PERFORMANCE OF MULTIPLE LINEAR REGRESSION ANALYSIS CONDUCTED UNDER RANDOMIZED COMPLETE BLOCK DESIGN

Abstract

In factorial experiments conducted under randomized block design, the multiple linear regression model fitting can be performed under different combinations of the quantitative levels of the two factors and the number of replications. To determine the best combination, considering the same number of levels per factor and the same number of experimental units, it was concluded through a simulated data study that the quality of the fit increases when regression is performed in experiments with fewer combinations of levels (treatments) and more replications. Therefore, if linearity is expected, using four treatments evaluated in a 2 × 2 factorial design for model fitting is recommended. Otherwise, nine treatments evaluated in a 3 × 3 factorial design are recommended. All of this is for experiments with coefficients of variation of 20%.
 Keywords: Treatments, Replications, Experimental precision.