Colonialidad del poder y clasificación social
F. Dominación / explotación, colonialidad y corporeidad
missing values reduced the total sample size from 93 to 77. The parent-subset regional sample checking procedure again verified the regional subset samples.
The samples were all left with a greater number of cases than variables (Mariko 8; Chimbu Gorge 44; Nebilyer-Kaugel 19; Erave 6), so it was possible to assess the homogeneity of the covariance matrices using Box's M. The value of the associated F ratio was not significant (p = .5366) and it is reasonable to assume that the dispersion of the group covariance matrices is equivalent.
One of the five variables, nasal breadth, displays a significant deviation from the normal distribution for the Chimbu Gorge sample based on the Shapiro-Wilk statistic (Table 10) . On the strength of previous statements concerning the robustness of discriminant analysis
to violations of the multivariate normality assumption, nasal breadth was maintained in the analysis.
The three discriminant functions computed by the analysis are given in Table 26. The first two functions are both significant (p <
.0001) and combined account for 93.2% of the total variance.
The interpretation of the discriminant functions is again based on the across-groups correlations between the discriminating variables and the discriminant functions (Table 27) . The variables most highly correlated with function 1 are glabella-opisthocranion and basion- nasospinale, indicating that the morphological features contributing most to group separation on the function are vault length and facial prognathism. Function 2 correlates best with cheek height and, to a lesser extent, nasal breadth.
A plot of the canonical discriminant function scores for all cases and the group centroids is given in Figure 15. There is an obvious degree of overlap in the dispersion of the individual group cases. The adjacent positions of the group centroids for Chimbu Gorge and Erave on both functions indicate a particularly strong similarity between these two regions for all of the morphological features identified by the across-groups correlations as significant. Mariko and Nebilyer-Kaugel are the two most disparate groups, primarily on the basis of function 1 which accounts for the major component (69.6%, Table 26) of group separation. Chimbu Gorge and Erave achieve intermediate positions on this function.
As a general interpretation of the plot, Mariko is distinguished by smaller vault length and greater facial prognathism, Nebilyer- Kaugel primarily by greater vault length. Chimbu Gorge (and Erave by virtue of its close association) is distinguished on the basis of
Table 26. Summary statistics for the performance of the discriminant functions produced by Analysis Two, Central Highlands Division discriminant analysis % o f C a n o n i c a l W i l k ' s F u n c t i o n E i g e n v a l u e V a r i a n c e C o r r e l a t i o n L a m b d a C h i 2 d . f . P 1 1 . 1 5 0 6 6 9 . 6 5 . 7 3 1 4 . 3 0 1 0 8 5 . 8 5 15 . 0 0 0 0 2 . 3 8 9 2 2 3 . 5 6 . 5 2 9 3 . 6 4 7 3 3 1 . 1 0 8 . 0 0 0 1 3 . 1 1 2 1 6 . 7 9 . 3 1 7 5 . 8 9 9 2 7 . 6 0 3 . 0 5 5 1
Table 27. Across-groups correlations between discriminating variables and discriminant functions 1 and 2, Analysis Two, Central Highlands Division discriminant analysis
Across-groups correlations
Variable Function 1 Function 2
Glabell a - o p i s t h o c r a n i o n .7006** -.2539 Nasal breadth -.2608 -.4557** Cheek height -.0906 .7339** B a s ion-nasospinale . 6605** .0709 B a s i o n -prosthion .1370 -.0742 N u mber of cases: 77 * p < .01 . /N 1 ★ * p < .001
Table 15. Plot of canonical discriminant function scores for all cases and the group centroids, Analysis Two, Central Highlands Division discriminant analysis
Function 1 4.0 + 0 + C C C C CCC NE C CCCCCC N M C C CENCNN C C MM CCCC NNN M M C M E N N M N N N + + Function 2 - 4.0 + + -6.0 -4.0 -2.0 0 2.0 4.0 6.0
Symbols u s e d in plot: M Mariko
C Chimbu Gorge
N Nebilyer-Kaugel
E Erave
greater cheek height from both Mariko and Nebilyer-Kaugel. These observations axe again derived via a direct inspection of the regional means for these variables (Tables 9 to 12) and by examination of the
signs of the correlation values from Table 29.
The group classification results (Table 28) indicate that Analysis Two has not been as successful as Analysis One in differentiating the four regions, with a success rate of 74%. Mariko is misclassif ied once (12%) , the individual being grouped within the Chimbu Gorge sample. Misclassifications for Chimbu Gorge (16%) are fairly evenly spread throughout the other three regions, whereas both Nebilyer- Kaugel and Erave are, like Mariko, only misclassified with Chimbu Gorge (32% and 100% respectively). These results accurately reflect the interpretation of the canonical plot in identifying Mariko as a separate group which has its only affiliation (albeit slight) with Chimbu Gorge. The very close similarity of Erave to Chimbu Gorge is also indicated, as is a relative similarity between these two regions and Nebilyer-Kaugel.
The general conclusion to be drawn from the results of Analysis Two is that the Division is relatively homogeneous, with only five variables contributing significantly to a pattern of regional variation that separates Mariko from the remainder, primarily on the basis of its greater degree of facial prognathism. It shares a minor degree of similarity with the geographically neighbouring Chimbu Gorge but is well separated from the more western regions. Although Nebilyer-Kaugel can be identified by its greater vault length, on the basis of the classification results it falls within a generally homogeneous subset of the Division which contains Chimbu Gorge, Nebilyer-Kaugel and Erave.
Table 28. Group classification results, Analysis Two, Central Highlands Division discriminant analysis
Predicted group membership
Actual group n MA CG NK ER Mariko 8 7 1 0 0 8 7.5 1 1 2 . 5 0 0 Chimbu Gorge 44 2 37 4 1 4 . 5 8 4 . 1 9.1 2 . 3 Nebilyer-Kaugel 19 0 6 13 0 0 3 1 . 6 68 . 4 0 Erave 6 0 6 0 0 0 1 0 0 0 0
Percent of grouped cases correctly classified: 74.03% 1 Numbers in italics are percentages.
5.5.1.5 Analysis Three
The design of Analysis Three is similar to that of Analysis Two in that it also uses the more statistically conservative criteria of pin and pout values set at .05 to control the stepwise selection procedure of Stage 1. It differs, however, in the variables initially selected for entry in Stage 1: only those variables with significant F values on the basis of the anova described in Section 5.4 are chosen. This is a logical step in the variable reduction procedure, for it excludes those variables that do not contribute significantly to group separation in univariate fashion. Reducing the number of variables initially entered into the analysis also has an added advantage in that it reduces the chances of a case being excluded from consideration due to missing values over a large number of variables.
Stage 1 The eleven variables significant for F in the anova were