Abstract
Six yield prediction models based on different combinations of plant characters in Hirnalyan primitive arleys were compared by means of (i) The R -adequacy test and (ii) The Residual Mean Square Ratio (R SR) test. Percentage Relat-ive Efficiency Estimate (PRE) was derived for each of the 15 possible pairs of regression models. The study indicated the superiority of regression model involving two independent variables namely area of flag leaf and total number of grains per plant.
INTRODUCTION
A major problem in model building studies is the choice of the independent regressors that are of real value. The reliability, of course, increases by increasing the number of independent variables but this causes much more increase in the volume of work, time and cost. To avoid complexity and minimize the effort, it is desired to have fewer regressors in the model that can serve the purpose of prediction. Stepwise procedures and all possible regression methods, based on repeated significance tests, as discussed by Draper and Smith 0981) are commonly used for the purpose of selecting such variables. A functional model having a minima! subset of regressors with a minimum mean square error estimate or high predictability for deriving a suitable optimal is considered to be the best one.