Behavioral Ecology
doi:10.1093/beheco/arp067
Advance Access publication 14 May 2009
Personality traits in wild starlings: exploration
behavior and environmental sensitivity
Jeroen Minderman,a Jane M. Reid,b Peter G.H. Evans,c and Mark J. Whittinghama
School of Biology and Psychology, Ridley Building, Newcastle University, Newcastle upon Tyne, NE1
7RU, UK, bSchool of Biological Sciences, Zoology Building, University of Aberdeen, Tillydrone Avenue,
Aberdeen, AB24 2TZ, UK, and cSchool of Ocean Sciences, University of Bangor, Menai Bridge, Anglesey,
LL59 5AB, UK
a
Animal personalities, defined as consistent and correlated individual differences in behavioral traits, are suggested to be common
in the animal kingdom and can have important fitness consequences. Individual differences in sensitivity to environmental cues
are predicted to be part of animal personalities and are important because they will affect an individual’s ability to respond to
environmental change. Such environmental sensitivity as a personality trait needs further study because existing studies have
rarely directly related environmental sensitivity to well-established personality traits such as exploration behavior and have
focused on captive animals of specific model species. Using standardized assays of exploration behavior, we show that individual
variation in 1) the speed of exploration behavior and 2) the parts of the environment that are explored are repeatable in juvenile
wild starlings (Sturnus vulgaris). Environmental sensitivity was measured in separate assays and was not correlated with the speed
of exploration behavior. Instead, environmental sensitivity was strongly predicted by what part of the environment was used
during the preceding exploration behavior assays. Thus, in juvenile wild starlings, behavioral traits other than the speed of
exploration behavior better predicted environmental sensitivity. These results suggest that the relevance of exploration behavior
as a personality trait may not be easily generalized across species. Furthermore, although unrelated to exploration speed, this
study illustrates how environmental sensitivity correlates with well-known personality traits and thus further highlights how
animal personalities can limit behavioral phenotypic plasticity in wild populations. Key words: behavioral plasticity, behavioral
syndromes, coping styles, Fair Isle, routine formation, temperament. [Behav Ecol 20:830–837 (2009)]
ndividual differences in the behavioral response to novel or
adverse stimuli are suggested to be widespread in the animal
kingdom (Wilson et al. 1994; Sih, Bell, and Johnson 2004;
Dingemanse and Réale 2005; Réale et al. 2007) and to have
important fitness consequences (Smith and Blumstein 2008).
Such intraspecific behavioral variation can be quantified
along axes such as aggressiveness (e.g., Benus et al. 1991;
Verbeek et al. 1996), boldness–shyness (e.g., Verbeek et al.
1994; Wilson et al. 1994), or exploration–avoidance (e.g.,
Benus et al. 1987; Verbeek et al. 1994). Individuals are often
consistent in their behavior, and within-individual variation
can be correlated across multiple axes allowing individuals
to be classified as more ‘‘active’’ (being on average more
aggressive, bolder, and faster explorers) or more ‘‘passive’’
(being less aggressive, shyer, and slower explorers) in their
response to novel or adverse stimuli (Koolhaas et al. 1999;
Sih et al. 2004; Réale et al. 2007). Such consistent and correlated behavioral variation has been referred to as coping
styles, behavioral syndromes, temperament or animal personalities; can be correlated to physiological traits (Koolhaas et al.
1999; Carere et al. 2003; Korte et al. 2005; Ruiz-Gomez et al.
2008); and can have an additive genetic component (van
Oers, Drent, de Jong, et al. 2004, Fidler et al. 2007) and can
therefore show an evolutionary response to selection (Drent
et al. 2003; van Oers, de Jong, et al. 2004).
Variation along the exploration–avoidance axis has been
studied in both captive and wild great tits (Parus major), quan-
I
Address correspondence to J. Minderman. E-mail: Jeroen.Minderman@
newcastle.ac.uk.
Received 13 January 2009; revised 30 March 2009; accepted
2 April 2009.
The Author 2009. Published by Oxford University Press on behalf of
the International Society for Behavioral Ecology. All rights reserved.
For permissions, please e-mail: [email protected]
tified by counting the number of perches used or the number
of movements made by a single bird in a standardized behavioral assay. In these studies, exploration behavior was found to
be repeatable and heritable (Dingemanse et al. 2002; Drent
et al. 2003) and correlated with novel object approach time
(Verbeek et al. 1994), risk taking during foraging (van Oers,
Drent, De Goede, et al. 2004), and aggressiveness (Verbeek
et al. 1996). Also, slow explorers showed a greater increase in
fecal corticosteroid metabolites (higher reactivity of the hypothalamo–pituitary–adrenal or HPA axis) after a social challenge than faster explorers (Carere et al. 2003). This
combination of behavioral and physiological variation provides a good parallel to studies on coping styles in rodents,
in which individuals are classified along a proactivity–reactivity
axis based on differences in aggressiveness (Koolhaas et al.
1999). Reactive individuals show high HPA axis reactivity
and are behaviorally analogous to slow exploring great tits,
whereas proactive individuals show lower HPA axis reactivity
and are behaviorally analogous to fast exploring great tits
(Groothuis and Carere 2005).
Although this analogy between coping styles and personalities is a strong indication that such consistent individual differences in behavioral response to novel or stressful stimuli
provide a general framework to study intraspecific variation
in animal behavior, there are some behavioral traits that are
understudied in the personality framework. For example, proactively coping house mice (Mus musculus) make more mistakes when running a familiar maze in which elements are
altered, compared with more reactive individuals (Benus
et al. 1987, 1990). This was interpreted as proactive individuals being more prone to form routines and less sensitive to
environmental cues. It has been suggested that such individual differences in sensitivity to environmental cues reflect
an individual’s ability to adapt to a changing environment
Minderman et al.
