8

In a binomial experiment, if we observe $x=0$ positive individual among $n$ individuals, then the proportion of positive individuals is significantly lower than $3/n$ with a type 1 error less than and very close to $5\%$. This fact, sometimes called the "rule of three", is a consequence of the inequalities $$\exp\left(-\frac{np}{1-p}\right) \leq \Pr(X=0) \leq \exp(-np).$$

Do you know other such basic easy rules for statistics? I find them very interesting and useful. This principle is not really a "rule of thumb" because it has a reliable theoretical foundation, but I don't see another tag for this question (I hope it is not off-topic)

whuber
  • 281,159
  • 54
  • 637
  • 1,101
Stéphane Laurent
  • 17,425
  • 5
  • 59
  • 101
  • "Normally, more than two-thirds of the people are average" (meaning within one standard deviation of the mean)? – Dilip Sarwate Jun 08 '12 at 11:42
  • One very simple one that comes to mind is how the variance of a sample proportion of successes out of $n$ Bernoulli trials is no more than $1/4n$ (which is achieved when the success probability is $1/2$) – Macro Jun 08 '12 at 12:11
  • Many "rules of thumb" are based on theoretically rigorous analyses or approximations. – whuber Jun 08 '12 at 13:02

1 Answers1

12

Check out Gerald van Belle's book "Statistical Rules of Thumb" a very nice little paperback text loaded with examples of rules of thumb and explanations including the "Rule of three" that you mention above.

Michael R. Chernick
  • 39,640
  • 28
  • 74
  • 143