460 
THE WILSON JOURNAL OF ORNITHOLOGY • Vol. 124. No. 3. September 2012 
N 8 
>. 
o 
c 
CD 
3 
cr 
<D 
lL 0 - 
3 sec 
FIG. 5. Six examples of pairs of syllable types classified as the same from songs of Woody Island (W) and Esperance 
(E) Silvereves. 
syllables continuously appearing in individuals' 
songs there could be population-level acoustic 
constraints (e.g., the physical and/or biological 
sound environment) that would maintain popula¬ 
tion "identity' across time. Population compari¬ 
sons would have to be done as a ‘snapshot’ in time 
if that is not the ease, and in comparisons any 
population differences might be illusory, as could 
only be measured by repeat sampling over time. 
I turned to a multivariate statistical approach 
with these caveats in mind, and given the above 
problematical typological approach to character¬ 
izing song characteristics. First. I used the values 
of the six acoustic features of the syllables Iron 
three randomly-selected birds (as replicate mea 
sures per bird) trom each ol the two population: 
and applied discriminant analysis to represen 
them in multivariate space to learn if individual: 
are distinguishable. The correlation matrices 
among the acoustic variables had KMO value: 
of 0.652 (Woody Island) and 0.724 (Esperance 
indicating adequate samples, and Bartlett’s sphe¬ 
ricity test resulted in P < 0.001 for both samples 
indicating the matrices were well conditioned anc 
neither was an identity matrix. The syllable 
characteristics of the three Woody Island birds 
(Fig. 6A) were significantly different by discrim¬ 
inant analysis: (Wilks’ lambda = 0.635. F = 9.57. 
P < 0.001) with syllables correctly assigned to 
their respective individuals at better than chance 
probability (bird I = 70%, bird 2 = 50%, bird 
3 - 69%: random expectations 33.3% of syllables 
ass.gned to each bird). The syllable characteristics 
ot the three. Esperance birds (Fig. 6B) also were 
significantly different (Wilks’ lambda - 0,216, 
P — 48.4, p < 0.001) with syllables correctly 
assigned at better than chance probability (bird 
1- 56.5%, bird 2 - 91.4%, bird 3 = 70 7%) 
Distributions Of syllable features in both popula¬ 
tions exhibited considerable scatter and overlap 
etween birds as indicated in two-dimensional 
plots (Fig. 6). but the results suggest the birds 
exhibit a degree of individually identifiable 
features in their song syllables. I then addressed 
the question of whether song syllables of birds 
from different populations also are distinguishable. 
1 averaged the scores for each acoustic feature 
across syllables for each bird in a sample. Thus, 
lor a given bird there were six averaged values 
(one for each acoustic feature), and consequent!): 
the population samples for statistical treatment 
were: Woody Island n - 19, Esperance ri = 10. 
These values were examined with discriminant 
analysis. Evaluating the correlation matrix with 
die KMO test highlighted the small sample size 
with a KMO value of 0.477; although the matrix 
was well conditioned by Bartlett's lest: P < 
0.001. 1 proceeded with the discriminant function 
analysis recognizing the KMO value below bin 
tairly near an acceptable 0.5, together with the 
significant Bartlett’s test. The classification func¬ 
tion resulted in 89.5% of syllables correctly 
assigned to the Woody Island sample and 60.0^t 
ot syllables correctly assigned to the Esperance 
sample. Individual variable correlations (loadings) 
with the discriminant function were significant 
lor frequency modulation, amplitude modulation. 
Wiener entropy, and goodness of pitch (Fig. 7). A 
greater proportion of Esperance syllables than 
those of Woody Island loaded on the discriminant 
function over the lower range of values of am¬ 
plitude modulation whereas Woody Island sylla¬ 
bles predominated in the higher range ol amplitude 
modulation values. The result in the amplitude 
modulation comparison was reversed for the other 
three acoustic features with Wiener Entropy having 
the most marked pattern. Values of the single 
canonical variate differed for the two samples 
(ANOVA: F = 13.8, P = 0.001). and the values for 
Woody Island were less variable than those of the 
Esperance sample (Fig. 8). 
