Bland-Altman comparison of two methods for assessing severity of Verticillium wilt of potato S.K.R. Yellareddygari, N.C. Gudmestad * Department of Plant Pathology, North Dakota State University, Fargo, 58105, USA article info Article history: Received 24 April 2017 Received in revised form 19 July 2017 Accepted 23 July 2017 Available online 29 July 2017 Keywords: Canopeo Correlation Mixed effects model abstract The agreement between two disease assessment approaches is important to know prior to replacing or interchanging the use of an established method with a recently developed method of measurement. Frequently used statistical methods to compare two different disease rating methods is the Pearson correlation coefficient or the ordinary least square regression (OLS), but they have their shortcomings. Bland-Altman proposed an alternative method for studying agreement between methods using simple graphs and basic statistics. Traditionally, when disease management strategies are being evaluated in the field, the severity of the disease is estimated using a visual assessment. Canopeo, designed by the Oklahoma State University app center, is a smart phone app designed for measuring green canopy cover. Thus, the aim of this study was to explain the Bland-Altman method with examples of visual and Canopeo methods of wilt measurement. Symptoms of Verticillium wilt in potato were estimated (repeated measures) in two trials using Canopeo and a traditional visual assessment method. Complete wilt data (repeated measures) were considered for studying the agreement between visual and Canopeo assessments. A preset cutoff limit of 5% bias (total allowable) between rating methods was considered acceptable prior to using the Bland-Altman comparison. The Bland-Altman method for determining the agreement in wilt severity methods in trial 1 and trial 2 estimated that the mean difference between rating methods were 5.10 and 5.91%, respectively. A mean difference greater than five indicates that the methods of measuring wilt are not in agreement. The study reported here demonstrates that Pearson correlation and OLS regression are inappropriate for assessing the agreement between two methods of measurement. © 2017 Elsevier Ltd. All rights reserved. 1. Introduction Plant pathology research often encounters two methods of measurement that assess the same quantity. For example, studies where pustules are counted with the naked eye and magnification hand lens or estimating cell concentration using hemocytometer and spectrophotometer methods. One method is traditionally used before the introduction of another method for replacement or interchangeable use. In this scenario, the new method is compared with an established method rather than the measured variables of each subject (Bland and Altman, 1999). Some level of disagreement is allowed between methods because each method is subjected to random measurement error, knowing the amount of agreement between methods is important prior to a researcher replacing an old with a new method (Bland and Altman, 1999). Agreement is quantified by appraising the differences and cause of these differ- ences between two quantitative methods of measurement (Giavarina, 2015). Most commonly a Pearson correlation coefficient (r) is used to compare two analytical methods (Altman and Bland, 1983; Ludbrook, 2002) and based on the magnitude of r the de- gree of association between two methods is determined. For example, correlation coefficient was used to compare the Assess (software) disease quantification method to that of visual method for counting maize rust lesions (Bade and Carmona, 2011). Also, correlation was used to compare common bean leaf area index (LAI) measurements by a LAI-2000 plant canopy analyzer to that of central leaflet width method (de Jesus Junior et al., 2001). However, correlation coefficient measures linear association rather than agreement between methods (Bland and Altman, 1986, 2010; Hopkins, 2004). Simply stated, correlation is used to measure the strength of the relationship between variables and it is inappro- priate for quantifying systematic differences between two * Corresponding author. E-mail address: neil.gudmestad@ndsu.edu (N.C. Gudmestad). Contents lists available at ScienceDirect Crop Protection journal homepage: www.elsevier.com/locate/cropro http://dx.doi.org/10.1016/j.cropro.2017.07.019 0261-2194/© 2017 Elsevier Ltd. All rights reserved. Crop Protection 101 (2017) 68e75