Detection of abnormal values in statistical panels

Abstract

This article is devoted to solving the problem of data research, the values of which are presented in the form of statistical space-time panels with point anomalous changes in values that may occur, for example, due to the influence of socio-economic, external and/or internal political factors (for panels describing legal statistics data), or due to the influence of natural and man-made emergencies (for panels describing the process of functioning of production or economic facilities or systems). The use of classical mathematical methods for analyzing panel data with abnormal data values, as a rule, leads to distorted results that do not allow, in particular, their use to predict future values of statistical indicators with acceptable accuracy. In this regard, the task of identifying abnormal values in statistical panels arises. To solve this problem, a method has been developed based on a retrospective analysis of the values of the indicators contained in this panel and the use of correlation analysis methods. At the same time, it is assumed that the appearance of anomalies leads to a violation of the established relationships between statistical indicators. The method involves the implementation of two consecutive stages. The first stage involves identifying the moment or time interval in which there are abnormal values of the sample elements in the statistical panels. At the second stage, specific elements that have abnormal values at a given moment or time interval in statistical panels are identified. The presented methods are illustrated by an example for legal statistics data for the North Caucasus Federal District, which allowed us to confirm information about an abnormal change in the number of terrorist acts in the Chechen Republic in 2011.