The third, or meteorological hypothesis, which makes barometric pressure the main intermediary between solar activity and earthquakes, seems at first sight almost as improbable as the thermal and electrical hypotheses. Nevertheless, it has a certain degree of observational support of a kind which is wholly lacking in the other two cases. Among the extensive writings on the periodicity of earthquakes one main fact stands out with great distinctness: earthquakes vary in number according to the season. This fact has already been shown incidentally in the table of earthquake frequency by months. If allowance is made for the fact that February is a short month, there is a regular decrease in the frequency of severe earthquakes from December and January to June. Since most of Milne's earthquakes occurred in the northern hemisphere, this means that severe earthquakes occur in winter about 20 per cent oftener than in summer. The most thorough investigation of this subject seems to have been that of Davisson.[128] His results have been worked over and amplified by Knott,[129] who has tested them by Schuster's exact mathematical methods. His results are given in Table 8.[130] Here the northern hemisphere is placed first; then come the East Indies and the Malay Archipelago lying close to the equator; and finally the southern hemisphere. In the northern hemisphere practically all the maxima come in the winter, for the month of December appears in fifteen cases out of the twenty-five in column D, while January, February, or November appears in six others. It is also noticeable that in sixteen cases out of twenty-five the ratio of the actual to the expected amplitude in column G is four or more, so that a real relationship is indicated, while the ratio falls below three only in Japan and Zante. The equatorial data, unlike those of the northern hemisphere, are indefinite, for in the East Indies no month shows a marked maximum and the expected amplitude exceeds the actual amplitude. Even in the Malay Archipelago, which shows a maximum in May, the ratio of actual to expected amplitude is only 2.6. Turning to the southern hemisphere, the winter months of that hemisphere are as strongly marked by a maximum as are the winter months of the northern hemisphere. July or August appears in five out of six cases. Here the ratio between the actual and expected amplitudes is not so great as in the northern hemisphere. Nevertheless, it is practically four in Chile, and exceeds five in Peru and Bolivia, and in the data for the entire southern hemisphere. TABLE 8 SEASONAL MARCH OF EARTHQUAKES AFTER DAVISSON AND KNOTT A: _Region_ B: _Limiting Dates_ C: _Number of Shocks_ D: _Maximum Month_ E: _Amplitude_ F: _Expected Amplitude_ G: _Ratio of Actual to Expected Amplitude_ A B C D E F G