during oceanographic summer and winter, since in order for the 

 model to fit the data taken during the seasonal extremes, it must, 

 by nature of the model, fit the data taken during the periods of 

 spring warming and autumn cooling. Thus, the taking of addi- 

 tional data during the latter seasons neither adds to nor detracts 

 from the results obtained. Similar reasoning applies to the spa- 

 tial distribution of data. If a third-degree polynomial describes 

 the longitudinal variation of surface temperature, then it is neces- 

 sary only to have a few observations distributed over the area to 

 determine the shape of the polynomial. Again, the taking of addi- 

 tional observations is unnecessary. In support of this observation, 

 an additional fit to the data taken in 1954 was made to the same 

 data shown in figure 6, except that only 22 observations picked 

 randomly from the original 87 observations, taken in the 1 -degree 

 square under consideration, were used. Tlie results follow: 



Area C (1954) 



Data Set 1 



Data Set 2 



286 



221 



75.0 



74.9 



0.87 



0.87 



1.8 



1.7 



Number of observations 

 lOOff", percent variance explained 

 R, multiple correlation coefficient 

 CT, standard deviation in degrees F 



The almost identical results of the two regression analyses sug- 

 gest that the additional 65 observations used in the first analysis 

 did not contribute any additional information, and that the abnormal 

 spatial distribution of data did not distort the statistical analysis. 



69 



