Exploring ANCOVA models with bimodal error structures

Abstract

Analysis of covariance is a frequently employed statistical technique in experimental and quasi-experimental research. A key assumption in this analysis is that the error terms follow a normal distribution. This paper investigates parameter estimation and hypothesis testing within covariance analysis models when the error term distribution deviates from normality and instead follows an alpha skew-normal distribution. We consider a one-way deterministic analysis of covariance, a one-way deterministic analysis of covariance with two covariates, and a stochastic analysis of covariance. The unknown model parameters are estimated using the maximum likelihood method. Based on these estimators, new test statistics are proposed to assess both the treatment effect and the significance of the slope parameter. A Monte Carlo simulation study is conducted to compare the efficiency of the proposed estimators with traditional least squares estimators. The simulation results demonstrate that the maximum likelihood estimators exhibit greater efficiency compared to the least squares estimators. Furthermore, the test statistics derived from maximum likelihood estimators are found to be more powerful than those based on least squares. In the application section, two real-world datasets are analyzed to illustrate the proposed method.

Keywords

• alpha skew normal distribution
• analysis of covariance
• iteratively reweighting algorithm
• stochastic covariate

References

1. [1] A. Agresti, Statistical Methods for the Social Sciences (5th ed.), Pearson, 2018.
2. [2] S. Acitas and B. Senoglu, Robust factorial ANCOVA with LTS error distributions, Hacettepe J. Math. Stat. 47 (2), 2018, pp. 347–363.
3. [3] P. J. Bickel and K. A. Doksum, Mathematical Statistics: Basic Ideas and Selected Topics (2nd ed.), Springer, 2015.
4. [4] J.B. Birch and R.H. Myers, Robust analysis of covariance, Biometrics 38 (3), 1982, pp. 699–713.
5. [5] N. Celik, B. Senoglu and O. Arslan, Estimation and testing in one-way ANOVA when the errors are skew-normal, Colomb. J. Stat. 38 (1), 2015, pp. 75–91.
6. [6] D. R. Cox and P. McCullagh, Some aspects of analysis of covariance, Biometrics 38 (3), 1982, pp. 541–561.
7. [7] D. Elal-Olivero, Alpha-skew-normal distribution, Proyecciones (Antofagasta) 29 (3), 2010, pp. 224–240.
8. [8] D. Elal-Olivero and H. W. Gómez, Bayesian modeling using a class of bimodal skewelliptical distributions, J. Stat. Plann. Inference 139 (5), 2009, pp. 1821–1832.

9. [9] L. Fahrmeir and G. Tutz, Multivariate Statistical Modelling Based on Generalized Linear Models (2nd ed.), Springer, 2001.
10. [10] R. C. Geary, Testing for normality, Biometrika 34 (1947), pp. 209–242.
11. [11] H. Hassani, E. S. Silva and R. Gupta, Forecasting bifurcations and multimodality in economics and finance, Physica A 428 (2015), pp. 270–279.
12. [12] P. J. Huber, Robust Statistics, John Wiley, New York, 1981.
13. [13] I. Kiss and E. Zsoter, Bimodal distributions in financial time series, Comput. Econ. 25 (3), 2005, pp. 215–230.
14. [14] M. L. T. Lee, F. C. Kuo and G. A. Whitmore, Skewed distributions arising from mixtures of normal distributions, Can. J. Stat. 26 (4), 1998, pp. 747–763.
15. [15] G. Li and Y. Wang, Mixture models for heterogeneity in biomedical data analysis, J. Biopharm. Stat. 22 (5), 2012, pp. 1050–1063.
16. [16] E. Limpert, W. A. Stahel and M. Abbt, Log-normal distributions across the sciences: Keys and clues, BioScience 51 (5), 2001, pp. 341–352.
17. [17] V. Mameli and M. Musio, Some New Results on the Beta Skew-Normal Distribution, Topics in Theoretical and Applied Statistics, 2016, pp. 33–45.
18. [18] D. C. Montgomery, Design and Analysis of Experiments (5th ed.), John Wiley & Sons, New York, 2000.
19. [19] E. S. Pearson, The analysis of variance in cases of nonnormal variation, Biometrika 23 (1932), pp. 114–133.
20. [20] M. F. Schilling, A. E. Watkins and W. Watkins, Is human height bimodal?, Am. Stat. 56 (3), 2002, pp. 223–229.
21. [21] S. R. Searle, G. Casella and C. E. McCulloch, Variance Components (2nd ed.), Wiley, 2009.
22. [22] B. Senoglu, Estimating parameters in one-way analysis of covariance model with short-tailed symmetric error distributions, J. Comput. Appl. Math. 201 (2007a), pp. 275–283.
23. [23] B. Senoglu, Robust estimation and hypothesis testing of linear contrasts in analysis of covariance with stochastic covariates, J. Appl. Stat. 34 (2007b), pp. 141–151.
24. [24] B. Senoglu and M.D. Avcoglu, Analysis of covariance with non-normal error terms, Int. Stat. Rev. 77 (3), 2009, pp. 366–377.
25. [25] B. Senoglu and S. Acitas, statistiksel Deney Tasarm: Sabit Etkili Modeller, Nobel Yayn, Ankara, 2010.
26. [26] H. S. Sazak, M. L. Tiku and M. Q. Islam, Regression analysis with a stochastic design variable, Int. Stat. Rev. 74 (1), 2006, pp. 77–88.
27. [27] W. Y. Tan and M. L. Tiku, Sampling distributions in terms of Laguerre polynomials with applications, New Age International (formerly Wiley Eastern), New Delhi, 1999.
28. [28] M. Tiku and A. Akkaya, Robust Estimation and Hypothesis Testing, Age International, New Delhi, 2004.
29. [29] D. Van den Bergh and G. Molenberghs, Stochastic covariates in repeated measures models: A comparison of methods, Biometrics 56 (4), 2000, pp. 1211–1218.