Congratulations.We're about halfway through the course.Welcome to Class 6 on Sampling.Although sampling is only one of many facetsthat we've considered in this course, it's critical.Sampling can make or break your study.And you can be sure that if you'redefending a thesis or a dissertation,you'll be asked about your sampling strategy in detail.
Sampling requires hard decisions.And these decisions may be beyond your total control.So you'll need to consider what you want to do,what you can do, and what you're allowed to do.And sampling strategies provide the rules to shape the study.In a complicated but very informative 1967 book by Kish,
he reminds that sampling is not easy.As he says, "sampling theory assumes populations thatare more simply and clearly defined than thosewe need for subjects of representation and inference."And "for most surveys, it is difficult or impossible to makesamples entirely representative of the desired populations."
These are really important points.To begin, we're going to start with the architectureof constructing a sample.And we start with populations.To get a grasp of this, it's very importantto distinguish between four types of populations.First is the inferential population.
So this is the entire universe.You want to try to extend your findings to the broadestpopulation possible-- the inferential population.But this isn't easy.So we have to take it down a notchand consider the target population.
Although we might be interested in studying grade 12 students,and we can find grade 12 students around the world,we have to lower our expectations a bitand consider the group for whom we would liketo make general statements.So we might like to make general statements about grade12 students in Canada, or the US, or perhaps just in BritishColumbia or Ontario.
Once we define the target population,the next step is to consider the frame.This is about the list.And the list becomes really important.What's the list?It's the list from which the sample is selected.So think about it.If you're interested in studying physicians,
then you'd probably like to talk to the Canadian MedicalAssociation to see if you can get a list of practicingphysicians in Canada.This is an example of a list.Take a moment and think about the list that mightbe relevant to your research.
From this list we develop the survey sample.And this is comprised of those selected from the samplingframe following a probabilistic or non-probabilistic strategy.We'll talk about these strategies in a moment.From the sample, survey respondents are generated.And those are the people who actually respond to the survey.
I've just mentioned the words "probability"and "non-probability."And this probably sounds a bit frightening.But once we're through with this class,you'll have a good grasp on what these terms mean.We're going to start with non-probability samplingdesigns.Most texts on survey research don't reallydeal with non-probability sampling designs.
And if they do, they're not very encouraging.These are often not considered legitimate sampling designs.But I argue the opposite.If constructed carefully and correctly,they can be very powerful sampling designs.So what's a non-probability sampling design?Here, the researcher uses subjective judgmentsto determine the units of the population that
are contained in the sample.In other words, you have to make the call about whois included in the sample.But how you make this call is critical.We're going to cover many non-probabilistic samplingdesigns.And we're going to start with the convenience sample.A convenience sample is called a convenience sample
because it's convenient.It's comprised of a group of individualswho are readily available to participate in a study.And it is part of an existing group where itis easy to obtain volunteers.A classroom makes a great convenience sample.Here we have an intact group.And it's easy to access the group.
However, just because a convenience sample exists,it doesn't necessarily mean it's a good choice.If you're interested in studying the relationshipbetween violence in movies and aggressive behaviorby the public, a classroom of first-year university studentsas volunteer subjects would not be the best choice.They're a very select group.And the topic doesn't really relate to their location
as first-year university students.So the advantage of convenience samples is again, convenience.However, the disadvantage is the lack of generalizability.So use convenience samples only when they're warranted.The next type of sample is a volunteer sample.And this is sort of a generic term.
We're going to talk about more specific types of volunteersamples in a moment.So how would you construct a volunteer sample?Well, one way is to stand outside a bar, a pollingstation, or a shop, and wait for people to come out.Here you would ask volunteers to participate in your survey.However, you must ask yourself, is this the best choice
in terms of a sample?So think a moment about phone-in radio programs,and the types of people who phone in,and the types of comments they make.Are these comments representativeof any population?All samples by nature are volunteer,because we can't force people to participate.
However, in this type of volunteer sample,individuals are self-selected, and therefore not verylegitimate.We can sharpen this design a bit by having volunteer sampleswithin a convenience context.For example, if we recruited volunteersfrom a community center, they're bounded by a place.
The next type of sample is an intercept poll.At first this will sound like a volunteer sample.But it has a bit of a twist.Potential survey participants are approachedin a particular location.Have you ever been in a shopping malland been asked to participate in a study?I have.And it's usually market researchersasking for my participation.
To make this approach a bit different,you can build in a sort of randomnessin this type of design.That means surveying every ith person.For example, every 10th person leaving a polling stationduring an election.You could even employ a probabilistic sampling strategyby choosing the sampling locations randomly.
And this would lead to a more systematic sampling approach.Another non-probabilistic sampling strategyis an in-person delivery sampling.This is called the foot in the door approach.It involves both the convenience and an intercept sample.So how do we do this?First, pick a convenient location.
Let's choose Oktoberfest in Bavaria.So we as researchers wait outside the beer tentuntil potential study participants exit.And like an intercept sample, we choose every ith personand ask them to participate in a survey.In this sampling design, we have the advantageof having the interviewer present to explain the study.
If a group of people exit the beer tent together,one person is asked to serve as the designated surveyparticipant.Then either the survey is given to the person whowill complete it on the spot, or she or hewill mail it in later.And hopefully our respondents will be sober enough to do so.The next design to consider are volunteer opt-in panels.
This is becoming more and more popular.And you may have actually been recruited to such a sample.In volunteer opt-in panels, individualsare recruited to participate in online surveys.The sample is recruited through banners, pop-up advertisements,or other types of online recruiting.I'm sure you've seen some of these.
This can be done in two ways, either single opt-inrecruiting, or double opt-in recruiting.Let's start with single opt-in recruiting.Here, potential panel participantsrespond to recruitment methods such as bannersjust by clicking on this banner.He or she will be redirected to a specific panel recruitmentportal, and then will be asked to supply
demographic information, including an email address.For double opt-in recruiting, it starts out the same way.Potential participants are sent a confirmation emailwith a link.If you have clicked on this link,you become a part of an active panel.And you'll be called on from time to timeto participate in a survey.
Now we're going to talk about cases.And you'll just have to bear with me for a moment,because this is a bit complicated.There are a couple of ways of thinking about cases.The first way is that all studiesare a case of something.So it might be a country, three organizations, or hospitalswithin a health board system.So going back to grade 12 students,
our case might be grade 12 students in British Columbia.So this is a very broad use of the term "case."The second way of thinking about caseis as a research strategy in and of itself,which might employ survey research as one data collectiontool.In the ensuing examples, either definition may apply.
But you must be clear to specify howyou're using the term "case."The first type of case is the typical case,sometimes described as the weakest form of a case study.And here we try to choose just that-- a typical caseof something.This is one that's representative, normal,or usual.
And this approach might not be as weak as it seems,because it may be preferable over a random approach.So if you're going to choose a typical city in Canada,would it be wiser to make this choice explicitly,or would it be better to draw a random city out of a hat?But here we start to see the problem with "typical."
Of what is Toronto typical?Regina?Winnipeg?Halifax?Next is a critical case study approach.And here we set out to test a well-formulated theory.The chosen case must meet all of the conditionsto test, confirm, challenge, and extend the theory.
For example, we may want to test theory regardingtracking students in school.The literature tells us that there are two approaches--early versus late tracking.So the cases could include a sitein Germany, which would be a case of early tracking,versus Canada, a case of late tracking.
Next are extreme or deviant cases.And here we focus on a rare or extremely unusualinstance of a phenomenon of interest.An example would be a case of a site with earthquake survivors.This type of case isn't used very often.Next are similar/dissimilar cases.
Here we need at least two cases.And we choose them for their maximal similarity or maximaldissimilarity to each other.In both instances, we're looking for maximum variationto examine important differences and commonalitiesbetween the cases.Or we can take another approach and focus
on the extent to which cases serve to confirm or disconfirmcertain phenomenon, or highlight best or worst policy strategiesand practices.So what do case studies have to do with survey research?Once you've identified one or more cases,many different survey research sampling strategiesmay be employed.
