The Matrix Cookbook

(If you go “Lying with statistics is unethical - what is this?” then read the intro)

A comonly used way to misuse statistics, is to claim causality between two entities of which a relation has been demonstrated. For example

a) If people eating apples have longer live expectancy, then present the results as if it was because of the apples. The truth may be that people eating apples are more likely to do more exercise, but that concern won’t sell your apples, right?

b) If people wearing a tie have higher average income, then you can claim it was because of the excelent silk ties you are selling. The truth probably is that people wearing a tie have made other carrier choices from early on in their life, but never mind.

In case you want to counter a causality claim, remember that NO data, NO statistics and NO mathematics can  formally imply causality. Causality is the result of scientific theories or everyday comon sense. Thus, if the reason for one entity causing another is not clear, it is probabily just a wild guess of the authors.

(If you go “Lying with statistics is unethical - what is this?” then read the intro)

A very simple trick to fool the crowd is to conduct your survey under “the right circumstances”, e.g., select carefully when or who to ask in order to get the answer you are looking for. For example:

• If you want to show that the political conservatives are getting ahead of the socialists, then just ask people outside the church og financial district instead of the supermarket or subway.
• If you want to demonstrate that many people are frustrated with delays in the airport - then ask them on a day with heavy rain and many delayed airplanes.
• If you want to prove that your garden product gives fantastic results, make sure to make the test in a year of good conditions.

If your want to counter the results of a survey, investigate carefully how the data was collected and how this affects the outcome. Everybody making a survey has a preferred outcome, and if you have identified the preferrences, you can start think of how this most likely have influenced the sampling - conciously or not.

It is not just a phrase - there is actually a book with the title “How to lie with statistics”. It is by Darrell Huff, was first published in 1954 and according to Wikipedia the best selling statistics book in the second half of the 20th century. The examples are from back then, but the points maken are amazingly relevant still today. Huff shows some of the most frequent flawed and fraudulent statistics and it is worth reading for the basic knowledge - as a manual on how NOT to get fooled by the tricks of sales people, lobbyists, evil statisticians or undereducated civil servants.

As a celebration, and because its pretty good fun, I’ll run a series of postings on how to lie with statistics - the tricks of the trade. Most of them will be from the book by Huff, but some of my own experience.