Akademik Veri Yönetim Sistemi
1ST INTERNATIONAL CONFERENCE ON MATHEMATICAL AND RELATED SCIENCES (ICMRS), Antalya, Türkiye, 30 Nisan - 04 Mayıs 2018, cilt.1991
In this study, three different X1, X2 and X3 data sets with a mean of 0, variance of 1 and 1000 samples, which were obtained by simulation and suitable for Gaussian distribution, were used. All three data sets are designed as two levels of A and B (A) in nested design. The variance components of simulated X1, X2, and X3 were assigned as sigma(2)(A):0.025, sigma(2)(B(A)):0.025, and sigma(2)(E):0.95 in X1, sigma(2)(A):0.25, sigma(2)(B(A)):0.25 and sigma(2)(E):0.50 in X2, and sigma(2)(A):0.375, sigma(2)(B(A)):0.375, and sigma(2)(E)0.25 in X3, respectively. The single chains of 200000 iterations were considered with the 20000 cycles of burn-in periods and, different thinning intervals of 18, 36, 72 cycles to result 10000, 5000, 2500 posterior samples of each parameters of interest in total for XI, X2 and X3. In result, small bias values (less than 5%) are detected only in all three chains of X3. It is revealed that the range values are well established m chains diluted from 180000 to 10000, 5000 and 2500, due to no autocorrelation was detected in any of the different variable by thinning interval combinations used in this study. Furthermore, it is important that the lowest bias values are calculated in the variable X3 where 25% of the total variance belongs to the residual. Thus, it was determined that the shares of variance components influenced estimations than thinning intervals in Bayesian analyses.