การเปรียบเทียบวิธีการประมาณค่าพารามิเตอร์สำหรับแผนแบบสุ่มตลอดในบล็อกสมบูรณ์อิทธิพลคงที่กรณีข้อมูลระยะยาว / วิลาสินี จันทราวุฒิ = A comparison of parameter-estimation methods for a fixed-effect randomized complete block design with longitudinal data / Wilasinee Chantrawut
The objective of this research is to study and to compare the parameter-estimation methods for fixed-effect randomized complete block design with longitudinal data by the Ordinary Least Square estimation (OLS), Two-Stage estimation (TS) and Maximum Likelihood estimation (MLE). The model is Yijk = tau i + beta j + alpha k + tau alpha ik + epsilon ijk, which epsilon ijk are independently distributed and epsilon ijk follow the first-order autoregressive model, epsilon ijk = phi epsilon ijk-1 + u ijk, The comparison is done when data were generated with phi are 0, 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95 and 0.99. The variances are 1, 25, and 100 with the 3x3, 4x4, and 5x5 designs. The data are simulated by Monte Carlo technique and repeated 500 times for each situation to calculate for the average of Euclidean distance (Eu) of parameter estimator in design and autoregressive parameter and Mean Square Error of variance. The conclusions of this research are 1. Case of independently distributed errors. Euclidean distance by Ordinary Least Square estimators and Two-Stage estimators are almost the same as Maximum likelihood estimators with no estimation in autoregressive arameter in all cases. The value of Eu and MSE(sigma^2e) will decrease when collect the data for more replicate and size of design increasing. But Eu and MSE(sigma^2e) will decrease when the variance is increasing. 2. case of first-order autoregressive errors. The Maximum likelihood estimators have minimum values of Eu, MSE(phi^) and MSE(sigma^2e) in all cases. Teh error of estimation will increase when the autoregressive parameter is increasing or the number of replicate is decreasing or the sample size is decreasing.
