ANALYSIS AND DATA PROCESSING SYSTEMS

It is argued that in most cases two reasons underlie the incorrect application of nonparametric goodness-of-fit tests in various applications.The first reason is that when testing composite hypotheses and evaluating the parameters of the law for the analyzed sample, classical results associated with testing simple hypotheses are used. When testing composite hypotheses, the distributions of goodness-of-fit statistics are influenced by the form of the observed law F(x, q) corresponding to the hypothesis being tested, by the type and number of estimated parameters, by the estimation method, and in some cases by the value of the shape parameter. The paper shows the influence of all mentiomed factors on the distribution of test statistics. It is emphasized that, when testing composite hypotheses, the neglect, of the fact that the test has lost the property of “freedom from distribution” leads to an increase in the probability of the 2nd kind errors. It is shown that the distribution of the statistics of the test necessary for the formation of a conclusion about the results of testing a composite hypothesis can be found using simulation in an interactive mode directly in the process of testing.The second reason is associated with the presence of round-off errors which can significantly change the distributions of test statistics. The paper shows that asymptotic results when testing simple and composite hypotheses can be used with round -off errors D much less than the standard deviation s of the distribution law of measurement errors and sample sizes n not exceeding some maximum values. For sample sizes larger than these maximum values, the real distributions of the test statistics deviate from asymptotic ones towards larger statistics values. In such situations, the use of asymptotic distributions to arrive at a conclusion about the test results leads to an increase in the probabilities of errors of the 1st kind (to the rejection of a valid hypothesis being tested). It is shown that when the round-off errors and s are commensurable, the distributions of the test statistics deviate from the asymptotic distributions for small n. And as n grows, the situation only gets worse. In the paper, changes in the distributions of statistics under the influence of rounding are demonstrated both when testing both simple and composite hypotheses. It is shown that the only way out that ensures the correctness of conclusions according to the applied tests in such non-standard conditions is the use of real distributions of statistics. This task can be solved interactively (in the process of verification) and rely on computer research technologies and the apparatus of mathematical statistics.

1. Kolmogoroff A.N. Sulla determinazione empirica di una legge di distribuzione. Giornale del Istituto Italiano degli Attuari, 1933, vol. 4, no. 1, pp. 83–91.

2. Bol'shev L.N., Smirnov N.V. Tablitsy matematicheskoi statistiki [Tables for mathematical statistics]. Moscow, Nauka Publ., 1983. 416 p.

3. Anderson T.W., Darling D.A. Asymptotic theory of certain “Goodness of fit” criteria based on stochastic processes. The Annals of Mathematical Statistics, 1952, vol. 23, pp. 193–212.

4. Anderson T.W., Darling D.A. A test of goodness of fit. Journal of the American Statistical Association, 1954, vol. 29, pp. 765–769.

5. Kuiper N.H. Tests concerning random points on a circle. Indagationes Mathematicae (Proceedings), 1960, vol. 63, pp. 38–47.

6. Lemeshko B.Yu. Neparametricheskie kriterii soglasiya: rukovodstvo po primeneniyu [Nonparametric goodness-of-fit tests. Guide on the application]. Moscow, Infra-M, 2014. 163 p. DOI: 10.12737/11873.

7. Stephens M.A. The goodness-of-fit statistic V_N: distribution and significance points. Biometrika, 1965, vol. 52, no. 3–4, pp. 309–321.

8. Watson G.S. Goodness-of-fit tests on a circle. 1. Biometrika, 1961, vol. 48, no. 1–2, pp. 109–114.

9. Watson G.S. Goodness-of-fit tests on a circle. 2. Biometrika, 1962, vol. 49, no. 1–2, pp. 57–63.

10. Zhang J. Powerful goodness-of-fit and multi-sample tests: PhD Thesis. York University. Toronto, 2001. 113 p. Available at: http://www.collectionscanada.gc.ca/obj/s4/f2/dsk3/ftp05/NQ66371.pdf (accessed 02.05.2021).

11. Zhang J. Powerful goodness-of-fit tests based on the likelihood ratio. Journal of the Royal Statistical Society: Series B, 2002, vol. 64, no. 2, pp. 281–294.

12. Noughabi H.A, Arghami N.R. General treatment of goodness of fit tests based on Kullback–Leibler information. Journal of Statistical Computation and Simulation, 2013, vol. 83, pp. 1556–1569.

13. Noughabi H.A. A new estimator of Kullback–Leibler information and its application in goodness of fit tests. Journal of Statistical Computation and Simulation, 2019, vol. 89, no. 10, pp. 1914–1934.

