8235

• UCREFRP Catalog Number: 8235
• Author: Douglas, M. E., R. R. Miller and W. L. Minckley
• Title: Multivariate Discrimination of Colorado Plateau Gila spp.
• USFW Year: 1998
• USFW - Doc Type: The "Art of Seeing Well" Revisited
• Copyright Material: YES

DISCRIMINATION OF COLORADO PLATEAU GILA SPP.

165

they, and many other state biologists, could not see
the clear-cut differences suggested by taxonomists.

Holden (1991) further noted that

this situation exemplified a fairly common phenom-
enon of the time. Many fishery managers either could
not or would not identify the more difficult nongame
fishes to species, especially closely-related minnows
and suckers, or else they misidentified them. Because
these species had little prominence in management
decisions, they were often identified only to family
even though a multitude of kinds were represented.
Taxonomic concerns were the realm of taxonomists,
usually housed in universities, rather than fish and
game departments.

This study was undertaken to clarify morpho-
logical variability within and among big-river Gila
of the Colorado Plateau. The initial hypothesis un-
der test is that the study species are morphologi-
cally indistinguishable. This aspect (i.e., the
swamping of between-species by within-species
variance) baffled fishery biologists for years and
led to the management philosophy described
above. Three different data sets were employed in
this study (i.e., continuous morphometric mea-
surements, discrete meristic characters, and an
amalgam of the two suitable for field use). Results
were compared and contrasted to determine the
efficacy of each character set in discriminating
species. In addition, a discriminant function gen-
erated for all three species from field data was used
to classify juveniles (< 180 mm standard length,
SL) and putative hybrids. Two hypotheses are test-
ed here. The first argues that characters which dis-
criminate adults will also discriminate juveniles.
This issue, an addendum to the original problem
of within-species variability, also plagued early
researchers (Vanicek and Kramer 1969:195). The
second hypothesis is that putative hybrids are in-
termediate in their phenotypes (as per Hubbs et al.
1943) and cannot be unambiguously assigned to
one or another of the parental species.

Methods

#'

Specimens, measurements, and data sets.-Mu-
seum specimens of the three Gila species (to 1978)
were used in this study. These included the fol-
lowing adults and juveniles: G. robusta (81 adults,
106 juvniles), G. cypha (58 adults, 30 juveniles),
and G. elegans (28 adults, 6 juveniles). An addi-
tional17 "hybrids," were examined, as were three
G. seminuda (a hybrid species of G. robusta X G.
elegans ancestry; DeMarais et al. 1992). A series
of 25 morphometric and 10 meristic variables were
collected from each specimen (Appendix 1). A list-

ing of specimens examined is provided in Appen-
dix 2.
Very few values were missing from the original
data (for G. robusta, 50 out of 2,835 metrics
[1.76%]; for G. cypha, 20 out of 2,030 [0.98%];
and 18 out of 980 [1.83%] for G. elegans). Vir-
tually all missing data were morphometric; only
four individuals lacked either the left pharyngeal
bone or its accompanying teeth. Missing values
were estimated from other specimens of the same
species by linear regression of the character under
consideration onto the character that explained the
greatest proportion of its variance (G. D. Schnell,
University of Oklahoma, Missing Data Estimator
program, unpublished). This approach has been
used successfully in other morphometric studies
in which missing values were encountered (see
Douglas et al. 1984, 1992; Schnell et al. 1985).
Morphometric characters were then size-adjust-
ed by subtracting loglo of each measure from 10glO
of the same individual's standard length. Ratio ad-
justment is supported for two reasons. First, the
null hypothesis of collinearity and zero-intercept
could not be rejected in any of the comparisons
between ratio elements, based upon Bonferroni-
adjusted criteria. When such null criteria are met,
ratios correctly remove effects of overall size
(Jackson and Somers 1991). However, the most
important reason for using this technique is that,
when coupled with a discriminant function (be-
low), it provides field workers with a simple meth-
od for classifying Gila specimens in the field, one
that requires only a caliper and a calculator. The
complicated, computer-driven methods of multi-
variate size correction do not allow this (noted also
by McElroy et al. 1997).
The total data were partitioned into three sub-
sets. The 10 meristic characters formed one subset,
and the 25 morphometric characters were a second.
The third subset was composed only of those vari-
ables that could be measured in the field. As such,
it contained 3 (of 10) meristic characters and 19
(of 25) morphometric characters (Appendix 1).
Discriminant analyses andfunctions.-For a lin-
ear discriminant analysis to be optimal (i.e., to
provide a classification that minimizes probability
of misclassifications), certain data assumptions
must be met. Each group must be a sample from
a multivariate normal population; population co-
variance matrices must all be equal; and group
sample sizes must be similar. To avoid these dif-
ficulties, a nonparametric discriminant analysis
(PROC DISCRIM; SAS 1989) was used to eval-
uate specimens for their inclusiveness in desig-