Wednesday, October 18, 2006

Sampling: When less is good enough

Why do we use a subset?

We sample for one of three reasons: to prove, project, or probe.

Prove: provide credence for a given point of view or confirm a suspicion. Surveys are often done by companies or politicians to justify a position. Not bad in and of itself, but taken for what it is. They may include two steps: First, is something important and second are you doing something about it. Often known as the "gap" approach. In customer analysis it is often common to find a sample used as a matter of convenience for understanding behavior. Many companies have a sandbox for asking questions of the form: Do customers {fill in hunch here}?

Project: when you need the "true number" and asking everybody costs too much in terms of time or money. In this case care is taken in the sampling and weighting of that sample to ensure that the results mirror the larger population from which it is taken. Market share from panel data, media ratings, etc. are good examples. Daily political surveys may or may not be good.

Probe: when you need to understand how different groups behave or respond to marketing ideas. The compare and contrast approach rests on relative differences more than absolute numbers. The champion-challenger approach to message testing is a classic example.

Sampling is a powerful tool, but it is important to use the methods appropriate with the objective. For instance A/B testing does not need static samples and complex projection schemes.

No comments: