Predicting the Sex of Dovekies by Discriminant Analysis

Author(s): Dariusz Jakubas and Katarzyna Wojczulanis

Source: Waterbirds, 30(1):92-96.
Published By: The Waterbird Society
DOI: http://dx.doi.org/10.1675/1524-4695(2007)030[0092:PTSODB]2.0.CO;2
URL: http://www.bioone.org/doi/
full/10.1675/1524-4695%282007%29030%5B0092%3APTSODB%5D2.0.CO
%3B2

BioOne (www.bioone.org) is a nonprofit, online aggregation of core research in the

biological, ecological, and environmental sciences. BioOne provides a sustainable
online platform for over 170 journals and books published by nonprofit societies,
associations, museums, institutions, and presses.
Your use of this PDF, the BioOne Web site, and all posted and associated content
indicates your acceptance of BioOne’s Terms of Use, available at www.bioone.org/
page/terms_of_use.
Usage of BioOne content is strictly limited to personal, educational, and non-
commercial use. Commercial inquiries or rights and permissions requests should be
directed to the individual publisher as copyright holder.

BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofit
publishers, academic institutions, research libraries, and research funders in the common goal of maximizing access to
critical research.
 Predicting the Sex of Dovekies by Discriminant Analysis
DARIUSZ JAKUBAS AND KATARZYNA WOJCZULANIS
Department of Vertebrate Ecology and Zoology, University of Gdansk,
´ al. Legionów 9, 80-441 Gdansk,
´ Poland
Internet: biodj@univ.gda.pl

Abstract.—We examined sexual dimorphism of Dovekie (Alle alle) from the breeding colony in Hornsund
(south Spitsbergen). Each bird was sexed using DNA extracted from blood. Body morphometrics including flat-
tened wing, head-bill, tarsus length, bill width and body mass were recorded in 331 birds caught during the breed-
ing season in 2003-2006. Analysis of measurements of mated pairs showed that males are usually bigger than females
in case of head-bill length, bill width and body mass. Only head-bill length was significantly correlated within a pair
(r69 = 0.354, P < 0.005). A forward stepwise discriminant function analysis (DFA) was used to evaluate sexual size
dimorphism in Dovekie. The first function, applied to 4 measurements, identified head-bill length and bill width
as the best measurements for sexing and correctly classified 70% of our sample (but only 38% better than chance).
The second function, which was applied to only 3 parameters (excluding bill width), identified head-bill length as
the best measurement for sexing and correctly classified 65% of our sample (30% better than chance). In Dovekie,
a species having a high degree of overlap between the sexes, discriminant function should be used with caution. It
is recommended that a combination of morphometrics, behavior observation and genetic analysis be used to obtain
the highest accuracy in sexing individuals correctly. Received 30 January 2006, accepted 9 August 2006.
Key words.—Dovekie, Alle alle, sex determination, discriminant function, genetic analysis, sexual dimorphism.
Waterbirds 30(1): 92-96, 2007

Correct sex identification in monomor- mentioned studies where those differences

phic bird species is crucial for interpretation
of ecological and behavioral data. Sexing by
dissection is easy but is controversial ethical-
ly and is not appropriate for majority of stud-
ies because it eliminates the main object of
examination. For many species sex can be
determined by other, humane methods
based on measurements, behavior and DNA
analyses. Discriminant analysis has been
used widely to identify sex of monomorphic
species of birds (e.g., Evans et al. 1993; Phil-
ips and Furness 1997; Jodice et al. 2000; Dev-
lin et al. 2004). This method generates the
function based on a combination of several
morphometric variables that best distinguish
individuals of known sex. This function al-
lows the identification of sex of unknown in-
dividuals (Sokal and Rohlf 1981). Simulta-
neous use of various methods allows the ver-
ification of accuracy and applicability of dis-
criminant analysis to sexing.
In the Dovekie (Alle alle), like in other Al-
cid species, sexes are generally similar in size
and indistinguishable in plumage (Stemp-
niewicz 2001; Montevecchi and Stenhouse
2002). A few studies showed intersexual dif-
ferences in mean body mass (Roby et al.
1981; Stempniewicz et al. 1996), tarsus and
culmen length (Bradstreet 1982). However,
were described based often on small sample
size of sex-determined individuals, ranges of
measurements were not given and/or strong
intersexual overlap occurred. Only in one
paper (Taylor 1994), discriminant analysis
was used to sex Dovekies. However, equation
of the discriminant function was not present-
ed. In this context, there are no clear criteria
for sexing this species in the field using mor-
phometrical data.
In the present study a discriminant func-
tion was developed to sex Dovekie nesting in
Hornsund (South Spitsbergen) from a sam-
ple of birds that were sexed using molecular
methods. The aim of this study was to esti-
mate the accuracy of using discriminant
function to sex this species in the field. DNA-
based techniques were used to verify identifi-
cation based on discriminant function.
METHODS
Dovekies were caught in their breeding colony at
Ariekammen slopes (77°00’N, 15°33’E) in Hornsund
(South Spitsbergen) in 2003-06. Adult breeding individ-
uals were removed from the nests during the incubation
period. In total, 144 birds, which formed 72 pairs were
caught. Apart from paired individuals, 187 adult birds
(78 females and 109 males) from unknown nests were
caught in mist net standing on the surface of the colony
during the incubation and chick rearing period. Indi-

