İSTATİSTİKTE FREKANSÇI VE BAYESYEN YAKLAŞIMIN ÖRNEKLEM BÜYÜKLÜĞÜ ÜZERİNDEKİ ETKİLERİNİN KARŞILAŞTIRILMASI: METODOLOJİK ÇALIŞMA

Abstract

Amaç: Bu çalışmada, frekantist ve Bayesyen yaklaşımlar kullanılarak örneklem büyüklüğünün araştırma sonuçları üzerindeki etkilerinin araştırılması amaçlanmıştır. Gereç ve Yöntem: Çalışmamızda küçük ve büyük örneklem büyüklüğünde istatistiksel farklılıkları karşılaştırmak amacı ile küçük ve büyük olmak üzere iki örneklem oluşturulmuştur. Çalışmaya alınan tüm hastalar erkek ve 40 ile 50 yaş aralığındadır. Küçük örneklem için iskemik kalp hastalığı (İKH) olan 32, İKH olmayan 37 kişi çalışmaya dahil edilmiştir. Büyük örneklem için İKH olan 355, olmayan 545 kişi çalışmaya alınmıştır. Tüm hastalarn glukoz, trigliserid (TG), total kolesterol (TKOL), yüksek yoğunluklu lipoprotein kolesterol (HDL), düşük yoğunluklu lipoprotein kolesterol (LDL), üre, kreatinin, hemoglobin, hematokrit (HCT), kırmızı kan hücresi dağılım genişliği (RDW), lökosit (WBC), trombosit (PLT), ortalama trombosit hacmi (MPV), nötrofil (NÖT), lenfosit (LYM) değerleri kaydedilmiştir. Küçük ve büyük örneklemler frekansçı ve Bayesyen yaklaşımla karşılaştırılımıştır. Bulgular: Küçük örneklem büyüklüğünde frekantist yaklaşım ile yapılan analizde tüm biyokimyasal veriler İKH olan ve olmayan kişilerde karşılaştırılmış ve glukoz seviyeleri dışında diğer parametrelerde anlamlı fark saptanmamıştır. Yine grupların Bayesyen yaklaşımla yapılan karşılaştırmalarında parametreler arasında anlamlı istatistiksel fark elde edilmemiştir. Buna karşın büyük örneklem büyüklüğünde frekantist yaklaşım ile yapılan karşılaştırmalarda glukoz, TG, TKOL, HDL, LDL, üre, kreatinin, hemoglobin, HCT, WBC, NÖT ve LYM değerleri her iki grup arasında anlamlı olarak farklı çıkmıştır. Aynı şekilde Bayesyen yaklaşım ile yapılan karşılaştırmalarda glukoz, TG, TKOL, HDL, LDL , üre, kreatinin, hemoglobin, HCT, WBC, NÖT ve LYM değerleri iki grup arasında istatistiksel olarak anlamlı çıkmıştır. Sonuç: Büyük örneklem büyüklüğünde ve yüksek bir güçte çalışmada verinin frekansçı ya da Bayesyen istatistik ile değerlenirilmesi açısından fark bulunmamamaktadır.

Keywords

• Frekantist
• bayesyen
• örneklem büyüklüğü

Project Number

PROJE NUMARASI YOKTUR

COMPARISON OF FREQUENTIST AND BAYESIAN APPROACHES ON SAMPLE SIZE: METHODOLOGIC STUDY

Abstract

Objective: In the present study, we aimed to evaluate the effects of sample size on results of study by using frequentist and Bayesian approaches. Material and Methods: The small sample consisted of 32 patients with ischemic heart disease (IHD) and 37 control subjects. In order to compare the statistical differences between small and large sample sizes, two samples were constituted. All the patients included in the study were male and between 40-50 years old. The large sample consisted of 355 IHD patients and 545 controls. Patients’ biochemical variables including glucose, triglyceride (TG), total cholesterol (TC), high density lipoprotein cholesterol (HDL), low density lipoprotein cholesterol (LDL), urea, creatinine, hemoglobin, hematocrit, (HCT), red cell distribution width (RDW), White blood cell (WBC), platelet (PLT), mean platelet volume (MPV), neutrophil (NEUT), lymphocyte (LYM) were recorded. Patients in the small and large samples were compared with both frequentist and Bayesian approaches. Results: Except for glucose levels there were no statistical differences with respect to the biochemical variables of two groups in a small sample size when the variables were analyzed by the frequentist approach. Similarly, we did not find any differences between biochemical variables when the data were analyzed by the Bayesian approach. When the large sample size data were analyzed by the frequentist approach, glucose, TG, TC, HDL, LDL, urea, creatinine, hemoglobin, HCT, WBC, NEUT, LYMP levels were found to be statistically significantly different between patients who had IHD and the controls. Similarly, there were significant differences between two groups with respect to glucose, TG, TC, HDL, LDL, urea, creatinine, hemoglobin, HCT, WBC, NEUT, LYMP levels when the data analyzed by Bayesian approach. Conclusion: Our study results suggested that there were no differences between the frequentist and Bayesian approach results when the sample size is large and the power of the study is high.

