430 Mr Udny Yule, Fluctuations of sampling 
than that calculated in each case, and for yellows and greens the | 
difference is nearly three times the probable error. Speaking | 
generally, however, the agreement is fairly close, and the data hardly | 
the number of seeds per plant is, however very high—from a | 
mere half-dozen seeds up to two or three hundred—and it was 
suggest any fluctuations of physical significance. The range of 
suggested to me by Mr Darbishire that the poor plants might be | 
exceptional, and that it might be worth while to try the effect of 
their exclusion. Excluding plants with less than 50 seeds in the | 
case of Tables I and II of the original, and plants with less than’ 
100 seeds in the case of Table III, the standard deviations obtained © 
are given in the last four lines of Table C. Here the agreement is | 
distinctly closer than before. Finally use may be made of the | 
fact that Table III of the Report deals with the two characters | 
yellowness and roundness in combination. The expectation of | 
yellow-rounds is 9/16 or 56:25 per cent. and the mean percentage 
given by all the plants of Table III of the Report is close to this. 
} 
Including all plants the actual standard deviation (Table D) is 
lower than that calculated by 2°8 times the probable error: 
omitting plants with less. than 100 seeds the agreement is 
extremely close. In the case of the combination of characters as | 
in the case of the single character there is no evidence of any 
significant fluctuation. 
TasLe D. (Data from Table IIT, loc. cit. above.) Standard devia- | 
tions of percentage of the pair of dominant characters, yellow- | 
round seeds, in F',. Hapectation 9/16 or 56°25 °/,. 
| 
| 
| Number | Percentage of | Standard deviation pq] 
| of plants | yellow-rounds} and probable error | ‘?4 
\Alltplamts Mee 86 56-51 5°36 + -28 6:13 
Excluding plants with 
| less than 100 seeds... 43 56°77 4:27 + 32 4:25 
| 
I turn now to Mr Lock’s paper on his experiments on maize | 
(Annals of the Botanic Gardens, Peradeniya, 11. 1906). Mr Lock | 
himself worked out for one of his most extensive tables (Table 33) | 
for a DR x RR cross the probable error and its theoretical value 
for simple sampling, calculating the actual value by a graphic) 
method: he also pooled together all the cases of expectation 
50 per cent. and made the same comparison. In both cases the 
actual somewhat exceeded the expected value of the probable 
error though the difference was not great. The results of my 
own work on Mr Lock’s data are shewn in Table E. For three 
