Reading a number honestly

Most bad decisions made from data come from a small number of misreadings, and none of them require advanced statistics to avoid.

The average hides the distribution. An average order of Rs 4,000 might be most orders at Rs 1,500 and a few at Rs 40,000. Always ask for the median alongside the mean, because when they differ substantially the data is skewed and the average is describing something that does not exist.

The average also hides the spread. Two branches both averaging 40 sales a day are different businesses if one ranges from 38 to 42 and the other from 5 to 90.

A percentage without the base is meaningless. "Conversions up 200 percent" is three orders instead of one. Always ask how many.

A change without a comparison is meaningless. Sales rose in March, and they rise every March.

Small numbers are noise. Two conversions from forty visitors does not establish a five percent rate. It establishes almost nothing, and acting confidently on a handful of observations is the commonest analytical error in business.

Ask three questions of every number you are given. How many observations is it based on? Compared with what? And who was excluded from the count? The third one catches a surprising amount, because a survey of people who replied is a survey of people who reply.