CORRELATIONS IN NOTHOSCORDUM AND ALLIUM. Ill 



The solid dots and the circles show the actually observed 

 mean numbers of flowers per inflorescence for the several 

 grades of stem length in the two series. The agreement 

 between observation and theory is not very close, but N is 

 not very large in either series and consequently the means 

 of some of the arrays are based upon an inadequate number 

 of observations. While the observed means do not fall so 

 near the theoretical lines as one might like to see them, it 

 seems quite idle to seek for a better equation than that for 

 the straight line. Regression seems to be, so far as we can 

 judge from the present data, linear. 



The material for Allium stellatum was taken in the fall of 

 1907 from the hills at Meramec Highlands, Mo. The plants 

 were not collected from such restricted habitats as those of 

 Nothoscordum, but were drawn from a considerable range of 

 territory, perhaps four square miles altogether. The method 

 of gathering and measuring the stalk§ was the same as for 

 Nothoscordum, In this case the range of variation is so 

 wide that it is convenient to group in classes of three units 

 for purposes of calculation. The data are presented in Table 

 III; the calculated variation constants, in Table B. The 

 coefficient of correlation between length of stalk and number 

 of flowers per inflorescence is .619±.018. Here again the 

 variation constants require no particular comment and are 

 given merely for the convenience of the reader. The co- 

 eflacient of correlation is shghtly higher than that for Nothos- 

 cordum, but whether any significance is to be attached to 

 the difference could only be determined by the collection of 

 further scries of material. 



The equation to the regression straight line is 



y=— 15.480 + 1.311 x, 



where x= length of stalk and ?/ = number of flowers per in- 

 florescence. The graph. Diagram II, shows that the empirical 

 means fall very close to the theoretical fine for all but the 

 last three classes. Here there appears at first to be some 

 indication of non-linear regression, but a glance at the cor- 

 relation table shows that the means are based on only 4, 3, 

 and 1 observations respectively and hence cannot be givea 



