SENSITIVITY ANALYSIS OF STATISTICAL CRITERIA TO PSEUDORANDOM VARIABLES GENERATION PARAMETERS

This paper examines the influence of the random number generator initialization parameter (seed) on the reliability of statistical conclusions derived from the Kolmogorov-Smirnov and Froncini tests. The object of the study is statistical goodness-of-fit tests, while the subject is the influence of the seed parameter on the outcomes of these tests for different sample sizes. The research aims to determine the role of the seed parameter in shaping statistical conclusions and to develop recommendations for ensuring the robustness of statistical testing results. The methodology is based on a series of numerical experiments with samples of various sizes (n = 10, 20, 30, 40, 100, 200), generated using different seed values ranging from 0 to 100. For each sample, the Kolmogorov-Smirnov and Froncini tests were applied, and the probability of correct statistical decisions was evaluated. The results demonstrate that the Kolmogorov-Smirnov test is sensitive to changes in the seed parameter, particularly for small sample sizes: for n = 10, the probability of a correct decision does not exceed 67%, while for n = 200, this probability increases to 94.5%. In contrast, the Froncini test shows complete independence from the seed parameter, regardless of sample size, confirming its robustness and reliability. Therefore, the Froncini test is recommended for scenarios where the stability of statistical testing outcomes is critical. The scientific novelty of the study lies in quantifying the influence of the random number generator's initialization parameter on the results of widely used statistical criteria and in demonstrating the advantages of the Froncini test for ensuring reliable statistical conclusions. The practical significance of the study is the formulation of recommendations for selecting the minimum necessary sample size and appropriate statistical criteria to enhance the stability of results when working with random data.