256 ANDERSON 



little to conventional Statistics. His own method of path coefficients is a 

 semi-graphical method for exploring factor interaction. It is in part a device 

 for making pattern data out of pointer readings.) In Ecology, Cytology, 

 and Haemocytology where Statistics has dealt with frequencies of occurrence 

 using Poisson techniques, it has been effective though scarcely revolutionary. 

 With pattern data, in all the fields with which I am personally conversant, 

 it has always been inefficient and sometimes positively misleading. 



In my own work I have spent the last thirty years in an attempt to 

 measure evolution-in-progress directly from its results, that is, from the varia- 

 tion in living populations of plants and animals. Early in this attempt I 

 worked directly with one of the world's greatest statisticians, Sir Ronald 

 Fisher. I went to him with data on variation in populations of wild irises. 

 He took a lively interest in the problem; with his facile mind he worked 

 out an extension of some of his other techniques which could be applied to 

 such problems, pointed out the kind of data which would be decisive for 

 such a problem, and encouraged me in getting them. When I turned them 

 over to him he worked out his multiple-discriminant function, using them as 

 an illustration (1936), and in his gracious way gave me more credit than I 

 deserved. 



Working with Fisher was a stimulating experience. He cleared many cob- 

 webs out of my thinking. He would say, "Now what do you mean by this 

 statement? // you mean so-and-so then obviously such-and-such follows, but 

 if on the other hand you mean thus-and-so, why then just as obviously this- 

 and-that should follow." On the other hand, he did not in the end sell me 

 on using his methods for my problems. I had been too strongly influenced 

 by E. M. East and his insistence on the exhaustive examination of the phe- 

 nomenon itself. I was muddleheaded compared to Fisher, but my ideas were 

 soundly set on a broad observational base; though I was greatly impressed, 

 I was not bowled over. After all, as Whitehead remarked (Price, 1954), some- 

 times "muddleheadedness is a condition precedent to independent thought, 

 may actually be independent thought in its first stage." Eventually I worked 

 out a simple graphical method for analyzing my iris data. It showed virtually 

 everything I could have gotten out of the problem by analysis-of-variance 

 methods and some things I could not have. A decade later, after various 

 trial-and-error attempts, I originated the pictorialized scatter diagram 

 (1954), a simple, precise, semi-graphical, semi-mathematical way of dealing 

 with multiple-sense-impression problems. It has been rapidly adopted by 

 various workers in my own field, soundly damned as heresy by a few 

 agronomo-statisticians, and roundly praised by a few mathematicians. These 

 experiences have led me to examine critically the relationships between pat- 

 tern data on the one hand and statistics and applied mathematics on the other. 



The point of view exemplified by de Loor's generalizations is common to 

 nearly all the statisticians I have dealt with. It is not shared by the applied 