14. Kac M., Kiefer J., Wolfowitz J. On tests of normality and other tests of goodness of fit based on distance methods. The Annals of Mathematical Statistics, 1955, vol. 26, pp. 189–211.

15. Lilliefors H.W. On the Kolmogorov-Smirnov test for normality with mean and variance unknown. Journal of the American Statistical Association, 1967, vol. 62, pp. 399–402.

16. Lilliefors H.W. On the Kolmogorov-Smirnov test for the exponential distribution with mean unknown. Journal of the American Statistical Association, 1969, vol. 64, pp. 387–389.

17. Martynov G.V. Kriterii omega-kvadrat [Omega-square criteria]. Moscow, Nauka Publ., 1978. 80 p.

18. Tyurin Yu.N. O predel'nom raspredelenii statistik Kolmogorova–Smirnova dlya slozhnoi gipotezy [On the limit distribution of Kolmogorov–Smirnov statistics for a composite hypothesis]. Izvestiya AN SSSR. Seriya matematicheskaya = Mathematics of the USSR-Izvestiya, 1984, vol. 48, no. 6, pp. 1314–1343. (In Russian).

19. Tyurin Yu.N., Savvushkina N.E. Kriterii soglasiya dlya raspredeleniya Veibulla–Gnedenko [Goodness of fit for the Weibull–Gnedenko distribution]. Izvestiya AN SSSR. Seriya: Tekhnicheskaya kibernetika = Soviet journal of computer and systems sciences, 1984, no. 3, pp. 109–112. (In Russian).

20. Lemeshko B.Yu., Postovalov S.N. O zavisimosti raspredelenii statistik neparametricheskikh kriteriev i ikh moshchnosti ot metoda otsenivaniya parametrov [On the dependence of the distributions of statistics of nonparametric tests and their power on the parameter estimation method]. Zavodskaya laboratoriya. Diagnostika materialov = Industrial laboratory. Materials diagnostics, 2001, vol. 67, no. 7. pp. 62–71. (In Russian).

21. R 50.1.037–2002. Rekomendatsii po standartizatsii. Prikladnaya statistika. Pravila proverki soglasiya opytnogo raspredeleniya s teoreticheskim. Ch. 2. Neparametricheskie kriterii [Recommendations for standardization. Applied statistics. Rules for checking the agreement of the experimental distribution with the theoretical one. Pt. 2. Nonparametric tests]. Moscow, Standards Publ., 2002. 64 p.

22. Lemeshko B.Yu., Lemeshko S.B. Distribution models for nonparametric tests for fit in verifying complicated hypotheses and maximum-likelihood estimators. Pt. 1. Measurement Techniques, 2009, vol. 52, no. 6, pp. 555–565.

23. Lemeshko B.Yu., Lemeshko S.B. Models for statistical distributions in nonparametric fitting tests on composite hypotheses based on maximum-likelihood estimators. Pt. 2. Measurement Techniques, 2009, vol. 52, no. 8, pp. 799–812.

24. Lemeshko B.Yu., Lemeshko S.B., Postovalov S.N. Statistic distribution models for some nonparametric goodness-of-fit tests in testing composite hypotheses. Communications in Statistics. Theory and Methods, 2010, vol. 39, no. 3, pp. 460–471.

25. Lemeshko B.Yu., Lemeshko S.B. Models of statistic distributions of nonparametric goodness-of-fit tests in composite hypotheses testing for double exponential law cases. Communications in Statistics. Theory and Methods, 2011, vol. 40, no. 16, pp. 2879–2892.

26. Lemeshko B.Yu., Gorbunova A.A. Application of nonparametric Kuiper and Watson tests of goodness-of-fit for composite hypotheses. Measurement Techniques, 2013, vol. 56, no. 9, pp. 965–973.

27. Lemeshko B.Yu., Lemeshko S.B., Postovalov S.N., Chimitova E.V Statisticheskii analiz dannykh, modelirovanie i issledovanie veroyatnostnykh zakonomernostei. Komp'yuternyi podkhod [Statistical data analysis, simulation and study of probability regularities. computer approach]. Novosibirsk, NSTU Publ., 2011. 888 p.

28. Lemeshko B.Yu., Lemeshko S.B., Rogozhnikov A.P. Interactive investigation of statistical regularities in testing composite hypotheses of goodness of fit. Statistical models and methods for reliability and survival analysis. Ed. by V. Couallier, L. Gerville-Reache, C. Huber-Carol. Hoboken, NJ, Wiley-ISTE, 2013, ch. 5, pp. 61–76.

