The Art of Statistics: Learning from Data

by David Spiegelhalter

One way around this problem is to demand a very low P-value at which significance is declared, and the simplest method, known as the Bonferroni correction, is to use a threshold of 0.05/n, where n is number of tests done.