•
Personality traits in wild starlings
(Koolhaas et al. 1999) and could affect life-history traits (e.g.,
mate choice; Sih and Bell 2008). However, with the exception
of one study of captive great tits that showed that faster explorers were slower to adapt to a change in feeder location
after a period of training (Verbeek et al. 1994), the correlation between exploration behavior and environmental sensitivity has never been explicitly studied in any species. More
generally, although the repeatability, heritability, and correlation among personality traits (such as exploration behavior)
have been extensively documented in laboratory studies of
captive animals (e.g., Verbeek et al. 1994; Drent et al. 2003),
to our knowledge only 2 studies have quantified variation in
exploration behavior in a wild population (Dingemanse et al.
2002; Drent et al. 2003), and the number of studies doing so
for personality traits generally has been growing only slowly
(reviewed by Dingemanse and Réale 2005, see also, e.g., Boon
et al. 2007; Dochtermann and Jenkins 2007; Martin and Réale
2008). Because of this lack of data from the wild, further
study of the individual consistency of and correlation among
behavioral traits in wild animals is essential. Furthermore, because (as outlined above) it is largely unclear how specific
personality traits affect the ability to respond to environmental change, there is a need to test the hypothesis that wellestablished personality traits such as exploration behavior
predict individual differences in environmental sensitivity, in
particular in wild animals.
We therefore aimed to 1) quantify variation in the behavior
of wild starlings (Sturnus vulgaris) using behavioral assays,
2) estimate the individual repeatability of such behavior,
and 3) explicitly test the hypothesis that variation in exploration behavior is related to an individual’s sensitivity to an
environmental cue. Specifically, we test whether, as suggested
by studies on captive rodents and birds, faster explorers will
be less sensitive to a change in a familiar environment than
slower explorers.
METHODS
Study site
To study individual consistency in behavioral traits in wild
animals, a study system is needed in which individuals can
be retrapped at a high rate within a short period of time.
A population of approximately 150–250 breeding pairs of
starlings on Fair Isle, Shetland, provides this opportunity.
We focused on recently fledged juveniles because due to
the small size (ca. 750 ha) and isolation of the island, large
numbers can be repeatedly trapped during the first month
after fledging. Birds were trapped using unbaited Heligoland
traps, mist nets, or baited drop traps of different sizes.
Biometrics were recorded immediately after trapping, following a set protocol.
831
The holding protocol
Captured starlings were transported to holding cages (0.5 3
0.5 3 0.75 m), placed in an indoor facility at ambient outside
temperature. Two to three birds were housed in the same cage.
Two perches constructed out of natural wood, food (a mix of
organic chick feed and mealworms), and water were provided
ad lib. Although birds were held for a minimum of 0.5 h, actual
holding time varied (0.6–14.2 h, mean ¼ 4.08, median ¼ 3.42)
because of logistical constraints on running the trials. This variation was controlled for statistically.
The assay aviary
Before starting an assay, a bird was transported to an outside
assay aviary (Figure 1) in a white cloth bird bag. The assay
aviary was a wooden structure (2 3 2 3 1.5 m) with blind walls
and either an open (wire mesh) or closed (opaque polythene
sheet) roof. The bird was placed in a completely blinded
(darkened) ‘‘start cage’’ to minimize immediate handling
stress. This cage was connected to the aviary by the ‘‘entry
hatch’’ (0.25 3 0.25 m) on floor level that allowed for release
of the bird into the aviary without further handling. Inside the
aviary, a total of 11 perches were provided (Figure 1g,h),
which were placed so that a bird could only change its perch
by either actively flying or hopping between them. Observations were made from a hide attached to the aviary by a viewing
window (0.5 3 0.3 m, covered by wire mesh). The hide was
darkened so that the observer could not be seen from inside
the aviary. All behavior was entered into an event recorder by
a single observer (J.M.).
Exploration behavior assays
All exploration behavior assays took place in the hours of daylight, between 20 June and 14 July 2006, 8 June and 9 July
2007, and 15 June and 14 July 2008; lasted 8 min; and were
started by opening the entry hatch after the bird had been
in the start cage for 10 min. The time taken for a bird to enter
the aviary (latency to enter) was recorded for later use as a behavioral measure. If a bird had not entered the aviary on its
own accord, after 1 min it was encouraged to do so by opening the start cage briefly, after which all birds entered immediately. These birds were given a latency to enter of 60 s.
Seven other behavioral parameters were recorded (all expressed as number per 8 min): the 1) number of perches used
(unique perches used, 0–11), 2) the time to reach the last new
Behavioral assays
The behavioral assay procedure was derived from the ‘‘open
field test’’ commonly used in animal psychology (Walsh and
Cummins 1976) and similar to that used in other studies to
quantify exploration behavior in birds (e.g., Verbeek et al.
1994; Dingemanse et al. 2002; Drent et al. 2003). By keeping
individuals under standardized conditions before running
the assays, the effects of variation in state (e.g., hunger
and/or stress level) on behavioral responses were minimized. Although achieving such standardization when working with wild birds is difficult, holding conditions were kept
as standardized as possible (cf. Dingemanse et al. 2002), and
remaining variation (e.g., time of day) was controlled for
statistically.
Figure 1
The assay aviary setup. (a) The main aviary, (b) observation hide,
(c) viewing window covered in wire mesh, (d) start box, (e) start
hatch, (f) exit hatch, (g) 6 perches along walls, (h) 5 perches along
linear structure in center of aviary. All 4 walls of the aviary were
completely blind (partly transparent here for presentation
purposes).
832
perch, 3) total time spent perched, 4) number of perch bouts, 5)
number of movements while perched (steps and hops), 6) time
spent on the ground, and 7) number of flights. After the assay
ended, birds were released by opening the roof of the aviary.
