Use of Stepwise Regression in Developing a Prediction Model for Seed Yield in Flax

The stepwise regression analysis was used to determine a prediction model for seed yield in flax. The data studied were from a randomized complete block experiment in which 200 breeding families were replicated twice in each of two localities. Seeds and capsules per area (S/A and C/A) gave the highest correlation with yield followed by tillers per area (T/A) and capsules per tiller (C/T), but 100-seed weight (S W) and seeds per capsule (S/C) gave negative or insignificant values with yield. Using all six independent variables, the prediction model accounted for 97.79 and 99.62% of the variability in yield for the two localities. Almost the same R²values were obtained with the five significant variables determined by the stepwise regression analysis. Confining the analysis to the preharvest characters (T/A, C/T and C/A) resulted in R²values of .8020 and .6923. Since CA is the product of T/A and C/T, the predictive model on the basis of C/A alone gave an R²of .7235 indicating that C/A is a better predictive variable than its two components (T/A and C/T).