The com position o f our male sample set did not allow the separation o f geographic groups based on each sample locality, as many o f them are represented by only one specimen (Figure 4 .13B).
An alternative approach w as to define impala groups according to their geographical proximity. A criterion o f proxim ity w as established by applying by one degree radius to each sample locality. W hen these radius areas touched or overlapped, the males from these sample locations w ere pooled.
As a result, a total o f 37 males w ere pooled into four geographic groups (Figure 4.14A). G roup EA F com prises 11 males from Tanzania and Kenya; G roup SAP 13 males from South A fiica and Swaziland; G roup ZA M 10 males from Namibia, B otsw ana and Zambia; and G roup A N G 5 males fi"om Nam ibia and Angola. O f the remaining 14 individuals, 12 males w ere geographically to o isolated to take part in a group and 2 males (M 40 and M 32) w ere know n only by their country o f origin.
To visualise the position o f these groups and individuals in a principal com ponent graphic space, the PC A scores w ere plotted and the 51 males labelled accordingly (Figure 4.14B). The application o f proxim ity criteria generated th e separate labelling o f E ast and South A fncan males. In agreem ent w ith the results fi"om the prior analysis these tw o geographical regions share similar morphotypes.
V - 3 - 2 8 EAF A n g o la 26-31 ^ " B o tsw a n a S o u th A f r ic a SAP 2.5 • EAF A SAP ■ ZAM + AN6 ® no g roup
FIGURE 4.14 - (A) African map showing the localities where A. melampus males were collected. The one degree radius of each locality is mariced by a black circle. Geographic groups generated by the application of proximity criteria are marked by red circles. Boxes shows males^ identification numbers. (B) Scatterplot of PCI vs PC2 scores from skull and horn data. Male were labelled according to geographic group illustrated on map.
D ata observation also shows a possible geographical d in e within G roup Z A M . evidenced by males M 14, M 16 and M 25 from B otsw ana and M 39 from Namibia (see Figure 4 .13A and 4.13B). A lthough theses males w ere included in G roup ZA M by geographical proximity criteria (Figure 4.14A), the PC A plot (Figure 4.14B ) shows that these specimens stay slightly apart from the cluster formed by other individuals from this group, corresponding also to their peripheral geographical location.
O nce g rou ps’ members w ere defined, ANO V A w as perform ed to test assum ptions regarding th e differences am ong them. PCA scores and results showed significant differences am ong geographic groups on P C I and PC2 (p<0.05) but not on PC3 (p=0,69). T o identify which pair(s) o f groups w ere involved, the p o s t hoc LSD test w ere perform ed (Table 4.10).
There w as no significant difference betw een G roup EAF and SAF on both P C s (Table 4.10). Considering P C I, G roup ZAM w as significantly different from all other groups. O n PC2 G roup A N G is m ore close to G roup ZA M and significantly different from G roup EAF and SAF.
TABLE 4.10 - Analysis of variance among the four geographic groups defined by criteria of proximity using PCA scores. Differences are significant at p<0,05.
Orniin
Group EA F SAF ZAM
PCI EAF SAF 0,489 ZAM p<0,0001 0,003 ANG 0,348 0,056 p<0,0001 PC2 EAF SAF 0,469 ZAM 0,003 0,103 ANG p<0,0001 0,003 0,171
N ext, a Discriminant function analysis (D FA ) w as carry out to determ ine w hether or not specimens w ould be correctly identified as belonging to their designated group, and try to assign unknowns. D FA w as undertaken using PC A scores. W ilks’ 1, w hich
denotes the statistical significance o f the discriminatory pow er o f the model, and ranges betw een 1.0 (no discriminatory pow er) to 0.0 (perfect discriminatory pow er), w as low enough to imply high case predictability (W ilks’ X = 0,197).
M isclassification cases and groups assigned to unknow n individuals are shown in Table 4.11. The correct self-classification rate w as 54.5% for G roup EA F. 84.7% for G roup SAF. 90% for G roup ZA M and 100% for G roup A N G . The low value for G roup E A F . followed by G roup SAF. is due to the fact that seven out o f the eight misclassified males belong to these tw o groups. In all cases, D FA analyses suggested males fi^om G roup EA F to be classified as G roup SAF (n=5) or vice-versa (n=2). These results confirm that animals from these tw o groups share similar m orphotypes.
TABLE 4.11 - Misclassifîed specimens (asterisked) and classification of originally unassigned specimens (no asterisk) from the Discriminant analyses of A. melampus males. Original group of misclassified specimens are in observed classification column. The column under the header 1 contains the first classification choice, that is, the group for which the respective case had the highest posterior probability, and so on. Correctly classified specimens not shown.
MjiIm OhM^rved 1 Classif.
2 3 4
M3 — ZAM SAF EAF ANG
♦M4 SAF EAF SAF ANG ZAM
M7 — ZAM SAF ANG EAF
M8 — EAF SAF ANG ZAM
M9 — EAF SAF ANG ZAM
MIO — ANG EAF SAF ZAM
*M12 EAF SAF ZAM EAF ANG
♦M14 ZAM SAF ZAM ANG EAF
*M15 EAF SAF ZAM EAF ANG
M26 — ZAM SAF EAF ANG
M28 — ZAM SAF EAF ANG
M31 — ZAM SAF EAF ANG
M32 — SAF EAF ZAM ANG
M35 — ZAM SAF ANG EAF
•M36 EAF SAF EAF ZAM ANG
M40 — SAF EAF ZAM ANG
•M41 EAF SAF EAF ZAM ANG
*M44 SAF EAF SAF ZAM ANG
M50 — EAF SAF ZAM ANG
♦M51 EAF SAF EAF ZAM ANG
From the 12 impalas w ith no group (Table 4.11), 3 males (MS and M 9 from U ganda and M 50 from Tanzania) w ere assigned to G roup E A F . M ale M 32 from S.
A frica and M 40 from A ngola w ere assigned to G roup SA F . Six m ales (M 3, M28, and M3 5 from Zam bia; M 7 from Zim babw e; and M 26 and M31 from M alawi) w ere assigned to G roup Z A M . M ale MIO from T anzania w as assigned to G roup A N G . From these m ales only M 40 and M IC w ere not assigned to groups n ear to their locality o f origin (Figure 4.14A). The m ale M 40 w as captured in A ngola and has no inform ation about locality origin. Results from discrim inant analyses did not classify this m ale as belong to G roup A NG , but instead to G roup SA F. This m eans that source o f this specim en can only be guessed by our analyses. M aybe it cam e from south-east A ngola and belonged to the south-w est end o f the G roup ZA M d in e .
The view taken in this study is that both size and shape m ay play an im portant role in distinguishing A epyceros m elam pus geographical variation and described subspecies. To investigate this w e analysed the position o f the geographic groups and individuals in a principal com ponent graphic space. To do so, the PC A scores from both size-in and size-out analysis w ere plotted (Figure 4.15) and the 51 m ales w ere labelled according to the geographic groups resulting from the proxim ity criteria.
The P C I scatter from size-in data (Figure 4.15A) indicates general skull and horn size. G roup positions show ed that sm aller individuals com prise the ZAM group, follow ed by the SAF group and finally the groups A N G and EAF w ith bigger individuals. The PC2 (Figure 4.15A and B) can be interpreted as a shape axis and show s that group ANG and ZA M share shape sim ilarities and differ from EA F and SAF groups. PC3 gathers few extra shape differences (Figure 4.15B).
The P C I scatter from size-out data (Figure 4.15C ) show s the same patterns presented by PC2 from size-in data (Figure 415.B ), i.e., groups A N G and ZAM sharing shape sim ilarities. A s size-in PC2 still has som e size inform ation, the observation o f the
A - Skull and horn c h a ra c te rs : size-in d ata
B - Skull and horn c h a ra c te rs : size-in d a ta
C - Skull and h orn c h a ra c te rs : size-o u t d a ta
D- Skull and horn c h a ra c te r : size-o u t d a ta
^ FIGURE 4.15 - (A) PCI vs PC2 and (B) PC2 vs PC3 scatterplots corresponding to male skull and horns characters using size-in data. (C) PCI vs PC2 and (D) PC2 vs PC3 scatterplots corresponding to male skull and horns characters using size-out data.
size-out PCA scatterplot confirm s m ore accurately the shape sim ilarities betw een ANG an d ZA M groups and revealed that SAF is som ew hat m ore sim ilar to those groups than to the EAF group, w hich shows a quite distinct position in relation to all others.
As m entioned before, the investigation o f geographic variation should take into account any possible confounding effect o f ontogenetic age variation across samples. T he establishm ent o f groups based on the proxim ity criteria provided new geographical units (see Figure 4.14A ) w hich allow s us to com pare their age distribution. This shows no significant differences across the sam ples (X^ test, p ^ . 7 ) . H ow ever sam ple sizes are sm al (Table 4.12).
TABLE 4.12- Number of skulls of Aepyceros melampus used in the present study sorted age categories and groups based on geographic proximity criteria. YAD= young-adult, ADU= adult, OLD= old and VOL= very old.
