COMPARING FREQUENTIST AND BAYESIAN APPROACHES FOR MIXED DESIGN ANOVA IN REPEATED MEASUREMENTS: A SIMULATION STUDY WITH EXPONENTIAL DISTRIBUTIONS


Özel Z., Kaya Başar E., Tekindal M. A.

12th International Conference of the International Biometric Society's Eastern Mediterranean Region, EMR2023, İzmir, Turkey, 8 - 11 May 2023, pp.48

  • Publication Type: Conference Paper / Summary Text
  • City: İzmir
  • Country: Turkey
  • Page Numbers: pp.48
  • Akdeniz University Affiliated: Yes

Abstract

Mixed design analysis of variance in repeated measurements allows the examination of data obtained from the same measurement performed multiple times. Bayesian and frequentist approaches are two different methods used for estimating a parameter for repeated measurements. In this study, the results obtained from bayesian and frequentist approaches for repeated measurements in balanced and unbalanced samples were investigated. The aim of this study is to compare the results of both frequentist and bayesian approaches using the mixed design ANOVA method for data obtained from four repeated exponential distributions with different coefficient of variations and balanced/unbalanced designs (4x4). Datasets with coefficient of variation of 0.5 and 1 were generated from the exponential distribution for the study design, which consists of balanced and unbalanced samples. The inclusion of unbalanced design aims to achieve positive heterogeneity by increasing the variance. The simulation study was repeated 1000 times. Evaluations were made using both frequentist and bayesian approaches, and the results were evaluated according to confidence intervals, and the approaches were compared. Bayesian and frequentist approaches offer different methods for estimating a parameter in repeated measurements. Both methods have advantages and disadvantages. The choice of appropriate method may depend on the availability of prior information; Bayesian approach may yield more accurate results if prior information is available, while frequentist approach may be more appropriate if no prior information is available. In conclusion, both approaches can be used as alternatives to each other in repeated measurements.