置回抽樣與水稻產量穩定性

Abstract

目前常用於區域試驗資料的穩定性分析方法，依照統計概念分為有母數分析方法 (例如聯合回歸分析)、無母數分析方法 (例如Nassar’s method等)，以及採取與對照組比較的分析方法 (例如優勢分析)。由於不同分析方法的使用時機與統計概念不同，容易在選用上出現疑慮，或在分析結果上產生判讀的困難，因此本研究選用不同類型的五種穩定性分析方法，包含Finlay and Wilkinson regression analysis、Nassr’s method、Kang’s method、Wilcoxon signed rank test、Superiority analysis，運用三筆臺灣水稻區域試驗的產量資料，進行實際的穩定性分析運算，發現不同方法在同樣資料的情況下，會產生不同結論而產生判讀的困難，因此後續再搭配置回抽樣的使用，取得不同的環境組合後再利用穩定性分析，發現置回抽樣後，原先結果互有差異的穩定性分析方法產生趨同的結論。本研究最後也將針對不同的穩定性分析方法進行討論，並對於現行的區域試驗產量資料分析提出建議，期望能協助未來水稻育種在區域試驗上有正確的分析判讀，而能選拔出高產、高品質的新品種。

The stability analysis method which were adopted frequently divided into parametric statistics, such as the joint regression analysis; nonparametric statistics, such as the Nassar’s method, and the method needs to take comparison with the check, such as the superiority analysis. According to the different statistical principle and timing of usage, breeders will confused by the choice of analysis methods or feel difficult to identify the result correctly. Thus, this study find five stability analysis method from different statistical principles, involve Finlay and Wilkinson regression analysis, Nassr’s method, Kang’s method, Wilcoxon signed rank test, and superiority analysis; use three datasets of multiple environments trial practiced in Taiwan to identify the stability properties of the potential lines. We found that the results from different methods truly lead to different conclusion and generate the confusion. As a result, we get one thousand bootstrap samples from the environments combination to make a larger scale of datasets, then do the same analysis by five analysis methods. The result of stability analysis with bootstrap become similar, and have an easier-identified conclusion. Finally, this study discuss about the stability analysis as well and submit the suggestion of data analysis by currently multiple environments trial, expect can help the rice breeder to get the correct conclusion to improve the yield and quality of rice.