456 JENNINGS— HEREDITY IN PROTOZOA. [April 24, 



comes a long period in which both dimensions remain nearly the 

 same — the length increasing slowly, while the breadth fluctuates. 

 Different growth stages during this period have little marked effect 

 on the coefficient of correlation between length and breadth; they 

 tend to prevent its reaching i.ooo, but this it would not reach for 

 other reasons. 



Now, for a certain period before fission (taken as two hours, in 

 our curves) , the length decreases while the breadth increases. Greater 

 breadth will then be associated with less length, tending to produce 

 again a negative correlation. If we make a collection of individuals 

 representing various stages in this process, we should, therefore, 

 expect to find the correlation much less than in collections taken ( i ) 

 either before these processes have begun, or (2) after they are 

 ended. We can realize this, in the main, by taking from a large 

 random sample all the largest specimens (which are, of course, the 

 older ones) and combining these into a single correlation table with 

 specimens from the same culture that are beginning fission (the 

 oldest specimens of the culture). I performed this operation for 

 lot I of Table X. This collection contains 131 specimens beginning 

 fission (row 25, Table X.), and 134 specimens (not dividing) that 

 are 196 microns, or more, in length (row 28, Table X.) ; throwing 

 these together, we have a collection of 264 of the oldest specimens in 

 the culture (row 29, Table X.). For the 131 specimens beginning 

 fission the coefficient of correlation is -]- .6546; for the 134 large 

 specimens it is -j- 4681. When the two are taken together the corre- 

 lation disappears. The computation gives us a coefficient of -|- .0350, 

 but this is less than its probable error (.0415), so that the figures 

 have no significance; no correlation appears. 



The effects of the inclusion of various growth stages on the 

 observed correlation shows itself in many other ways, which will 

 become evident to anyone who carefully examines the data of Table 

 X.. in connection with our curves of growth (Diagram 5), and the 

 relations brought out in the foregoing paragraphs. Note, for exam- 

 ple, the coefficients of correlation for lot 9 (rows 16-18, Table X.). 

 For the specimens 3 to 4 hours old the coefficient is but .3201, and 

 for those 4.20 to 5 hours old it is .5557. When we throw these two 

 lots together, so as to include a much greater proportion of the 



