Open AccessOvercoming collinearity in path analysis of soybean [Glycine max (L.) Merr.] grain oil content
Author(s) -
Murilo Viotto Del Conte,
Pedro Crescêncio Souza Carneiro,
Marcos Deon Vilela de Resende,
Felipe Lopes da Silva,
Luíz Alexandre Peternelli
Publication year - 2020
Publication title -
plos one
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.99
H-Index - 332
ISSN - 1932-6203
DOI - 10.1371/journal.pone.0233290
Subject(s) - collinearity , path analysis (statistics) , culling , multicollinearity , statistics , mathematics , residual , path coefficient , correlation , trait , biology , computer science , zoology , regression analysis , algorithm , herd , geometry , programming language
Path analysis allows understanding the direct and indirect effects among traits. Multicollinearity in correlation matrices may cause a bias in path analysis estimates. This study aimed to: a) understand the correlation among soybean traits and estimate their direct and indirect effects on gain oil content; b) verify the efficiency of ridge path analysis and trait culling to overcome colinearity. Three different matrices with different levels of collinearity were obtained by trait culling. Ridge path analysis was performed on matrices with strong collinearity; otherwise, a traditional path analysis was performed. The same analyses were run on a simulated dataset. Trait culling was applied to matrix R originating the matrices R 1 and R 2 . Path analysis for matrices R 1 and R 2 presented a high determination coefficient (0.856 and 0.832, respectively) and low effect of the residual variable (0.379 and 0.410 respectively). Ridge path analysis presented low determination coefficient (0.657) and no direct effects greater than the effects of the residual variable (0.585). Trait culling was more effective to overcome collinearity. Mass of grains, number of nodes, and number of pods are promising for indirect selection for oil content.