Wednesday, January 25, 2012

You want more drama? Use less data!

Beware of making statements on averages or proportions based on small samples. Take the example of a 10% response rate for your customer base. Now consider a particular customer segment - let's say, customers who bought hiking boots. And you find out that the response rate within this group is 30%. Before you consider that as something to share with your clients or before assuming/concluding that this segment will be a significant driver in your response model - look at the size of this segment. Chances are, that segment will be too small to get any business interest or there is an interaction effect you should explore and identify (in your model).

In many instances, I have come across interesting findings and patterns that turn out be nothing but occurrences in extremely small samples.

The title of this post and the idea of writing something on this came from "Deceptive Data and Statistical Skullduggery" by Gordon H. Bell.

No comments: