Systematic review of Nyctimene ceplialotes and N. albiventer 
127 
STATISTICS: RESULTS AND DISCUSSION 
Univariate statistics 
Mean, standard deviation, minimum and 
maximum values and sample size for each island 
are presented in Table 1, this paper, and Kitchener 
et al (1993: Table 1) for all characters examined. 
Multiple regressions 
Multiple regressions were run for skull, dentary, 
dental and external characters on sex and nine 
island populations. The islands of Buru, Banda 
Neira, Halmahera, Ternate and Moratai were 
excluded because of their small sample size and/ 
or absence of either males or females (Table 2). 
Sex 
Metacarpal 3-5 lengths and postorbital breadth 
were significantly influenced by sex (F ]% = 6.11, P 
= 0.015; F i96 = 4.67, P = 0.033; F ]% = 4.653, P = 
0.033; and Fj 96 = 4.639, P = 0.034, respectively. 
Island 
Only postorbital breadth was not highly (<.01) 
significantly related to island. Clearly there was 
considerable difference in morphology between 
these island populations. 
Interactions 
There were no significant interactions. 
Multivariate analyses 
Because the multiple regression analysis 
involved testing a large number of interactions, 
some of these tests will be significant at 
0.05>P>0.01 by chance alone. For this reason, none 
of the characters measured were considered to be 
markedly influenced by sex. As a result both males 
and females are combined for all 23 characters 
examined in the following DFA. 
All populations 
The DFA was first run using all 23 skull, dental, 
dentary and external body characters and the 
islands Aru, Kai Besar, Kai Kecil, the Halmahera 
group (Halmahera, Ternate and Morotai) Papua 
New Guinea, Selaru and Yamdena (Tanimbar 
Islands), Seram, Ambon and Sulawesi. Banda 
Neira and Buru were ungrouped because there 
were only two individuals from each of these 
islands. The above analysis was then repeated 
using only five characters (forearm length, FA; ear 
length EAR; C ( M 2 length, C,M,; width across M'M' 
from labial side, M'M 1 ; and rostral length, RL). 
This analysis extracted four significant functions 
which together explained 99.6% of the variation, 
with 68.0% of individuals classified to their correct 
island group. 
From Figure 2, three broad groups were defined: 
Figure 2 Canonical variate analysis on the following 
island populations: Aru, □ ; Kai Besar, □; Kai 
Kecil, O; the Halmahera group (Halmahera, 
Ternate and Morotai), ■; Papua New Guinea, 
▲; Selaru 9; and Yamdena, A (Tanimbar 
group); Seram, ♦; Ambon, •; and Sulawesi, 
X; Banda Beira +, and Buru *, were 
ungrouped. The DFA plots of functions 1 and 
2 were based on a selection of five characters 
(skull, dentary, and external body), with 
males and females combined. 
the Ambon group (Ambon-Seram-Buru-Sulawesi); 
the Kai group (Kai and Tanimbar islands); and the 
Halmahera group (Halmahera, Aru and Papua 
New Guinea). When a DFA was again run using 
the reduced set of five characters and these three 
island groups, the analysis extracted two 
significant functions (Figure 3). Function 1, which 
explained 78.9% of the variance, separated the 
Ambon and Kai groups from the Halmahera 
group. The character loading heavily on the 
standardised canonical variate coefficients (>0.6) 
on this function was CjMj length (Table 3). 
Function 2, which separated the Ambon group 
from the Kai group, explained 21.1% of the 
variation. The characters loading heaviest (>0.6) on 
Function 2 were forearm and ear length (Table 3). 
A total of 99.2% of individuals was correctly 
classified to their appropriate island group. One 
specimen of the Ambon group classified to the Kai 
