Sampling - YouTube

All right, so for today's lesson we're gonna introduce you to different types

of sampling methods and we're gonna see how those methods can get rid of any

inherent bias. So I thought I'd start off today's lesson with an example: a

political researcher is trying to estimate who's going to win the next presidential

election. So to study this matter they thought to themselves well I need to

take a sample. So they went to a large shopping center where a lot of different

type of people congregate and they randomly sampled a thousand people by

asking them who are they going to go for. So now what we need to consider is, is

this a good sampling method? If it is comment on why it's a good method and if

it's not coming on while you feel that way. So pause the video and spend them

out a minute trying to figure this one out.

We're back, hopefully after you've had a few moments to think about that problem

you've come to the conclusion that this was not a good sampling method and

that's true for a multitude of reasons. Now while they did randomly sample

people at that shopping center, we need to realize that that shopping center was

in a specific city, or a specific town or specific state. And only

people who went to that center at that particular day or time could have been

selected. This sampling method is biased, we don't want to have a biased sampling

method otherwise it's not going to yield accurate results for what we're

interested in. This may have given us a good representation of who people at

that shopping center would vote for but not a good representation of the whole

United States. A bias is any systematic overrepresentation or under-

representation of certain characteristics or traits from your

population of interest. In this case we overrepresented people who went to that

center and underrepresented people who didn't now. We should think about what

are the different ways we can get rid of inherent bias because bias is bad. We're

going to spend a lot of time in this course thinking about ways to get rid of bias

or at least in this next lecture. If you have a biased sample, you're not going to

accurately estimate the population that you're interested in. How we are going to

get rid of bias, how bias is is eliminated is by introducing randomness.

Randomness is key, randomness favors no one, it has no inclination. When you

introduce randomness into a sample you'll get rid of bias and so what we're

going to do is we're going to talk about three different sampling methods that

introduce randomness. So the first sampling method

that we're going to learn about today is what's called a simple random sample or

SRS for short and this gives every single member of the population the same

chance of being selected. This is getting rid of any inherent bias for us, every

single person, every single u.s. voter rather has the same chance of being

selected. Regardless of where they live, what their ethnicity is, what is their

income level, everybody has the same chance of being selected. So one way we

could do this is to put everybody into a big hat and, you know everybody's name

into a big hat ,and randomly pull out let's say a thousand names. Or we can

assign everybody a number, all US voters a number, and randomly pull out or select

a thousand numbers. Now there's some benefits to this sampling method and

some downsides. Right, one that we talked about is everybody can be

selected, everybody has the same chance of being selected and it's not biased in

and of itself. However though, there are some things we need to consider; imagine

we did this in practice right. We were trying to estimate or predict who is

going to win the presidential election and we randomly pulled out a thousand US

voters. Is it possible for instance that we pulled out a thousand people all from

California? It technically is, it's not likely but it's possible. You can

still end up with unintentionally a biased sample when you perform a simple

random sample by sheer chance. You didn't intend it to be biased but it's possible

you can end up with a biased sample. Maybe you selected 500 people from

California and 500 people for Texas and didn't represent the other 48 states. We

also need to consider that how are we going to get a list of all, I don't know, 250

million potential U.S. voters. That's going to take

a lot of time, effort, money and resources that we probably don't have. So while

this is a completely random method there are some downsides into performing this.

So another type of sampling method that helps fix some of the issues that an SRS

has is what's called a stratified random sample. And what this does is it breaks

up your population into groups which it refers to as strata; but you can still

think of this as groups and then it performs a simple random sample from

each group. So for instance, if we were trying to predict the U.S. election, the

winner of the U.S. presidential election rather, we could separate United States

voters into East, Central and West. And so then what we would then do is we would

randomly select a certain number of people from those three regions and in

doing so we would be able to better represent the United States as a whole.

Now you don't need to select the same number of people from each group or each

strata, you can select a different amount of people based on the size of those

strata. So for instance if there's more people living on the west side of the

United States compared to the central, you would want to represent that in your

sample. So the benefit of this sampling method compared to a simple random

sample is that it can guarantee you represent observations of a certain type.

A simple random sample cannot guarantee that you would represent all the groups

that you're interested in based on your population. So there are a lot of

advantages of working with a stratified random sample in that you break up

these observations into certain groups and you randomly select observations

from every group. However, there are still some downsides to a stratified random

sample. Again this method can take a lot of time, effort, money and resources.

I mean how are we gonna contact random people from the central United States,

random people from the east eastern United States, and random people from the

western United States. I mean that's going to take a lot of time, effort, money

and resources. You'll also run into issues where if you don't select the

correct percentage of observations from each group; for instance if there's more

people in the western United States than there central and you selected the same

amount of people from the West and the central. Well in that case you would be

biased, you would be over-representing central United States people and

under-representing Western United States people. So if you're going to perform a

random, excuse me, a stratified random sample you want to make sure the

percentages of each group match up with what you're sampling from each group. So

the two sampling methods that we cover at simple random sample and stratified

random sample both of those methods can take a lot of time, effort, money and

resources. So the last one that we'll cover called a random cluster sample

that's going to be the easiest for us to accomplish. And what this does is similar

to a stratified random sample but it breaks up your population into groups

which in this case it calls clusters. So you can think of strata, clusters and

groups as synonyms, they all essentially mean the same thing. And so for a random

cluster sample though, how it's different from a stratified random sample, is that

it doesn't select observations from every group. So if we were trying to do

the predict the presidential election of the United States and we had our three

regions; west, central and east for instance. We could randomly select people

from one of those three regions maybe from the the west, maybe from the east. We

can select all people from a certain group, we can select a subset of people

from a certain group, we could even select two groups, two whole groups,

simple random samples from two groups. The only real difference between a

stratified random sample and a cluster random sample is that you're not

selecting members from every single group in a cluster sample. A stratified

random sample you select from every single group. Now this is going to be

beneficial for us because it's easier to perform. Right if you're a researcher in

the western part of the United States it's probably easier to get observations

or take a survey from people who live in that region. It breaks up these these

population you're interested in two groups. It's gonna be your easiest way to

get a sample. However though, the big downside to this, maybe you kind of came

across it, is it's going to be biased. There's no way to avoid that because

you're not selecting observations from every single group, you're guaranteed to

end up with a biased sample. The only real reason you would want to do something

like this is because it's the only thing you have

available. You just don't have the time, effort, money, resources to take or to get

an unbiased representative sample. In reality companies that try to predict

the presidential election they'll spend millions of dollars to get a sample of

just about 3,000 US voters that's unbiased and representative. It can be

very difficult to get that, so a lot of the times the only option that some

people are left with or some researchers are left with is taking a random cluster

sample. Is it ideal? Absolutely not, but it is the only thing that, it is

sometimes the only thing that's possible. Yeah so these were the three sampling

methods that we'll cover in this course: simple random sample, random stratified

sample and a random cluster sample.