176 



"Pending completion of a calibration tank, it was felt that an estimate of 

 the precision of the CTI could be secured from field data analyzed according to 

 the method given by Grubbs (1948). During, a cruise when fairly honnogeneous 

 conditions were to be expected, the heads of instruments No. 2 and No. 3 were 

 lowered together and 678 and 679 pairs of simultaneous measurements of tem- 

 perature and conductivity, respectively, were made. The nneans of the meas- 

 ures (see Table I) were taken as indicating a systematic difference of 0.02 in the 

 instruments in both temperature and conductivity. The data were corrected by 

 subtracting 0.01 from each reading from CTI No. 2 and adding 0.01 to each read' 

 ing from CTI No. 3. Analysis by the Grubbs method for partitioning the vari- 

 ance into that attributable to the environmental variation and that due to instru- 

 ment variation gave the estimates shown in Table la ^^^ is an estirnate of the vari 

 ance of the environment had there been no instrument errors 



2 

 !^ e#2 

 are estimates of the variances of the random errors inherent in the instruments 



and a 2^ 



"Grubbs (1948) says: 'It is to be noted that the variance of the estimate, 

 est. (al^) depends on (a)^ x^ > the variance in the characteristic measured, (b) 

 ael ' t^he variance of the errors of measurement of instrument Ij , (c) a eZ^' ^^^ 

 variance of the errors of measurement of instrument Iz, and (d) a n , the number 

 of observations in the sample. It is seen, therefore, that in order to obtain a 

 precise estimate of c el^ when using only two instruments, the variation in the 

 characteristic measured, i. e. , a ^ » should be held to a reasonable minimum 

 or the sample size, n, should be sufficiently large. ' He also says: 'The effi- 

 ciency with which the absolute values xi , X2, x^ are measured depends on 



(1) the relative size of Se^ and Sx^ . . . . ' So far we have been able to determine, 

 the correct method of comparing Sg^ and Sx^, i.e., est. (jg^) and est. (t;x^)> is 

 not known. 



"It would seem, however, that a ^ for the temperature is sufficiently 

 small and n is sufficiently large so tha^est. (je2^) ^^^ ^^^- (oe3^) should be 

 good estimates for the temperature errors. Unfortunately, the conductivity 

 measures do not exhibit a comparable homogeneity. Still, the orders of mag- 

 nitude of esto (-'x^) ^^^ est. (-e^) for the conductivity are relatively the same as 

 those for temperature. It would seem, then, that had the fluctuations in the en- 

 vironment for conductivity been as restricted as those for temperature the esti- 

 mates of precision for conductivity would have been of the same order as those 

 obtained for temperature. 



"From this analysis we estimate the standard deviation of the errors of 

 measurement for the CTI as not greater than O.O5OC and 0.05 CI °/oo." 



There appears to be common agreement among all who have used conduc- 

 timetric methods for in situ salinity measurements that many problems connec- 



