This week, in a series of three posts, I’m explaining how random samples of 500 or 1,000 can be useful.
- What is sampling error? (this post)
- How do you interpret the margin of error?
- Is a sample of 500 or 1,000 really enough?
Surveys are used to estimate something in a ‘target population’
A political poll, for example, might set out to estimate how eligible New Zealand voters feel about a new policy that ex-students should pay interest on their Student Loans. In this instance the target population is ‘eligible New Zealand voters.’ Any survey designer should be able to tell you who the target population is for their survey.
One way to measure whether eligible New Zealand voters think students should pay interest is to take a census (ask them all).
Sampling error is what you get for not taking a census
For obvious reasons a census would not be practical for this question, so survey researchers will ask a random sample of eligible New Zealand voters, and calculate a survey estimate of what they think.
A result based on a sample can never be as precise as a result based on a census, because the sample does not include all members of the population. The sampling error (or margin of error) tells you the price paid for not taking a census. The fewer people you randomly sample from the target population the less precise your survey estimate will be (the larger the margin of error), and vice versa.
Note: Sampling error is just one kind of survey error. There are many others (see related post).