92
 SEXING DOVEKIES
viduals were ringed, measured and weighed. We mea-
sured the flattened wing length (from the bend in the
wrist to the tip of the longest primary with flattening
wing), head-bill length (distance from the back of the
head to the tip of the bill), bill width (from left to right
edges of the bill at the base; measured in 2004 and
2006), tarsus (the diagonal distance from the middle of
the joint between tibiotarsus and tarsometatarsus to the
junction of the tarsometatarsus with the base of the mid-
dle toe). All measurements were taken to the nearest 0.1
mm with dial callipers except for wing length, which was
measured to the nearest 1.0 mm using a stopped wing
ruler. Body mass was measured to the nearest 1.0 g using
300 g Pesola spring balance. All birds were measured by
the same worker.
A forward stepwise discriminant function (DFA) was
used to determine the best measurements to identify
the sex of Dovekie. The effectiveness of the DFA was as-
sessed first in terms of the proportion of adults of
known-sex that were classified correctly using all indi-
viduals in the analysis (self-test), and second by cross-val-
idation (each case is classified by the functions derived
from all cases other than that case; SPSS, Inc., Chicago).
Due to unequal sample size for males and females
chance-corrected procedure (Cohen’s kappa statistic)
was used to determine if the classification was better
than random or chance alone (Titus et al. 1984). Statis-
tical analyses were done using SPSS for Windows, Ver-
sion 14.0 (SPSS, Inc., Chicago). Cut-off points were
calculated for each analysis so as to later assign sex to in-
dividuals in the predictive group. Cut-off points were
calculated as the weighted average of values of discrimi-
nant scores (i.e., the value calculated by summing up
the constant and all coefficients once the data are en-
tered) for each sex (means were weighted by number of
males and females; Hair et al. 1995). Adults with discrim-
inant scores greater than the cut-off point were classi-
fied as male and those with lower scores as female with
probability computed from group sizes (Fig. 1). Since
new individuals were caught each year, we assumed that
interseasonal differences in measurements were a result
of variability in the Dovekie population. Variances of
each variable for males and females (for all studied sea-
sons combined) were not significantly different (Lev-
ene test, wing length: W329 = 1.131, n.s.; head-bill length:
W324 = 0.003, n.s.; bill width: W142 = 0.0004, n.s.; tarsus:
W229 = 1.122, n.s.). Also body mass was consistent within
sexes (W320 = 1.714, n.s.). However, this parameter varies
in the different periods of the breeding season (Stemp-
niewicz 1981; Taylor 1994), so it was excluded in this
study in order to create a function that could be used to
Figure 1. Probability of being Dovekie male in relation
to discriminant score from Function 1. Note: the major-
ity of data points overlap.
93
sex adults caught during the whole breeding season.
The covariance matrices do not differ between groups
for two applied sets of data with different sample size:
wing, bill-head, tarsus length, bill width (Box’s M test –
1.289, F3, 1248851 = 0.423, n.s.) and for wing, bill-head and
tarsus length (Box’s M test – 1.841, F1, 148671.6 = 1.833,
n.s.). Data did not exhibit multicollinearity (r < 0.29 for
all pairwise correlation). All data met the assumptions
of normality (Shapiro-Wilk test, P > 0.05) except for
males wing length (W = 0.977, P < 0.005). However, data
were not transformed, because discriminant analysis is
robust with respect to this assumption (Tabachnick and
Fidell 1996).
Blood was used for DNA-based sex identification. A
small blood sample (50 µl) was taken out from the bra-
chial vein of each bird, immediately suspended in 1 ml
lysis buffer and stored until molecular analysis in the
laboratory at the University of Gdansk ´ (Poland) and in
Natural History Museum at the University of Oslo (Nor-
way). DNA was extracted from 100 µl blood and buffer
solution, using the Blood Mini kit (A&A Biotechnology,
Gdynia, Poland). The protocol described by Griffiths
et al. (1998) was optimized to amplify the CDH-W and
CDH-Z genes located on the avian sex chromosomes us-
ing two sets of primers: P2 and P8 (for samples from
2003) and 2550 F and 2718 R (for samples from 2004
and 2005). Gel electrophoresis revealed one band in
the male and two in the female.
RESULTS
Mean head-bill length, bill width and
body mass differed significantly between
males and females (Table 1). However, the
majority of those values strongly overlapped
between sexes (81-91%). The mean length
of flattened wing and tarsus were similar in
both sexes (Table 1).
Due to the small sample size of bill width
(data for only two seasons), two separate dis-
criminant analyses were carried out. The
first function was applied to 4 measurements
and the second to 3 measurements (exclud-
ing bill width). When a stepwise discrimi-
nant function was applied to 4 morphologi-
cal measurements, one equation best sepa-
rated the sexes. This function identified bill-
head length and bill width as the best mea-
surements to identify the sexes.
Function 1 = Head-bill * (0.55) + Bill width *
(1.30) – 46.63
This equation assumes a priori probability of
being male of 0.58 and female of 0.42 (com-
puted from group sizes) and results in the
cut-off point of D = -0.09 (Table 2, Fig. 1).
This function correctly classified 70.2% of
141 adults of known sex (60.0% of 60 fe-
94 WATERBIRDS

