Good data collection is built on good samples. But the samples can be chosen in many ways. Samples can be haphazard or convenient selections of persons, or records, or networks, or other units, but one questions the quality of such samples, especially what these selection methods mean for drawing good conclusions about a population after data collection and analysis is done. Samples can be more carefully selected based on a researcher’s judgment, but one then questions whether that judgment can be biased by personal factors. Samples can also be draw in statistically rigorous and careful ways, using random selection and control methods to provide sound representation and cost control. It is these last kinds of samples that will be discussed in this course. We will examine simple random sampling that can be used for sampling persons or records, cluster sampling that can be used to sample groups of persons or records or networks, stratification which can be applied to simple random and cluster samples, systematic selection, and stratified multistage samples. The course concludes with a brief overview of how to estimate and summarize the uncertainty of randomized sampling.
1.3 - Why Sample At All? - Part 1
Loading...
From the course by University of Michigan
Sampling People, Networks and Records
40 ratings
University of Michigan
40 ratings
Course 4 of 7 in the Specialization Survey Data Collection and Analytics
From the lesson
Module 1: Sampling as a research tool
Meet the Instructors
James M LepkowskiResearch Professor
Survey Research Center, Institute for Social Research
All this stuff about sampling.
It would be a whole lot easier if we could just collect our observations and
our survey without having to worry about much besides getting enough subjects to
have a large enough sample to be able to say reliable things about the population.
If it were only so easy.
Welcome back.
I'm continuing on our discussion about sampling people, records, and networks.
And in doing sort of a reflection on what
would be nice about samples if we could just take anybody.
And just get enough observations to draw our conclusions.
Yet what we're going to do is a bit more formal than just that.
It's going to have the same object, but it's more formal in this process.
They're just aspects of the process that lead us into more and
more complexity as we do it.
And that's unfortunate, and yet
I think it's a consequence of thinking carefully about these kinds of things.
In this lecture, we're going to cover a couple of diverse topics that can again
provide context for what we'll see later about specific sampling methods.
So, we're going to talk about surveys and sampling in this lecture, but
I just wanted to highlight something.
If you look ahead to the lectures three,
four, five, six, seven, those next lectures, there
are five questions that we're going to address once we have this background.
We're going to talk about, why sample at all?
Maybe it's not necessary to do a sample.
Maybe we can just do something
completely different in terms of drawing our conclusions.
Secondly, if we're going to do a sample, why do we need to randomize it,
and how do we do it, if we're going to be forced into doing it or
if we think it's necessary to do it?
Thirdly, what happens when we randomize?
What happens to our ability to draw conclusions about the population?
And so, we're going to look into what the consequences of randomization are.
Fourthly, how do we evaluate the result, how good it is?
And what role can randomization play in that?
And then, our fifth and final question in Lecture 7,
what kinds of things can we sample?
And that will be fairly brief.
It's going to be people, records and networks.
So, here though, we're going to concentrate on why sample at all.
And in order to do that, let's look in this lecture
at a set of topics here that I've got outlined in the back.
We're going to talk a little bit about a census or a sample.
A little bit about accuracy.
What we mean by accuracy here?
We're going to talk up a little bit about probabilities and something called frames.
We're going to address techniques, the processes that we're going to use.
And then, a little bit about some deficiencies that happen when
we attempt to apply those techniques.
And then we'll mention, briefly, things about complex design.
So, that's what we're going to be addressing here.
These are kind of scattered, but related topics that are foundational background
to understanding what we're going to be doing as we go ahead with this.
Okay, so let's look at this first one about a census or a sample.
So, many times in a textbook,
you'll see this kind of thing where as the drawing in the lower right shows,
there's a population of a large, diverse group of people, many colors.
And from that population, a sample is drawn.
And we want that sample to be miniature of that population.
That's kind of one perspective on it.
That's a classical perspective that comes about from
19th century studies of the human condition in northwest Europe.
I want to make sure that my approach leads to good accurate conclusions about
the population.
I want to be able to tell people, what is the condition of the people living in
London, in this neighborhood in London, at this time with respect to sanitation or
with respect to crime, with respect to working environments and so on.
I just want to make sure that I've got good accurate conclusions
that I can draw on.
And one way I can do that is to just do the whole thing.
Draw a sample and take all.
