With this formula, let us figure out what size sample
we need to get a plus or minus error of two percent. Two
divided by .02 (percents will appear in hundredths)
equal 100. Square 100 and you get 10,000. You multiply
this by .25 and your answer is 2,500. You need 2,500
survey responses to get a +/2 percent margin of error.
Notice that you did not need to have information on
the size of the total population in your audience.
If you wanted to achieve a plus or minus margin of
error of five percent, you would calculate as follows:
Two divided by .05 equals 40. Square 40 and you get
1,600. You multiply this by .25 and your answer is 400.
For all practical purposes, a plus or minus error of
five percent will be the lowest margin of error needed.
For professional pollsters, a sample of 1,000 is capable,
statistically, of reflecting the opinions of 250 million
people. As George Gallup was quoted earlier in this
chapter as saying, the size of the sample itself is not as
large of a factor in determining the accuracy of a survey
as conventional common sense would seem to indicate.
For the purposes of most surveys, you will develop
and implement sample sizes from 100 to 400, which
should be adequate. We mention 100 as the low end of
sample size because our sample size determinate
formula will work with total populations of 4,000 or
more. For lower total populations, we suggest
10 percent of the total population until a minimum of
100 sample respondents has been reached. Additionally,
below 400, categorizing by various demographics
becomes less and less reliable. If you intend to break the
survey results down by age or sex, you are also breaking
the sample size down into smaller numbers.
For example, if the question was, When do you
watch television the most? and the response scale
covers five time blocks with 100 plus respondents, you
can get a percentage that will show general population
vie wing trends. If you try to break those respondents out
into categories based on rank sex or marital status, your
sample size for those separate categories will no longer
be 100 plus. You cannot say most women watch
television from 3 to 4 p.m. because only 32 of your
sample respondents were women and 32 is not a reliable
sample size. With a small sample such as 100, the
demographic analysis of survey results is likely not to
have any usable degree of certainty. You would need a
larger sample size in order to make each category
separated out from the general sample to at least total
100. With a survey of 400 respondents, it is likely that
breaking out various categories would be possible
without producing noticeable bias.
With the above statistical jargon in mind, this rule
of thumb makes sense: use a sample size of 400 for
populations of 4,000 or more, and 10 percent to a
minimum of 100 for populations under 4,000.
Remember, if you choose to use a smaller sample size
than 400, be careful of breaking your total sample into
smaller categories of respondents. With the appropriate
sample size and careful attention to the details of survey
construction and implementation, you will have a low
enough chance of error to make intelligent decisions
based on facts.
Much more important than the size of the sample,
beyond a reasonable minimum, is how that sample was
selected. The only way you can be sure that bias has not
occurred in the selection of a respondent is by random
selection. This means everybody in your total possible
audience has an equal chance at being selected to
complete one of your surveys.
How often have you seen a survey printed on the
back of a base newspaper? This type of survey is fine if
you understand that the respondents are only people who
read your product to begin with. These same
respondents are also the only people who care enough
about whatever the question was to bother answering it.
If that small slice of your audience is that important to
you, then by all means run the survey on the back page
of the base paper.
However, let us say you were trying to find out how
many people like the garden section in your weekend
paper. You printed the survey in the weekend edition and
the results show that your audience loves it. You might
be getting responses only from weekend readers who,
in fact, do enjoy the garden section. In reality, it could
just as well be that they are the only ones reading the
weekend edition because the garden section turns off the
rest of your potential audience. If you want to know what
all of your possible readership thinks on a certain topic,
then you must select respondents by some method that
enables a true cross section or representation of that
One possible random sampling method used in the
military that is easy, effective and previously mentioned
in this chapter is to select survey respondents from the
base alpha lists or personnel rosters. Again, work with
the administration section of your command to be sure
personal privacy is not violated. Choose a randomly