29. Lemeshko B.Yu., Lemeshko S.B., Blinov P.Yu., Veretel'nikova I.V., Novikova A.Yu. Statisticheskii analiz interval'nykh nablyudenii odnomernykh nepreryvnykh sluchainykh velichin "Interval'naya statistika 5.4" [Statistical analysis of interval observations of one-dimensional continuous random variables "Interval statistics 5.4"]. The certificate on official registration of the computer program. No. 2018666213, 2018. Available at: https://ami.nstu.ru/~headrd/ISW.htm. (accessed 03.05.2021).

30. Lemeshko B.Yu. Lemeshko S.B., Semenova M.A. K voprosu statisticheskogo analiza bol'shikh dannykh [To question of the statistical analysis of big data]. Vestnik Tomskogo gosudarstvennogo universiteta. Upravlenie, vychislitel'naya tekhnika i informatika = Tomsk State University Journal of Control and Computer Science, 2018, no. 44. pp. 40–49. DOI: 10.17223/19988605/44/5.

31. Lemeshko B.Yu., Blinov P.Yu. Kriterii proverki otkloneniya raspredeleniya ot ravnomernogo zakona: rukovodstvo po primeneniyu [Tests for checking the deviation from uniform distribution law. Guide on the application]. Moscow, Infra-M Publ., 2015. 183 p. DOI: 10.12737/11304.

32. Lemeshko B.Yu. Kriterii proverki otkloneniya raspredeleniya ot normal'nogo zakona: rukovodstvo po primeneniyu [Tests for checking the deviation from normal distribution law. Guide on the application]. Moscow, Infra-M Publ., 2015. 160 p. DOI: 10.12737/6086.

33. Lemeshko B.Yu., Blinov P.Yu. Kriterii proverki otkloneniya ot eksponentsial'nogo zakona: rukovodstvo po primeneniyu [Tests for checking the deviation from exponential distribution law. Guide on the application]. Moscow, Infra-M Publ., 2021. 352 p. DOI 10.12737/1097477.

34. Lemeshko B.Yu. Kriterii proverki gipotez ob odnorodnosti: rukovodstvo po primeneniyu [Tests for homogeneity. Guide on the application]. 2nd ed., rev. and add. Moscow, Infra-M Publ., 2021. 248 p. DOI 10.12737/986695.

35. Pearson E.S., D’Agostino R.B., Bowman K.O. Tests for departure from normality: comparison of powers. Biometrika, 1977, vol. 64, pp. 231–246. DOI: 10.1093/biomet/64.2.427-a.

36. Tricker A.R. The effect of rounding on the significance level of certain normal test statistics. Journal of Applied Statistics, 1990, vol. 17, no. 1, pp. 31–38. DOI: 10.1080/757582644.

37. Tricker A.R. The effect of rounding on the power level of certain normal test statistics. Journal of Applied Statistics, 1990, vol. 17, no. 2, pp. 219–228. DOI: 10.1080/757582833.

38. Deidda R., Puliga M. Sensitivity of goodness-of-fit statistics to rainfall data rounding off. Physics and Chemistry of the Earth, 2006, vol. 31, pp. 1240–1251. DOI: 10.1016/j.pce.2006.04.041.

39. Lemeshko B.Yu., Lemeshko S.B. Effect of the roundoff on the properties of criteria for testing statistical hypotheses. Optoelectronics, Instrumentation and Data Processing, 2020, vol. 56, no. 3, pp. 35–45. DOI: 10.3103/S8756699020030103. Translated from Avtometriya, 2020, vol. 56, no. 3, pp. 35–45. DOI: 10.15372/AUT20200305.

40. Lemeshko B.Yu., Lemeshko S.B. O vliyanii oshibok okrugleniya na raspredeleniya statistik kriteriev soglasiya [About the influence of rounding errors on distributions of statistics of the goodness-of-fit tests]. Vestnik Tomskogo gosudarstvennogo universiteta. Upravlenie, vychislitel'naya tekhnika i informatika = Tomsk State University Journal of Control and Computer Science, 2020, no. 53. pp. 47–60. DOI: 10.17223/19988605/53/5.

41. Lemeshko B.Y., Lemeshko S.B. About the effect of rounding on the properties of tests for testing statistical hypotheses. Journal of Physics: Conference Series, 2021, vol. 1715, p. 012063. DOI: 10.1088/1742-6596/1715/1/012063.