noc18-mg33-Lec 17

IIT Kanpur July 201
20 followers
4 Video Views·Dec 3, 2023
[Music]
[Music]
very good morning good afternoon good
evening my dear friends welcome this to
this data analysis and decision making
one course on the NPTEL MOOC series and
this is a 12 week course of 30 hours
each week as you know we have 5 lectures
and each lecture being off half an hour
so we are in the 17th lecture which is
we have just started the the 4th week
and we are in the second class for the
4th week and I am Ragunan and Sengupta
from the IME Department IIT Kanpur so if
you remember we were just in the in the
last class we were discussing about
interval estimation and I said that to
wrap it up I said that for the mean mean
value for the interval estimation given
the parameter estimate is known from the
sample we will use the Z distribution or
the T distribution and if for the
standard deviation to be found out we
will use the the chi-square distribution
in the F distribution and the other
distributions considering can also be
used using the concept of central limit
theorem and all these things so we will
try to do a problem and I'll then then
discuss the so the problem is basically
what we left in the discussion in the
last class which is the 16th the last
three two slides so you will first cover
the concept then do a problem and in
case if it is needed we will just
discuss the concept and then go further
on continue but with the promise that
will utilize those concepts where for
where we have not utilized the problems
to basically create them in more details
will clear them in a much more bigger
scope where things would be much much
holistic in their concepts or that I am
sure the students would understand it
much better so this is the example for
the mean with standard deviation
variance known the height X of boy of
boys studying in class 2 of any school
in city of Bhopal is normally
distributed with mean
125 centimeters and variance Sigma
squared is 100 cm centimeter square so
remember the variance for the whole
population for all the students who are
studying in in in the city of Bhopal in
class to the whole population variance
is given as Sigma square which is thier
hundred I am only talking about the
values so I am NOT going into the unit's
we also know the height in centimeters
of 5 such boys who have been selected as
the sample are given as 120 100 110 140
and 130 then we want to find out that
what is the probability there are two
questions what is the probability that
the height of any boy selected at random
number one from the whole population and
from this been selected will be between
120 and 130 so basically we have between
120 and 130 we have to find out that the
boy being selected from the whole
population would be in that group and
boy selected from the sample would be in
that group so what is that group that
group is basically interval 120 to 130
so you have to basically solve it before
I give you the solution first let us see
the problem what are the informations
given so the information is given and
basically the population mean and the
population variance so population mean
and population variance are given and
you want to find out something to do
with the mean so obviously we know that
as the population variance is given we
have to use the Z distribution so
obviously it will be T if the population
variance was not known no not given so
for population mean we they either we
use the Z or the T so the condition is
the first bullet point says then X is
basically normally distributed with 125
as the mean and the variance is 100 now
we need to find out that what is the
distribution of 5 observations taken in
a group and what is the distribution of
their sample mean
so here the suffix x mean means and
comma 5 means that we are taking a bunch
of 5 finding or they mean and basically
trying to find out what is the
distribution of that mean so x mean
comma 5 means
bar 5 so this is also obviously we know
that will be normally distributed but
125 mean value but now in this case the
total variance would be C if you
remember the Sigma square by n so it
will be 100 by 5 would be 20 so where N
is equal to 5 and the sample mean is
given by v 125 the mean yeah as we found
out and now from the information what we
need to know is that what is the
probability that the value of x would be
lying between 120 and 130 and in the
second case that what is the probability
that the sample mean would be lying
between 120 and 130 so let us do the
problems accordingly in the first case X
is the random variable so let us ask
that what is the distribution of X we
know the distribution of X is normal
what is the mean I my answer is 125 and
Sigma square is 100 so we you need to
convert that into a standard normal and
proceed with our calculations so the
probability that X is between 120 and
130 is given by this let me highlight so
I need to convert X into the standard
normal as I do that this is basically
the x1 and x2 which I have the
corresponding LCL and UCL for the X they
would also be converted into LCN and UCL
for the Z distribution so they would
also be transformed accordingly so here
how do we convert the X into a standard
normal deviate it will be X minus the
expected value of x divided by the
standard deviation exactly what we do is
is basically this this X is now
converted no Z so we have not written it
my apologies for that so this would
basically be converted into assets so
technically so what we are doing is X
minus mu by Sigma so mu is basically
given as 125 which is here the Sigma is
basically 10 we convert that into a Z
value the capital is at standard
and these are basically z1 and this is
also they - so these z1 and z2 are
corresponding to 120 and 130 so hence
for the Z it is basically bounded
between minus 5 to plus 5 the value
basically comes out with the probability
so here you have the distribution the
mean value is 0 this is 0 points minus
0.5 plus 0.5 the whole area bounded
