The computer program for the screening procedure is similar to that 

 presented by R. G. Miller (1958). Analysis of the data will proceed 

 without the inclusion of variables X3 and X5 (L^ and K^/Lq) , in part 

 because of their redundancy with each other and with T. (The reader is 

 referred to the list of symbols in Table 1 for definitions of the various 

 notations used for the variables). 



METHOD OF ANALYSIS 



As in the main part of this report, the procedure adopted for select- 

 ing predictors involves expression of a predictand Y as a linear function 

 of a number of predictors X^ (n=l,...,N). 



Thus: 



Y = A„ + A^X^ + A2X2 - ...A„X^ + ... A^X^ 



where the coefficients Aj^(n=o, ...,N) are determined using the method of 

 least squares. 



Because of the large numbers of predictors involved, the screening 

 procedure required the use of a high-speed, large-memory computer. The 

 IBM 7030 was used. Basically, the manner in which the predictors were 

 screened is shown below: 



1) Y = A^ + B^X^ 



2) Y = A^ -^ B^X^ . C^X^ 



3) Y = A + B X, + C ,X^ NX 



n n 1 n-1 2 n 



where Y is a predictand, the As are constants, the Xs are predictors , and 

 B , B C^ , C_, etc. are regression coefficients. 



The procedure is to first select the best single predictor (X ) for 

 regression equation 1. The second regression equation contains the first 

 predictor (XI) and the predictor (X2) that contributes most to reducing 

 the residual after the first predictor is considered, regardless of its 

 lag position. This means that the second regression equation contains the 

 best set of two predictors that includes the X selected at the first 

 screening step. The procedure will not necessarily select the best set 

 of r predictors out of the original set containing P predictors. In the 

 closely analogous field of meteorology, however, it has been shown that 

 the screening procedure can select a highly reliable set of predictors when 

 applied to problems that involve redundant, interrelated variables. 



A-2 



