,9o8.] JENNINGS— HEREDITY IN PROTOZOA. 395 



(1902) ; alsoby AlcClendon ( 1908). But we have at present nothing 

 like a thorough analysis of the matter, based on extensive data. 



I. General Methods of Work; Statistical Treatment and 



Its Uses. 



Before we can study experimentally the nature and causes of the 

 existing variations, we must, of course, know their extent, character 

 and distribution. To this end I have made a statistical study, con- 

 structed frequency polygons, and determined the more important 

 constants of variation and correlation. This has. of course, not been 

 done because of belief in any occult virtue in mathematical treat- 

 ment. Statistical methods have been used in this preliminary survey 

 merely because they form the most natural and direct way of discov- 

 ering and displaying the problems on which we wish to work ; I doubt 

 whether the most determined critic of the use of such treatment in 

 biolog}' could suggest any other way for our material. But I am 

 fully convinced that " crucial evidence is always individual in the 

 last analysis '' (Whitman) ; that the preliminary statistical examina- 

 tion of the facts requires development as soon as possible into precise 

 experimental knowledge. It is valuable to know just how many 

 men out of a thousand will die in a given period, but it is infinitely 

 more valuable to know which ones will die if the conditions are not 

 changed, and why; and the latter knowledge includes the former. 

 I have therefore advanced at once from the descriptive statistical 

 work to experimental treatment. A curve or polygon of variation 

 (such as Diagram i ) or a correlation table (such as Table I.) is to 

 be looked upon as a mass of problems. The place occupied in the 

 polygon or table by any individual is due to certain causes, and it is 

 these causes that we seek. 



In seeking these causes by experimental niethods, statistical 

 treatment is again found to be of the greatest value for detecting 

 and registering the effects of single factors, under complex condi- 

 tions. This method may be compared to a microscope ; it enables 

 us to detect and deal with causes and eff'ects which we could not 

 handle without it. I am convinced that it is a great mistake to hold 

 that the onlv or the main use of statistical treatment is for " dealingf 



