462 BOTANICAL GAZETTE [DECEMBER 
and distinguish real differences. This can best be accomplished 
by a more frequent use of checks and replications. Replications, 
however, require a liberal seed supply. Although a single plant 
ordinarily produces a limited quantity of seed, there are means 
of increasing the normal yield considerably. A root divided into 
halves and planted at the customary distances will produce nearly 
twice as much seed as a whole root. Moreover, seed yield is greatly 
influenced by soil fertility. A rich soil stimulates under a proper 
temperature the production of flowering shoots and enables the 
plant to mature a large crop. If these means are employed to 
increase the average seed production and only the heaviest seed- 
yielding plants are used for the preliminary tests, a sufficient 
quantity of seed may be obtained to make a large number of plant- 
ings from each individual. When a sufficient number of replica- 
tions are made to distinguish real differences, only the few best 
families should be saved. As roots of a single family are apparently 
of equal breeding value, it is advisable to plant all the roots of 
each family in an isolated group, where distance or intervening 
obstructions will prevent their crossing with plants of a different 
family. There is apparently no gain from harvesting the plants 
of each family separately. Moreover, the combined yields of the 
individual plants of a family are necessary to make a satisfactory 
test of the relative merits of the families the following year. There- 
after, no further selection would appear to be necessary. If all 
but 3 or 4 of the best families are discarded and the remainder 
mixed and prevented from crossing with poorer stock, they probably 
will maintain their vigor and are not likely to deteriorate in the 
absence of further selection. 
The possibility of further improvement apparently lies in the 
isolation of more valuable mutants, but whether the original 
stock or the new variety should be used as a source of material 
can be determined only when it is known in which the mutants 
will occur with the greater frequency. 
