FROM ORDINARY DAYS OF THE ELEVEN YEARS 1890 TO 1900. 
235 
not fall at the centre of the month, and a question may arise as to whether some 
allowance should be made for this. The desirability of selecting the 5 quiet days so 
that their mean shall fall near the middle of the month is recognised, but such a 
choice is not always possible. 
The practice in vogue at Kew from 1890 to 1900 was to treat the base value of 
each element as invariable throughout each calendar month, except in so far as it was 
influenced in the case of H and Y by departures of the temperature of the magneto¬ 
graph magnets from the mean temperature of the month. Using these base values, 
it would have been possible to deduce mean values depending on all the days, or on all 
but the highly disturbed days, in the usual way. Having regard to the special circum¬ 
stances, however, a different method—presently to be described—has been adopted as 
equally satisfactory and much simpler. It recognises that it is really on the absolute 
instruments that the accuracy of the annual inequality ultimately depends. The 
magnetographs serve only to show what allowance has to be made to any individual 
absolute observation to bring it to what it would have been if taken at an hour when 
the element in question has its mean value for the day. To take a simple example, 
suppose we observed the declination daily at noon, and that in the diurnal inequality 
for a particular month the entry under noon is + 3 /- 0. Then to get the mean value 
of D for that particular month all we have to do is to subtract 3 ; '0 from the 
arithmetic mean of the absolute observations. The correction — S r '0 applied to 
individual noon readings will not in general give exactly the mean value for the day, 
but so far as the monthly mean is concerned that is immaterial. 
If we observe not daily but weekly, the 4 or 5 monthly days of observation at noon 
may all happen to depart from the average day of the month in the same direction, 
so that the method if applied to the observations of a single year could hardly claim 
to be satisfactory. If, however, we deal with the observations not of one but of a 
number of years, omitting days of large disturbance, accidental features must largely 
disappear, and this is the course that has been adopted. 
The ordinary absolute observation of H consists of two parts, the vibration 
experiment and the deflection experiment. The mean times of these two experiments 
were found, and departures at each of them from the mean value for the day were 
deduced from the diurnal inequality, and the arithmetic mean of the two was applied 
with appropriate sign as a correction to the observed value of H. The inequalities, 
it will be remembered, were calculated for each individual month for H (and Y). 
When the curves were considerably disturbed at the time of an absolute observation— 
not, of course, a frequent event—the observation was simply omitted. As highly 
disturbed days had also been omitted from the ordinary day diurnal inequalities, we 
may reasonably regard these inequalities as appropriate for the correction of the 
absolute observations. 
It had not been the practice to commence the H observation at a fixed hour, and 
the time required for the deflection experiment varies somewhat according as 
VOL. CCXVI.-A. 2 K 
