Statistics provide data for action. A statistic which will not be used was not worth the trouble of collecting. I consider this a part of the definition of the subject and will not undertake to support it here. The action of course need not follow immediately; it may not be taken until the results of many other observations are available.

Specifically, statistics deal with random variation. In any practical situation, after we have accounted for every variable which can be brought under control, we still cannot say exactly what will happen. In manufacturing a coil spring, we may use metal of fixed chemical specification, even from the same heat, have a fixed setting on the machine which coils the spring, the same time and the same temperatures in tempering it, and still the tensions of the successive springs which are produced will vary. Given a variety of wheat whose seed is the purest strain we can procure, we plant a number of plots in soil of the same chemical composition, apply the same manurial treatment, see to it that the slopes and the irrigation are the same, plant the seed and cultivate the grain in the same way, and yet we find that the yields vary. In no practical case can theory provide an exhaustive account of the relevant variables, our physical control of real situations is never perfect, and consequently we are never in a position where the result will come out exactly to order.