Table 1. Body measurements of all caught Dovekie, Student t-test, level of significance and overlap (%) of measure-
ments between both sexes.

Female Male Comparison

Overlap
Measurement Mean N SD Mean N SD t-value P (%)

Wing [mm] 125.2 150 3.18 125.6 181 2.87 -1.06 n.s. 99.4
Head-bill [mm] 52.9 148 1.09 53.7 178 1.07 -6.97 <0.0001 88.3
Bill width [mm] 12.5 60 0.63 13.1 84 0.66 -5.23 <0.0001 81.3
Tarsus [mm] 20.2 102 0.67 20.3 129 0.67 -1.74 n.s. 93.5
Body mass [g] 161.1 148 11.45 166.2 174 10.38 -4.15 <0.0001 90.7

males and 77.7% of 81 males). The result of
the cross-validation test produced the same
classification results. Chance-corrected pro-
cedure showed that classification was 38%
(kappa = 0.383, SE = 0.087, P < 0.001) better
than chance.
The second function applied to 3 mea-
surements (excluding the bill width) identi-
fied head-bill length as the best measure-
ment to identify the sexes.
Function 2 = Head-bill * (0.99) – 52.92
This equation assumes a priori probability of
being male of 0.55 and female of 0.45 (com-
puted from group sizes) and results in the
cut-off of D = -0.03 (Table 2). This function
correctly classified 65.3% of 323 adults with
known sex (62.2% of 148 females and 68.0%
of 175 males). A cross-validation test also cor-
rectly classified 65.3% of the sample.
Chance-corrected procedure showed that
classification was 30% (kappa = 0.302, SE =
0.056, P < 0.001) better than chance.
Each of the discriminant functions tested
was significant. However, Function 1 ex-
plained more (51%) of the variability in the
data than Function 2 (32%; see squared ca-
nonical correlation values, Table 2).
Analysis of measurements of mated pairs
showed that males are significantly bigger
than females in case of head-bill length, bill
width and body mass (Table 3). However,
percentage of males bigger or the same size
as females was high (29-49%; Table 3). Only
head-bill length was significantly correlated
within a pair (r69 = 0.354, P < 0.005). The re-
lationship in case of the rest parameters was
not significant (wing: r70 = -0.118, bill width:
r32 = 0.275; tarsus: r63 = -0.063, body mass: r66
= 0.013, n.s.).
DISCUSSION
There are overall size differences in at
least part of measurements between males
and females of Dovekie, but there is much
overlap, which reduces discrimination be-
tween the sexes. In 60-71% of the pairs mea-
sured, the larger head-bill length, bill width
and body mass alone could correctly identify
males within pairs. The discriminant func-
tion calculated in the present study allows
classifying anonymous birds with unknown
partner. Sexing Dovekie according to the
generated Function 2 based on head-bill
length allows correct identification of 65%
of birds. Adding bill width to the model im-
proved its efficiency to 70%. In fact, calculat-
ed kappa value indicated that presented
functions were classifying significantly better
than chance (38% for Function 1 and 30%
for Function 2), but not nearly as well as self
test and cross-validation would indicate. Due
to low value of kappa, Function 2 although

