Generally, the significance of the improvement attained at each step 

 of the screening is tested and the screening discontinued when the amount 

 of improvement is found not to be significant. Near that point the addi- 

 tion of many more predictors usually lowers the predictive ability of the 

 system on independent data. As pointed out by Panofsky and Brier (1958), 

 however, objective standard significance tests may be misleading on data 

 such as those of this study, because the underlying assumptions may be 

 violated. The predictors used here are certainly interdependent in time 

 and space. Often the most practical and convincing test of significance 

 can be an application of the result to an independent set of data. This 

 will be the method used by the present investigators for "significance" 

 tests in future work with predictor equations for data of this sort. 



RESULTS 



Predictors Selected 



As mentioned earlier, variables X3 and X5 (Lq and Hq/Lq) were not used 

 when screening for predictor equations, because of their redundance with 

 each other and with T (wave period). The results of the screening procedure 

 for 1 through 4 predictors are given in table Al , where the order of the 

 first four selected predictors is shown along with the lag and correlation 

 coefficients. For example, in run 1 the first predictor selected by screen- 

 ing was (T) with a lag of 0-4 hours and a correlation of 0.47. The second 

 predictor selected (H^, with lag of 0-4 hours) increases the correlation to 

 0.64. Four predictors bring the correlation to 0.72. 



Predictor equations, each containing the four predictors of table Al , 

 are presented below for the five screening runs. 



(1) V = 1.73 -0.26(T) , + 0.18(H ) , + 0.14 (S^) , + 0.02 (U ^) , 

 -1 o-l r-1 of-1 



(2) S = -0.63 -1.58 (Mz) ^ + 0.10 (p) . + 0.03 (°^) ^ + 0.64 (C) . 



(3) (M^) = 0.64 -0.17 (Sg)_;^ + 0.40 (C)_^ -0.02 (p)_^ + 0.03 (h)_ 



s 



(4) J = -0.03 -0.18 (S,) ^ + 4.70 (7? ) , + 0.31 (V) , + 0.09 (T) ^ 



X r-1 'r -1 -1 -6 



(5) K^ = 0.12 + 0.02 (U ^) ^ -0.10 (S^) ^ + 0.01 {.<^) , + 0.20 (V) , 



X ox -2 f -4 -1 -6 



3 

 where p = (p-1) x 10 and the other variables and their units are as defined 



and explained in Appendix A. The numbers subscripting the variable symbols 



designate lag periods, each of which is of four hours duration. 



A- 3 



