FACTORS INFLUENCING FLOWER SIZE IN NICOTIANA 339 
the more significant differences in variability under these same condi- 
tions. Thus we see that although the mean for ''different dates," 
26.69 =t: .20 mm., does not differ sensibly from the ''mean for totals" 
of 26.83 =t -09 mm., yet the coefficient of variability for the former 
distribution is over twice as great as in the latter case. The difference 
between the two is 5.32 db .59 per cent, which is certainly large enough 
in comparison with its probable error to acquire considerable signifi- 
cance. The proportionate increase in variability for the corresponding 
length arrays is very slightly higher than that for width arrays, but 
for all practical purposes the same. Here again the difference between 
the two coefficients of variability, 3.05 =b .28 per cent is large enough 
to be of unquestionable significance. In figures i and 2 are given a 
graphic representation of the facts brought out in these tables. 
It seems fair to assume that to determine the constants for flower 
size of a large series of plants, an F2 population for example, the flowers 
on the plants would naturally be measured either all of them during 
the beginning of the flowering season, all toward the end of the season, 
or most probably in some such fashion as is expressed in the distribu- 
tion of "different dates." Peculiarly enough published reports give 
us no account of the way in which the measurements of flower size 
were collected. It seems fair, however, to assume that in field work 
measurements of 25 flowers should represent the minimum required 
to give even an approximately fair mean flower size for a given plant. 
On this basis to any one familiar with the taking of such measurements 
the number of operations necessary to obtain the mean distribution of 
flower size in a representative F2 population represents a stupendous 
amount of time. With anything like the normal amount of assistance 
available it hardly seems possible that any such number of measure- 
ments could be accumulated in less than a month and a half; and, if 
so, some of the plants would be measured at the beginning, others at 
the middle, and still others at the end of the period of flowering. Such 
a system would merely give a representative mean and would not 
give a body of data suitable for the calculation of other constants. 
The only other alternative involves the measuring of a few flowers 
on each plant during the first week of the season and repeating this 
process each week throughout the season. Even this, however, would 
involve the necessity of taking a larger number of measurements than 
is ordinarily possible during such a time. We feel that a distribution 
based upon means derived from measurements taken throughout the 
