In 1903 I carried out an analysis—referred to here for brevity as A—of the results given by the Kew magnetographs on “quiet” days during the 11 years 1890 to 1900. This investigation brought to light various novel phenomena. It was subsequently shown—in a paper to be referred to as B—that these phenomena are equally true of “quiet” days at Falmouth. Some of the phenomena suggested the possibility of differences of a certain kind between quiet days and other undisturbed days, and between ordinary days and disturbed days. To prosecute this enquiry, it was necessary to make an analysis of the data from all days at Kew from 1890 to 1900. Declination being the element of most practical interest, and least open to instrumental uncertainties, it was decided to treat it thoroughly in the first instance. In measuring the quiet day curves at Kew—a practice instituted in 1890—it has been usual to smooth them when any little irregularities occur, by drawing a free- hand pencil trace following the general trend. It was decided, with some hesitation, to continue the practice, so that the results from all days should be as strictly comparable as possible with those from quiet days. The nature of the difficulties will be understood from the accompanying diagram. The continuous line curve ABCDEFQRS represents a hypothetical photographic record. The dotted line AB'C'R' represents the smoothed curve. When the object aimed at is the regular diurnal inequality, it will probably be generally conceded that the method of smoothing adopted is satisfactory so far as the wave-like portion ABCDE is concerned, at least so long as the interval of time corresponding to this portion is under an hour. If, however, the times from A to B and from B to C were each an hour, the procedure would be disapproved by some authorities, who would argue that the free-hand curve should always be drawn so that its ordinate at any particular hour should represent the arithmetic mean of an infinite number of ordinates, uniformly distributed in time throughout the preceding and succeeding 30 minutes. It should, however, be remembered that the exact instant when an hour falls is really arbitrary. One observer may use Greenwich time, another local, and if the smoothing were carried out in accordance with the view last mentioned, it might make all the difference which choice happened to have been made. A disturbance such as QBS presents difficulties of another kind. If the time interval from Q to S is only a few minutes, and the general trend of the curve is very clearly shown, and closely similar to that of the average day, there can, I think, be little doubt that the best plan—at least when diurnal inequalities are concerned—is simply to disregard the disturbance altogether. If, however, the time from Q to S is considerable, and the general trend of the curve not clearly shown, the appropriate treatment is difficult to determine.
Epistemic values of quantity and variety of evidence in biological mechanism research