Table 2. Parameters of the discriminant functions developed to identify the sex of Dovekie.

Discriminant Squared canonical Wilks’

Function correlation Lambda F df P Cutting score

Function 1 0.512 0.738 24.11 2,136 <0.001 D > -0.09 = male

Function 2 0.322 0.896 26.11 1,226 <0.001 D > -0.03 = male
 SEXING DOVEKIES
Table 3. Size differences between males and females in pairs of Dovekie nesting in Hornsund.
Females larger Both sexes the
Measurement than males [%] same size [%]
Wing [mm] 38.9 9.7
Head-bill [mm] 25.4 7.0
Bill width [mm] 23.5 5.9
Tarsus [mm] 41.5 4.6
Body mass [mm] 36.8 2.9
basing on bigger sample size, should be
treated with more caution.
In both applied functions, parameters
related to head and bill were the best fea-
tures for distinguishing the sexes. Similarly,
in Taylor’s study (1994) bill length and
depth were the two most discriminating
variables and correctness of applied func-
tion was bigger than in the present study
(above 80%). However, it is hard to inter-
pret the classification result of mentioned
discriminant analysis because sample sizes
were unequal, details of discriminant func-
tion was not presented and kappa value was
not calculated. It is possible that additional
measurements of bill better differentiate
males and females of Dovekie.
Caution should be used when applying
any discriminant function to a new group of
birds. The effectiveness of discriminant
analysis depends partially on the stability of
characters employed in the function. Such
morphometric variables as wing length or
bill dimension may change with age. In
Dovekie, subadults from Baffin Bay and
Bjørnoya differed significantly from adults
in the majority of body measurements
(Bradstreet 1982; Stempniewicz et al. 1996).
Other traits, e.g., mass may change season-
ally. Body mass of Dovekie was not applied
in the present study because it varied clearly
in the different periods of the breeding sea-
son (Stempniewicz 1981; Taylor 1994; own
data). Mass could be used if all birds are
weighed at the same stage of the season and
there is limited variation among years (Phil-
ips and Furness 1997). The consistency be-
tween workers running measurements is an-
other factor that can limit the effectiveness
of discriminant analysis. It is best to mea-
sure all birds with one worker with the same
95
Males larger than Paired
females [%] N pairs t-test P
51.4 72 -0.40 n.s.
67.6 71 -4.70 <0.0001
70.6 34 -3.10 <0.005
53.8 65 -1.73 n.s.
60.3 68 -2.72 <0.01
equipment. In order to apply a proposed
discriminant function to an independent
sample of birds measured in another place,
it is important to recognize geographical
size variation. In Dovekie, geographical
variation has been found among birds from
Spitsbergen, Bjørnoya and Franz Josef Land
(Stempniewicz et al. 1996). In this context
the discriminant function presented here
should be used only for Spitsbergen birds.
However, the problem of geographical vari-
ation could be solved by calculating a popu-
lation-specific cut-off value directly from
discriminant scores using a maximum-like-
lihood method, without the requirement
for any birds of known sex in the sample
(van Franeker and ter Braak 1993).
The DFA presented in this study allows
the sex identification of more than half of
studied group of birds. It is an easy and non-
invasive method and can be applied to data
gathered in the past (when genetic identifi-
cation of sex is impossible). However, it
