Statistical thinking changed in a striking way since its birth. From the descriptive statistics, linked up the research of statistical regularity, until the investigative statistics in the first part of XX century. Successively, the change was much quicker. First of all, statisticians put sampling and inference as the central core of the matter. Fisher came up with a revolutionary idea to solve the problem of inductive reasoning, by preliminarily designing the experiments in the process of data generation. However, Fisher's inference was no longer probabilistic, involving conclusions and choices which are outside the scope of probability theory. Then, Neyman substituted inductive behavior for Fisher's inductive reasoning, who he considered as a basic concept of philosophy of science. Later on, Neyman's ideas were further developed by Wald, who, in turn, transformed statistics in decision theory. Finally, Savage logically supported the Bayesian paradigm, prominent in all Statistics. Recently, statisticians promoted some pedagogical reasons to hesitate in introducing Bayesian paradigm to beginners. In particular, they placed variability (or "variation") as the essential mark in statistical teaching and in empirical enquiry, linked up a revaluation of the causal research and even of the old investigative statistics. The present paper compares the development of statistical thinking with respect to the similar change in philosophy of science from Galileo and Bacon until nowadays. Besides, it suggests coming back in placing phenomena considered from a non-deterministic viewpoint (collective phenomena) as the study subject of the matter, with a revaluation of all phases of statistical method, from the initial definition of a phenomenon until the final decision. Likewise, the essential mark in statistical teaching should be the method of scientific research rather than the pure and simple study of variation.