🔍
Selecting a Representative Sample – When Does a Sample Accurately Match a Population? (7-2) - YouTube
Channel: Research By Design
[6]
We select our sample from a population.
We want our sample to be representative
[12]
of the population, so that after we
conduct our well-designed research, what
[18]
we learn from the sample can tell us
something valuable, by generalizing back
[23]
to the population. A representative
sample has similar physical or
[29]
psychological characteristics as the
population. The best way to get a
[34]
representative sample is to use a random
sample.
[38]
However, a random sample does not
guarantee a representative sample. A
[44]
sample is random when every member of
the population has an equal chance of
[50]
being selected. The random sample is
randomly selected from a larger
[57]
population of subjects. Later, we may use
random assignment, for example, when we
[63]
split our randomly selected sample into
two randomly assigned groups. But because
[69]
of the way that randomness works,
sometimes despite our best efforts, our
[75]
randomly selected sample differs from
our population in some important way.
[81]
Random selection is the best way to get
a representative sample, but it does not
[87]
guarantee a representative sample. Truly
random sampling is difficult to do,
[94]
especially when we are studying human
beings. Sometimes we don't have the time
[100]
or money or availability to do truly
random sampling. Even if we were studying
[106]
college students at our University, a
truly random sample would have to
[110]
include all college students across the
country, so many times we are left
[116]
studying the people whom we have
available. They become our participants
[121]
because they are convenient, and we call
this a convenience sample. The convenient
[127]
sample uses participants who are readily
available and this is not necessarily a
[133]
bad practice, but we should be honest with
[136]
ourselves about what we are doing. Before we begin
[140]
generalizing, applying the results from
our convenience sample to the population,
[144]
we should replicate the same study at
other universities, with other college
[149]
students to see if others find the same
results. Most research in the behavioral
[156]
sciences is done using convenient
samples, and because of this, one joke
[161]
goes that psychology is the scientific
study of college sophomores. Overstated,
[168]
yes, but something to keep in mind when
we are using convenience samples.
[174]
Convenience samples are more likely to
lead to sampling error, and sampling
[181]
error occurs when samples are not random or representative. Sampling error happens
[188]
when we ask too many of the same people, or we ask the wrong people. The time is
[201]
World War 2 in England. In a dimly lit
Quonset hut, the Royal Air Force crews
[207]
gather having just returned from a
bombing run over Germany. The debriefing
[211]
begins with a moment of silence for the
flight crews who did not return, then the
[217]
lieutenant or "leftenant" asks the
returning flight crews, from which
[222]
direction did the fatal attacks come? The
pilots respond to the man "We were
[228]
attacked from above and behind." It's
unanimous. The leftenant scribbles
[234]
this on a piece of paper, hands it to a
corporal and instructs, "Take this
[238]
information to our departing flight
crews, this may save lives. But as the
[245]
Corporal is leaving the dimly lit
Quonset hut, a hand reaches from the inky
[250]
shadows and says stop! No. That
information may cost lives so what is
[258]
wrong with that information? Stop the
video now if you want to ponder or
[262]
discuss. This is an example of sampling
error.
[268]
The lieutenant or leftenant is asking
the wrong people. What does he really
[274]
want to know? From which direction did
the fatal attacks come, but who is he
[279]
asking? All we know is that those who
survived and returned were attacked from
[286]
above and behind. Those who died may have been attacked from a different direction,
[292]
which is why those attacks were fatal.
Here is another famous illustration of
[299]
sampling error. In 1936, the magazine
literary digest which had a long track
[306]
record of picking presidential winners,
mailed 10 million postcards asking
[311]
people their choice for president. Of the
two million responses, 57% preferred the
[318]
Republican Alf Landon. However Landon
lost by a wide margin to Democrat
[325]
Franklin D Roosevelt. So what went wrong? Again, stop the video if you'd like to
[331]
ponder or discuss. The answer this time
is sampling error by asking too many of
[339]
the same people. The magazine selected
names and addresses from phone books and
[345]
automobile registration lists, but what
year was this? 1936 and what was
[353]
happening in America in 1936? The Great
Depression. Who could afford cars and
[361]
personal telephones? People with money or jobs. Literary digest was asking too many
[368]
of the same people. Most poor and
formerly middle class people, those
[374]
without cars or phones voted for the
Democratic Party and FDR, and by the way
[379]
the picture of FDR was done by my 15
year-old niece Abby. So let's summarize
[386]
what we've learned about populations and
samples. Results from the sample should
[392]
generalize to the population, that is
what is true about the sample should
[397]
apply in general to the population.
A sample generalizes better when the
[403]
sample is more representative of the
population. So let's use an example of
[409]
cooking stew. You want to know if the
stew is ready to serve. Are the carrots
[415]
fully cooked? Does it need more salt?
Might it contain an ill-cooked bit of
[420]
beef or underdone potato? In this example, what is your population? It's the entire
[428]
pot of stew. So how large of a sample do
you need to determine if the stew is
[434]
ready? Do you need to eat an entire bowl?
No. You can tell everything you need to
[442]
know from a single spoonful, as long as
what is true of that spoonful? As long as
[451]
the spoonful is representative. The
spoonful must contain a carrot, a pea,
[458]
some beef, the potato, etc. As long as the
elements of the stew are represented in
[464]
the spoonful, what we learned from the
spoonful can tell us about the entire
[469]
kettle. We only need a little taste to
know if the stew needs more salt.
[476]
Of course, if your spoonful lacks carrots then you cannot answer the question of whether
[482]
carrots are fully cooked from your non-representative spoonful. What is the best
[489]
way to get the representative sample? The best way to get a representative sample
[495]
is from a random sample. That means you stir the pot first so that all of the
[502]
elements in the stew have an equal
probability of being selected. However,
[507]
because a random sample can still be non-representative, we need inferential
[513]
statistics to determine how well our
sample represents our population.
Most Recent Videos:
You can go back to the homepage right here: Homepage