And of course, a case study designcan be strengthened by using a comparative approach.That's it for case studies.Now on to snowball samples.Envision a snowball and think about what a snowball samplingstrategy would look like.Here the ball rolls from one individual to the next.
And the researcher relies on the previously identified groupmembers to identify other group members.This is used when a population listing is not available,usually with difficult to reach populations.For example, drug users or power elites in certain countries.And this is how it works.
You would locate and interview one person,and then ask this person to identify others who would fitinto your sampling strategy.However, this raises an ethical red flag,because that person can't give you the name or phone numberor address of the other person.The person you have interviewed must approach that other person
and be invited to participate in the study,and to contact you as the researcher.Recruiting for this type of samplecan be complicated and problematic.In the case of drug users, for example,you might want to put up posters in safe injection sitelocations to recruit for your study.
In the case of power elites, it'sprobably through networking and word of mouth.Our last non-probabilistic sampling strategyis a quota sample.Here are the goal is to gather informationon individuals in the same proportionas they are represented in the population.
So let's pick males and females.That's easy.And let's assume that the population is comprisedof 50% males and 50% females.And let's say we want 500 of both in our study.So we'll have to narrow down the study by, for example,geographic location and age of the sample.
But once this is done, the potential respondentsare contacted.And you keep contacting individualsuntil you get 500 females and 500 males.So think about going door-to-doorto generate your sample.You keep knocking on doors until you get 500 of each.
This sort of sounds like a probability sample.But it's very different in one respect.Quota sampling allows interviewer discretionin the selection of individuals for the sample.Because interviewers have this discretion,it's critical to provide them with explicit instructionsregarding the characteristics of the sample.The advantage of a quota sample is that you always end up
with the desired sample size.However, there are several disadvantages.The first is interviewer freedom, whichmay lead to selection bias.So think about it.We're knocking on those doors.And in one of the yards there's an angry dog.Will we try to approach this potential participant,
even though this is a house that's designatedin our sampling strategy?Probably not.The problem of bias is compounded because non-responseis concealed.We just go on to the next house until weget 500 males and 500 females.So we may be excluding a particular type of person.
But this isn't evident.Also, because the preexisting conditions of the sampleare unknown, we have problems with generalizability.We might think we can generalize a quota sample to a givenpopulation.But this is not the case.We've spent a lot of time on non-probability samples.
And they certainly have their advantages.They're useful, efficient, and expedient.They're less demanding in terms of resources--that means time and money.And they're much more realistic in termsof real world research.And your research is real world research.As always, there are disadvantages.
The main disadvantage is that it involves judgmentsin terms of inclusion.And you as the researcher make these judgments.Your study may lack validity because of biasin the selection process.So you must be very careful to check such biases.If you don't, the credibility of your findingswill be challenged.
That's it for non-probability sampling.Now onto probability sampling.Probability sampling gets much more attentionin books than survey research design.Sometimes non-probability samplingisn't mentioned at all.With probability sampling, each unit-- that is,each person-- has a known nonzero probability
of being included in the sample.So think of a lottery.There are a finite number of balls in the lottery.And every one has a chance of being drawn.It might surprise, but a census is a probability sample.It just involves 100% of the sample.Everybody is included.
So now instead of a lottery, thinkabout a hat, and drawing names out of a hat.What makes probability sampling so different?One phrase-- a random selection mechanism.Through random selection, subjective bias is eliminated.Also, random selection underlies the theories
used to infer the sample results to the population.So how do we construct a random sample?The first way is a lottery.So think of a hat full of the names of your population.You will have predetermined the proportion of the populationyou want for your sample.And you simply draw names until you reach this proportion.
A second way is a table of random numbers.But this is very old-fashioned.Today we would use instead a computer-generated randomsample.A statistical software package will allow you to do this.For example, SPSS or Stata.Random samples are not at all arbitrary or haphazard.
The selection of each unit is independent of the selectionof every other unit.That is, the selection of one population memberdoes not influence the selection of another population member.Each is independent.With probability sampling, the probability of selectionis not always the same.
