are not very good for further investigation. They require too much space
and too much care. The calculation of a single percentage requires the
counting of some hundreds of individuals, taking many square meters for
their cultivation, and this, as my best races are biennial, during two
years. For this reason the countings must always be very limited, and
selection is restrained to the most perfect specimens.

Now the question arises, whether this mark is the best upon which to
found selection. This seems to be quite doubtful. In the experiments on
the heredity of the atavists, we have seen that they are, at least
often, in no manner inferior to even the best inheritors of the race.
This suggests the idea that it is not at all certain that the visible
characters of a given individual are a trustworthy measure of its value
as to the transmission of the same character to the offspring. In other
words, we are confronted with the existence of two widely different
groups of characters in estimating the hereditary tendency. One is the
visible quality of the individuals and the other is the direct
observation [414] of the degree in which the attribute is transmitted.
These are by no means parallel, and seem in some sense to be nearly
independent of each other. The fact that the worst atavists may have the
highest percentage of varietal units seems to leave no room for another
explanation.

Developing this line of thought, we gradually arrive at the conclusion
that the visible attribute of a varying individual is perhaps the most
untrustworthy and the most unreliable character for selection, even if
it seems in many cases practically to be the only available one. The
direct determination of the degree of heredity itself is obviously
preferable by far. This degree is expressed by the proportion of its
inheritors among the offspring, and this figure therefore should be