VAD Male
ADU OLD VOL
Total
Zam bia (ZAM) 2 4 2 2 10
South A frica (SAF) 1 7 2 3 13
E ast A frica (EAF) 1 10 0 0 11
Angola (ANG) 0 3 2 0 5
Total 4 24 6 5 39
To overcom e this problem we com pared geographical groups in a analysis restricted to a single m ost w ell represented age category am ong geographic groups - the prim e adults (A DU ), by generating categorised box and w hisker plots to illustrate the location and dispersal o f the data (Figures 4.16 to 4.18). R esults indicated that, in univariate analysis at least, size differences am ong the geographic groups are n ot a reflection o f age category effect but do have geographical origin. The patterns o f size variation correspond to those identified in the w hole PCA analysis (all age categories included).
Box Plot f o r VoTKibk 51 (only a d u lt moles)
118
112 C3
SAP ZAM EAF
Box Plot f o r V ariable 5 4 (only a d u lt moles)
Box Plot f o r V ariable 5 6 (only a d u lt moles) 42
SAP EAP
90
78
Box Plot f o r Vomoble 5 2 (only a d u lt moles)
86
CD
ZAM ANG EAF
Box Plot f o r Vorioble 5 5 (only o d u lt moles)
Box Plot f o r Vomoble 5 6 (only o d u lt moles)
M eoivSD M eon-5D AAean-5E Mean 286 284 272 ZAM EAF C 3 Meon*SE M ean-SE - D - Mean 130 SAP
Box Plot f o r Vorioble 5 3 (only o d u lt moles) 78
72
70
ZAM ANG
I , of
— FIGURE 4 . 1 6 - Box and whisker plots showing mean ( □ ) , standard error (I— I) and standard deviation ( —1— ) of measurements SI, S2, S3, S4, S5, S6 and S7 from adult males and by ^ groups based on the geographical proximity criteria (SAF= South Africa group; ZAM= Zambia group; ANG= Angola group and EAF= East Africa Group). See Figure 4.2 for
Box Plot f o r Variable 5 8 (only o d u lt m ales)
52
Box Plot f o r V ariable 511 (only a d u lt m ales)
79 75 73 (ZD ^^eon♦SE M ean-5E SAP ZAM
Box Plot f o r V ariable 51 3 (only o d u lt m ales)
28.5
2 27,5
Box P lot f o r \,toriable 5 9 (only o d u ft m a tes)
75
67
Meon-SE
Box Plot f o r V ariable 512 (only o d u lt m ales)
132
Box Plot f o r V ariable 514 (only o d u lt males)
53
51
SAP ZAM ANG EAF
Box Plot f o r V ariable 5 1 0 (only o d u lt m ales) 278
274
270
254
— FIGURE 4.17 - Box and whisker plots showing mean ( D ), standard error (
(ZZl
) and standard deviation (—I—
) of measurements S8, S9, SIO, S ll, S12, S13 and S14 from adult males and u. by groups based on the geographical proximity criteria (SAF= South Africa group; ZAM= Zambia group; ANG= Angola group and EAF= East Africa Group). See Figure 4.2 forBox Plot f o r Wonoble H I (only o d u lt moles)
g
SAF ZAM ANG EAF
Box Plot f o r Vorioble H4 (only o d u lt moles)
EAF
Box Plot f o r Vorioble H2 (only a d u lt moles)
320
ZAM ANG EAF
Box Plot f o r V ariable H5 (only o d u lt moles)
Box Plot f o r V w io b k H 6 (only o d u lt moles)
ZAM ANG GROUPS Meon-SO I I Meon*SE Meon-SE —O - Mean
Box Plot f o r Vorioble H 7 (only o d u lt moles)
Meon-SD a Meon*SE Meon-SE 34 26 22 1=1 520 400 SAF EAF
Box Plot f o r V ariable H3 (only a d u lt moles)
170
SAF
Meon*SD Meon-SD
Meon-SE
FIGURE 4.18 - Box and whisker plots showing mean ( □ ), standard error (
CZl
) and standard deviation (_ L
) of measurements H I, H2, H3, H4, H5, H6 and H7 from adult males and by groups based on the geographical proximity criteria (SAF= South Africa group; ZAM= Zambia group; ANG= Angola group and EAF= East Africa Group). See Figure 4.3 for 04 measurement description.ANOVA was performed to investigate skull and horn characters contributing to the detected differences among the groups. In Table 4.13 we can see the results o f ANOVA using size-in and size out data. Marked characters are significant at p<0.05, and a simple metric has been devised to show the percentage o f characters which contribute to the differences among geographic groups.
TABLE 4.13 - Analysis of variance among geographic groups defined by criteria of proximity using size-in and size-out data. Percentage of characters contributing to the differences among the groups is given. Marked differences among groups are significant at p<0,05 and represented by a cross in size-in analysis and by red circles in size-out analysis. The simultaneous presence of red circle and a cross, indicates those characters contributed to significant differences among groups in both type of analysis.