The study was conducted to identify plant characters associated with seed yield in close soybean plant spacings. Lines selected from two F6 soybean populations in the F10 generation segragating for degree of stem termination were grown in two locations. Traits measured included lengths of developmental stage. plant and canopy height, number of nodes, lodging at different stages, raceme lengths, and number of nodes and branch number of nodes.
One problem in analyzing data and drawing conclusions from such a study is related to the complex nature of the interrelationships of a large number of traits. Another, not unrelated problem, involves the calculation of multiple-regression equations with multicolinearities. Because of these problems, factor analysis was used to study the correlation matrix to identify sets of variables related to the same biological concept or function. The variables within sets were determined for each stem termination type, and measurements of these variables were standardized. Means were calculated by using the sum of the standardized variables for each set. Multiple regression was used to study the relationships between sets as the independent variables and yield as the dependent variable. Individual traits were selected from within each set and substituted in the multiple regression equations calculated with the mean value for each set. With use of this technique to determine multiple regression equations, the resulting equations involved measurements of different biological functions instead of repeated measurements of the same function in the plant. It was found that, in the determinate stem type, the measurement of the seed-filling period and branch number of nodes could possibly be used to predict yield while, in the semideterminate stem type, the measurement of the ‘fixed capital’ function and terminal raceme length were possibly useful. Finally, in the indeterminate stem type, the measurement of the ‘fixed capital’ could be used, but it had a very low predictive value.