338 
J. A DETLEFSEN AND E. ROBERTS 
males as two of the four most frequent classes, which was very 
convenient, since these were used again for mating in the next 
generation. Writing the form of all matings for every genera- 
tion in tables 1 to 6 in the usual Mendelian terms we have: 
w M 
red long 9 
white miniature cf 
+ 
w m 
+ 
w m 
w m W M W m 
white miniature 9 + red long 9 + red miniature 9 
+ ^. 
w M 
+ white long 9 
+ 
W M 
+ 
W m 
4- 
w M 
white miniature cf + red long cf + red miniature cf + white long c? 
Non-crossovers 
Crossovers 
THE DATA 
Series A; low selection 
Table 1 and text figure 1 give the main facts of this selection 
experiment. The Fi generation consisted of twenty-eight pairs 
whose total progeny showed 27.11 per cent crossovers. This is 
a little lower than might be expected in a general population, but 
the difference between this ratio and Sturtevant's (given in 
Morgan and Bridges, '16) ratio of 32.8, based on 41,034 progeny, 
is no greater than that recorded by Bridges (JNIorgan and Bridges, 
'16), who gives data showing 38.3 per cent crossover. This 
same stock has repeatedly given crossover ratios close to 33 
per cent. 
The crossover values in this and other similar tables are 
treated as variables and classified in frequency distributions in 
which the class interval is 3 per cent. The average of each class 
is placed at the head of the columns, e.g., 1.5, 4.5, 7.5, etc., which 
means that the class ranges were 0-3, 3-6, 6-9, and so forth. 
The crossover values in the Fi ranged from 10 per cent to 36.8 
per cent. There is no doubt but that some of these ratios have 
little meaning, for they are based upon small totals. We have 
