Open AccessCanonical correlations in agricultural research: Method of interpretation used leads to greater reliability of results
Author(s) -
Maria Inês Diel,
Alessandro Dal’Cól Lúcio,
Darlei Michalski Lambrecht,
Marcos Vinícius Marques Pinheiro,
Bruno Giacomini Sari,
Tiago Olivoto,
Oscar Valeriano Sánchez Valera,
Patrícia Jesus de Melo,
Francieli de Lima Tartaglia,
André Luís Tischler,
Denise Schmidt
Publication year - 2020
Language(s) - English
Resource type - Journals
ISSN - 2411-2933
DOI - 10.31686/ijier.vol8.iss7.2464
Subject(s) - canonical analysis , canonical correlation , mathematics , multicollinearity , variance inflation factor , non canonical , interpretation (philosophy) , statistics , canonical correspondence analysis , canonical form , variance (accounting) , econometrics , regression analysis , pure mathematics , computer science , ecology , accounting , abundance (ecology) , business , biology , programming language , microbiology and biotechnology
Canonical correlations analyzes are being used in the agrarian sciences and constitute an important tool in the interpretation of results. This analysis is performed by complicated mathematical equations and it is only possible to use it thanks to the development of computational software, which allow different interpretations of results, and it is up to the researcher to choose according to his knowledge. Canonical correlations can be interpreted using canonical weights, canonical loadings, or canonical cross-loadings. In Brazil, most of the works that use these analyzes interpret the canonical weights. Therefore, this study aims to show, through an analysis of canonical correlations, the best way to interpret the results, so that they are presented in the most reliable way possible. Data from an experiment with two cultivars of biquinho pepper seeded in 5 light spectrums were performed. The variables were root length and volume, plant height, number of leaves, fresh shoot and root mass, shoot dry mass. Two groups of variables were organized, the multicollinearity was determined through the condition number and the inflation factor of the variance. Canonical correlations analysis was carried out, and weights, loadings, and canonical cross-loadings were estimated for the interpretation of the results. After the interpretations, it was defined that the canonical cross-loadings should be preferred for the interpretation of the canonical correlations. Weights or canonical coefficients provide dubious results of relationships between groups of characters and should be avoided.