Whenever a bird was recaught within the same year, the
above protocol was repeated, but the same birds were not
retested in different years.
Environmental sensitivity assays
Between 15 June and 14 July 2008, a subsample of birds underwent an environmental sensitivity assay. Immediately after the
exploration behavior assay, birds were removed from the aviary
and placed back in the start cage. The procedure for the environmental sensitivity assays was identical to the exploration
behavior assays, with the exception of a small ‘‘exit hatch’’
(0.25 3 0.25 m) on floor level directly opposite the entry hatch
in the far wall (Figure 1f). This hatch was opened before
starting the assay, introducing both a novel element to the
aviary as well as a direct escape route from it. An environmental sensitivity assay lasted until the focal bird left the aviary via
the exit hatch or for a maximum of 8 min. The main measurement was a binomial variable describing whether or not
a bird had left before the end of the assay, which was taken as
a measure of an individual’s environmental sensitivity. All
birds that underwent an environmental sensitivity assay had
experienced only one exploration assay immediately previously and underwent an environmental sensitivity assay only
once.
All experiments were performed under license from Scottish
Natural Heritage (no. 7024), and the handling and holding
protocol was similar to those used successfully and without
harmful effects to the birds in other studies using starlings
(e.g., Devereux et al. 2006).
Statistical analysis
Overall approach and justification
Our statistical analysis consisted of 3 steps.
1. A principal component analysis (PCA) of measures
from the exploration behavior assays. By using a PCA,
we 1) avoided a priori (and arbitrary) decisions about
which behavioral measure to focus on, 2) were able to
include behavioral measures that would be uninformative when interpreted in isolation, and 3) retained all
variation present in the data collected.
2. Linear mixed effects models (LMMs) to 1) analyze
within- and between-individual sources of variation in
scores on extracted PC axes and 2) estimate individual
consistency in scores on these axes. LMMs are robust to
unbalanced data and allow for the simultaneous inclusion of fixed and random effects (Pinheiro and Bates
2004). By including individual as a random effect, a single variance component for individual is estimated, conserving degrees of freedom for estimation of fixed
effects. Thus, random effect estimates represent individual effects, given the effect of included confounding
variables.
3. A test of the relationship between individual estimates of
exploration behavior and measures of environmental
sensitivity.
This approach has been pioneered in recent studies of the
behavioral ecology of personality traits (e.g., Boon et al. 2007;
Dingemanse et al. 2007; Martin and Réale 2008), allows for
the retention of all data collected and a full analysis of withinand between-individual sources of variation, and is therefore
the most parsimonious.
Behavioral Ecology
Principal components analysis
All 8 measures from all exploration behavior assays were
scaled and centered to zero and entered into a PCA. Although
the full data set included variable numbers of repeat measures
per individual (individuals were measured 1–5 times),
restricting the PCA to the first measurement of each individual only did not alter the number of axes extracted or the
factor loadings on them. Also, although detrended correspondence analysis (DCA) was designed to handle nonnormal
variables better than a PCA (Fry 1994), the interpretation
of extracted DCA factors was similar, and we therefore
retained the more commonly used PCA. Only extracted
axes with eigenvalues larger than 1 were retained for further
analyses.
Linear mixed effects models
Individual was tested as the only random effect in addition to
the fixed factors year, sex, the type of aviary roof used (fixed
factor ‘‘covered,’’ wire mesh, or polythene), and trapping
method (mist net, drop trap, or Heligoland trap). Trial number (the number of assays the focal individual had undergone),
time since the last trial (days), time of day (sine transformed),
Julian date, time spent in the holding cage (hours), body mass
index (residuals of a linear regression of body mass over wing
length, using higher order regressions yielded similar results),
and the number of birds in the holding cage were tested as
covariates. Models were fit using maximum likelihood estimators; likelihood ratio tests (LRTs) were used to assess significance of fixed and random effects (Faraway 2005, 2007);
and minimum adequate models (MAMs) were produced
by stepward elimination of nonsignificant terms (Crawley
2007). Although not recommended for descriptive models
(Whittingham et al. 2006), MAMs can be used to predict
the individual effect on each PC axis given the effects of confounding variables but excluding ‘‘noise’’ of nonsignificant
factors.
To provide a measure of how much variation in behavioral
scores was accounted for by differences between individuals,
we calculated the repeatability (R, the intraclass correlation
coefficient, Falconer and MacKay 1996; McGraw and Wong
1996) by fitting the MAMs to the subset of data that only
included individuals that were measured repeatedly using restricted maximum likelihood estimation. We then calculated
R by dividing the estimated random (individual) effect variance component by the sum of the estimated random effect
variance and estimated error variance. Significance (testing
H0 that R was not different from 0) was assessed by transforming R to a Z score and testing against a standard normal
distribution, correcting for unequal numbers of measures
per individual (McGraw and Wong 1996; Galwey 2006; and
Shinichi Nakagawa personal communication). Because our
within-individual sample size was small (2–5 observations per
individual, median 2), we did not include random slopes in
the model.
To obtain individual scores on repeatable behavioral axes,
we extracted the best linear unbiased predictors (BLUPs) of
the random effects from the above MAMs fitted to the full data
set. BLUPs represent the individual scores on extracted PC
axes ‘‘controlling for’’ included confounding effects. To test
whether exploration behavior predicts the sensitivity to environmental cues, BLUPs for PC axes were included as a covariate in a model for whether or not an individual used the exit
hatch in the environmental sensitivity assay. This binomial
variable was modeled using a generalized linear model
(GLM) with a binomial error structure and a logit link function. In addition to the individual PC scores (BLUPs), we included Julian date, time of day (sine transformed), and sex
as potential confounding variables.
Minderman et al.