So there's a relevant population that were interested in.
And then, we're going to be drawing a sample from these and
thinking about how good are the outcomes that I've got, based on this sample.
Okay, census involves the competed enumeration of population when
the population is large, that's a very costly thing to do and
very time consuming.
Censuses are not done routinely for most of these subject matter.
For example, almost every country in the world,
I think every country in the world today at least aspires to doing
a census of the population, but those censuses are done only every five years,
or only ever ten years, because it's such an expensive undertaking.
And a limited number of variables, remember,
in the most recent census of population for
the US, there were just a limited set of questions and they advertised it that way.
You only have to answer ten questions.
There's nothing more that you have to do to try and
reduce burden and encourage participation in response.
The problems that are run into,
with respect to the census, is the timing on it is a problem.
They're only done every ten years.
The census in the US is done In years ending in zero, every ten years.
Going back to 1790, the first in the census of population.
And that means the data get rapidly out of date.
There are some problems with non-response that occur.
The US census is at first a process of
self administrative forms being mailed out to people through the postal service.
And then people are supposed to fill out those forms and return them.
But about 25% of the US population alone,
more than that, doesn't return those forms.
And then, there's a sequential process in which individuals go door by door
to those households that have not returned the forms, in order to collect the data.
So, it's a complex undertaking because it's so large and vast in its operation.
The cost per completed unit and the total cost are a little misleading.
It may not cost very much per person to do a census, but when you've
got a large population of hundreds of millions of people, the total cost can be
something that only could be attempted periodically by a government system.
And so, it's not feasible to do censuses all the time.
In between censuses, people can do samples.
They're going to get snapshots and try and
get as good a representation of that population as possible.
So the census and the sample working together.
Generally, the census represents what the sample aspires to be
measuring everybody and getting a good accurate picture of that.
But the sample is more cost effective and timely and can be done more often,
and filling the gaps in terms of time, filling the gaps in terms of population,
filling the gaps in terms of variables that are being measured.
The census can only do a limited number at a time.
[SOUND] So, the sample is far less costly and far less time-consuming to conduct.
It's only dealing with a subset of the population, but we want to make that that
sample somehow represents the population, a miniature of it as I've said before.
And if we can't do the whole and
thereby get perfection, we're going to have some error.
We'd love to have it so there's no error,
but there's going to be some error because we only have a subset of the individuals.
How large is that error?
And just what can we do with it?
Now actually we think that in the case of samples,
we can get more accurate results than with the census.
Because we can concentrate on a more limited set of subjects in getting
higher quality information for each of those subjects, those in the sample, than
we can with a census, where we've gotta spread our resources across vast numbers.
So, we can actually get more accurate results with the sample than we can
with a census.
But maybe we ought to talk a little bit about what that means.
And we'll do that when we talk about accuracy in a second.
Before we get there, though, [COUGH] let's talk about how samples get selected.
Now, we can do recruitment directly.
By that we're just going to go out and
assemble a set of volunteers to participate in our studies.
It's more of a recruitment activity than a sampling.
The idea is that we're going to find people willing to participate.
There's no issue about a list and a sample selection.
That's the kind of selection technique that's used more
in more formal government in academic service.
But we can just recruit a group of people and assemble them and
collect data from them.
So, the idea of using volunteers for the sample, people who are self-motivated,
they're the ones who initiate the decision to participate.
It's not that we approach them.
In contrast to the mechanisms that we have that we'll talk about the middle one here,
in which we actually have a list of the elements in the population.
We select the ones that we want from that list.
Now we're going to do that using randomization in what we're talking about,
but that selection is something that's instituted by the scientist,
the statistician researchers, and then we go to recruitment.
We only recruit among the sample, the selected group.
And so, there's a interjection of another aspect to this.
A list and a selection rather than just going to recruiting.
But there's a consequence of that.
And we'll talk a little bit about this in this course.
And that is, we may have that list, we may have the selection.
And then, we may do our recruitment, but there are some who are not going
to participate, we're going to have non-response.
And that is a deficiency that creates the necessity of a set of assumptions about
what we're doing that isn't there when we just consider the pure list selection and
recruitment kind of an approach.
So for our purposes, I'll talk a little bit about recruitment,
of recruiting these things directly, volunteer samples,
but mostly we're going to talk about having lists or
some way of assembling lists, artificially assembling a list,
and making selections, and then doing the recruitment.
