Lecture 12

IIT Kanpur July 201
20 followers
Aug 3, 2024
[Music]
hi welcome to the statical
inferencing so far whatever you have
learned is more or less part of
descriptive statistics using these we
have been able to answer questions for
the sample but not really about the
whole
population in fact in most most cases
you don't have exact answers to the
questions related to population majorly
because the population is huge and
collecting data for all of them is
really really difficult and cost
intensive so how do we proceed in such
cases we'll try to discuss that some of
that here let us start with a simple
example from manufacturing industry that
most of us might have heard of a company
that manufactures noodles is banned from
producing noodle packets with excessive
amounts of lead the quantity of lead
that is banned is greater than 2.5 PPM
parts per million which is the accepted
threshold in the
histry let's say that you are working as
part of the food regulator company and
you need to validate that the noodles do
have large amounts of lead obviously you
Cann go to each and every Factory and
check each packet for the amount of lead
as it will take up too much time money
and resources there has to be some more
efficient way so what if you do not
check every packet but instead take a
small sample find the average Le content
in them and conclude the results for the
whole bunch of noodle packets that are
manufactured consider that you have
around 30,000 packets of noodles that
this company has manufactured the mean
lead content of which is unknown to us
out of these 30,000 packets we select
around 100 packets randomly and measure
the lead content in each of them and
after this process we find out that the
mean lead content in these 100 packets
is 2.2 parts per million PPM with a
standard deviation of 0.7 PPM well do
you think these parameters would be same
for the 30,000 packets of population how
sure can you tell that it might have
happened that the sample may be selected
coincidently had the packets which
contained higher amounts of lead or it
could be the case that these packets had
lower amounts of lead as well we just
don't know that yet one thing is for
sure that we cannot at this point say
any anything decisively about the whole
population of the noodle packets but
we'll be able to by the end of this
discussion now I would like to bring
your focus to one key concept which is
sampling sampling is one of the key
things that help us make better
inferences as you know the Crux of
inferential statistics is that you need
to infer the population from the
parameters of the sample and the way you
create this sample plays a very
important role this is also used a lot
while building machine learning models
when you split them into train test and
validate remember that in our problem as
we mentioned earlier we have to select
100 packets randomly suppose that we
select these 100 randomly chosen packets
from any one of the single factories it
could be possible that the issue with
lead content was specific to that
factory only and as a result we end up
Banning the entire company to avoid such
problems food Regulators need to ensure
that the sampling procedure used is fair
and unbiased so that later on no one can
question the accuracy of the inference
after the startat skill analysis is
done so let us understand some of the
different sampling methods that are
commonly used in the industry let's take
a noodles example only and see what are
the different ways in which we can
select a sample that has 100 noodle
packets one way is that I get all the
noodle packets from the population of
30,000 and then just randomly select 100
packets so essentially each packet has
an equal probability of being selected
this method of sampling is known as
simple random
sampling one simple way of looking at
simple random sampling is that you
assign a number each element in the
population and place all the numbers in
a back and then bring a blindfolded
person who selects how many ever units
you want to be present in your sample
you mostly use simple random sampling
within a sampling method to generate the
final sample let's learn about these
sampling methods
suppose the company has two factories
and we know that 70% of the noodles are
manufactured in one Factory a and 30% of
the noodles in Factory B now you want to
select 100 packets from our sample so we
would select 70 packets from Factory a
using random sampling and 30 packets
from Factory B again using random
sampling within the factory specific
packets thus now again you would have a
sample of 100
packets but this time your sample would
be more representative of the population
as we have packets from both factories A
and B and the proportions would be the
same as that of the population this
approach is known as stratified random
sampling here the units are divided into
subgroups and they're selected randomly
but this is done in such a way that the
final sample has the same proportion or
approximately same of these subgroups as
does the population now let's look at
another
approach let's say that you are
collecting these noodle packets from
warehouses across all the country let's
suppose that there are around 20
warehouses in the country so what you do
in this approach is that you consider
each Warehouse as a
cluster this means that there are 20
clusters in total now as part of your
sampling obviously you will not analyze
each cluster separately instead of these
20 clusters you use simple random
sampling and select any three clusters
and consider all the packets within
these three clusters as your
sample here you saw that we divided the
population into clusters with where each
cluster is a warehouse and then we only
selected a sample of these clusters
using a simple random sampling since we
are dividing the population into
clusters this sampling plan is known as
cluster
sampling cluster sampling is usually
used when you see that the population
can be divided into different groups or
clusters that we have different
characteristics once you create the
Clusters you then use the similar ones
or you choose dissimilar ones within
which you can select all the members or
do simple random sampling
now let's say that in order to select
your sample what you did was to select
every third packet starting from the
second packet on a list of the
population so what did we do here we
selected a random starting point and
started picking out sample units at some
fixed or periodic interval such a kind
of sampling is known as systematic
sampling so in this discussion we
learned about four types of sampling
methods namely simple random sampling
stratify dandom sampling cluster
sampling and systematic sampling which
type of sampling techniques seems most
suitable for choosing our noodle packets
so that it represents our
population for many of the food
Regulators stratified sampling is
considered as one of the ideal sampling
techniques that is usually most
representative of the parent population
however in many cases simple random
sampling is also used when the parent
population is not heterogeneous in
nature what could be those cases where
the parent population is not
heterogeneous in nature for instance if
all the noodle packets are of the same
nature and manufactured in a single
Factory then simple random sampling
would be the most straightforward but if
the packets came in different flavors
and were manufactured in different
factories then ideally stratified
sampling would be the recommended method
one important thing to note is that all
the sampling techniques learned until
now are categorized under random
sampling or probability sampling method
so probability samples are types of
samples when every unit of the parent
population has a known chance of being
included in the sample types of sampling
non-random
sampling there is another category of
sampling when it comes to sampling
methods and that is known as non-random
sampling non-random sampling techniques
are those where the odds of any sample
unit being selected for a sample cannot
be really calculated we discuss
different non-random sampling methods so
the first type of non-random sampling we
look at is convenience sampling so let's
say you had to select 100 noodle packets
in our previous example and in order to
do this you simply choose the 100
packets that were around you and measure
their Le content as the name suggests
this sampling method is based on the
convenience of the person who is
selecting the sample clearly this
sampling method has the potential to
become extremely biased the only good
thing here is probably is that this is
relatively convenient sampling method
the other type of sampling method that
we'll look at is called judgmental
sampling again as the name suggests
judgmental sampling is purely judgmental
in which the sample units are chosen
only on the basis of the knowledge and
Judgment of the person who is selecting
the sample often the survey questions
and responses require highly specialized
skill set so for example if you wish to
study the implication of blockchain on
banking industry then it is advised that
you use judgmental sampling to select
people having relevant skill set to
respond to the questions let us
understand the difference between
stratified sampling and cluster sampling
now some of us might get confused
between stratified sampling and cluster
sampling as in both cases we are
dividing the population into
subpopulations in stratified sampling we
divide the population into subpopulation
and select the sample units in the same
proportion as the subpopulations so that
the sample is as representative as the
parent population however in the case of
cluster sampling we also divide the
population into subpopulation or
clusters but unlike stratified sampling
we only study selected clusters and not
all the Clusters
[Music]
[Music]
[Music]
