Data Science

Sampling

Learning about a big group by looking closely at a smaller part of it.

What's really going on

It’s usually impossible to ask everyone or measure everything, so we look at a smaller part, called a sample, and use it to learn about the whole group. Polls, product tests, and science experiments all depend on samples. A good sample looks like the whole group. A sample that leaves people out, or is too small, can give answers that are confidently wrong.

For educators

Background

A sample is a subset of a population used to estimate something about the whole. Random sampling gives every member an equal chance of being chosen, which reduces bias; convenience samples (whoever is easiest to reach) and voluntary-response samples often skew results. Larger samples reduce random variation but do not fix a biased sampling method. Repeated trials in experiments, such as dropping a paper helicopter several times, are another form of sampling that shows natural variation in measurements.

Misconceptions

  • A bigger sample is always a better sample. A large biased sample can be worse than a small random one.
  • A sample should give exactly the same answer as the whole population. Samples vary; the goal is an estimate that is close and unbiased.
  • One measurement is enough. Repeated measurements reveal variation and make results more reliable.