And we'll add on a few things about that response as we go along.
But because we do samples,
they may cost us less because it's smaller numbers that we're going to be doing, but
they also can lead to better data because we may have more resources to devote to
better measurement and even better coverage in response.
There's always the ever present concern for the fact that we have a sample and
not a census.
And so, we can be wrong in our sample result because
there are four possible outcomes that might arise.
And the bullseye image that you see here,
this is sometimes referred to in the survey realm as the total survey error
perspective, and you've probably seen this in the earlier courses
in our specialization help us to understand this more fully.
Now, I'll comeback and talk about accuracy a little bit more in the context of what
we're doing, but I just wanted you to understand that when we do samples,
we're not claiming that we get perfectly accurate results,
mainly because we know we got a sample, but not the full population involved.
We've mentioned that the samples can involve probability selection or
no probability sampling.
We've mentioned the kind of volunteer recruitment to get a sample,
sometimes for more convenient sources than we may use otherwise.
But there are lots of ways that sample can be drawn that don't involve
chance selection.
Matter of fact, even in the social sciences and
the studies that are done, randomization is not the norm.
The norm is non-randomized studies, non-randomized selection.
[COUGH] For
example, the sample may be selected through expert choice or judgment.
We rely on the expert to determine who falls into the sample.
We have systems in which we can fill quotas based on interviewer judgment.
We get down to the point where an interviewer is told that, look,
you need to get 10 interviews from this neighborhood,
from this voting precinct, from this voting district.
And we want half of the interviews that you do to be female.
We want two of those interviews to be done with people under the age of 30.
And we want four interviews to be done with people of this particular background.
That kind of thing though is in contrast to the non-response and
substitution that is sometimes done.
So, we may actually get a list to make a careful sample selection.
And then, when we go out into the field, we allow for the possibility that there be
non-response and we allow the interviewer to substitute another unit or
someone else to substitute for a carefully chosen unit.
So, there are three forms of non-probability sampling that I mentioned
there.
There are expert choice or judgement.
Filling quotas by interviewers or others, and then [COUGH] substitution.
There are also such things as online panels.
These are very widely used,
increasingly widely used today in studies of political opinion and attitudes.
And they basically recruit the sample in advance, recruit a panel of individuals
that you can draw a sample for a particular problem from.
Now they're going to participate on the basis of their voluntary
decision to be a member of the panel,
in contrast to river samples that draw the sample continuously, as needed.
Rather than having a panel already assembled and then draw a sample from it,
for a particular client or for a particular application.
We're going to do this through some kind of a flow process that's drawing people
through mechanisms on the web or through other kinds of recruiting mechanisms.
All sorts of things there are non-probability sampling, very inventive
kinds of things, but they're beyond the scope of this course for us to deal with.
What we're going to deal with is in the lower right hand corner there.
The image of the lottery, choosing something at random.
Something that comes up as in a lottery,
a set of numbers that have some payoff for us.
But here we're going to use that randomized selection probability sampling.
And here there are a range of techniques on the probability side
that we're going to be dealing with.
And these involve as I mentioned randomization only design,
simple random selection, that we'll do in our next unit.
Stratified selection, the idea of grouping and then getting greater efficiency in
our operation by drawing selections from each of the groups.
Grouping and choosing the groups themselves, this I'll call cluster
samples, which are designed to improve the efficiency by reducing our costs.
So they can reduce costs substantially.
There are also accounting methods, like systematic selection,
that make the selection process much easier to do and it could be done by
clerical trained individuals, rather than highly trained professionals.
And then more complex techniques that involve probabilities that deal
with particular real world problems in drawing these samples, but
using again probabilities in the selection process.
So, different realms of operation.
One, mainly recruiting, focusing on recruiting a sample and
non-probability techniques that fall under that.
And the other that have techniques that involve probability methods, and
techniques such as outlined here.
Well, let's take a little break here and then come back to Lesson 3 about why we
sample at all, to wrap up this particular lecture.
Thank you.
Explore our Catalog
Join for free and get personalized recommendations, updates and offers.
Coursera provides universal access to the world’s best education,
partnering with top universities and organizations to offer courses online.
© 2018 Coursera Inc. All rights reserved.
