CONTROL INFORMATION OF STATISTICAL DATA BASED ON DISPERSION ANALYSIS

Abstract

Problem statement. The accuracy and reliability of the source statistical material is the most important task of statistical observation. Even with a good organization of observations or conducting an experiment, for one reason or another, inaccuracies, errors, that is, registration errors, may appear. Analysis of the existing method of information control shows that it is too complicated, uses different criteria depending on the sample size, critical criteria values are selected from tables, calculation is performed using formulas and statistical functions of the Excel function master. The purpose of the article is to develop a method of information control based on dispersion analysis of observation data. Results. A single database of analogue objects is created. To correctly create a combined database, you need to be sure that the combined data belong to the same general population. The complexity of dispersion analysis depends on the size of the samples. If several samples of the same volume are combined, then it is easy to check their homogeneity with the help of the “One-factor dispersion analysis” tool, in the output of which the calculated and critical value F- Fisher's test - are given. The examples show that with the help of dispersion analysis it is possible not only to establish the homogeneity of samples, but also the reason for its violation. A method of dispersion analysis of the homogeneity of samples of different volumes was created using the “Descriptive statistics” tool of the analysis package. The adequacy check of the regression model of the active experiment was performed. The experiment plan determines the accuracy of the regression model. Some point is selected in the factor space and many points in its neighborhood are considered. An experiment is being conducted in this neighborhood, on the basis of which the first model is being built. The main requirement for the model is the ability to predict the direction of further experiments with the required accuracy. And the accuracy of this prediction in all search directions should be the same. A model that satisfies this requirement is called adequate. Checking the feasibility of this condition is called model adequacy analysis. In the process of conducting the experiment, it is necessary to make sure that the measured response values belong to the same general population and the technological process does not require regulation. For this purpose, is carried parallel experiments. After conducting 4 experiments according to the experiment planning matrix, it is necessary to make sure of the received response samples’ homogeneity and the reproducibility of the experiments. Conclusions. The performed calculations prove that: the existing method of information control is too complex and has significant disadvantages − different criteria are used depending on the sample size, critical values of the criteria are selected from tables, the calculation is performed using formulas and statistical functions of the Excel function master; the method of information control based on dispersion analysis does not have these disadvantages and is universal, because there is one criterion for small and large samples, the calculated and critical value of F-criterion are given in the initial information of the tool “One-factor dispersion analysis”, it can be used to create a single database analogue objects and to check the adequacy of the regression model at the active experiment.