534 
NATURE 
[APRIL 4, 1907 
On the Extinct Emeu of the Small Islands off the 
South Coast of Australia and probably Tasmania. 
Some of my colleagues in Australia, as I gather from 
““ Notes” in Narure (vol. Ixxv., pp. 228, 467), have lately 
been at work on the identification of the small emeu of 
the islands in Bass Strait and Tasmania, now extinct. 
Prof. Baldwin Spencer, of Melbourne, having examined 
the bones of the emeu which once lived on King Island 
and found them smaller than those of Dromaeus ater of 
Kangaroo Island, has felt justified in proposing a name 
for that bird, and has called it D. minor. Colonel Legge, 
an old colonist, has also been working on the King Island 
emeu, and proposed for it a name, which, however, he 
withdrew in a postseript to his paper in favour of Prof. 
Spencer’s one already published. From memory, having 
seen a pair in his boyhood, Colonel Legge considers the 
Tasmanian emeu a distinct small species. 
Now I believe that the question of the emeus of small 
size which about a century ago yet lived in Tasmania and 
on the small islands off the south coast of Australia can 
only be settled by a careful comparison of their bones, 
and then, and then only, shall we know whether one or 
more species lived on those islands. I do not know of the 
existence in museums of specimens, either mounted skins 
or skeletons, of well authenticated Tasmanian emeus, but 
we possess two authentic skeletons and two mounted speci- 
mens of Dromaeus ater (Peron), which in the first years 
of last century was abundant on Kangaroo Island: two 
of these four specimens are in Paris, one is in Florence, 
and one in Liverpool. Mine is a skeleton, and is one of 
the three brought alive to France by Peron in 1803 from 
V'Ile Decrés (Kangaroo Island) (Nature, vol. Ixii., p. 102; 
Ibis, 1901, p. 1); the Liverpool specimen is, I think, not 
located; it is undoubtedly D. ater, but might hail from 
King Island or even from Tasmania; it may be the lost 
“lesser emea’’ of the Bullock Museum, dispersed in 1819. 
I may now add that last summer my friend Mr. 
Alexander Morton, director of the Tasmanian Museum at 
Hobart, sent me some bones of the small emeu which he 
had collected on King Island, in Bass Strait, asking me 
to compare them with the corresponding bones of the 
skeleton of D. otery in this museum. I did so at once, 
aided by Prof. E. Regalia, a high authority on ornithic 
osteology; the result of our careful comparison was that, 
barring some slisht differences of purely individual value, 
the remains of the three specimens from King Island ex- 
amined were absolutely identical with the corresponding 
bones of Peron’s specimen from Kangaroo Island. I there- 
fore wrote to Mr. Morton’ (from whom I have not heard 
since) that I had not the slightest doubt that D. ater 
(Peron) once lived on King Jsland, and unless new evidence 
should show the contrary, I am much inclined to favour 
the hypothesis that the samé diminutive emeu once lived 
in Tasmania. Henry H. GIGiiort. 
Royal Zoological Museum, Florence, March 29. 
Mean or Median. 
THE two applications of the median, suggested in Mr. 
Galton’s letter (Nature, February 28) and his article 
(March 7) respectively, seem to me to be somewhat 
distinct. : 
In the case of a jury or committee voting as to a sum 
of money to be given, there is no question of truth, but 
only of expediency. If any amount be proposed and put 
to the vote, the proposition will (by the ordinary way of 
voting) be defeated so long as that amount is above the 
median; the. process of voting tends, therefore, to give 
an amount not greater than the median. Mr. Galton’s 
suggested procedure is in this case, it seems to me, 
quite correct, and a saving of time would be effected if 
the problem were consciously approached from his stand- 
point. 
The case of averaging a series of estimates with the view 
of arriving at objective truth appears to be on a different 
footing. If there is a considerable sprinkling of fools or 
knaves amongst the estimators, or of persons with a 
tendency to bias—as the buyers and sellers might be in 
judging the weight of cattle, according to the suggestion 
of Mr. Hooker—the question as to choice of means is 
one that is difficult to answer. The important question is, 
NOw 1O53.. VOL. 75] 
‘ 
in fact, not the ‘* probable error,’’ but the probable bias, 
for the whole frequency distribution may centre round an 
entirely erroneous value. If, on the other hand, the 
observers are honest and unbiassed, the choice of average — 
turns on the form of the frequency distribution; we 
require that average which is (1) least erroneous, as a 
rule, (2) least subject to fluctuations of sampling—two 
conditions which may very well conflict. As regards (1), 
psychologists, following Fechner, suggest the geometric 
mean, I believe, as the best. But the distribution of 
guesses given by Mr. Galton does not appear to follow 
the law of the geometric mean; if it did, the median 
should be less, not greater, than the arithmetic mean. 
Further, so far as one can judge, the geometric mean 
would give a value as much too low as the median is too 
high. Looking at the distributions in Prof. Pearson’s 
memoir on errors of judgment (Phil. Trans., 1902), there 
seems very little to choose between the mean, the median, 
and the mode; sometimes one is the best and sometimes 
another. 
As regards (2), the probable error of the median has 
been discussed on several occasions by Prof. Edgeworth 
(Phil. Mag., 1886, 1887; Camb. Phil. Trans., xiv., 1885). 
The value is 0-674... /2h/n, where h is the true 
ordinate of the frequency distribution at the median, i.e. 
1/./27.0 for the normal curve. For the normal distribu- 
tion, therefore, the probable error of the median is greater 
than that of the mean in the ratio of 1-25: 1, approxi- 
mately. For a flatter topped curve with more curtate 
tails the ratio of probable errors is greater than 1-25: 1, 
and accordingly for all such distributions the arithmetic 
mean is the better form of average. But for a curve with 
a high central peak and long tails, the probable error of 
the median may be less than that of the mean, and it 
will be the more stable form of average. As an_ illus- 
tration, Prof. Edgeworth has taken the case of a dis- 
tribution compounded of two superposed normal curves 
with the same means and numbers of observations; if the 
standard deviation of the one is to that of the other in 
ratio greater than 2-236: 1, the median has a lower prob- 
able error than the mean. The figure shows the critical 
distribution for which the probable errors of mean and 
median are the same. 
In the absence of definite knowledge as to the frequency 
distribution of estimates in any specific case, it does not 
seem to me that any confident judgment as to choice of 
means can be given. G. Upny YuLe. 
March 26. 
Golden Carp attacked by a Toad, 
Tue following account of a toad attacking a golden carp 
may be of interest to some of your readers from its bearing 
on an ancient belief that frogs and toads are at enmity 
with carp, and kill them by destroying their eyes. Izaak 
Walton in the ‘“‘ Compleat Angler’’ refers to this belief, 
