Skip to main content
ARS Home » Southeast Area » Byron, Georgia » Fruit and Tree Nut Research » Research » Publications at this Location » Publication #301420
Title: Disease severity estimates - effects of rater accuracy and assessments methods for comparing treatments

Author
item Bock, Clive
item EL JARROUDI, MOUSSA - University Of Liege
item KOUADIO, LOUIS - Agriculture And Agri-Food Canada
item MACKELS, CHRISTOPHE - University Of Liege
item CHIANG, KUO-SZU - Chung Hsing University
item DELFOSSE, PHILIPPE - University Of Luxembourg

Submitted to: Plant Disease
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/13/2014
Publication Date: 8/1/2015
Citation: Bock, C.H., El Jarroudi, M., Kouadio, L.A., Mackels, C., Chiang, K., Delfosse, P. 2015. Disease severity estimates - effects of rater accuracy and assessments methods for comparing treatments. Plant Disease. 99:1104-1112.

Interpretive Summary: Assessment of disease is fundamental to the discipline of plant pathology, and estimates of severity are often made visually. However, it is established that visual estimates can be inaccurate and unreliable. In this study estimates of Septoria leaf blotch on leaves of winter wheat from non-treated and fungicide treated plots assessed by four raters were compared to true values. Raters generally showed good to acceptable agreement with true values, although one rater was particularly poor. Consequently, estimates of mean disease severity on control and treated plots differed among raters and image analysis, and although a comparison of disease severity estimated on the two treatments demonstrated significant effects of fungicide, the estimates by two of the raters showed that the fungicide treated plots actually had more severe disease compared to the control plots. Furthermore, comparing mean estimates of disease on control and treated plots demonstrated type II errors. Regression analysis of yield against disease severity showed a significant effect of disease on yield, but variability in data was rater-dependent and analysis confirmed significant differences between slopes and intercepts of the relationships among raters. These results demonstrate the need for accurate and reliable disease assessment to minimize errors and to ensure disease severity/yield relationships are faithfully represented, and where multiple raters are deployed, they should each assess specific replicates of all treatments.

Technical Abstract: Assessment of disease is fundamental to the discipline of plant pathology, and estimates of severity are often made visually. However, it is established that visual estimates can be inaccurate and unreliable. In this study estimates of Septoria leaf blotch on leaves of winter wheat from non-treated and fungicide treated plots assessed by four raters were compared to true values measured using image analysis for accuracy and precision using Lin’s concordance correlation (LCC, 'c) in both 2006 and 2007. The mean estimates of disease severity were analyzed using general linear modeling for the estimates by each rater and the true values from image analysis. The relationship between disease severity estimated by each rater or images analysis and yield (2007 only) was explored with linear regression. Raters 1 and 2 showed mostly good agreement with true values ('c = 0.038 to 0.994), while rater 3 and 4 had less good agreement ('c = 0.047 to 0.718). Consequently, the estimates of mean disease severity on control and treated plots differed among raters and image analysis (F=11.0 to 22.6, P=<0.0001), and although a comparison of disease severity estimated on the two treatments by image analysis and raters 1 to 4 all demonstrated significant effects of fungicide (F=7.5 to 37.8, P=<0.0001 to 0.006), the estimates by rater 3 and 4 in 2006 showed that the fungicide treated plots had more severe disease compared to the control plots. Furthermore, all combinations of pairwise analysis of image analysis measured and rater estimates of treated and control plots demonstrated type II errors among raters. In 2007, regression analysis of yield against disease severity showed a significant effect of disease on yield for all assessment methods/raters (F=17.0 to 145.6, P=<0.0001). The coefficient of variation was least for image analysis (CV=13.4), and greatest for the regression based on rater estimates (14.0 to 20.5). A dummy variable regression analysis confirmed significant differences between slopes and intercepts of the relationships based on severity data from image analysis or raters. These results demonstrate the need for accurate and reliable disease assessment to minimize errors and to ensure disease severity/yield relationships are faithfully represented, and where multiple raters are deployed, they should each assess specific replicates of all treatments.