at increasing rate
and finally the red one would mean that
is increasing at a decreasing rate
because the slopes are changing if the
blue one constant slope dydx or D DW and
the green one is
increasing rate is increasing and the
red red one it is decreasing that is Du
DW is
decreasing now let us consider the
diagram in more details and if we
consider the the decreasing return to
scale diagram in details other things
would come out very
easily consider I'm plotting the bundle
of inputs all the inputs which are there
along the
x-axis which I'm just marking for I'll
erase the color but I'm marking for
Simplicity for all the viewers and
readers to see this IP which is input
I'm marking along the x-axis the output
I'm again marking in red is along the y-
axis and the graph which is
here which I will highlight using the
violet color which I'll again erase is
this
one is increasing but the slope is
slowly
falling and this is what I want to now
highlight now notice
here
notice here there are dmus
are on this side of the curve depending
on all the dmus to be maximum efficient
they should lie on the curve of the
production possibility Frontier because
it's decreasing return to scale so if a
dmu is with suffix one and considered is
inefficient because it is not on the the
efficient Frontier line so it can reach
the efficient Frontier line in many ways
it can go either vertically up which is
the Blue Line it can go horizontally to
the left which is the pink one and apart
from the infinite ways it can go I'll
only consider one and why I will do that
I'll come to that later is the one which
is shown in green which is moving
radially in the direction till it touch
is the curve now what I mean by radially
is basically if I consider the green
line the angle which is
there the angle which is there which I'm
drawing here is 90° because that would
be considered as the shortest possible
way in trying to reach the point of the
curve such that the dmu1 which was
inefficient can be made efficient now
when I say the word efficient and I have
marked the blue green and the pink one
what does that
mean now that means the following if I
consider the pink one and if you notice
the slide the pink one means the input
was at
ip12 which I'm now highlighting
here the output remains same which is
op11
and the input decreases from ip12 till
IP C for example here that means at the
same level of output the inputs are
being decreased that means I want to
maintain the same output but my total
input is being
decreased example I may hire less number
of people all the contractual labors I'm
just going to the other example of say
for example the steel companies which I
mentioned or the car companies which I
mentioned I want to produce the same
amount of output but the more I produce
the revenues would decrease at at would
start decreasing at at at a increasing
rate and I want to basically maintain my
output but use less input less input
maybe one example can be I recruit less
number of people do not hire the
temporary workers give some extra over
time to the permanent workers and do
that
work another way can be I want to
basically maintain my input and increase
the output till it reaches the vertical
height which is basically dmu1 efficient
which is blue in color this dmu one
efficient input based is the horizontal
movement onto the left hand side if
you're seeing the slides
yourself and in the case when I'm
considering the blue one which I just
mentioned inputs are being fixed outputs
are being
increased and in the case of the dmu
efficient which is the green one where
it moves
radially there what we are doing is we
are trying to based on dmu 1 which was
inefficient we are trying to decrease
the input as well as increase the output
that means doing a balancing act on
based on that
now which means that the dmu1 which was
inefficient can reach the efficient
Frontier in different ways and these
three are the three different scenarios
based on which dmu1 which was
inefficient can reach now if you have
noticed one thing in the graph the blue
one is marked as dmu1 efficient and then
in the bracket in the suffix i written
op which is output oriented because
there input is not being changed output
is being increased because it is
vertically up in the case when I'm going
horizontally onto the left it is written
in the pink one dmu1 efficient and in
the suffix is written i/p which is input
oriented which means that output is
being kept fixed and input is being
changed so this input oriented model and
output oriented model are the ideas of
input oriented uh formulation of the
problem and output oriented formulation
of the problem will be coming up within
few
minutes now similarly if I consider the
decreasing return to scale I have spent
quite a time a lot of time in trying to
explain the decreasing return to scale
but there is one small change in the
graph if you consider the graph here
which we considered there the output
were all considered along the y- AIS and
the inputs were all considered along the
x-axis so in this graph they are just
interchanged that means output is along
the x-axis and input is along the y-
axis actually it it is the first diagram
is actually I spend time because that
would make much better sense so all the
detailed discussions which I mentioned
and that how the dmu which is
inefficient can go vertically up go
horizontally out to the left or or going
go radially would all be repeated in
this diagram
also now consider the increasing return
to scale now in the increasing return to
scale again I'm considering the idea of
input along the xaxis output along the
y- AIS ideas are
same the curve as it is increasing
return to scale I'll highlight it once
first and then do the detail discussion
so this is the curve which is increasing
return to
scale and all the dmus are on this side
of the curve and they want to reach the
efficiency which is along the curve
which is marked in
blue now here is the
analysis same thing but I'll just repeat
and highlight the important facts here I
consider the dmu VIS is inefficient dmu1
inefficient and in order that this
inefficient dmu which is
one it can be efficient and can reach
the efficient Frontier again there are
innumerable ways but I'll consider only
these three different ways of trying to
move to the efficient Frontier one is
moving vertically up which is the pink
line going horizontally left with the
blue line
and going radially which is the green
one and if you notice the pink one is
mentioned as dme1 efficient IP which is
input
oriented one moving on to the left which
is the blue one horizontal left is in
the as I mentioned the color is blue
dmu1 efficient op which is output
oriented and if I consider the green one
again it is stand and 90° onto the curve
this angle is
90°
this angle is
90° so if I consider
this
the the movements are the green one
would be trying to balance both of them
now let us see how the balance comes up
for the pink blue and the green if I
consider the pink one it means that I'm
keeping the input fixed and I'm
increasing the
output exactly the same way we would
have considered in the first diagram
which is decreasing return to scale
keeping input fixed increasing the
output now when I consider the output
being fixed input being decreased it is
moving along the horizontal left and
when I consider the radial Direction it
is basically a balance of trying to
adjust the input and adjust the output
now the reason why I have drawn a dmu1
efficient input based dmu1 efficient
output based in different colors pink
and blue here in the increasing return
to scale and also in the similar way
dme1 efficient input oriented and dme1
efficient output oriented which are
marked in pink and blue color
respectively we'll come to the model
formulation when we pay specific
attention to the decreasing return to
scale model and the increasing return to
scale model where input oriented model
or output oriented model which is
important and which is more posible and
one more practical will be analyzed with
this I'll end this class and we'll
continue discussion on the DM model
later on thank you very much and have a
nice
[Music]
day
[Music]
[Music]
