1048 THE STRUCTURE OF EVOLUTIONARY THEORY
and bivariate techniques, claimed that intrasample variation in mean shell size
swamped all other factors. Moreover, they could locate no clear evidence at all for
Baker's interregional distinctions.
By using multivariate methods to study the influence of covariance sets,
particularly the jigsaw constraint, upon geographic variation, I was able to resolve
this question (Gould, 1984b) in a way that honored (as partial) the findings of all
these excellent researchers. The Dutch scientists correctly noted the strong influence
of variation in mean shell size among samples. But I was able to show: (1) that this
variation can be isolated on a single factor axis; (2) that size ranges among samples
are effectively equal, and influence the shells in essentially the same way in each of
the four regions; and (3) that these intraregional differences in size almost surely arise
for ecophenotypic reasons (see argument and documentation in Gould, 1984b), based
on more vigorous and continuous growth of shells in moist and well vegetated
microhabitats.
But I also discovered that each of Baker's four regions could be clearly
identified by the evolution of differences that may be small in a genetic sense (a
common situation for geographic variation within species), but that produce
substantial effects upon the adult phenotype by altering several key characters in
tandem through constraints of ontogenetic channels identified by covariance sets. For
example, shells from Bonaire (see Fig. 10-8) grow a distinctively jutting apertural lip,
a consequence of conjoined modification in characters building the third allometric
phase.
Effectively all other geographic variation could be ascribed to the jigsaw
constraint. For reasons that I could not resolve, Cerion develops virtually no variation
in average adult shell size (measured as height plus width) within local populations in
each of Baker's four regions—with a range from 29.79 mm in Eastern Curasao to
30.69 mm for Aruba, giving a maximum interregional difference of only 1.6 percent.
This contingently evolved (and not, obviously, geometrically necessary) invariance of
size triggers a maximal effect for the jigsaw constraint—that is, so long as substantial
variation exists in the sizes of whorls.
Cerion uva does, in fact, exhibit extensive and geographically distinctive
variation in whorl sizes, with regional means spanning almost a full whorl, and
ranging from 8.56 whorls in Western Curasao to 9.35 in Aruba. The maximal "play"
thus accorded to the jigsaw constraint then establishes the interregional distinctions
that Baker had correctly noted but could not adequately characterize. Figure 10- 9
shows minimum convex polygons drawn around the multivariate means for samples
in each region (in a study based on 135 samples of 19 measures on each of 20 snails).
The corners of the triangular diagram represent the first three axes of a factor analysis
for mean vectors of samples. The three axes hold nearly equal explanatory power
(30.5, 34.2, and 32.6 percent respectively, for a total of 97.3 percent of all
information in the 19 measures among samples).
The second axis absorbs the intersample differences in size that led Hummelinck
and de Vries to miss the regional distinctions. The extensive variation on this
dimension does not differentiate the four regions, as indicated by the