Keywords

• Frequentist
• bayesian
• sample size

Project Number

PROJE NUMARASI YOKTUR

References

1. Rindskopf D. Overview of Bayesian statistics. Eval Rev 2020;44(4):225-37. google scholar
2. Beerenwinkel N, Siesbourg J. Probability, statistics, and computational science methods. Methods Mol Biol 2019;1910:33-70. google scholar
3. 3. Sami W, Alrukban MO, Waqas T, Asad MR, Afzal K. Sample size determination in health research. J Ayub Med Coll Abbottabad 2018;30(2):308-11 google scholar
4. Sacco W. Statistical power considerations in the use of cost-effectiveness analysis. Professional Psychology 1982;13(5):752-8. google scholar
5. Mazen AM, Graf LA, Kellog KE, Hemmasi M. Statistical power in contemporary management research. The Academy of Management Journal 1987;30(2):369-80. google scholar
6. Kim HY. Statistical notes for clinical researcher: Type I and type II errors in statistical decision. Restor Dent Endod 2015;40(3):249-52. google scholar
7. Mozaffarian D, Benjamin E, Go A, Arnett DK, Blaha MJ, Cushman M, et al. Heart disease and stroke statistics-2016 update. A report from the American Heart Association. Circulation 2016;133(4):e38-360. google scholar
8. Roth GA, Johnson C, Abajobir A, Abd-Allah F, Abera SF, Abyu G, et al. Global, regional, and national burden of cardiovascular diseases for 10 causes, 1990 to 2015. J Am Coll Cardiol 2017;70(1):1-25. google scholar

9. Sweetnam PM, Thomas HF, Yarnell JW, Baker IA, Elwood PC. Total and differential leukocyte counts as predictors of ischemic heart disease: The caerphilly and speedwell Studies. Am J Epidemiol 1997;145(5):416-21. google scholar
10. Weijenberg MP, Feskens EJ, Kromhout D. White blood cell count and the risk of coronary heart disease and all-cause mortality in elderly men. Arterioscler Thromb Vasc Biol 1996;16(4):499-503. google scholar
11. 11Gupta R, Gupta VP, Sarna M. Bhatnagar S, Thanvi J, Sharma V, et al. Prevalence of coronary heart disease and coronary risk factors in urban Indian population: Jaipur Heart Watch -2. Indian Heart J 2002;54(1):59-66. google scholar
12. anjith MP, Divya R, Metha VK, Krishnan MG, KamalRaj R, Kavishwar A. Significance of platelet volume indices and platelet count in ischaemic heart disease. J Clin Pathol 2009;62(9):830- 3. google scholar
13. Pizzulli L, Yang A, Martin JF, Lüderitz B. Changes in platelet size and count in unstable angina compared to stable angina or non-cardiac chest pain. Eur Heart J 1998;19(1):80-4. google scholar
14. Houghton DE, Koh I, Ellis A, Key NS, Douce DR, Howard G, et al. Hemoglobin levels and coronary heart disease risk by age, race, and sex in the reasons for geographic and racial differences in stroke study (REGARDS). Am J Hematol 2020;95(3):258-6. google scholar
15. Stefan AM, Gronau QF, Schönbrodt FD, Wagenmakers EJ. A tutorial on Bayes Factor Design Analysis using an informed prior. Behav Res Methods 2019;51(3):1042-58 google scholar
16. Biau DJ, Jolles BM, Porcher R. P value and the theory of hypothesis testing: an explanation for new researchers. Clin Orthop Relat Res 2010;468(3):885-92 google scholar
17. Pernet C. Null hypothesis significance testing: a short tutorial. F1000Res 2015;4:621. google scholar
18. Greenland S, Senn SJ, Rothman KJ, Carlin JB, Poole C, Goodman SN, et al. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations Eur J Epidemiol 2016;31(4):337-50. google scholar
19. Domenech RJ. The uncertainties of statistical "significance" Rev Med Chil 2018;146(10):1184-9. google scholar
20. Morey RD, Romeijn JW, Rouder JN. The philosophy of Bayes factors and the quantification of statistical evidence. J Math Psychol 2016;72:6-18. google scholar
21. Kelter R. Analysis of type I and II error rates of Bayesian and frequentist parametric and nonparametric two-sample hypothesis tests under preliminary assessment of normality. Computational Statistics 2021;36:1263-88. google scholar
22. Kelter R. Power analysis and type I and type II error rates of Bayesian nonparametric two-sample tests for location-shifts based on the Bayes factor under Cauchy priors. Computational Statistics Data Analysis 2022;165(21):107326. google scholar