XVII.—Tests for Randomness in a Series of Numerical Observations

Abstract

It is often a matter of considerable importance to decide whether the fluctuations exhibited by a series of observations are random in character, or whether some factor operating according to a definite law must be assumed to exist. One general method of testing a statistical hypothesis is to work out results which would be expected theoretically, and then to compare these with the observations. A statistic such as the mean or standard deviation to be observed on a particular hypothesis may be calculated and compared with the actual mean or standard deviation as evaluated from the observed figures. Agreement does not necessarily imply that the hypothesis is a true one, or even that it is consistent with the observations, because it might be found that when another statistic was calculated, the agreement was no longer satisfactory. Obviously the greater the number of tests applied and found satisfactory, the greater the confidence which can be placed in the conclusions.