•
Personality traits in wild starlings
833
Statistical software R (v. 2.7.0) (R Development Core Team
2008) was used for all analyses. The standard stats package was
used for the PCA (prcomp) and the analysis of environmental
sensitivity (glm). The DCA was performed using package vegan
v.1.8.8 (Oksanen et al. 2007). The mixed effects models were
run using package lme4 v.0.99875.9 (Bates 2007). For all tests,
a significance level of 0.05 was used.
cluding these factors. Repeatability estimates of PC1 and PC2
scores from their respective full models were also significant
and very close to the estimates from the MAMs (results not
shown).
PC3 scores increased with trial number but were unaffected
by any of the other covariates nor were they different between
individuals. Repeatability of PC3 was close to zero (R , 0.001,
P ¼ 0.500, in MAM for PC3 fitted to repeated data set only).
RESULTS
Exploration behavior assays
Environmental sensitivity assays
The full (N ¼ 376) data set was used in the PCA. The sample
size used in the LMMs was reduced to N ¼ 360 (102 assays
from 76 individuals in 2006, 129 from 113 individuals in 2007,
and 129 from 115 individuals in 2008) because body mass
index was not available for 16 individuals. The ‘‘repeated’’
data set consisted of N ¼ 98 assays on 42 different individuals
that were measured more than once. Descriptive statistics for
the (untransformed) behavioral measures from the assays are
presented in Table 1.
Three of a total of 27 environmental sensitivity assays were of
birds of unknown sex and therefore excluded from the analysis, reducing the sample size to N ¼ 24 assays. Correcting for
the effects of Julian date, time of day, and sex, an individual’s
PC1 score (its BLUP for PC1) from the preceding exploration
assay did not predict the probability of leaving during the
environmental sensitivity assay (Figure 2a and Table 4). In
contrast, individuals with lower PC2 scores in the exploration
assays were significantly more likely to leave the aviary during
the environmental sensitivity assays (Figure 2b and Table 4).
Variation in environmental sensitivity in relation to PC3 scores
was not examined because PC3 scores were not repeatable.
Principal components analysis
The PCA on the exploration assay data produced 3 axes with
eigenvalues .1 (Table 2). Higher scores on the first axis (PC1,
39% of the total variation) implied more perch bouts, more
flights, more movement while perched, and a larger number
of different perches used. Higher scores on the second axis
(PC2, 24% of the total variation) implied more time perched
and less time on the ground. Higher scores on the third axis
(PC3, 14% of total variation) implied that birds reach their
last new perch relatively quickly (shorter time to last new
perch) and enter the aviary sooner (short latency to enter).
Explaining variation in exploration behavior scores
Table 3 shows the full mixed effects models for each of the 3
extracted PC axes. PC1 scores increased with trial number,
were different between years (higher scores in 2007), decreased slightly during the day, increased with Julian date,
were higher for birds that were caught in drop traps, and
tended to be higher for birds that had been housed with more
conspecifics. In addition to these factors, the MAM for PC1
included ‘‘covered,’’ but the effect of time of day was not
retained. When this model was fitted to the repeated data
set only, PC1 scores were significantly repeatable (R ¼ 0.309,
P ¼ 0.007).
PC2 scores also differed between years (lower scores in
2008), decreased with Julian date, increased with body mass
index, and were different between individuals. PC2 scores were
significantly repeatable (R ¼ 0.400, P , 0.001) in a MAM in-
DISCUSSION
We show that the behavior of juvenile wild starlings in standardized behavioral assays 1) can be classified along 3 distinct behavioral axes and 2) was repeatable within individuals for the
first 2 of these. Measures that loaded on the first (PC1) axis
(e.g., the number of perches used or the number of flights)
are similar to those used in other studies on birds to quantify
the speed of exploration behavior (Verbeek et al. 1994;
Dingemanse et al. 2002; Drent et al. 2003). Thus, we suggest
that repeatable variation in PC1 scores reflects personality
trait variation in starlings, describing individual differences
in the speed of exploration behavior. In contrast, variation
in PC2 scores describes whether the perches or the ground
was used predominantly. Thus, we argue that whereas PC1
scores describe the ‘‘quantity’’ of exploration behavior in
terms of its speed, PC2 scores describe differences in the
‘‘quality’’ of exploration behavior in terms of which areas of
the environment are explored.
In addition to significant and repeatable between-individual
differences, PC1 scores (exploration speed) also increased
with trial number. Dingemanse et al. (2002) interpreted this
as a potential decrease in anxiety with increased habituation.
Similarly, we found an increase in exploration scores with
Julian date, probably reflecting an age effect: Breeding is
Table 1
Descriptive statistics for the raw data of the 8 behavioral parameters measured in the behavioral assays and the first 3 extracted PC axes
Number of perches used
Time to last new perch (s)
Latency to enter (s)
Time spent perched (s)
Number of perch bouts
Moves while perched
Time spent on ground (s)
Number of flights
PC1
PC2
PC3
a
Inter-quartile range.
Mean
Median
Minimum
Maximum
IQRa
Standard deviation
4.95
255.00
16.46
339.70
28.42
54.53
58.50
40.95
0
0
0
5
279.50
3.93
369.90
23
46
1.99
31
20.232
0.303
20.069
0
1.41
0
0
0
0
0
0
24.051
24.671
23.114
10
525.70
60
480
133
329
479.10
179
5.350
2.170
3.299
4
267.40
23.41
137.90
33
51
67.88
46.25
2.575
1.424
1.403
2.30
148.26
21.78
114.43
24.74
42.58
108.90
33.00
1.770
1.379
1.064
Behavioral Ecology
834
Table 2
Factor loadings, eigenvalues, and proportion of the total variance
explained by the first 3 axes extracted by PCA of 8 behavioral
measures
Eigenvalue
Proportion of variance explained
No. of perches used
Time to last new perch
Latency to enter
Time spent perched
Number of perch bouts
Moves while perched
Time spent on ground
Number of flights
PC1
PC2
PC3
3.133
0.392
0.396
0.194
20.227
0.035
0.519
0.418
20.228
0.507
1.901
0.238
20.126
20.264
20.184
0.683
20.066
20.085
20.630
20.076
1.133
0.142
20.364
20.676
20.499
20.254
0.115
0.121
0.174
0.197
Factor loadings with an absolute correlation of greater than 0.3 are
shown in bold. Five further extracted axes (not shown) cumulatively
explained less than 10% of the total variation in behavioral measures.