should be used in the field with caution. To
minimize a risk of wrong sex identification,
individuals with canonical values close to cut-
off points should be sexed by other methods
(DNA-based analyses and observation of
copulatory position).
ACKNOWLEDGMENTS
The authors are very grateful all staff of the Polish
Polar Station at Hornsund (Institute of Geophysics, Pol-
ish Academy of Science) for all help. We thank Dr. L.
Wennerberg for help with molecular sex identification
and Prof. L. Stempniewicz and an anonymous reviewer
for critical comments on the first version of the manu-
script. This project was supported by a grant from the
University of Gdan´sk (BW 1140-5-0162-3) and Ministry
of Education and Science (0640/PO4/2005/29). All
birds were handled under the permission of the Norwe-
gian Animal Research Authority and the Governor of
Svalbard.
96 WATERBIRDS
LITERATURE CITED
Bradstreet, M. S. W. 1982. Pelagic Feeding Ecology of
Dovkies, Alle alle, in Lancaster Sound and Western
Baffin Bay. Arctic 35: 126-140.
Devlin, C. M., A. W. Diamond and G. W. Saunders. 2004.
Sexing Arctic Terns in the Field and Laboratory,
Waterbirds 27: 314-320.
Evans, D. R., E. M. Hoopes and C. R. Griffin. 1993. Dis-
criminating the sex of Laughing Gulls by linear mea-
surements. Journal of Field Ornithology 64: 472-
476.
Griffiths, R., M. C. Double, K. Orr and R. J. G. Dawson.
1998. A DNA test to sex most birds. Molecular Ecol-
ogy 7: 1071-1075.
Hair, J. F., Jr., R. L. Anderson, R. L. Tatham and W. C.
Black. 1995. Multivariate data analysis. Prentice
Hall, Upper Saddle River, NJ.
Jodice, P. G. R., R. B. Lanctot, V. A. Gill, D. D. Roby and
S. A. Hatch. 2000. Sexing Adult Black-legged Kitti-
wakes by DNA, Behavior and Morphology. Water-
birds 23: 405-415.
Montevecchi, W. A. and I. J. Stenhouse. 2002. Dovekie
Alle alle. In The Birds of North America, No. 71
(Poole, A. and F. Gill, eds.). The Birds of North
America, Inc., Philadelphia, PA.
Philips, R. A. and R. W. Furness. 1997, Predicting the
Sex of Parasitic Jaegers by Discriminant Analysis.
Waterbirds 20: 14-23.
Roby, D. D., K. L. Brink and D. N. Nettleship. 1981. Mea-
surements, chick meals and breeding distribution of
dovekies (Alle alle) in north-west Greenland. Arctic
34: 241-248.
Sokal, R. R. and F. J.Rohlf. 1981. Biometry: the princi-
ples and practice of statistics in biological research.
2nd ed. Freeman and Co., New York.
Stempniewicz, L. 1981. Breeding biology of the Little
Auk Plautus alle in the ornsund region, Spitsbergen.
Acta Onithologica 18: 1-26.
Stempniewicz, L. 2001. Alle alle Little Auk. The Journal
of the Birds of the Western Palearctic. Oxford Uni-
versity Press. BWP Update, Vol. 3: 175-201.
Stempniewicz, L., M. Skakuj and L. Iliszko. 1996. The lit-
tle auk Alle alle polaris of Franz Josef Land: a compar-
ison with the Svalbard Alle a. alle populations. Polar
Research 15: 1-10.
Tabachnick, B. G. and L. S. Fidell. 1996. Using Multivariate
Statistics. Harper Collins College Publishers, New York.
Taylor, J. R. E. 1994. Changes in body mass and body re-
serves of breeding Little Auks (Alle alle L.) Polish Po-
lar Research 15: 147-168.
Titus, K., J. A. Mosher and B. K. Williams. 1984. Chance-
corrected classification for use in discriminant anal-
ysis: ecological applications. American Midland Nat-
uralist 111: 1-7.
van Franeker, J. A. and C. J. F. ter Braak. 1993. A gener-
alized discriminant for sexing fulmarine petrels
from external measurements. Auk 110: 492-502.