Karl Pearson 
101 
relationships of A and B (P < 000,000,1) and A and C (P < -000,001) are very 
low, the origin of the second hump in the guinea-pig distribution for A requires 
much more analysis and the certainty by control experiments, that it always 
repeats itself, and is not the result of hitting a " pocket." 
It seems to me that any statistical analysis by modern methods of the trypano- 
some data compels us to confess that either statistical methods must be discarded 
entirely in these trypanosome investigations, or they must be pushed to their 
logical conclusion, and used as the fundamental instrument of research which can 
guide our enquiries by inference and suggestion when, and when only, it is handled 
by the trained craftsman. Thus far the use made of statistical methods seems 
merely to have confused the issues, and brave would be the man who would venture 
to say after reading this section of our present paper that any two strains discussed 
by the commission are definitely "same" or certainly differentiated. 
(5) On the Probability that the Animal in luhich the Trypanosome is cultivated 
makes essential Differences in the Distributions of Frequency. 
But the very method which casts apparent discredit on the lesults at present 
reached seems able to lead us to definite conclusions provided we start with it as 
the fundamental mode of investigation. Really very little inspection seems to indi- 
cate that not only the host but the period of infection materially influences the 
frequency distribution. These points have not been wholjy disregarded by the in- 
vestigators in this field, but they have had no quantitative measure by which they 
could appreciate the relative influence of the various environmental factors. Nor 
indeed could the method be fully applied without experimental observations on 
trypanosomes of the same strain subjected to differential treatment. Knowing in 
such cases the quantitative divergence produced, we should be in a position to infer 
whether two strains from different sources were separate species or merely modified 
by differential environment. Until we have such quantitative measure no hypothesis 
of sameness or difference can flow from statistical treatment ; nobody as yet knows 
how much to attribute to environment, how much to attribute to individuality 
of strain. 
In endeavouring to throw light on this matter we are, however, checked at 
the very start by the absence of effective material. In some cases the period of 
infectivity is not given ; in others we are not always able to break up the total 
frequency by reference to the host, or to a single host. And even when we merely 
classify by one type of animal as host, we may have reduced our n:iaterial to such 
small numbers that samples may be "same," which on larger numbers would 
show the marked divergence due to the emphasis of smaller ditferences'"". Some 
suggestive points can, however, be effectively dealt with and they are treated in 
the following paragraphs. 
* It may not be possible to differentiate Bavarian from Wiirtemberger on samples of 50 crania, 
although quite possible on samples of 400. 
