But the deviation may be enlarged by a choice of extremes. In sowing
their seed, the average of the strain is seen to be changed, and in
repeating the experiment the change may be considerable. It is not
clear, whether theoretically by such an accumulation, deviations might
be reached which could not be attained at once in a single sowing. This
question is hardly susceptible of an experimental answer, as it would
require such an enormous amount of seed from a few mother plants as can
scarcely ever be produced.

The whole character of the fluctuations shows them to be of an opposite
nature, contrasting [18] manifestly with specific and varietal
characters. By this method they may be proved to be inadequate ever to
make a single step along the great lines of evolution, in regard to
progressive as well as to retrograde development.

First of all fluctuations are linear, amplifying or lessening the
existing qualities, but not really changing their nature. They are not
observed to produce anything quite new, and evolution of course, is not
restricted to the increase of the already existing peculiarities, but
depends chiefly on the continuous addition of new characters to the
stock. Fluctuations always oscillate around an average, and if removed
from this for some time, they show a tendency to return to it. This
tendency, called retrogression, has never been observed to fail, as it
should, in order to free the new strain from the links with the average,
while new species and new varieties are seen to be quite free from their
ancestors and not linked to them by intermediates.

The last few lectures will be devoted to questions concerning the great
problem of the analogy between natural and artificial selection. As