Skip to main content
ARS Home » Plains Area » Manhattan, Kansas » Center for Grain and Animal Health Research » Hard Winter Wheat Genetics Research » Research » Publications at this Location » Publication #273539

Title: Multivariate mixed linear model analysis of longitudinal data: an information-rich statistical technique for analyzing disease resistance data

item VETURI, YOGASUDHA - University Of Delaware
item KUMP, KRISTEN - North Carolina State University
item WALSH, ELLIE - The Ohio State University
item OTT, OLIVER - North Carolina State University
item Poland, Jesse
item KOLKMAN, JUDITH - Cornell University
item NELSON, REBECCA - Cornell University
item Balint-Kurti, Peter
item Holland, Jim - Jim
item WISSER, RANDALL - University Of Delaware

Submitted to: Phytopathology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/20/2012
Publication Date: 10/28/2012
Citation: Veturi, Y., Kump, K., Walsh, E., Ott, O., Poland, J.A., Kolkman, J., Nelson, R., Balint Kurti, P.J., Holland, J.B., Wisser, R. 2012. Multivariate mixed linear model analysis of longitudinal data: an information-rich statistical technique for analyzing disease resistance data. Phytopathology. 102(11):1017-1025.

Interpretive Summary: A parameter called Area Under the Disease Progress Curve (AUDPC) is often used to quantify host resistance in crop plants using repeated measurements of disease levels during the season. One limitation of this method is that changes in resistance rankings during the season are not accounted for by AUDPC. A new method of analysis was developed by incorporating advanced statistical models called longitudinal mixed linear models. The new method of analysis offers a more sensitive way of detecting effects that vary over the season such as adult plant resistance or temperature-sensitive resistance. This will help researchers more precisely characterize disease resistance in crop cultivars and identify genes that condition that resistance.

Technical Abstract: The mixed linear model (MLM) is currently among the most advanced and flexible statistical modeling techniques and its use in tackling problems in plant pathology has begun surfacing in the literature. The longitudinal MLM is a multivariate extension that handles repeatedly measured data, such as ratings of disease resistance taken across time, which could be useful to plant pathologists. In this study, using an example dataset from a multi-environment trial of northern leaf blight on maize, longitudinal MLM analysis was performed and its utility was examined. For this dataset, 290 maize lines with diverse levels of resistance exhibited highly correlated disease response profiles and daily resistance responses followed an autoregressive pattern of correlation decay. Because it is common practice to convert repeated measures data to area under the disease progress curve (AUDPC), univariate analysis of AUDPC was compared to longitudinal analysis of repeated measurements. Given the high temporal correlation among maize lines studied here, similar results were obtained using both approaches. Based on side-by-side comparisons of the modeling approaches and computer simulation analysis, using AUDPC was an effective way of circumventing the application of a more complex modeling procedure but longitudinal analysis was an information-rich statistical technique providing broader applicability. To aid in the application of longitudinal MLMs, annotated program syntax for model fitting is provided for the software ASReml.