We can have equal probability samples or unequal probabilitysamples.In the case of probability samples,every member of the population has the same probabilityof selection.So we have a hat.We have the entire population in the hat.And we draw out a sample.
For unequal probability samples, some membersare more likely than others to be selected for the sample.So here we would want to separate out men and women,for example.And we know that men are less likely to participatein surveys.So we might want to draw a larger sample of menthan women.So from a given population, we might
want to draw a sample of say, 40% men and 30% women.Hopefully in the end we'll have equal sample sizes of both.Like non-probability samples, thereare different ways of approachingprobability samples.The first is simple random sampling.And I've already described this.
Each member of the population has equal probabilityof selection.So you must start out with a listof all members in the population,and then select at random until a previously specifiednumber of members or units has been selected.A second technique is systematic sampling.And this is very useful.
It's also called pseudo simple random samples.And it's very easy to do.Like simple random sampling, you need a population listing.Once you have this list, you choose a random start,and then select every ith unit.For example, you might choose every fourth personon the list.You can sharpen this approach by choosing a random start
in your list.That is, not starting at person number one,but say, person number 47, chosen at random.However, with systematic samplingit's important to pay attention to the nature of the list.If the list is constructed in some sort of cyclical fashion,
it could build in bias to your sample.For example, if your list is constructedof classes of students, and the classesare sorted by age, if you choose every 25th person on the list,your sample's going to be older.Next is a stratified sample.And we've already touched on this.
With a stratified sample, you divide the sampleinto groups called strata before the sampling begins.We've already done this with our males and females example.Each unit-- that is, person-- is assignedto only one stratum based on prior knowledge about a unit.So with our male and female example,
the person would either have to be male or female.Next we would draw independent samples from each stratum,using simple random sampling or systematic random sampling.We can keep it simple by using a proportionate stratifiedsampling strategy.
That is, we use an equal probability for each stratum.For example, we draw a 30% sample from each stratum.However, because strata can have different characteristics,we may want to get a little more sophisticatedand use a disproportionate sampling strategy.This is where we use unequal proportions.As I said earlier, males are less likely to participate
in survey research.So we might want to over-sample males in our study.This will increase the precision of the overall sampleestimates.Next and finally is cluster sampling.Here we're dealing with groups.So what's a cluster?These are naturally occurring groupings such as schools,
households, or city blocks.Let's use school districts as an example.In British Columbia, there 60 school districts.Using a cluster sampling approach,we would draw a random sample from the list of these 60school districts.So the cluster-- the school district--and not the individual, is the sampling unit.
And this is great, because you don't need listsfrom every individual from each of the 60 school districts.This would be a simple random sample approach.However, we can employ a stratified random samplingapproach.We might groups rural schools together,urban rural schools together, and urban schools
together, and draw from each of the strata.However, compared with stratified sampling,cluster sampling usually decreases the precisionof the statistics.So this is something to be aware of.Here's how a cluster sample would look graphically.
Probability sampling has one major advantage.It's the sampling strategy to choosewhen you want to generalize to a target population.However, the major disadvantage isthat it's employed when it's not the best samplingstrategy at hand.And this is often due to ignoranceabout non-probability sampling strategies.
Let's sum up.We started by considering the architecture of the sample.Then we went on to consider non-probability samplingstrategies.These included volunteer, intercept, in-person delivery,volunteer opt-in, case studies, snowball, and quota samples.Then we looked at probability sampling strategies,
including simple random, systematic, stratified,and clustered.Now you have a sound basis to start constructing your sample.
Series Name: Designing and Doing Survey Research
Publisher: University of British Columbia
Publication Year: 2015
Segment Num.: 1
Professor Lesley Andres discusses probabilistic and non-probabilistic sampling, delineating the different types of each and describing their advantages and disadvantages.
Looks like you do not have access to this content.
Professor Lesley Andres discusses probabilistic and non-probabilistic sampling, delineating the different types of each and describing their advantages and disadvantages.