highly synchronous, and thus, at any given date birds are of
similar age. Finally, birds that were caught in drop traps had
significantly higher PC1 scores, suggesting differences in propensity to enter (baited) traps between birds with different
personality traits, similar to those found in, for example,
pumpkinseed sunfish (Lepomis gibbosus, Wilson et al. 1993).
PC2 scores were also affected by Julian date and corrected
body mass, so that birds spent more time on the ground later
in the season and when their body mass was low. Crucially,
however, controlling for these effects and also the slight differences between years, we find that individual differences still
explain a significant part of the variation in both PC1 and PC2
scores, providing important further evidence for the generality of personality traits in wild animals. Finally, we validated
our analysis of PC scores by analyzing 2 univariate measures
loading on PC1 and PC2, respectively. Both the number of
perches used (R ¼ 0.223, P ¼ 0.037) and the probability of
using the ground (time on ground expressed as a binomial
variable; R ¼ 0.378, P ¼ 0.001) showed significant and repeatable between-individual variation. Also, both PC1 (R ¼ 0.480)
and PC2 (R ¼ 0.379) scores were repeatable when estimated
from 1-way analyses of variance with individual as the only
fixed effect (Lessells and Boag 1987). Thus, both our PCA
and analysis of its scores are robust.
A key element of animal personalities or behavioral syndromes is the potential correlation among behavioral traits.
We tested the prediction that faster explorers are slower to respond to an environmental cue (a change in a familiar environment). Our data do not support this prediction: The
probability of leaving the aviary during the environmental sensitivity assays was unrelated to our measure of exploration
speed (PC1 scores). This absence of a correlation between
the speed of exploration behavior and sensitivity to the particular environmental cue tested here for starlings raises questions about the generality of the predicted relationship
between exploration behavior and environmental sensitivity.
In contrast, environmental sensitivity was lower for birds with
higher PC2 scores, that is, for those that spent more time on
perches and less time on the ground during the exploration
behavior assays. The absence of an effect of PC1 scores, but
presence of a strong effect of PC2 scores, on environmental
sensitivity suggests that the latter is not related to the speed
of exploration but rather to which parts of the assay aviary were
Table 3
Estimated effect sizes (b), LRT x2 test statistics and their associated degrees of freedom, and the significance of the LRT for all fixed and
random effects included in 3 LMMs for the PC axes (PC1, PC2 and PC3) with eigenvalues .1
PC1
PC2
b 6 SE
Fixed effects
Trial number
Time since last trial (days)
Yeara
2007
2008
Time of day
Julian date
Time in cage (hours)
Sexb
Male
Unknown
Corrected body mass
Coveredc
Trapping methodd
Heligoland
Mist net
Number of birds in
cage
Random effect
Individual
a
b
d
LRT
0.556 6 0.188
20.012 6 0.021
0.606
20.179
20.05
0.048
20.026
6
6
6
6
6
0.335
0.313
0.024
0.016
0.038
0.220
0.007
20.001
20.588
6
6
6
6
0.180
0.306
0.018
0.304
v2df
8.521
0.323
1
8.025
2
4.184
9.423
0.462
1.686
0.006
3.711
1
1
1
1
2
1
1
PC3
b 6 SE
P
LRT
0.004 20.073 6 0.153
0.570 20.013 6 0.017
v2df
1
5.642
2.002
1
1
0.018
0.157
6
6
6
6
6
0.225
0.212
0.017
0.011
0.026
1.711
2
0.425
1
0.001 20.222
20.274
0.174 20.015
0.014 20.006
0.330
0.039
5.203
2
0.743
0.406
2.176
1
0.023
0.131
6
6
6
6
0.119
0.208
0.012
0.208
0.816 20.099 6 0.151
20.021 6 0.345
0.140 20.034 6 0.148
0.018 20.746
21.057
0.041
0.029
0.002 20.034
0.497
0.032
6
6
6
6
6
0.297 14.103
0.278
0.021 6.019
0.014 0.947
0.033 0.595
0.431 20.123
20.114
0.940
0.036
0.054
0.409
6
6
6
6
0.162
0.273
0.016
0.270
0.015
P
0.635
0.326 6 0.137
0.444 20.022 6 0.015
0.226
0.585
20.870 6 0.216 17.197 2 ,0.001 20.011 6 0.187 12.155
20.061 6 0.495
0.254 6 0.430
0.050 20.275 6 0.185 0.757
0.423 6 0.213 3.85 1
5.973
LRT v2df
b 6 SE
P
12.155
1
1
2
1
1
1
2
1
1
,0.001
0.121
0.472
0.003
0.384
0.826
0.275
2.155
1
0.363
0.600
0.142
5.171
2
0.075
0.064
3.406
1
1
0.801
0.065
0.456
2
0.796
0.052
1
0.819
,0.001
1
1
1
.0.9
Parameter estimates from the MAM for PC1 were b ¼ 0.496 (trial number), b2007 ¼ 0.573 and b2008 ¼ 20.258 (year), b ¼ 0.048 (Julian date),
b ¼ 20.648 (‘‘covered’’), bHeligoland ¼ 0.8 and bmist net ¼ 0.08 (trapping method), b ¼ 0.433 (number of birds in cage). Parameter estimates
from the MAM for PC2 were b2007 ¼ 20.608 and b2008 ¼ 20.757 (year), b ¼ 20.032 (Julian date), b ¼ 0.044 (body mass index). SE, standard
error.
