How to Lie with Statistics

There is a very common belief that numbers do not lie. Darrell Huff wrote a short book in 1954 to show the exact opposite: a number can be accurate and still lead you to the wrong conclusion, depending on who chose it, how it was measured, and how it was presented.

The title is a provocation. The book does not aim to teach anyone how to deceive, but rather how to recognize when you are being deceived. According to the author, an honest thief is the one who explains the burglary so you can lock your window.

The book remains useful because today's world is even more packed with charts, percentages, and polls than the world of 1954. Evening news, commercials, social media, work meetings, and campaign flyers: in all of them, someone is trying to win you over with a number.

The first trick in the book is the sample with a factory defect. Nearly every statistic starts with a small group meant to represent a large group, and everything depends on whether that small group truly resembles the large one. If it does not, the sample is biased and the result comes out crooked from the start.

Huff shows how this happens even without bad intentions. An income survey of college alumni depends on who can be tracked down and who replies, and those who have done well in life are much more eager to answer. The result sounds wealthier than reality.

Think of a poll taken outside a shopping mall to find out how Americans spend their paychecks. It only hears from people who went to the mall at that hour, which already leaves out a lot of people. The book argues that a good sample must give every type of person a similar chance of being chosen, and that many published surveys fail that test.

When someone says the average, which average are they talking about? Huff reminds us that there are three: the mean (the arithmetic average), the median, which is the middle value when everything is lined up in order, and the mode, which is the most frequent value. Each one tells a different story.

Picture a small company where ten employees earn two thousand dollars and the owner takes home fifty thousand. The arithmetic mean is over six thousand, so a job ad can claim that pay there is excellent. The median, however, shows the two thousand that almost everyone actually gets.

The book teaches a simple rule of self-defense: whenever the word average shows up, ask which one was used and whether the values are widely spread out. That applies to neighborhood income, home prices, and class grades.