Premium
Combining Experiments to Predict Future Yield Data 1
Author(s) -
Cady Foster B.,
Allen David M.
Publication year - 1972
Publication title -
agronomy journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.752
H-Index - 131
eISSN - 1435-0645
pISSN - 0002-1962
DOI - 10.2134/agronj1972.00021962006400020025x
Subject(s) - statistics , mathematics , regression analysis , regression , stepwise regression , least squares function approximation , partial least squares regression , variables , linear regression , econometrics , estimator
Data front a series of fertility experiments including uncontrolled environmental variables are analyzed so that future yields may be predicted. Existing regression procedures for selecting variables in an estimated prediction equation had been found to perform poorly when used for new sets of similar data. A new criterion, the prediction sum of squares, based on the performance of the estimated equation for predicting observations not included in the least squares estimation, is developed for selecting the best predictor variables. The procedure gives an estimated prediction equation with a minimal number of predictor variables, including few interaction variables. Agronomically reasonable estimates of the regression coefficients also are obtained. Using the new procedure, a 30% reduction in the sum of squared deviations between the observed and the predicted observations compared with the stepwise regression method is found. Response curves are constructed for use in making soil test recommendations.