igos.j JENNINGS— HEREDITY IN PROTOZOA. 483 



on that relative proportion of breadth to length that belongs to 

 well-fed specimens. 



Thus, we find almost throughout that an increase in the ratio of 

 breadth to length is accompanied by an increase in the coefficient of 

 correlation ; a decrease in the ratio of breadth to length by a decrease 

 in the coefficient of correlation. Examining these two constants, in 

 the last two columns of Table XVIII. , we find this relation to hold 

 in every case of experimental procedure save one. (In the change 

 from row 3 to row 4 it does not hold ; this is due to another factor, 

 to be taken up later.) If without regard to experimental pro- 

 cedure, we merely compare the mean index (or ratio of breadth to 

 length) with the coefficient of correlation, we find the relation a little 

 less general, though still marked ; a large mean index is usually 

 accompanied by a high coefficient of correlation. 



Since, as we have previously seen, greater breadth is usually 

 accompanied by a higher mean index, it follows that greater breadth 

 is likewise usually accompanied by a higher correlation between 

 breadth and length. This is, on the whole, evident on inspection of 

 Table XVIII. , though since other factors are involved, the relation 

 is not without exception. But in general, broader specimens tend 

 to show a more constant proportion of breadth to length than do 

 thin ones. 



(/) In poorly-fed cultures, as we have just seen, the breadth is 

 apt to be variable in proportion to the length (giving low correla- 

 tion) because some of the individuals get more food than others. 

 But if all are reduced to an actually starving condition, then this 

 source of variation is removed, and we may again get high corre- 

 lation between breadth and length. This condition appears to be 

 realized in row 26 of Table XVIII. Here a large culture had been 

 reduced by starvation to a population of but '^y, and these give the 

 very high correlation of .8018 zb -0396. 



(^) When a given agent causes rapid multiplication, so that the 

 sample taken includes many different stages of growth, with their 

 diiterent proportions of breadth to length, the correlation becomes 

 low. This is the reason for the marked decrease in correlation in 

 changing from row 3 to row 4 in Table X\TII. 



All together, it is clear that no particular coefficient of correlation 



