Journal of Taiwan Agricultural Research

Author:Hsiu-Ying Lu

Abstract:

    Linear regression analysis is commonly used to assess the relative stability of varieties grown at different regions in Taiwan. However, the conventional least squares (LS) regression is susceptible to the occurrence of outliers (or unusual observations), which may have a deleterious effect on estimates of regression coefficients and on homogeneity of residual mean squares from the regression. Thus, outliers may have a significant impact on the precision of stability analysis. Many diagnostic statistics have been designed to detect outliers. They are classified into two approaches: classical regression diagnostics and recently developed robust regression. However, they are never applied in stability analysis of regional trial data. To investigate their applicability, demonstration of several diagnostics for the outlier detection using the barley data of Yates and Cochran (1938) was performed. The results revealed that the residuals from LS fits are not useful as outlier diagnostics, whereas the robust regression is useful in screening data sets and identifying outliers.

Key words:Regional trial, Stability, Outlier Diagnostics, Robust regression, Reweighted least squares regression

Download: PDF Links