318 



SCIENCE 



[N. S. Vol. XXXVI. No. 923 



subspecies or varieties, or to determine how 

 many of the named and recorded species will 

 ultimately have to be relegated to the scrap 

 heap and be listed only as synonyms. Still 

 less is the role of prophet assumed and an at- 

 tempt made to go beyond present returns and 

 indicate how many vertebrate species yet re- 

 main to be described, although it is believed 

 that in the case of some orders (as, for in- 

 stance, birds and mammals) reasonably good 

 guesses might be made. The estimate is as 

 follows : 



1. Mammals 7,000 



2. Birda 20,000 



3. Crocodiles and turtles 300 



4. Lizards 3,300 



5. Snakes 2,400 



6. Frogs and toads 2,000 



7. Salamanders 200 



8. Fishes 12,000 



Total : Jjjm 



H. W. Henshaw 

 Washington, D. C, 

 August 5, 1912 



SPECIAL ARTICLES 



ON THE SIGNIFICANCE OF VARIETY TESTS 



In the United States, considerable money 

 and effort have been devoted to' " variety 

 tests." This has been done upon the assump- 

 tion that the relative yield of a given variety 

 in one year is a reasonably good criterion of 

 its relative value in a subsequent year. But 

 to some of us, the value of variety tests as 

 they have been carried out by many of our 

 state agricultural experiment stations seems 

 very doubtful.' 



' There are several difficulties which have been 

 but poorly met in the problem of variety testing. 

 The identity of the variety must be beyond ques- 

 tion, but in many cases there may be grave doubts 

 as to the authenticity of the identifications, and 

 in the absence of herbarium records, it is impos- 

 sible to correct errors. The organization of scien- 

 tifically managed seed growers' associations may 

 be expected to overcome this difficulty in large 

 part. Again, varieties differ in their edaphic and 

 climatic requirements. Tests made in one place 

 may give results not at all applicable to other 

 localities with different conditions. Where the 



The utility of a test of n varieties is meas- 

 ured by the correlation between the yields of 

 the individual varieties for different years. 

 If the correlation be 0, the yield of a variety 

 in 1912 furnishes no criterion of its probable 

 productiveness as compared with others in 

 1913. If the correlation be high, then the 

 prediction of yield from one year's test may 

 be made with great certainty. 



Let us apply this test to a series of data 

 given by Half for eleven years' test of a num- 

 ber of varieties of wheat at Eothamsted. I 

 presume we can look upon these tests as not 

 only more extensive but more trustworthy 

 than many or most of those in experiment 

 station records. 



We may assume that, aside from the errors 

 of sampling, two kinds of influences deter- 

 mine observed yield: the innate capacity of 

 the variety and the conditions of grow^th to 

 which it is exposed — that is, the influences at- 

 taching to soil and season. We may correct, 

 in part at least, for the influence of season by 

 determining the mean yield of all the varieties 

 for each year to the nearest tenth bushel and 

 expressing the yield of each variety for that 

 year as a deviation from the general yearly 

 mean. These deviations with their signs 

 show in concrete terms the relative superior- 

 ity or inferiority of a variety for a given 

 year. Its value agriculturally, of course, de- 

 pends upon the consistency with which it 

 maintains its superiority from year to year. 



Table I. has been prepared from Professor 

 Hall's (which is arranged according to the 

 superiority of varieties as judged at Eotham- 



tests are made by wide cooperative experiments, 

 this difficulty may be overcome, but when work is 

 confined to a central station its value for a diversi- 

 fied state is o priori doubtful. Third, any test is 

 subject to the probable errors of random sampling, 

 and for the most part we have been given no 

 means of estimating the magnitude of this measure 

 of possible untrustworthiness. If the empirical 

 measure of the desirability of a given variety is 

 misleading in a particular year, it is of little 

 value for predicting the probable yield of the 

 variety in a subsequent year! 



= Hall, A. D., "The Book of the Eothamsted 

 Experiments," p. 66, 1905. 