between them is about 38% at is given
and hence the probability is point 0.38
now go let us go back to finding out
that what is the probability of the mean
value being between that so obviously
again we find out the same same sand
standard formula probability that x mean
of that corresponding to the sample size
of 5 you will be between 120 and 130 we
know the sample mean is distributed with
normal distributed with a mean value as
given as 125 and the standard deviation
or let us talk about the variance Sigma
square by by the sample numbers which is
5 which comes out to be 20 so hence it
will be it will be bounded between 120
and 130 again convert this x mean 5 into
a standard normal deviate which is
basically the the Z value and the
corresponding lower control and the
upper control will be calculated
accordingly which will be in the first
case the lower control will be 130 minus
120 minus 125 that is and divided by
square root of Sigma bus by by root of n
which is technically square root of
Sigma square by n that which comes out
to be square root of 20 put that values
and on the right hand side again you
will have that value of 130 minus 125
which is plus 5 divided by square root
of Sigma square by n so put those values
and that value comes out to be point
zero point 7 3 2 in this case that means
this is deviating more on to the left
more on to the right hand over
coverage would be much more than 0.8 0.3
and it is coming out to be almost almost
double so these values which I have seen
I am just giving the stand standard
normal table also in order to understand
so minus 5 and plus 5 so minus 5 value
is basically 0.195 so double of that
because you are going equally disposed
onto the left hand side on the right
hand side it comes out to be 0.38 as the
value for the case when you are using
0.7 so let me go back through the value
it is coming out to be zero point seven
three two so it is plus minus one point
one two so one point one two would come
out to be this so this would make about
this so point zero point three six eight
six so both zero point three six eight
zero on the left-hand side and the
right-hand side add them up or multiply
them by two comes out to be zero point
seven two which is seven to 2% now
consider the case of the statistical
inference interval estimation for the
mean with the standard deviation being
unknown so none the first question comes
is that if you are interested to find
out something with this with the mean
value for the population so what
distribution you should be used to
answer as we know is basically the data
said or the T now why we are using T
because the case is that the standard
deviation of the variance of the
population is not known if it is not
known mean we know we have to use one of
the SS and when we use this one of the
SS I am NOT going to come into that
detail what I am basically highlighting
the main points is that we will
basically lose one degrees of freedom
hence you'll basically be utilizing the
T distribution hence T distribution with
me n minus 1 degrees of freedom
but remember one thing here if I draw
the T distribution let me draw it so
this is not a normal this T so this is
the mean value and we want to find out
this limits so this is basically so
called
n minus one and this would be T n minus
1 but the and this whole area as we know
is equal to 1 minus alpha by 2 alpha
this ADL if we changed the color let me
change it to blue or green here this
will be basically alpha by 2 this idea
is also on 4 by 2 so hence some of alpha
by 2 alpha by 2 and 1 minus alpha would
basically be me the total area under
that curve is 1 now remember T
distribution have I mentioned is
symmetric so if it is symmetric the
values on to the left hand side and
right hand side would be CD given by and
4 by 2 here alpha by 2 here but only
remember the mean value being zero so
any value on to the right hand side
would be positive anywhere you want to
the left hand side would should be
negative so we use the T distribution
but in this case remember also we would
not be using my s - because s - is only
used for the case when the sample the
population mean is known but in our case
the population mean mean is not known so
you will be utilizing s without the - so
hence it will be divided by 1 by n minus
1 y1 n minus 1 please bear with me I
will be repeating these things time and
again because we lose one degree of
freedom considering that x1 to xn have
been obtained you'll use them for the
first time to find out what is the best
estimate of the population mean given
the sample mean is the best estimate
which we know so again we will basically
formulate the problem in this way where
xn bar - this term and xn bar plus this
term would basically with the
corresponding lower control and the
upper control corresponding T so we'll
do again a problem small simple problems
food inspector examines and this is the
case for the population variance being
unknown a food inspector examines
dangers of certain bread of butter brand
of butter and obtain the following
percentage of the impurities the values
of which are given as 2.31
92.1 so on and so forth you want to
basically form a 90% level of confidence
LLC for the estimate of the mean of the
impurity levels such that where you can
assume the population distribution of
the level of impurity is basically given
in my normal and we want to basal
comment intelligently according to that
so your values or impurities are given
the level of confidence ninety is given
level of confidence means the overall
area between the lower control limit and
the upper control me so if I basically
have and the distribution again I am
trying not trying to specify with this a
normal distribution T distribution
Christ is to be chi square distribution
or F distribution so these are the
limits the overall area we in between is
basically given by there will a
confidence which is one minus alpha so
from this n value is given as 10 the
mean value x x mean off with with the
