DISTRIBUTIONS OF REAL STATISTICAL DATA ARE NOT NORMAL | Распределения реальных статистических данных не являются нормальными

In the training courses on the theory of probability and mathematical statistics there are various parametric families of distributions of numerical random variables considered. Namely, we have been studying the families of normal distributions, log-normal distributions, exponential distributions, gamma distributions, Weibull-Gnedenko distributions, etc. All of them depend on one, two or three parameters. Therefore, for a complete description of the distribution it is sufficient to know or estimate one, two or three numbers. Parametric theory of mathematical statistics is widely developed, where it is assumed that the distribution of observations belong to one or another parametric family of distributions. This tradition comes from Karl Pearson, who in the early twentieth century proposed the use of four parametric family of distributions. The above families of distributions - are the subsets of a four-parametric family of Pearson. Unfortunately, parametric families exist only in the minds of the authors of textbooks on probability theory and mathematical statistics. In real life, they are not. Therefore, modern applied statistics and econometrics mainly use non-parametric methods, in which the distribution of observations can have arbitrary form. First, on an example of a normal distribution, we are discussing the impossibility of practical use of parametric families of distributions to describe specific statistical data. We give the results of research of metrologists and estimation of convergence in limit theorems. Then we discuss how the parametric methods can use for reject outlying observations. It is very unstable the significance levels for a fixed rejection rule and the parameter of the rejection rules for a fixed level of significance. Consequently, the rejection of the classic rules of mathematical statistics is not sciencebased