比例勝算模型在比較處理情境下病害嚴重度估計之運用

Abstract

在植物流行病學的研究領域裡，針對植物病害的探討時常需要估計植物的罹病程度，一般將罹病程度的高低(以百分比表示之)稱之為植物病害嚴重度(罹病度)；植物病害嚴重度(罹病度)是一種量化的數據資料，我們可使用這些量化的病害數據去推論相關之後續研究，如預測作物產量的損失、不同處理病害方式之間的比較及植物流行病學的監控與預測等研究。本研究目的是使用「比例勝算模型」來比較不同處理間(例如：品種、藥劑等)之差異，在此使用模擬方式來進行，並採用柑橘潰瘍病田間調查的數據資料來估計其參數，目的來比較以「比例勝算模型」與其他不同類別尺度之組中點計算，其標準為假設檢定之檢定力。結果發現「比例勝算模型」在低病害嚴重度時(小於30%)的檢定力表現最佳，而隨著病害嚴重度(罹病度)提升，「比例勝算模型」的檢定力也跟著下降，但其檢定力表現仍然優於以組中點計算的類別尺度；當真實病害嚴重度為30%時，使用「比例勝算模型」或組中點計算來處理修正之10％等距劃分病害尺度（即病害嚴重度在50％以下10％等距劃分，加上低病害嚴重度時再加以細分），其結果相差有限，但對於處理另一比較尺度Horsfall-Barratt (H-B)尺度時，「比例勝算模型」有較差的檢定力。相信此研究能提供較精確的植物病害嚴重度估計值之比較，冀望對植物病害研究領域的發展有所助益。

In the field of plant epidemiology, plant diseases severity generally needs to be assessed. Severity may be defined as the area of plant tissue affected by disease and may be expressed as quantitative data such data can be used to do research to predict yield loss, to compare treatments, and for monitoring and forecasting of plant diseases. The purpose of this study is to evaluate the performance of the proportional odds model in estimating disease severity for the purpose of comparing treatments (e.g., varieties, fungicides, etc.). A simulation method was employed to execute the study. The parameters of the simulation were estimated using the original data from the field based on a sample of citrus canker (Xanthomonas citri subsp. Citri)-infected grapefruit leaves. The proportional odds model was compared with the method using midpoint conversions of ordinal intervals. Here, the criteria for comparison is the power of hypothesis testing. The results of this study show that, at low disease severity (≤30%), the performance of the proportional odds model is superior to that of the midpoint conversion of the interval. However, as disease severity increases, the power of hypothesis testing by using the proportional odds model decreases. As the actual disease severity approaches 30%, the advantage of the proportional odds model will gradually disappear although it is still slightly better than using the categorical type of amended 10% scale, but is not better than the Horsfall-Barratt (H-B) scale. We hope that the results of this study will be helpful in improving the accuracy of disease severity assessment in plant epidemiology.