If, on the other hand, the men and women with statures considerably below the mean marry, their children will ultimately exhibit statures which are greater than that of their parents, but which are less than that of the whole population. Regression again occurs, but in the opposite direction, and such a case would be represented by the above diagram reversed. Continued selection of this kind would lead to an immediate increase in the mean stature (or the opposite, if the “sign” of the selection were reversed) in one or two generations, but after that the amount of change would be very small, while if the selection were to cease the race produced would slowly revert to the mean, which is characteristic of the whole population from which it arose. It is very important to grasp this result of the practical and theoretical study of heredity--the selection of the ordinary variations shown by a general population leads at once to a small change in the mean value of the character which is selected, but continued selection thereafter makes very little difference to this result, while the race slowly reverts to the value of that from which it arose on the cessation of the selection. Races which “breed true” do, of course, exist; thus the mean height of the Galloway peasant is greater than that of the Welsh. In the cases of “pure races”--that is, races which breed true with respect to one or more characters, we have to deal with another kind of variation, one which shows no tendency to revert to the value from which it arose. Let the observed variability of stature in a human population be represented by the frequency distribution _A_, and let the individuals at _N_--that is, those in which the stature was greater than the mean by the deviation _ON_--intermarry. It might then happen that the variability of the offspring of these unions would be represented by the frequency distribution _B_, in which the value of the mean is also that of the stock, at _N_, from which the race originated. It does not matter now from what variants in _B_ a progeny of the third generation arises: the mean height of the latter will be that of the pure race. In this case the individuals from which the pure race originated (those at _N_ in _A_) have exhibited a mutation. The stature of the individuals of this new race will continue to exhibit fluctuating variations, and the range of this variability may be as much as that of the stock from which it arose, _but the mean stature of the new race_ will continue to be that of the original mutants. [Illustration: FIG. 22.]