使用比例勝算模型在植物病害嚴重度比較處理上影響因子之探討

Abstract

病害的評估在植物流行病學的領域中是相當重要的議題之一，病害評估常被使用在預測產量損失之多寡、處理間之比較以觀察其效能、以及觀察植株抗病性育種等。進行上述評估之方式為病害嚴重度，其定義為病斑區域面積占該葉片總面積之百分比，常見於田間病害嚴重度之估計是使用目視估計的方式，但使用目視估計難以區分該葉片些微的差異，而造成近似百分比估計值難以取得且較為耗時，進而使用類別尺度作為估計值較為便利。現今大部分採用以組中點轉換之數值代表分組之區間，並進一步使用t檢定進行兩處理之比較，但使用此方法之假設前提為檢定之資料必須服從常態分布，而實際上資料型態之分布有很多種，並非全然均服從常態分布。因此本研究使用比例勝算模型進行兩處理之比較，並計算假設檢定之檢定力作為判斷方法之優劣標準，且加以探討影響此方法效能之影響因子及其相關特性，依其結果獲得在病害嚴重度小於等於40% 時，皆不遜於現今之方法；在低嚴重度 (≤10%) 進行病害評估時，若其分析之資料變動程度較大，使用比例勝算模型進行兩處理之比較皆有較佳之表現；且許多影響因子會影響其估計值，其中以資料整體的變動程度、估計之病害嚴重度的高低、使用之類別尺度、估計數值之計算方式以及數值之檢定方法等，均會影響結果。本研究之結果將可提供農業相關領域在分析資料時之參考。

Disease severity assessment is an important issue in plant epidemiology. The purposes of disease severity assessment are to forecast the yield loss, to compare the effects of treatments, and to evaluate the resistance of plants breeding. Severity is often used to estimate disease intensity. The definition of disease severity is the nearest percentage of lesions of all the area on the leaf. Because human's eyes can't easily discriminate disease severities, it is difficult to get the estimated values; also it will take more time to assess severities. Therefore, it is necessary to use the score of the categorical scale to estimate the severities. Nowadays, most previous studies have used the transformation of the midpoint of the interval scale to represent the corresponding interval. Furthermore, as two treatments compared, t-test of statistical methods is performed. When t-test is used, the assumption is the populations we are sampling have the shape of normal distributions. Otherwise, it will violate the assumption of t-test. However, in real life, there are many types of different data. There are not all data sampling from normal distributions. Base on the reasons, we used proportional odds model in this study to compare the difference of two treatments. The criterion of comparisons is the power of hypothesis testing. Our results show that, as the disease severity is ≤40%, this method of the proportional odds model is not inferior to the methods of the midpoint conversion. Even, there is a superior result by using proportional odds model, when there is a larger variation for the mean severity of the data ≤10%. Moreover, some impact factors affect the accuracy of estimates. For example, the variation of the data, actual disease severity, different categorical scale, computation of the estimated value, and the methods of hypothesis testing. Finally, the results of the study will be helpful for analyzing the data in agriculture science.