The correlation coefficient does not Indicate relative reliability 

 of the two BT's. The relative magnitude of the standard deviations must 

 he considered. If the instruments are reliable, standard deviations 

 should be approximately equal. For the most part, the equality represents 

 actual variations in the ocean. If the standard deviations are not approxi- 

 mately equal, then one of two conditions is evident: (1) a very small 

 standard deviation may Indicate that one measurement is being recorded sev- 

 eral times (reading error) or (2) one instrument is less accurate and less 

 reliable than the other. 



When the correlation coefficient is high and the standard deviations 

 are equal, the resulting standard errors of estimate are low and the two 

 instruments are reliable to the accuracy of the standard error of estimate. 

 The standard errors of estimate, S^ and Sy, can be computed as follows: 



(1 - r2) 1/2 

 - r2) 1/2 



When the correlation coefficient is low and standard deviations are 

 nearly equal but low, there may be negligible ocean variability. When 

 the correlation coefficient is low and standard deviations are high, at 

 least one of the two instruments is highly biased. If the standard devi- 

 ations are unequal, the Instrument with the higher standard deviation is 

 probably in error. An inspection of the statistics from the nine sets of 

 paired BT observations in table 3 indicates wide variability in Instrument 

 response. 



Whenever the correlation coefficient for one pair of instruments is 

 consistently high for all depths, the two standard deviations are expected 

 to be nearly equal. However, this was not true in slLL cases. In the nine 

 sets of paired observations, five coefficients (1,2,3,5,9) averaged more 

 than .70 for all depths. Only one of these five (2) has- a majority of the 

 differences between the paired standard deviations equal to or less than 

 .1°F. 



The first two sets of observations were taken dviring ECHO I during a 

 period (November 1958) when a definite seasonal thermocline existed. The 

 magnitude of standard deviations should increase sharply in the vicinity 

 of the thermocline with reliable Instruments, yet they should be nearly 

 equal for any given depth. This appears to occur between 320 and 360 feet. 

 If one Instrument responds faster than the other, there should be a marked 

 difference between the standard deviations. This appears to occur at 280 

 feet. The instrument common to both sets (1291^) appears to have a slower 

 rate of response. 



Bralove and Williams stated that "when two standard deviations are 

 unequal, probably the instruojent with the higher standard deviation is in 

 error." If this statement is assumed to be true, then the first two sets 

 of observations indicate that BT 1291^ is more accurate to a depth of about 

 360 feet and is less accurate at greater depths than the other instruments. 