Reference category is 2006.
Reference category is ‘‘female.’’ c Denotes the type of aviary roof used.
Reference category is ‘‘drop trap.’’
Minderman et al.
•
Personality traits in wild starlings
835
Figure 2
The probability of leaving the
assay during the environmental sensitivity assays, plotted
against (a) an individual’s
BLUP for PC1 (i.e., speed of
exploration behavior) and (b)
an individual’s BLUP for PC2
(higher values indicate more
time spent perched and less
time spent on the ground).
Each point represents whether
the individual left (1) or not
(0), and the line in (b) plots
the probability of leaving predicted by a binomial errors
GLM using the PC2 score as
a single significant predictor
(see table 4 for full model).
explored, specifically the ground. There are 2 potential problems with using the probability of leaving the aviary as a measure of environmental sensitivity. First, because PC2 scores
describe the parts of the aviary used during the exploration
behavior assay and the exit hatch was presented at a fixed position in the aviary (at floor level), differences in response to
the hatch may have been due to differences in space use rather
than differences in environmental sensitivity per se. However,
because there was variation among birds that left and those
that did not in terms of use of the ground (71% of ‘‘stayers’’
had at least spent some time on the ground during the exploration assays, whereas 15% of the ‘‘leavers’’ had not spent any
time on the ground at all), we argue that differences in response to the hatch cannot be accounted for by differences
in space use alone. Secondly, we cannot distinguish between
birds that failed to detect and those that detected but failed
to respond to the open hatch. For example, assuming that
using the hatch represents a ‘‘risk,’’ differences in the probability of leaving may be due to differences in the propensity to
take risks rather than sensitivity to (i.e., detection of) cues.
However, this is unlikely because risk taking is generally found
to be positively related to the speed of exploration (e.g., van
Oers, Drent, De Goede, et al. 2004), and we do not find
a correlation between our measure of exploration speed
(PC1) and the probability of leaving. Thus, we argue that
a higher probability of leaving (and correspondingly lower
PC2 scores) represents a good measure of environmental sensitivity and does reflect individual differences in space use or
risk taking alone.
Verbeek et al. (1994) originally suggested that the negative
relationship between exploration behavior of captive great tits
and their response to a change of feeder location might be
due to a difference in quality versus quantity of the exploration: Although fast explorers moved through the environment
more quickly, the information they gather would be more
superficial compared with the slow explorers. Based on our
results, we could extend this argument and speculate that individual differences in the speed of exploration behavior of
juvenile starlings might be both unrelated to the quality of the
information gathered during exploration and unrelated to an
individual’s environmental sensitivity and that there are separate behavioral axes (e.g., our second PC axis) that relate to
the latter.
More generally, our repeatability estimate of exploration behavior is on the lower end of the range found for great tits. This
suggests a more fundamental explanation for the absence of
a correlation between exploration behavior and environmental sensitivity. In the case of starlings, the speed of exploration
(defined as the perches used or number of flights made) may
not be very relevant biologically. The assay setup as used here
was pioneered by Verbeek et al. 1994 and was designed to
measure a behavioral trait relevant for great tits, a species that
forages in trees (Perrins 1979), moving between branches. In
contrast, starlings forage on the ground (Feare 1984), and it is
thus likely that behavioral traits relating to the propensity to
use the ground (i.e., our second PC axis) are both behaviorally and ecologically more relevant to starlings. Thus, measurements loading on PC2 could also be interpreted as
‘‘maintenance behavior’’ such as foraging. Although speculative, this interpretation appears to be in line with our observation that birds spent more time on the ground (i.e., had
lower PC2 scores) when their body mass was low (and may
Table 4
Estimated effect sizes (b), their associated standard errors (SE), Z scores, and P values from a binomial errors GLM for the probability that
a bird uses the exit hatch to leave the aviary during the environmental sensitivity assays, using a logit link function
PC1
PC2
b 6 SE
Intercept
PC score (BLUP)
Julian date
Time of daya
Sex
a
Sine transformed.
239.167
20.429
0.235
20.232
20.708
6
6
6
6
6
27.407
0.918
0.148
0.228
0.988
Z score
P
b 6 SE
21.429
20.467
1.588
21.015
20.717
0.153
0.640
0.112
0.310
0.474
28.646
23.052
0.048
20.021
20.337
6
6
6
6
6
25.727
0.910
0.141
0.237
0.956
Z score
P
20.336
23.355
0.343
20.089
20.352
0.737
0.001
0.732
0.929
0.725
836
have been more motivated to feed). In line with recent reviews
(Réale et al. 2007), this result stresses the need to tailor the
personality traits measured to the biology of the species in
question and to validate them in an ecological context.
Sih and Bell (2008) argue that differences in environmental
sensitivity is one of the aspects of animal personality research
that would benefit most from further work and outline how
individual differences in the capacity to detect differences in
signals (or between options) could play a fundamental role in,
for example, sexual selection and habitat and diet choice.
Indeed, in some contexts (e.g., foraging under predation
risk), we would expect such differences in sensitivity to environmental cues to have a direct effect on survival. In addition
to such direct affects, however, behavioral and physiological
traits related to an individual’s environmental sensitivity will
be crucial in setting the limits to its capacity to respond to and
adapt to environmental change. Although there is a general
agreement that such behavioral phenotypic plasticity on an
individual level can be crucial in determining the population
level response to environmental change (Nussey et al. 2007;
Gienapp et al. 2008), some studies find significant individual
variation in response (individuals can be plastic to different
degrees, e.g., Brommer et al. 2005; Nussey et al. 2005),
whereas others find the same degree of plasticity for all individuals (e.g., Reed et al. 2006; Charmantier et al. 2008). Given
these inconsistencies and the current occurrence of rapid
environmental change, it is crucial that we gain a better understanding of the limits of behavioral plasticity (DeWitt et al.
