AN ANALYSIS OF THE EFFECT OF SELECTION. 37 
Mendelian differences are still arising by mutation and may arise 
in a selection experiment as well as anywhere else; and those that arise 
in such an experiment are as likely to affect the character under ob- 
servation as are any Mendelian differences taken at random. It is 
therefore probable that selection sometimes makes use of variations 
that arise during the course of the experiment, or, rather, that varia- 
tions which may be available do arise. 
The question is, what is the relative frequency of the two kinds of 
available factor differences—those already present and those that arise 
de novo? The answer is found by investigation of the data on selection 
in inbred lines and in crossbred lines. In closely inbred strains there are 
not likely to be many factor differences present when selection is begun, 
while in crossbred lines these differences are likely to be numerous. 
That selection is usually effective in crossbred lines is a well-known 
fact, demonstrated many times with many different organisms. Not 
many experiments have been carried out on closely inbred material, 
but those of Johannsen (1903), MacDowell (1917), and the present 
paper (p. 11) show that selection may be without effect in such lines. 
In two of these cases selection was effective until the lines became highly 
inbred. But mutations influencing the characters under observation 
have been obtained in the selection experiments of Castle and Phillips 
(1914), Morgan (Morgan, Sturtevant, Muller, and Bridges, 1915, 
p. 205), Lutz (1911), and those reported in this paper (p. 31). 
Apparently, then, selection produces its effects chiefly through 
isolation of factors already present, but occasionally available muta- 
tions do arise during the course of the experiment. 
2. Does selection cause mutations, or influence their direction? 
The usual selection experiment consists in breeding from individuals 
that are extreme in some respect. This extreme character may be 
environmental in origin, or it may be caused by germinal differences. 
In the first case, no geneticist is likely seriously to maintain that selec- 
tion will have any effect whatever. In case the extreme character 
is germinal in origin, selection will of course be effective in eliminating 
certain genetic types. Moreover, given a combination of genes that 
produce the character in a certain degree, we are evidently in a better 
position to reach a further stage than if we have the character less well 
developed. For how long a tail will be when it gains an inch evidently 
depends on how long it was before it gained that inch. But it seems 
incomprehensible that selection of individuals of a constitution favor- 
1Evidence derived from forms that reproduce asexually is also available in studying this 
question, for such reproduction commonly prevents recombination, and therefore gives results 
comparable with those obtained from homozygous strains. Some of the evidence obtained from 
studies on asexually produced Protozoa (e. g., Calkins and Gregory, 1913; Jennings, 1916; Middle- 
ton, 1915) has shown that selection may be very successful in changing such forms. But it is 
very doubtful if these animals are comparable with the Metazoa in the method of distribution 
of their chromatin. It seems not improbable that in some cases recombination may here be 
possible in asexual reproduction. 