sample mean of 10 is given by two point
two to two and this s without the -
because we are utilizing the sample mean
as the best estimate of the population
mean based on which mean we calculate we
find out the standard error comes out to
be zero point five it's eight size seven
eight nine thus from this given
information we know that the lower
control limit and the upper control
limit are in such a way that X mean of
this ends observations and on the left
hand side if we remember to be s n
divided by the square root of n because
here again you will basically be
dividing by square root of n depending
on the set of observation which is there
for the sample into TN minus 1 into
alpha and T and on the right hand side
also it will be TN minus 1 into alpha
alpha by 2 but remember technically they
can be written later on we'll see for
the cry in the set own Chi in the F
distribution on the right hand side
values would be given by 1 minus alpha
by 2 depending on whatever area you have
to cover on the right hand side I'm
going to come to that later that's for
the given information we know that
probabilities being being such that that
the lower control limit which is x mean
for this n observations and SN being the
square root SN divided by square root of
n in
to 80 in the t-distribution estimate the
value is TN minus 1 alpha by 2 alpha by
2 is basically depending on the level of
confidence which you have so that and on
the right hand side the the upper
control limit would be x mean n and
again the same thing but with the same a
plus sign here so once you put this
values the lower control limit and upper
control limit you can check that comes
out to be 1 point 8 4 4 5 and upper
control has two point triple five six so
again I have added the the this this is
the first time you is seeing that a
distribution table on the T distribution
table on the leftmost column near the
degrees of freedom on the top most value
on the M the row you have the p-values
and the inside values are basically the
T values they remember one very very
interesting thing when he using the Z
distribution it gives you the actually
the overall coverage of the
probabilities hence if you add up all
the values starting from minus infinity
to plus infinity the probability values
for the Z destruction there would be one
but the values which are inside here are
the corresponding T values based on
which you are going to do with the the
calculation it doesn't give you the CDF
value so be careful about that that
would also be true for the F
distribution that should also be true
for the chi-square distribution so for a
value of 29 and if you see the value
being 29 because we have lost one degree
no this would be basically sorry my
apologies it would be 9 so basically we
go to 9 so the first we basically mark
this then let us see what is the degrees
of freedom degrees of freedom is
basically given by by 9 which we know 1
minus alpha is is known so if we have
so basically it was 90% so the overall a
they left and right is 10% so this is
would be 10% is the total 88 coverage so
it will be divided into two equal halves
5% 5% so if you see 5% 5% technically
0.05 so for the 0.05 value it comes out
to be one point eight three three one
three utilize that value to consider the
case when you want to find or do some
statistical inference but with respect
to finding or something further for the
variance of the standard deviation given
the mean value is known so this is the
first instance where we want to find out
the interval estimation or the
population variance
now if the mean value is known
immediately it will occur to your mind
that we should be utilizing s Dash and
not s because s dash we see means that
the population mean is known so we do
not use the sample mean and the degrees
of freedom remains as n 0.1 noted now
when we utilize that remember that the
chi-square distribution which is being
formed would also would not lose one
degrees of freedom so it will basically
be K Square and gamma alpha by 2 and the
right hand side it will be 1 minus alpha
by 2 here and this is the first time I'm
trying to basically explain why it is
like this don't see that this is the Chi
square so you will have I square on the
left hand side this is n this is alpha
by 2 this is the second one I square in
1 minus alpha by 2 this whole IDI inside
is 1 minus alpha now see how the
nomenclature is this if you see third
suffolk CBS gives me gives me alpha by 2
that means alpha by 2 alpha by 2 yes
alpha by 2 is the area on to the left of
the chi-square first value and 1 minus
alpha by 2 is the overall area which is
on to the left so we are basically
following the same nomenclature as
considering what is the overall area
covered on to the left see if I see this
so 1 minus alpha by 2 technically means
the whole idea we have covered from the
minimum value chi-square to that value
where it is chi-square this whole area
is 1 minus alpha by 2 and the right hand
side which 1 minus alpha by 2 would be
for the case wait wait wait yes this ad
should be also a 2 because the total sum
of the areas on the left hand side the
middle portion and the right hand side
should be alpha 1 so I'll basically
highlight this in more details because
let me do the problem I'll come to this
conceptual framework in more details
later on so in this case we use the
chi-square with without losing 1 degrees
of freedom to recap important things we
will use s with the - that means we are
not losing any degrees of freedom so
which is basically mu is mu not xn bar
because the sample mean is not to be
utilized and once we have that the
overall rule is probability of n s -
square by n divided by n means the
sample size divided by Chi square with
degrees of freedom n + alpha by 2 B
where the position and is depending on
the level of
confidence which we have and the right
and the right limited the UCL would be
again same thing but the chi-square
value would be n into 1 minus alpha by 2
depending on the nomenclature is and
