J. A. Harris 
459 
A constant calculated from a sample does not necessarily describe precisely 
the population from which it was drawn, for it may be too large or too small by 
an amount depending upon the chance errors of random sampling. As biologists 
we are interested in our samples only in so far as they indicate for us the charac- 
teristics of the population. The degree of significance to be attached to the 
constant calculated from a sample as a description of a population as a whole is 
determined by the probable error of the constant. The probable error of random 
sampling has nothing whatever to do with the care of the naturalist in doing his 
work, as some biologists seem to believe. Be the observer never so keen and 
conscientious, the significance to be attached to his conclusions is always to some 
extent dependent upon the number of observations before him on his work-table. 
This is merely a conclusion drawn by common sense from common observation. 
Fortunately in many fields of work the mathematician can give us formulae for 
determining the relative weight to be attached to constants calculated from 
different masses of data. These words of explanation are inserted for the benefit 
of fellow-biologists who may still be unacquainted with the true nature of the 
probable error of random, sampling so much used by biometricians, for throughout 
these pages I find it necessary to return continually to this fundamental safe- 
guard in the interpretation of the significance, as descriptions of populations, of 
constants drawn from samples of the population. 
In the investigation of the problem of selective elimination on the Cercis 
material, I was necessarily limited to the comparison of the constants for the 
original population with those for the eliminated ovaries, since a severe frost 
prevented any of the ovaries maturing. Here we are able to compare eliminated 
ovaries, the original population after partial elimination, and the matured fruits. 
Further details concerning problems and methods will be taken up under the 
individual problems, to which we now pass. 
IV. Discussion of Problems on Basis of Actual Data for 1908. 
Problem 1. Is there a difference in the mean number of ovules per locule, or per 
fruit, in the ovaries which ai'e eliminated and those which develop to maturity ? 
Seriations of the number of ovules per locule are given for these collections of 
material in Appendix, Tables XXVIII — XXX. The data for the total ovules 
per fruit are given for the three collections from the 28 individuals in Tables 
XXXI — XXXIII. The means, standard deviations and coefficients of variation 
for the number of ovules per locule, and the standard deviations and coefficients 
of variation of the total ovules per ovary are given for the 28 individuals of the 
three collections and their totals in Tables I — III. 
In comparing the means for the three series it is unnecessary to treat both 
ovules per locule and total ovules per fruit, for the latter is necessarily three times 
