Factors of the Data (un)Reliability in Surveys with Identical Samples (from the Same Population)

This work is devoted to the search for factors that could contribute to the appearance of differences between the two samples, and on this basis create an algorithm that will help researchers quickly find in which parts of the samples these deviations are observed, provided that the samples were lined up in an identical way and have the same structure. Based on the analysis of representativeness errors, we selected those that could be worked with in this study: sampling error. And then methods of its solution were proposed. The first chapter describes various approaches to dealing with errors, and also contains information that the influence of factors on the dependent predictor can be provided through the nonlinear interaction of features. Methods are also presented that help to identify which methods are more convenient for identifying such relationships between signs. The analysis of the data was carried out on the basis of the database provided by the company N, in which there were results on the approval of the activities of the leader of the country. From the available data, it was not possible to identify significant differences, since the difference between the samples was not large enough, only 7%. However, when using this algorithm on other values, it will be possible to determine by what parameters the displacements are observed. In addition, an attempt was made to determine which characteristics, nevertheless, in the aggregate, could make some contribution. Due to the model of classification trees, it was possible to determine which of the signs are interconnected and can jointly explain the dependent variable, that is, predict the likelihood of approval or disapproval of the President of the country. Based on the resulting model, interaction effects were determined, which needed to be tested for significance. Only one of the considered effects influenced the variable for both samples. Thus, despite the lack of clear meaningful results, we tested the algorithm for demonstrating the differences between measurements on identical samples from the same general population, and also considered the importance of considering nonlinear interactions of features that can have a greater impact than each variable individually.