1998; Lessells 2008). Thus, although it would provide a crucial
link between personality traits and the effect of phenotypic
(behavioral) plasticity on population-level responses to environmental change, empirical studies of how environmental
sensitivity per se forms a part of animal personalities are rare,
and the current study provides important further evidence in
this direction.
FUNDING
Fair Isle Bird Observatory Trust (2006, 2007 and 2008) (Ornithological Scholarship Grant); Newcastle University School Studentship (to J.M.); Royal Society University (Research Fellowship to
J.M.R.); Biotechnology and Biological Sciences Research Council
David Phillips Fellowship (BB/B502214/1 to M.J.W.).
This project would have been impossible without the help from staff,
volunteers, and research workers at Fair Isle Bird Observatory, and our
thanks go out to all of them. In particular, we would like to thank Deryk
Shaw, Phil Bell, and the kitchen staff for their endless supply of bait;
Paul Baxter, Mark Warren, Will Miles, Phil Knott, Graeme Cook, Simon
Boswell, Becki Rosser, Simon Davies, Mark Breaks, Adam Seward, Daisy
Brickhill, and Matthew Denny for their help with the catching, ringing,
and assays. Brian Wilson of Houll is single handedly responsible for
making this project possible by building the aviary and holding cages
to an excellent standard. Thanks to Shinichi Nakagawa for help with
the statistical analysis and to Claudia Garratt and Caroline Rhymer for
comments on the manuscript.
REFERENCES
Bates DM. 2007. The lme4 Package 0.99875-9; [cited 2009 April 14].
Available from: http://cran.r-project.org/web/packages/lme4/
index.html.
Benus RF, Bohus B, Koolhaas JM, Van Oortmerssen GA. 1991. Heritable variation for aggression as a reflection of individual coping
strategies. Experientia. 47:1008–1019.
Benus RF, Den Daas S, Koolhaas JM, Van Oortmerssen GA. 1990.
Routine formation and flexibility in social and nonsocial behavior
of aggressive and nonaggressive male mice. Behaviour. 112:176–193.
Behavioral Ecology
Benus RF, Koolhaas JM, Van Oortmerssen GA. 1987. Individual differences in behavioral reaction to a changing environment in mice
and rats. Behaviour. 100:105–122.
Boon AK, Réale D, Boutin S. 2007. The interaction between personality, offspring fitness and food abundance in North American red
squirrels. Ecol Lett. 10:1094–1104.
Brommer JE, Merilä J, Sheldon BC, Gustafsson L. 2005. Natural selection and genetic variation for reproductive reaction norms in a wild
bird population. Evolution. 59:1362–1371.
Carere C, Groothuis TGG, Möstl E, Daan S, Koolhaas JM. 2003. Fecal
corticosteroids in a territorial bird selected for different personalities: daily rhythm and the response to social stress. Horm Behav.
43:540–548.
Charmantier A, McCleery RH, Cole LR, Perrins C, Kruuk LEB,
Sheldon BC. 2008. Adaptive phenotypic plasticity in response to
climate change in a wild bird population. Science. 320:800–803.
Crawley MJ. 2007. The R book. Chichester, UK: John Wiley & Sons
Ltd.
Devereux CL, Whittingham MJ, Fernández-Juricic E, Vickery JA, Krebs
JR. 2006. Predator detection and avoidance by starlings under differing scenarios of predation risk. Behav Ecol. 17:303–309.
DeWitt TJ, Sih A, Wilson DS. 1998. Costs and limits of phenotypic
plasticity. Trends Ecol Evol. 13:77–81.
Dingemanse NJ, Both C, Drent PJ, van Oers K, Van Noordwijk AJ.
2002. Repeatability and heritability of exploratory behaviour in
great tits from the wild. Anim Behav. 64:929–938.
Dingemanse NJ, Réale D. 2005. Natural selection and animal personality. Behaviour. 142:1159–1184.
Dingemanse NJ, Wright J, Kazem AJN, Thomas DK, Hickling R,
Dawnay N. 2007. Behavioural syndromes differ predictably between
12 populations of three-spined stickleback. J Anim Ecol. 76:1128–1138.
Dochtermann NA, Jenkins SH. 2007. Behavioural syndromes in Merriam’s kangaroo rats (Dipodomys merriami): a test of competing
hypotheses. Proc R Soc Lond B Biol Sci. 274:2343–2349.
Drent PJ, van Oers K, van Noordwijk AJ. 2003. Realized heritability of
personalities in the great tit (Parus major). Proc R Soc Lond B Biol
Sci. 270:45–51.
Falconer DS, MacKay TFC. 1996. Introduction to quantitative genetics. New York: Longman.
Faraway JJ. 2005. Extending the linear model with R. Boca Raton (FL):
Chapman & Hall.
Faraway JJ. 2007. Changes to the mixed effects models chapters in ELM;
[cited 2008 September 21]. Available from: http://www.maths.bath.
ac.uk/;jjf23/ELM/mixchange.pdf.
Feare CJ. 1984. The starling. Oxford: Oxford University Press.
Fidler AE, van Oers K, Drent PJ, Kuhn S, Mueller JC, Kempenaers B.
2007. Drd4 gene polymorphisms are associated with personality
variation in a passerine bird. Proc R Soc Lond B Biol Sci. 274:
1685–1691.
Fry JC. 1994. Biological data analysis: a practical approach. New York:
Oxford University Press.
