DISCUSSION 



Several interesting observations are suggested by a study of 

 the 21 regression analyses considered above. 



Time/Space Consistency 



Perhaps the most obvious is the consistency in the statistical 

 parameters with respect to both year and area. In any analysis 

 technique this consistency is important. If the results from year 

 to year and area to area varied vi^idely, then the statistical model 

 would have little predictive potential. It is noted that: (1) The 

 inultiple-regression correlation coefficients vary from 0. 84 to 



0. 99, with 50 percent of the coefficients in the 0. 88-to-O. 91- 

 percent interval. (2) The percent of variance explained by regres- 

 sion varied from 71 to 97 percent, with 50 percent in the 79-to-86- 

 percent interval. (3) The standard deviations varied from 0. 9 to 



1. 9°F, with over 50 percent in the 1. 1-to-l. 5 F interval. The 

 median standard deviation was 1.2 F. Since the surface- 

 temperature data were obtained from bathythermograms, the data 

 have an instrumental error from 0.5 to perhaps 1. F, the 

 magnitude selected depending upon the reader's personal feeling 

 regarding bathythermogram accuracy. The instrumental error 

 represents the noise in the data. The difference between the 

 instrumental error and the standard deviation, about 0. 5 to 1. F, 

 could possibly be a systematic variation not considered in the 



67 



