102 MISSOURI BOTANICAL GARDEN. 



From these values it appears that the number of sepals and 

 the number of petals are very closely interdependent, all the 

 coefficients falling above .700*. The values obtained by 

 others for Ficaria ranunculoides show correlation ranging from 

 .099 to .291 for sepals and petals. The correlations for sep- 

 als and stamens, and for petals and stamens agree in having 

 the coefficients for 1903 and 1907 insignificantf while those 

 for 1904 lie in the neighborhood of .500, and are clearly sig- 

 nificant with regard to their probable errors. The correla- 

 tions for Ficaria range from .023 to .223 for sepals and sta- 

 mens, and from .224 to .382 for petals and stamens. I am 

 rather surprised to find such low values representing the 

 interdependence of the number of floral parts. 



The foregoing remarks are presented primarily as a contri- 

 bution of quantitative data on a species which has attracted 

 the attention of a number of well-known botanists, first 

 among whom was Darwin. General conclusions cannot be 

 drawn from them, but the disagreement of the constants 

 from the several series, coupled with the statement of taxo- 

 nomists concerning the condition of the dimorphic stamens in 

 different species, seems to indicate that there is a promising 

 field for the attention of some one who can take up the in- 

 vestigation on material growing under more normal condi- 

 tions. Some will say that the disagreements of these con- 

 stants show that biomctric methods are useless in their ap- 

 plication to questions such as this. To my mind quite the 

 reverse is true. They show that there are problems which 

 can only be solved by the delicate technique of higher sta- 

 tistics. And, most important of all, perhaps they indicate 

 the limits of trustworthiness of our conclusions. 



* The probable error of the first coefficient in the table is given as 0, 

 but while this is the value given by the usual formula, too great biological 

 significance must not be attached to it. A slightly larger collection of 

 material might have given r less than 1, and the probable error would then 

 have been rather high because of the smallness of N. 



t I have not considered the form of the regression line in these tables. 

 Possibly the correlation ratio would give a positive relationship between 

 these characters, but I have not felt it worth while to calculate this 

 constant on these series of material. What we need is more data I 



