Nature, 69: 149-150 (December 17, 1903)

Ueber Erblichkeit in Populationen und in reinen Linien. Ein Beitrag zur Beleuchtung
schwebender Selektionsfragen.
By W. Johannsen. Pp. 68. (Jena: G. Fischer, 1903.)

1 Francis Galton's Difference Problem
(Biometrika, vol. i. p. 390).

PROF. JOHANNSEN has set himself a hard task, namely, the reconciliation of the views of Prof. de Vries on mutations with those of the biometric school, particularly with the Galtonian theory of regression. We say a hard task, because to perform the task of reconciliation requires, on the one hand, an intimate knowledge of the mathematical theory of statistics, and on the other a power of clearly defining the exact biological points which are at issue. It is not an easy matter to distinguish between a so-called mutation and an extremely improbable variation; indeed, the utmost caution is needed when we remember that in every case of continuous variation it has been shown theoretically that the extreme variations in populations of even many thousands must be separated by wide intervals, the wider the more extreme the variations.1 Clearly it is practically impossible to distinguish straight off between a "mutation" and an extreme variation in the biometric sense. Both parties would probably agree that only observation of the results of propagating from the individual thus classified could serve as a criterion between the two views.

According to the biometricians, the type of the variation would regress in the offspring, either to the population mean if a "pure line" did not exist, or to the "type of the pure line" if such did exist; in the latter case a change in type from that of the "pure line" could then be produced by selective breeding within the line for a generation or two. According to de Vries, no further change could take place until a new "mutation" appears. Unfortunately, de Vries's own experiments are very far from conclusive in this respect. Thus in his experiments on clover he was not content with the discovery of a mutation, but went on stringently selecting year after year, in exactly the manner in which the biometrician would suggest that a "stock" should be formed from extreme variations. According to the biometrician, two or three generations of selection will form a stock which, while very variable about its type, will yet breed true, or with but small regression.

Prof. Johannsen seems to assume that this result of biometric theory (1898) is the view only of de Vries, who published his conception of the line "as perfectly constant and yet highly variable" three years later. Thus the criterion between the "Biometriker," as Johannsen calls them, and the "Mutators," as we may perhaps call their opponents, cannot be made to turn on the breeding true of "pure lines" or on the variability of such lines about their type. It can only turn on whether, within the "pure line," there exists regression and progression when we breed from variants which are not so extreme as to be at once classed by the "Mutators" as new mutations. Prof. Johannsen had a good opportunity for dealing with this problem in his experimental observations on the bean Phaseolus vulgaris, but he has unfortunately not provided the exact data on which it could be answered. He has shown that the population of bean seeds, as distinguished from bean plants, exhibits Galtonian regression; he may, more doubtfully, be held to have shown that "pure lines" breed true. But this is no reconciliation of the biometric and mutational theories, for both parties accept the breeding true of pure lines.

Unfortunately Prof. Johannsen seems to think that a single bean seed may be taken as typical of a plant, and thus the whole inner meaning of allowance for homotyposis escapes him. If his view—that pure lines show no internal regression—were correct, then the correlation between mother and daughter plants ought to be perfect, for either of them represents the "pure line," and that is "völlig konstant." Unfortunately Prof. Johannsen has not determined this correlation, but from his published material it can be indirectly worked out for the case of the mean weight of the beans produced by nineteen mother plants and their daughter plants. The correlation thus obtained is 0.59±0.13; this might be equal to the imperfect correlation of the biometricians, who find the value for man, horse and dog to be 0.5, but it is very far from the perfect correlation needed by those who assert that there is no regression within the pure line to its own type. Prof. Johannsen's own investigation of this problem (pp. 36-37) is quite fallacious; and this is owing, we think, to inexperience in the use of statistical methods. From this standpoint we should like to protest against any such crude process of determining goodness of fit as that of placing a normal curve down on seven or eight blocks forming a "histogram," and judging the look of the fit. No such test is valid, and, further, he has not yet shown that the normal curve of errors itself is suited to describe the phenomena referred to.

We hope Prof. Johannsen will continue his experiments, but at the same time in this, as in so many other cases, we hold that statistical methods cannot be safely used without proper training. Experiments of a most laborious character may be rendered nugatory because the observer has not started with a clear conception of what statistical processes he is going to employ to deduce his results, nor what observations are needful if any conclusions at all are to be reached by legitimate numerical arguments. The book shows the increasing interest in the problems of inheritance and in biometric methods; it is characterised throughout by a courtesy of tone which is very pleasing when contrasted with some recent controversial papers on heredity; but it fails, and fails badly, to prove any definite point, because the author has not clearly stated his problem, and had he done so has really not the knowledge needful to deal effectively with statistical data.