474 JENNINGS— HEREDITY IN PROTOZOA. [April 24, 



XA'III. As there appears, tlie mean breadth increased from 33.600 

 to 47.300 microns. The length, on the other hand, decreased from 

 123.666 to 114.033 microns. The mean ratio of breadth to length 

 thus increased very greatly, from 27.136 per cent, to 41.455 per cent. 

 The latter is the largest mean index I have ever observed in Para- 

 mecia not selected with relation to the age of the individuals ; it is 

 exceeded only by the mean index of the young halves during fission 

 (see Table X.). With the increase in the mean ratio of breadth to 

 length, there is as usual an increase in the correlation between the 

 two dimensions ; this reaches the unusually high value of .8152. The 

 nutritive fluid left the variation in length about the same, but con- 

 siderably decreased the variation in breadth. This is undoubtedly 

 due to the fact that before the hay infusion was introduced some of 

 the specimens were well fed, some poorly fed, as the chances of the 

 daily life determined; while after the infusion was introduced all 

 were well fed, so that there was less variation in breadth than before. 

 Characteristic forms after the infusion was introduced are shown in 

 Fig. 3, a to c (page 423). 



The facts in these cases are nearly, parallel with those observed 

 in the third series of experiments on the progeny of D (Table XVIII. , 

 rows 13-15). If we combine the two samples of c (row 20, Table 

 XVIII.), as we did those of D, the effect is, as in the case of D, to 

 decrease greatly the correlation between length and breadth But in 

 the present case the very high positive correlation of the two samples 

 taken separately is not entirely overcome by combining them, though 

 the correlation falls to .1758. The actual numerical coefficient just 

 given is the resultant of a number of conflicting factors. In the two 

 samples taken separately greater length is associated on the whole 

 with greater breadth, giving high positive correlation, which in pass- 

 ing from Table LVI. to Table L. an increase in breadth is associated 

 with a decrease in length, tending to diminish the correlation. The 

 facts show clearly that the observed statistical correlation does not 

 involve any necessary and constant relation of the one dimension to 

 the other; both dimensions depend on various factors, which some- 

 times act in the same way on both, sometimes differently. 



Combining the two samples of r (as in row 20, Table XVIII. ), 

 gives, of course, increased variation, illustrating, like most of our 



