METHODS 183 



might very well not agree with my selection. I habitually use at least 

 four statistical books for this reason, and my advice is that if difficulty 

 is experienced with one book, another should be consuked. I cannot 

 easily imderstand a new statistical topic by reading about it in an 

 annchair. Not until I sit down with paper, pencil and calculating 

 machine can I really see what the author is trying to tell me. 



For the ecologist, a book or books on general statistics will be 

 required. There is now a considerable number of these. Ecology is a 

 very practical science, and the best books for the purpose are probably 

 those with an industrial outlook, for the irregular data that must often 

 be used are most likely to be treated in such books, and may not be 

 adequately considered in books intended for the laboratory experi- 

 menter. For this reason, books that lay excessive emphasis on the 

 analysis of variance may not be useful, for this excellent method is an 

 experimenter's tool. It requires a high degree of symmetry, and if there 

 are serious gaps in the record, it cannot be applied at all. In the spawn 

 date investigation, for example, observers entered and left the scheme 

 all through its existence, no two areas were equally represented, no 

 two years had the same number of observations. Correlation analysis 

 is unaffected by these features, but the analysis of variance cannot be 

 appHed to data in tatters. Williams in suggesting that the analysis of 

 variance could be apphed was perhaps misled by the apparent sym- 

 metry of the published summaries. These looked temptingly like 

 data suitable for this technique, but the illusion does not survive 

 closer acquaintance. After what I have written about the necessity of 

 individual choice of a general book, it is of course absurd to make too 

 strong recommendations, but I would suggest that Davies (1949) and 

 Brownlee (1949) should be tried. Fisher (1932), the first book to be 

 written for research workers, is often thought to be difficult, but, in 

 fact, it is sometimes easy to fmd the right method there. In six months 

 or less, any person likely to need those methods should be able to apply 

 them by following these books. I personally make no attempt to 

 remember the details, but, unless the method is one I often use, I work 

 with the books open beside me. Intricate details need not be memorized 

 as if for an examination. 



No general book on statistics known to me contains an adequate 

 account of correlation analysis. Most do have a chapter or two on the 

 subject, but that is not enough, and the methods thus briefly explained 

 are too simple for ecology. The book by Ezekiel (1930) is probably the 