that overall area would be in between
would be level of confidence which was 1
minus alpha which is this so let us do a
problem and I will highlight it and and
and come to the basic concepts more more
in more details suppose a sample of 30
students school students are given an IQ
test if the sample has a standard error
of 12.25 so we are basically considered
to stand the sample standard error is
12.25 find a 90% confidence interval for
the population standard deviation I
assume that the population mean is known
so if the population mean is known then
this 12.25 is basically s - suffix n and
B 30 that means given the standard error
for the sample and provided the
population mean is known we know that is
the standard data such that we have not
lost any degrees of freedom so in this
case continuing from the table we'll
come to the table Chi square n + 1 minus
alpha by 2 this is beings on the right
hand side would basically be Chi square
what is n n is 30 because you are not
losing any degree freedom and 1 minus
alpha by 2 basically 0.95 that comes out
to 18 point 4 9 3 and Chi square n alpha
by 2 on the left hand side basically
comes out to be forty three point seven
five and then the confidence interval is
found out in the same way as we have
just noted down and the confidence trail
comes out to be 4 Sigma comes out to be
10 point one four five and fifteen point
six zero five so this is the the value
of 30 and this is the value of 29 we
come to that later on so here if it is
basically 30
go back what was one tenth and he wanted
a value at n is equal to 30 and 0.05 and
zero point nine five zero point nine
five is this forty three point seven
five and zero point zero five is basic
eighteen point four nine three check
these values and this is basically
eighteen point four nine three and forty
three point seven three three you
utilize these values to solve your
problems accordingly now let us consider
the problem for the same time but with
the population mean being not given so
consider that statically inference
problem is their interval estimation is
therefore the standard deviation of the
variance but with the population mean
being unknown so here again there we
lose one degrees of freedom because the
population mean is not known we use the
sample mean to find out the population
mean as we do that we have a chi-square
which has 1 degrees of freedom less
hence it is n minus 1 in the numerator
in place of s dash it now would be s
without the dash because you are
utilizing the different standard error
where mean values of the population are
not known and repeating things time and
again please listen to me carefully
and obviously in place of n which was
pre multiplied before s it will be n
minus 1 depending on the formula so you
one can check any good book for the
formulas so the formulas are basically
to be utilized not to be derived here
again the same problem suppose a sample
of 30 students exactly if the sample has
a standard deviation of 12 point two
three now it is not 12.25 find a 90%
confidence interval for the population
standard deviation again and now nothing
is mentioned about the population mean
so as no information is given we
implicit Li assume that the population
mean is unknown so here again we find
out Chi Square 1 minus alpha for N and
now being n minus 1 because it will be
your last one degrees of freedom so it
will be n minus 1 will be 29 so that
value comes out to be for a chi square
29
and 0.95 comes out to be 17 point zero
seven zero eight and chi-squared with
with degrees of freedom of twenty nine
but alpha by two now being zero point
zero five if you look into the table i'm
going to come to that again it comes out
to be forty two point five five seven
utilize that in the formula and solve
your problems accordingly so let me
check the table so here twenty nine this
is forty two point five five seven four
the value of 0.9 0.95 and the renew is
basically for 0.05 it is coming out to
be seventeen point seven zero eight so
let me check forty two point five five
seven four zero point nine four zero
point nine five depending on on the
total and level of confidence being
ninety because you have divided it
accordingly so you can use these values
to solve it strategical inference
estimation can be their confidence
interval we can be there when you are
trying to find out the difference of the
means provided sigma1 and sigma2 unknown
so if it is difference on the mean and
standard deviation obviously we'll use
the z-distribution
i'm going to be just repeated good at
points come back to the problem-solving
later on when you are trying to find out
the difference of the means provide a
sigma and sigma two are unknown both
being equal case an unequal case we will
use the t-distribution with degrees of
freedom being lost one from the first
and one from the second so be careful it
would not be now no more n minus one it
will be n minus two depending on a
number of odd degrees you're basically
lost from the first population and the
second population similarly you can use
the to find of the ratios of the
standard deviations or square which is
the variances provided mu1 and mu2 is
known we will use the f distribution
without dude is losing any degrees of
freedom it will be MN n and in the case
when you have basically the finding of
the ratios of the variances provided mu1
and mu2 are unknown
use one degrees of freedom for the first
population and 1 degrees of freedom from
the second population so it will be F
distribution with M minus 1 and n minus
1 I'll come to that in more details
later on so with this I will end this
seventeenth lecture and continue more
discussion about the hypothesis tting
and going to the multiple linear
regression and then definitely go into
the multivariate statistics which would
be to one of the main part on the
discussion how they can utilize for
different type of problem solving have a
nice day and thank you very much
[Music]
[Music]
[Music]
