A comparison of raters and disease assessment methods for estimating disease severity for purposes of hypothesis testing.

Assessment of disease severity is most often made visually, and estimates can be inaccurate. Nearest percent estimates (NPEs) of Septoria leaf blotch on leaves of winter wheat by four raters (R1-R4) assessing non-treated (NT) and fungicide-treated (FT) plots were compared to true values using Lin’s concordance correlation coefficient (ρc) on two dates in 2006 and 2007. Estimates were converted to Horsfall-Barratt (HB) mid-points and again compared for accuracy and precision. Estimates of severity from FT and NT plots were analyzed to ascertain effects of rater using both the NPE and HB values. Regardless of method, all raters showed a range of agreement with true values on FT and NT plots (ρc = 0 to 1). Use of the HB scale most often reduced agreement (84.4% of the time), and did not improve rater-associated bias of treatment mean severity estimates. Consequently, estimates of mean severity differed significantly among raters and from true values (F=126 to 1260, P=0.002 to<0.0001). However, a comparison of treatment effects showed that the true values and R1 to R4 all demonstrated significant effects of fungicide (F=101 to 1952, P=0.002 to <0.00001). Ranking of raters differed on one occasion when HB values were used. These results demonstrate the effect of the HB scale, and the need for accurate disease assessment to minimize over or underestimates compared to true severity so as to minimize the potential for type II errors.