Trying to find out truth from random phenomena . . .
__an impossible task a lot of people regrettably had failed to. I am speaking specifically on Null Hypotheses Significance Tests as they are understood by some kind of researchers. Some features should stay in mind: 1) The aim is not to find out whether H0 is true or not true (relative to a Population parameter value) Against all expectations because we start to suppose that H0 is true, and all algorithm follows if it was, our goal is in fact rather different: to get (or not) sufficient evidence that the Null is *very unlikely* then reject the hypotheses: the parameter value is different from the one H0 suppose. Precisely because NHST is a game of chance we can (Type I error) lead to wrongly reject in spite of H0 is true: it is enough to draw a sample from a Normal Distribution mu=0 (say) and the sample mean be sufficiently large (in absolute value) such that H0: mu=0 is incredibly (!). Is this really possible? Of course, if random draw was lacked. Can you get a wise idea of the mean American´s men height if you take the professional basketball players, or, in contrast, the weightlifters? About the point we are talking I use to compare the Statistician to a XIX century doctor: in almost total absence of clinical analysis diagnosis were totally obtained by intuition based on the patient symptoms shown.