288 KEPORTS ON THE STATE OF SCIENCE, ETC. 



or random sampling. A well-known case of this is afforded by the paper of Pearl 

 and Parshley (1913, Biol. Bull. 24, 205), who believed that they had clear evidence 

 that in cattle the relation of time of insemination to the cyclical events of oestrus 

 influenced the sex-ratio. Later investigation of a larger body of material, however, 

 convinced them that their first result had been wholly due to chance (Pearl, 1917, 

 Maine Agri. Exp. Station Bull. 261, (3), 130). 



C. Incomplete Record of Adequate Data. 



4. Very often the investigator is so much preoccupied with the solution of a particu- 

 lar question that he is content to record his data incompletely, provided that this 

 will suffice for his special problem. He fails to remember that complete record may 

 make it possible for later investigators to use his original data for the solution of quite 

 new problems. A good example of this is afforded by the classical paper of Bateson 

 and Brindley (Proc. Zool. Soc. 1892, 285) upon dimorphic organs. The authors were 

 concerned to prove that in the beetle Xylotrupes, while the frequency-curve for body- 

 length was of normal type, that for cephalic horn-length was bimodal ; but that in 

 the stag-beetle Lucanus both body-length and mandible-length showed normal 

 frequency distributions. The frequency distributions of these four characters were 

 therefore given singly. Since, however, two measurements were taken and recorded 

 for each individual, it would have been possible to present not only the information 

 immediately required, but also all information bearing on the correlation between 

 body-length and appendage-length, by means of two-way tables. This information 

 was later required by another investigator : luckily the original data had been pre- 

 served, and so new conclusions could be drawn (J. Genetics, 1927, 18, 45). A pre- 

 cisely similar failure to record by means of two-way tables is found in the paper 

 by Djakonov (J. Genetics, 1925, 15, 201) on the bimodality of forceps-length in male 

 earwigs (Forficula). Here again only lucky chance preserved the original data, which 

 were then found to yield new results (J. Genetics, 1927, 17, 309). 



5. Frequently not merely are data published in an incomplete way, but owing to 

 lack of space or for other reasons are not published at all. The danger of this pro- 

 cedure may be illustrated by the benefits accruing from its converse. Haldane 

 (J. Genetics, 1920, 10, 47) was able to demonstrate from Nabour's data on heredity in 

 the grasshopper Parateltix (J. Genetics, 3, 141, and 7, 1) that two factors which the 

 original investigator had thought to segregate independently were in reality linked. 

 He expressly states that this would not have been possible if it had not been for 

 the exceptional fullness of Nabour's records. 



6. Duncker (1903, Biometrika, 2, 307) re-analysed the figures of Yerkes (1901, Proc. 

 Amer. Soc. Arts and Sci. 36, 417), which involved the careful measurement of a number 

 of characters on eight hundred Fiddler-Crabs (Gelasimus). Neither author published 

 the data in full ; and since they were utilised only for certain special purposes, the 

 very fundamental growth-relations between the various organs were not brought out. 

 Duncker himself points out that asymmetry of all the organs on the side of the large 

 chela increases with absolute body-size, but does not tabulate the figures by size 

 classes. It is therefore impossible to arrive at the laws of growth underlying the 

 phenomena. Duncker calculates a number of correlation coefficients from which 

 he deduces certain conclusions. The conclusions would have been much more firmly 

 based, however, if the underlying growth-laws had also been established, as only by 

 so doing can we hope to understand the biological, as opposed to the statistical, 

 meaning of the coefficients. This therefore represents a failure not only to publish 

 the data in full, but also to analyse the data sufficiently fully even for the purpose 

 envisaged by the author. 



D. Failure to Choose the Most Suitable Measurements or Conventions. 



7. Sometimes data are less valuable than they should be because the points of refer- 

 ence used in making measurements are chosen arbitrarily instead of conforming to 

 an accepted standard, or of being chosen with reference to their biological significance. 



An example of the latter procedure is shown by two recent authors (Nomura, 1926, 

 and Sasaki, 1926, Sci. Report Tohoku Imp. Univ. 2, 57 and 197) who have made 

 elaborate measurements of a number of Molluscan shells, with a view to the analysis 

 of relative growth of parts. The results, however, would have been more valuable 

 if measurements had been made of the magnitudes needed for determining the mathe- 

 matical growth relations of a Molluscan shell, as set forth for instance in D'Arcy 

 Thompson's ' Growth and Form ' (Cambridge, 1917), chapters xi and xii. 



