These coefficients are all small, but the number of individual cases, 600 months, is so large that the probable error is greatly reduced, being only +-0.027 or +-0.028. Moreover, the nature of our data is such that even if there is a strong connection between solar changes and earth movements, we should not expect a large correlation coefficient. In the first place, as already mentioned, the earthquake data are not strictly homogeneous. Second, an average of about two and one-half strong earthquakes per month is at best only a most imperfect indication of the actual movement of the earth's crust. Third, the sunspots are only a partial and imperfect measure of the activity of the sun's atmosphere. Fourth, the relation between solar activity and earthquakes is almost certainly indirect. In view of all these conditions, the regularity of Table 7 and the fact that the most important correlation coefficient rises to more than four times the probable error makes it almost certain that the solar and terrestrial phenomena are really connected. We are now confronted by the perplexing question of how this connection can take place. Thus far only three possibilities present themselves, and each is open to objections. The chief agencies concerned in these three possibilities are heat, electricity, and atmospheric pressure. Heat may be dismissed very briefly. We have seen that the earth's surface becomes relatively cool when the sun is active. Theoretically even the slightest change in the temperature of the earth's surface must influence the thermal gradient far into the interior and hence cause a change of volume which might cause movements of the crust. Practically the heat of the surface ceases to be of appreciable importance at a depth of perhaps twenty feet, and even at that depth it does not act quickly enough to cause the relatively prompt response which seems to be characteristic of earthquakes in respect to the sun.