Hi All,
Ken Rogers wrote:
"A word of caution is in order. A significant difference does not mean the
same thing as a large difference. It is one thing to say that males are
larger on average when looking at large samples of each sex. It is quite
another to say what the probability is that a single bird of known size will
be a male. ".
I echo this word of caution, and in fact stated briefly what a stastical
difference between means implies:
"So while a significant difference doesn't preclude overlap in measurements
between the sexes (it indicates that the difference in the means of the two
or more groups being tested are greater than what could be expected by
chance)".
I did not state that any individual whipbird can be reliably sexed by itself
but rather when comparing a known pair (behavioural definition and ASSUMED
to be ONE MALE & ONE FEMALE) then the measurements seem to me at least to be
reliable, as in this case you are not viewing a single individual but rather
comparing two birds you assume to be of the opposite sex: "...... when you
have a pair that are inferred to be a pair from behavioural observations
then the larger bird can be reliably sexed as the male (and the
colour-banding allowed the pairs to be clearly identified through consistent
association such as foraging together, calling to each other with
vocalisations other than song, and also through duetting......"
Ken also wrote: "Also, despite the general excellence of HANZAB's
measurements of skins - much less
measurer error than with live birds - samples are too small to support any
firm conclusions - other, of course, than this one".
I would tend to disagree somewhat. Certainly the sample sizes can be viewed
as relatively small for EWhipbird (26 adult male, 18 adult female nominate
subspecies), however the results of the statistical analysis for a
comparison of adult male and female nominate (to which I will restrict
myself here) showed significance for wing, tail and tarsus at the P<0.01
level (the 'standard' accepted level of a two-way analysis is generally
P<0.05, which is what was obtained for bill). So, the results are highly
satistically significant, despite the relatively small sample sizes.
Statistical power (roughly equating to the ability of the analysis to detect
a difference between males: a smaller sample is more likely to fail to find
a difference between means that is real, than a larger sample, this is
termed a type I error) increases with sample size, so the fact that this
result was obtained with a small (but not inadequate) sample in my mind at
least indicates that the difference is real and a firm conclusion of sexual
size dimorphism (again, it is not absolute and can include overlap in the
measurements of the sexes) is supported. This is also despite any possible
confounding effects of geographical variation in size as specimens were
taken from NSW, QLD and VIC.
I also note that with most of the pairs that I worked with intensively
(behavioural observations) I could readily see the difference in size
between the birds and could confirm the male was the larger by knowing the
colour-bands. I realise this argument is somewhat circular as measurements
are the basis for the sexing. I also did not always know the colour-band
combination when first observing the bird (so this did not always bias my
scrutiny of size).
One last point I wish to make is in regards to the way measurements are
analysed themselves. Currently, most methods focus on univariate measures
(either using wing, tail or tarsus for example by themselves to conclude
with a degree of confidence the sex of any individual bird), it is likely
that multivariate methods will prove much more powerful in resolving
differences between the sexes, and that this method should be adopted more
often (for example, these types of analysis will combine a number of
univariate measures to maximise the separation between the samples, i.e.
male and female. Principal Components Analysis and Discriminant Function
Analysis are two examples of multivariate methods that can be used in such
cases, and have been used with some species with success, e.g. Little
Penguin DFA). Unfortunately, you have the catch 22 problem of needing to
know the sexes first in order to accurately test for differences (without
assuming, whether based on no data or a small amount, that the larger mean
belongs to the male or female sex).
Anyway, while I agree with most of what Ken says, and respect that he is
well-versed in the science of bird morphometrics, I felt the need to clarify
what I was saying.
Cheers, Dean
_________________________________________________________________
ninemsn Extra Storage is now available. 30MB of storage on ninemsn Groups -
great for sharing photos and documents. Go to
http://join.msn.com/?page=dept/home&pgmarket=en-au
Birding-Aus is on the Web at
www.shc.melb.catholic.edu.au/home/birding/index.html
To unsubscribe from this mailing list, send the message
"unsubscribe birding-aus" (no quotes, no Subject line)
to
|
