THE STATISTICAL STUDY OF VARIATION 49 



several or many different races. Variation in the number and propor- 

 tion of these races in the population would produce wide differences in 

 statistical data. 



4. Germinal diversity among the individuals of a population due to 

 hybrid ancestry. 



Analysis of the causes contributing to bi- and multimodal curves is 

 possible by means of experimental breeding. By testing individuals 

 typical of the various groups indicated by the statistical examination 

 and examining their progeny statistically, the elements composing the 

 original population can be differentiated. It should be noted that the 

 close proximity of two different races sometimes causes contamination 

 of material and consequent skewness of the variation polygon but not 

 necessarily a bimodal curve. 



Correlation. All of the biometrical principles considered in the pre- 

 ceding pages pertain to the analysis of variation in a single character. 

 One of the most striking facts of somatogenesis, however, is the physio- 

 logical interdependence of characters in multicellular organisms. From 

 the earliest stages of embryogeny it is possible to trace associations in 

 the development of various characters. This physiological correlation 

 of characters is one of the most important considerations in the modern 

 study of heredity and it is given due attention in Chapter VI. As regards 

 the statistical study of variation the question to be considered is whether 

 the continuous variations in adult somatic characters are in any cases 

 mutually related or interdependent. It is obvious that, if such a condi- 

 tion be found to exist, it will have an important bearing upon plant and 

 animal breeding inasmuch as selection with reference to a single character 

 would in all likelihood have a definite effect upon certain other characters. 

 The most satisfactory method of investigating this matter is to consider 

 the variation in two characters at a time. 



The Correlation Table. In preparing a correlation table the observed 

 data are transferred directly from the original record by the simple method 

 of tallying. In order to prepare a correlation table either the indi- 

 viduals to be examined must be labelled with permanent numbers or 

 else the observation on the two characters must be made for each indi- 

 vidual before passing on to the next. In either case the datum on each 

 character is recorded under the individual number. Next a table is ruled 

 off with a number of horizontal rows corresponding to the total number of 

 class values for one of the characters and a number of columns equal to 

 the total number pf class values for the other character. It is under- 

 stood that a frequency table for each of the two characters has been 

 previously prepared so that the range of class values is known. In 

 Fig. 21 the material examined consists of the same 400 plants of Sixty 

 Day oats that we have studied with reference to total yield of plant. 



