International Journal of Marine Science, 2017, Vol.7, No.5, 37-50
43
Continued Table 1
Species
Pectoral fin
count
No. of Gill
raker
No. of
the
lateral
line
scales
No. of bars on
caudal fin
Stripes on body
Color of dorsal
tip of 1st dorsal
fin
Color of
barbels
Color of body
U. vittatus
15-16
7-8 + 19-21
36-38
7-8 (4-5 in
upper lobe
and 3 (rarely
4) in lower
lobe
Two yellow or
pale brown
mid-laterally
stripes extending
from eye to the
base of caudal
fin; two
dorsolateral
brown stripes
extending from
operculum to
behind 2nd dorsal
fin
Dark
White
White to
silvery, with
dark brown
dorsally and
white
abdomen
Note: Information based on Uiblein and Gouws (2014, 2015) and Uiblein and Heemstra (2010)
1.2 Transforming measurements to remove size dependent effects
All morphometric distances were first log transformed and tested for normality (Jolicoeur, 1963). The outliers in
the morphometric data may distort the general tendency of the size distribution. Hence, all variables (distances)
were tested for the presence of outliers using Cook’s distance estimates (Cox and Small, 1978). However, no
observations were found significantly influential.
The overall variation of a morphometric distance is the compound effect of size and shape of the organism (Sajina et
al., 2011a; 2011b; Pazhayamadom et al., 2015). Since all the morphometric distances measured in this study are
linear in nature (thus giving high correlations with total length), it is important to remove the effect of size before
subjecting them to statistical (multivariate) tests. For this, an allometric approach was followed (Equation 1) where
the correlation of distances with size of the fish is reduced by converting them into size independent shape variables
(Reist, 1985).
M
trans
= log
M
–
β
(log
TL
–
log
TL
mean
) (1)
M
trans
is the transformed morphometric distance, M
is
the original distance measured,
TL
is the total length of the
fish,
β
is the within-group (location) slope of the regressions of log
M
on log
TL
and
TL
mean
is the location-wise
mean total length.
1.3 Statistical analysis
Correlation coefficients between the transformed distances (
M
trans
) and the total length of the fish (
TL
) were
estimated to validate the efficacy of allometric transformation i.e., of removing size dependent effects. Further, the
M
trans
distances were subjected to principal component analysis (PCA), a dimension reduction technique where most
variations in the data are reduced to a few principal components. The PCAwas employed using the
prcomp
function
in R statistical package (R Core Team, 2014). A scree plot was used to determine how many principal components
should be retained to explain most of the shape variations between the two fish populations. Further, a scratching
procedure as described by Hatcher (1994) was followed to identify distances that loaded heavily on each principal
component. These distances (with high loadings) were considered to be influential because such information alone
can classify the individuals to their respective populations or fish stocks.
A two-sample t-test was employed using the
t-test
function in R statistical package (R Core Team, 2014) to obtain
quantitative information on these influential distances and to determine whether their means are significantly
