Sampling in survey research
From the food we eat to the TV we watch, from political elections to
magazine advertising – much of life is affected by the results of sample
surveys. It's never been more important that the design and analysis
of sample surveys produce good data. It's critical for making sound business
decisions and recognizing questionable data, when it arises.
The purpose of survey research is to take a representative sample from
a population, and then use the data gathered to make inferences about
an entire population. Sampling saves money, time, and effort, and may
provide as much information as a study of an entire population. The best
results are often achieved by random sampling, which will tend to represent
groups within the population, proportionally.
In some cases, we do not want proportional representation. For example,
consider if we are studying voting patterns to determine how Republicans,
Democrats, and Independents will vote. A random sampling would reach
fewer Independents, and in the process of reaching enough Independents,
we would survey more Democrats and Republicans than we need to. In a
case like this, we would use quota sampling.
Quota sampling can be used either to ensure that the sample is proportionally
representative of the population, or to over-sample among smaller segments
of the population. Quota sampling requires that we know or have a good
estimate of the true percentage for each segment to which we apply quotas.
It is random within population segments, but with the constraint that
a specific number of surveys are conducted within each segment. With
quota sampling, the interviewers might be told to interview so many male
Democrats, so many female Republicans, and so on.
Quotas for gender and race could be used to ensure that the sample within
each political party is proportionally representative of the party's
population. And we might set quotas by party to ensure that an equal
(or nearly equal) number of surveys are conducted with members of each
party. In this way we would have as much confidence in our results for
Independents as we have for Democrats and Republicans. We would then
weight the results by party to ensure that the overall results represent
each party, proportional to their representation in the overall population.
The question of sample size arises early in the planning of any survey.
To take a larger sample than needed to achieve the
desired results is costly, whereas very small samples often lead to no
practical use in making good decisions. As discussed above, the size
of the sample and the use of quotas are largely determined by the level
of confidence desired for key segments within the population. The main
objective is to obtain both a desirable accuracy and a desirable confidence
level with minimum cost. ‡
|
