SIGNIFICANT ACCURACY IN RECORDING GENETIC DATA 219 
compare random frequency distributions of the corolla size of single 
plants with frequency distributions of the flowers when selected from 
vigorous branches and measured on the same day they have opened. 
This procedure gives a measure of the accuracy of single flower selec- 
tions. To illustrate this, data from two species with very different 
sized flowers are submitted. 
Table II 
Comparison of Random Samples of Corolla Length on Single Plants and Samples in 
which Constant Errors have been Largely Eliminated 
Name 
Class Centers in Millimeters 
20 
23 
.4 
25 
26 ! 27 
28 
29 
30 
A^. paniculata, Random 
3 
4 
14 
17 
t6 2 
3 
" " Selected 
18 
4 
3 
4 
3 
4 
3 
I 
Ran 
2 
4 
5 
Sel 
" . " Ran 
16 
20 
I 
5 
2 
Sel 
Ran 
2 
3 
15 
22 
Sel 
3 
... 
Name 
Class Centers in Millimeters 
70 
73 
76 
79 
82 
85 88 
90 
94 
97 
100 
I 
3 
2 
I 
16 
22 
6 
3 
2 
4 
I 
14 
18 
•2 
I 
" " " Sel 
" " " Ran 
3 
4 
T7 
I 
" " " Sel 
" " " Ran 
3 
" " " Sel 
2 23 
These plants are among the most uniform and the most variable 
respectively, and give an idea of the range of variability involved. 
The other test made was to select fifteen flowers on a plant at 
random, and determine the mean to the nearest millimeter; then to 
find the deviation from this mean when single flowers were selected. 
In 100 tests of flowers shorter than thirty millimeters 88 selections were 
made within the 3 millimeter class to which the mean belonged. 
The remainder were in contiguous classes. On flowers between 70 
and 100 millimeters long 82 out of 100 selections were within the 6 
millimeter class to which the mean belonged. The remainder with 
2 exceptions were in contiguous classes. 
From these tests it will be seen that the probable error of the 
selection (equal chances) is not over plus or minus 2 percent. If this 
