PAPERS ON ZOOLOGY AND PHYSIOLOGY 319 
regression) were straightened out it would be almost identical 
with the mean of the pure line. The mean of the 20 parents 
represented in this second score of generations is 1.750 mm. 
The mean for the 221 offspring is 1.538, or 0.006 mm. below 
the mean for the genotype or pure line. What more could be 
asked in demonstrating the ineffectiveness of selection! 
The mean lengths for the parents in these eight classes are 
here given reduced to the scale; mean of the line=100, and 
likwise under these in each case the mean of the offspring 
reduced to the same scale, 
1.50-1.54|1.55-1.59|1.60-1.64|1.65-1.69|1.70-1.74|1.75-1.79|1.80-1.84|1.85-1.89 
99.45 |103.35 {104.65 |109.85 |111.80 {115.70 {117,65 {119.60 
98.80. |100.75 91.65 {100.75 98.15 {102.05 96.85 |106.60 
If we now express the regression by a series of fractions as 
was done with all three sublines we get the following: 
—0.65+2.60+-13.00+-9.10-+-12.65+13.65-+-20.80-+- 13.00 
O55, 3.35) '))/4:65) ,. 9:85)) 11:80; 15.70) 17.65) 19:60 
These fractions reduced to decimals give the following 
series : 
—1.18+0.78-+-2.77-+.0.92-++ 1.08-+-0.87-+1.17-++0.66. 
In order to measure the regression as usual let us add these 
fractions and divide by their number. This gives the decimal 
0.884, which indicates that the regression is 884/1000 com- 
plete. However, if we leave out one parent in the 1.50-1.54 
class, the only one selected which was below the mean of the 
line, we get as the result for the nineteen parents above the 
mean the decimal 1.031, showing the regression to be more 
than complete. I think that the method of putting the parents 
into classes and computing the regression for each class is 
certainly open to serious objection, as it is not an accurate 
measure of regression in all cases, even though an enormous 
number of individuals be included. Of course more important 
than the number of either parents or offspring is the number 
of classes considered. If the results are to be plotted, here 
again we are limited to a relatively small number. Certainly 
the regression (digression in this case) of the offspring of the 
single parent in the 1.50-1.54 class should not count for as 
much as the regression of the five parents in the 1.80-1.84 
class. Would it not be better to multiply each fraction by the 
