Comparison of Piecewise Regression and Polynomial Regression Analyses in Health and Simulation Data Sets

Abstract

Objective: Piecewise regression, which one or more pieces are combined in breakpoints, is widely used as a statistical technique. It was aimed to compare piecewise regression analyses and polynomial regression analysis using both simulated data and real data sets.

Material-Method: In the application step of the study, algorithms were created by using R software for simulation practice. Polynomial and piecewise regression analysis methods were compared using data sets with n=100 units and 1000 times running simulation. In addition, estimation performances of piecewise and polynomial regression built by using the data sets which contained in the number of tuberculosis cases according to age in 2010 year and the number of measles cases from 1993 to 2015 years in Turkey were compared.

Results: It was found that there was a significant difference between all of the polynomial and piecewise regression models (p<0.001). The  values of piecewise regression models were higher than polynomial regression models; MSE, AIC and BIC values were observed to be lower. According to the result of both simulation and real data set applications, piecewise regression models which were generated according to optimal knots were found to have better estimation performance than polynomial regression models according to , MSE, AIC and BIC criteria.

Conclusions: This study revealed that data analysis with piecewise regressions having optimal knots provided statistically superiority, although polynomial regression methods are preferred in the field of health studies mostly.

Keywords

• Piecewise Regression,Simulation,Tuberculosis,Measles,Polynomial Regression

Comparison of Piecewise Regression and Polynomial Regression Analyses in Health and Simulation Data Sets

Abstract

Amaç: Bir veya daha fazla parçanın kırılma noktalarında birleştirildiği parçalı regresyon, istatistiksel bir teknik olarak yaygın bir şekilde kullanılmaktadır. Bu çalışmada hem simülasyon verisi hem de gerçek veri setleri kullanılarak tek değişkenli polinom regresyon analizi ile karesel ve kübik parçalı regresyon analizlerinin karşılaştırılması hedeflendi.

Materyal-Metot: Çalışmanın uygulama basamağında R yazılım programı kullanılarak simülasyon uygulaması için algoritmalar yazıldı. Polinom ve sürekli parçalı regresyon analiz yöntemlerinin karşılaştırılması n=100 birimlik veri setleri için 1000 tekrarlı simülasyon ile gerçekleştirildi. Ayrıca Türkiye’de 2010 yılındaki tüberküloz vaka sayılarını içeren tüberküloz veri seti ile Türkiye’deki 1973-2010 yılları arasındaki kızamık vaka sayılarını içeren kızamık veri setleri kullanılarak oluşturulan polinom ve parçalı regresyon modellerinin tahmin performansları; , HKO, ABK ve BBK değerlerine göre karşılaştırıldı.

Bulgular: Tüm polinom ve parçalı regresyon modellerinin , HKO, ABK ve BBK değerleri bakımından performansları istatistiksel olarak birbirinden farklı bulundu (p<0.001). Parçalı regresyon modellerinin  değerlerinin polinom regresyon modellerine göre daha yüksek; HKO, ABK ve BBK değerlerinin ise daha düşük olduğu gözlendi.  Gerçek veri setleri ile yapılan uygulamalarda en uygun dönüm noktalarına göre oluşturulan tüm parçalı regresyon modellerinin  değerlerinin polinom regresyonlardan daha yüksek; HKO, ABK ve BBK değerlerinin ise daha düşük olduğu belirlendi Oluşturulan parçalı regresyon modellerinin veri setlerini polinom regresyonlara göre daha iyi tahmin ettiği belirlendi.

Sonuç: Sağlık alanında yapılan çalışmaların çoğunda polinom regresyon yöntemlerinin tercih edilmesine rağmen bu çalışma ile en uygun dönüm noktalı parçalı regresyonlarla veri analizinin istatistiksel açıdan üstünlük sağladığı uygulamalarla ortaya konmuştur.

