Premium
Accuracy and Reliability of Severity Estimates Using Linear or Logarithmic Disease Diagram Sets in True Colour or Black and White: a Study Case for Rice Brown Spot
Author(s) -
Schwanck André Aguiar,
Del Ponte Emerson Medeiros
Publication year - 2014
Publication title -
journal of phytopathology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.53
H-Index - 60
eISSN - 1439-0434
pISSN - 0931-1785
DOI - 10.1111/jph.12246
Subject(s) - intraclass correlation , concordance , diagram , statistics , logarithm , mathematics , concordance correlation coefficient , reliability (semiconductor) , correlation , kappa , combinatorics , reproducibility , medicine , physics , mathematical analysis , geometry , power (physics) , quantum mechanics
Four disease diagram sets to aid in assessments of rice brown spot severity were developed and evaluated with regard to accuracy and reliability of the estimates. These sets had severity increments in a linear ( LIN ) or logarithmic ( LOG ) fashion with the diagrams depicted in true colour ( COL ) or black and white ( BW ). Ten inexperienced raters evaluated one of the sets, totalling forty raters. On each group of raters, the assessment was made first unaided and then aided with the respective diagram set. The minimum and the maximum severities of the diagram sets were 0.5 and 36%, respectively, and the five intermediate severity values were 6, 12, 18, 24 and 32% for the LIN diagram sets, and 1.8, 3.3, 6, 11 and 20% for the LOG sets. Overall agreement, measured by the L in's concordance correlation coefficient ( ρ c ), increased considerably when using the aid (unaided mean ρ c = 0.53, aided mean ρ c = 0.87) due to a strong reduction in the systematic bias, measured by the bias correction factor C b (unaided mean C b = 0.60, aided mean C b = 0.95). All diagrams led to similar accuracy and precision, but a consistent overestimation was still observed when using the LIN sets, and variability for the absolute errors was higher for the LOG sets, compared with the LIN sets. Estimates using the diagram sets were more reliable based on the intraclass correlation (mean ρ = 0.79–0.86) compared with unaided estimates (mean ρ = 0.51–0.67). Raters exhibited preference for specific values, such as the ‘knots’ (10, 20, 30%, etc.), and the severity values represented in the diagrams, especially when using the LIN sets. The diagram sets similarly helped to improve accuracy and reliability of estimates of rice brown spot epidemics.