ITS ERRORS AND CORRECTIONS, ETC. 487 
The low value of g' obtained in this case, compared with that by water experi- 
ments, resembles the value obtained for the magnet C, at the Cape (§ 25), and 
both, it seems to me, indicate some derangement of the wires. 
38. P.S.—I have discussed the observations made in 1846, 1847, and till June 
30, 1848 (Toronto Obs., vol. iii., received since this paper was written), for the 
temperature coefficient: the observations for 1846 gave g’ = 1:355 scale divisions, 
but those for 1847 and 1848 were so vitiated by large disturbances (in some cases 
the daily differences being upwards .of 100 scale divisions), that they were evi- 
dently unfitted for a good determination ; and some mode of rejecting the larger 
disturbances was desirable. In order, however, that the same rule should be 
followed throughout, the differences which after correction for temperature (by 
g = 1:33 scale divisions) were found to exceed 10 scale divisions, were rejected 
in all cases from November 1843 to June 1848. The following are the results :— 




1844, q’= — = 1-254 sc. div.; prob. error =0:081 sc. div.; diffs, rejected = 2-7 per cent. 
1845. q/= saan =1-354 _..,, ,  S0-088" ,. 4, = 20 , 
790-83 
1846. f= 7 =1263_,, sf TE -., 7 Gd hak 
1847. q— ae aa oS 29 ” =0:077 Ty 29 == 12-0 ” 
93% 
[1848]. = a SSS 1Gi0g; a = 0:05499 ;, Ma Soo as 
If we weight these results by the inverse squares of the probable errors, we obtain 
¢g = 1°378 sc. div.; but we obtain the most probable result by taking the sums 
of differences for the whole period, or 
,_ 4007-90 
1 = 9948-80 
This differs only 0:03 sc. div. from the result obtained for 1844 and 1845. 
There appears, however, to be an increase in the value of ¢ for 1847 and 1848, 
but whether this is a real increase (which is quite possible, if the value depends 
on the condition of the wires) would require a more careful discussion to deter- 
mine. ‘The daily means vary so similarly at all the stations that it only requires 
for such a discussion to have one station accurately corrected (or to take a station 
within the tropics where the temperature variations are small), when, by sub- 
tracting the corrected daily mean scale readings reduced to a common unit (re- 
jecting days of large disturbance) from those for the station whose temperature 
coefficient is required, the remainders will depend almost entirely on the differ- 
ences of temperature for the bifilar in question. 
= 1°359 sc. div.; probable error = 0:029 sc. div. 
* This includes the six months of 1848, and the two months of 1843. 
t Six months of 1848 only, no differences having been rejected in November and Decem- 
ber 1843. 
VOL. XXII. PART III. 6K 