Galwey NW. 2006. Introduction to mixed modelling: beyond regression and analysis of variance. Chichester, UK: Wiley Blackwell.
Gienapp P, Teplitsky C, Alho JS, Mills JA, Merilä J. 2008. Climate
change and evolution: disentangling environmental and genetic
responses. Mol Ecol. 17:167–178.
Groothuis TGG, Carere C. 2005. Avian personalities: characterization
and epigenesis. Neurosci Biobehav R. 29:137–150.
Koolhaas JM, Korte SM, De Boer SF, Van Der Vegt BJ, Van Reenen CG,
Hopster H, De Jong IC, Ruis MAW, Blokhuis HJ. 1999. Coping styles
in animals: current status in behavior and stress-physiology. Neurosci Biobehav R. 23:925–935.
Korte SM, Koolhaas JM, Wingfield JC, Mcewen BS. 2005. The Darwinian
concept of stress: benefits of allostasis and costs of allostatic load and
the trade-offs in health and disease. Neurosci Biobehav R. 29:3–38.
Lessells CM. 2008. Neuroendocrine control of life histories: what do
we need to know to understand the evolution of phenotypic plasticity? Philos Trans R Soc B-Biol Sci. 363:1589–1598.
Lessells CM, Boag PT. 1987. Unrepeatable Repeatabilities—a common
mistake. Auk. 104:116–121.
Martin JGA, Réale D. 2008. Temperament, risk assessment and habituation to novelty in eastern chipmunks, Tamias striatus. Anim Behav. 75:309–318.
McGraw KO, Wong SP. 1996. Forming inferences about some intraclass correlation coefficients. Psychol Methods. 1:30–46.
Minderman et al.
•
Personality traits in wild starlings
Nussey DH, Postma E, Gienapp P, Visser ME. 2005. Selection on heritable phenotypic plasticity in a wild bird population. Science.
310:304–306.
Nussey DH, Wilson AJ, Brommer JE. 2007. The evolutionary ecology
of individual phenotypic plasticity in wild populations. J Evol Biol.
20:831–844.
Oksanen J, Kindt R, Legendre P, O’Hara B, Stevens HH. vegan: community ecology package. 2007. R package version 1.8-8; [cited 2009
April 19]. Available from: http://cran r-project org/, http://r-forge
r-project org/projects/vegan/.
Perrins CM. 1979. British tits. London: Collins.
Pinheiro JC, Bates DM. 2004. Mixed-effects models in S. S-PLUS. New
York: Springer.
R Development Core Team. 2008. R: a language and environment for
statistical computing. Vienna (Austria): R Foundation for Statistical
Computing.
Réale D, Reader SM, Sol D, McDougall PT, Dingemanse NJ. 2007.
Integrating animal temperament within ecology and evolution. Biol
Rev. 82:291–318.
Reed TE, Wanless S, Harris MP, Frederiksen M, Kruuk LEB, Cunningham EJA. 2006. Responding to environmental change: plastic responses vary little in a synchronous breeder. Proc R Soc Lond B Biol
Sci. 273:2713–2719.
Ruiz-Gomez MD, Kittilsen S, Hoglund E, Huntingford FA, Sorensen
C, Pottinger TG, Bakken M, Winberg S, Korzan WJ, Øverli Ø. 2008.
Behavioral plasticity in rainbow trout (Oncorhynchus mykiss) with
divergent coping styles: when doves become hawks. Horm Behav.
54:534–538.
Sih A, Bell AM. 2008. Insights for behavioral ecology from behavioral
syndromes. Adv Stud Behav. 38:227–281.
Sih A, Bell A, Johnson JC. 2004. Behavioral syndromes: an ecological
and evolutionary overview. Trends Ecol Evol. 19:372–378.
837
Sih A, Bell AM, Johnson JC, Ziemba RE. 2004. Behavioral syndromes:
an integrative overview. Q Rev Biol. 79:241–277.
Smith BR, Blumstein DT. 2008. Fitness consequences of personality:
a meta-analysis. Behav Ecol. 19:448–455.
Tabachnick BG, Fidell LS. 2001. Using multivariate statistics. Boston
(MA): Allyn & Bacon.
van Oers K, de Jong G, Drent PJ, van Noordwijk AJ. 2004. A genetic
analysis of avian personality traits: correlated, response to artificial
selection. Behav Genet. 34:611–619.
van Oers K, Drent PJ, De Goede P, van Noordwijk AJ. 2004. Realized
heritability and repeatability of risk-taking behaviour in relation to
avian personalities. Proc R Soc Lond B Biol Sci. 271:65–73.
van Oers K, Drent PJ, de Jong G, van Noordwijk AJ. 2004. Additive and
nonadditive genetic variation in avian personality traits. Heredity.
93:496–503.
Verbeek MEM, Boon A, Drent PJ. 1996. Exploration, aggressive behavior and dominance in pair-wise confrontations of juvenile male
great tits. Behaviour. 133:945–963.
Verbeek MEM, Drent PJ, Wiepkema PR. 1994. Consistent individualdifferences in early exploratory-behavior of male great tits. Anim
Behav. 48:1113–1121.
Walsh RN, Cummins RA. 1976. Open-field test—critical-review. Psychol Bull. 83:482–504.
Whittingham MJ, Stephens PA, Bradbury RB, Freckleton RP. 2006.
Why do we still use stepwise modelling in ecology and behaviour?
J Anim Ecol. 75:1182–1189.
Wilson DS, Clark AB, Coleman K, Dearstyne T. 1994. Shyness and
boldness in humans and other animals. Trends Ecol Evol. 9:
442–446.
Wilson DS, Coleman K, Clark AB, Biederman L. 1993. Shy bold continuum in pumpkinseed sunfish (Lepomis-Gibbosus)—an ecological study of a psychological trait. J Comp Psychol. 107:250–260.