Keywords

• Parçalı Regresyon,Simülasyon,Tüberküloz,Kızamık,Polinom Regresyon

References

1. Freedman DA. Statistical models: theory and practice: cambridge university press; 2009.
2. Freund RJ, Wilson WJ, Sa P. Regression analysis: Academic Press; 2006.
3. Hartley HO, Booker A. Nonlinear least squares estimation. The Annals of mathematical statistics. 1965;36(2):638-50.
4. Seber G, Wild C. Nonlinear regression. 2003. Hoboken: John Wiley & Sons Google Scholar. 2003.
5. Park SH. Experimental designs for fitting segmented polynomial regression models. Technometrics. 1978;20(2):151-4.
6. Wainer H. Piecewise regression: A simplified procedure. British Journal of Mathematical and Statistical Psychology. 1971;24(1):83-92.
7. Eubank R. Approximate regression models and splines. Communications in Statistics-Theory and Methods. 1984;13(4):433-84.
8. Gallant AR, Fuller WA. Fitting segmented polynomial regression models whose join points have to be estimated. Journal of the American Statistical Association. 1973;68(341):144-7.

9. Berberoglu B, Berberoglu CN. Modeling the Structural Shifts in Real Exchange Rate with Cubic Spline Regression (CSR). Turkey 1987-2008. International Journal of Business and Social Science. 2011;2(17).
10. De Boor C, Rice JR. Least squares cubic spline approximation, II-variable knots. 1968.
11. Poirier DJ. Piecewise regression using cubic splines. Journal of the American Statistical Association. 1973;68(343):515-24.
12. Porth RW. Application of least square cubic splines to the analysis of edges. 1984.
13. Schwetlick H, Schütze T. Least squares approximation by splines with free knots. BIT Numerical mathematics. 1995;35(3):361-84.
14. Draper NR, Smith H. Applied regression analysis: John Wiley & Sons; 2014.
15. Harrell FE, Lee KL, Califf RM, Pryor DB, Rosati RA. Regression modelling strategies for improved prognostic prediction. Statistics in medicine. 1984;3(2):143-52.
16. Chan S-h. Polynomial spline regression with unknown knots and AR (1) errors: The Ohio State University; 1989.
17. De Boor C, Rice JR. Least squares cubic spline approximation I-Fixed knots. 1968.
18. Eubank RL. Nonparametric regression and spline smoothing: CRC press; 1999.
19. Wold S. Spline functions in data analysis. Technometrics. 1974;16(1):1-11.
20. Hawkins DM. On the choice of segments in piecewise approximation. IMA Journal of Applied Mathematics. 1972;9(2):250-6.
21. Ruppert D. Selecting the number of knots for penalized splines. Journal of computational and graphical statistics. 2002;11(4):735-57.
22. Agarwal GG, Studden W. An algorithm for selection of design and knots in the response curve estimation by spline functions: Purdue University. Department of Statistics; 1978.
23. Marsh LC, Cormier DR. Spline regression models: Sage; 2001.
24. Powell M. The local dependence of least squares cubic splines. SIAM Journal on Numerical Analysis. 1969;6(3):398-413.
25. Smith PL. Splines as a useful and convenient statistical tool. The American Statistician. 1979;33(2):57-62.
26. Wegman EJ, Wright IW. Splines in statistics. Journal of the American Statistical Association. 1983;78(382):351-65.
27. Genç A, Oktay E, Alkan Ö. İhracatın İthalatı Karşılama Oranlarının Parçalı Regresyonlarla Modellenmesi. Atatürk Üniversitesi Sosyal Bilimler Enstitüsü Dergisi. 2012;16(1).
28. Markov D. Information content in stock market technical patterns: A spline regression approach. 2003.
29. Marsh LC. Estimating the number and location of knots in spline regressions. Journal of Applied Business Research. 1986;3:60-70.
30. Studden WJ, VanArman D. Admissible designs for polynomial spline regression. The Annals of Mathematical Statistics. 1969;40(5):1557-69.
31. Hurley D, Hussey J, McKeown R, Addy C, editors. An evaluation of splines in linear regression. The 132nd Annual Meeting; 2004.
32. Mulla Z. Spline regression in clinical research. West indian medical journal. 2007;56(1):77-9.
33. Parkhurst A, Spiers D, Hahn G. Spline models for estimating heat stress thresholds in cattle. 2002.