Experimental demonstration of the insurance

Behavioral Ecology Vol. 12 No. 3: 340–347
Experimental demonstration of the insurance
value of extra eggs in an obligately
siblicidal seabird
L. D. Clifford and D. J. Anderson
Department of Biology, Wake Forest University, Winston-Salem, NC 27109, USA
A variety of organisms regularly produce more offspring than they raise. Despite the apparent energetic waste of such a reproductive tactic, overproduction may be favored by natural selection in some cases. One such case is when surplus offspring can
serve as replacements, or insurance, for failed siblings. We tested the Insurance Egg Hypothesis (IEH) as an explanation for
the overproduction of offspring in an obligately siblicidal seabird, the Nazca booby (Sula granti), which fledges a maximum of
one nestling regardless of its clutch size. We manipulated clutch sizes within the range of natural variation encountered in this
species (one–two eggs). The IEH predicts that parents with two-egg clutches should have higher reproductive success than those
with one-egg clutches because the second egg can provide a nestling when the first egg fails to hatch, or when the first chick
dies young. Consistent with the IEH, natural one-egg clutches that were enlarged to two eggs produced more hatchlings and
fledglings than control one-egg clutches did, and natural two-egg clutches that were reduced to one egg produced fewer
hatchlings and fledglings than control two-egg clutches did. We also evaluated aspects of the Individual Optimization Hypothesis,
which proposes that individual optimal clutch sizes differ, as an explanation for clutch size variation in this species. In Nazca
boobies, selection driven by replacement value appears to favor clutches larger than one even though final brood size is invariably
one. One-egg clutches may be produced by parents experiencing some proximate limitation, such as a lack of food. Key words:
clutch size, Individual Optimization Hypothesis, Insurance Egg Hypothesis, Nazca booby, siblicide, Sula granti, surplus offspring.
[Behav Ecol 12:340–347 (2001)]
ife history theory predicts that selection optimizes the
cost-benefit relationships involved in clutch size evolution. An apparent paradox is presented by a variety of organisms that regularly produce more offspring than they can raise
to maturity. For example, female coelacanths produce twice
as many embryos as can be housed in their ovary; many parasitoid wasps lay large broods in a host from which only a
single larva survives; and many female marsupials produce
more young than the number of teats they have, resulting in
the death of excess young (Mock and Parker, 1997).
A number of avian species have an obligate brood reducing
system (reviews in Anderson, 1990a; Mock et al., 1990) in
which surplus offspring are virtually always eliminated (⬎ 95%
of broods) through substantial, overt sibling aggression (Mock
et al., 1990). Offspring overproduction in these cases may
have evolved to provide surplus offspring as insurance against
early failure of ‘‘core’’ offspring (number of offspring that
parents actually raise; Mock and Forbes, 1995). The Insurance
Egg Hypothesis (Dorward, 1962) views production of excess
young as an adaptation to uncertain offspring viability (Anderson, 1990a).
Surplus offspring are thought to serve an insurance function in diverse taxa (both obligately siblicidal and others) including angiosperms (Ehrlén, 1991), beetles (Bartlett, 1987),
parasitoid wasps (Rosenheim and Hongkham, 1996), birds
(e.g., Anderson, 1990a; Aparicio, 1997; Cash and Evans, 1986;
Dorward, 1962; Forbes, 1990; Forbes et al., 1997; Graves et al.,
1984; Hunt and Evans, 1997; Mock and Parker, 1986; Wiebe,
1996), and mammals (Anderson, 1990b; Millar, 1973; Mock
and Parker, 1997). Theoretical studies provide robust support
for the insurance idea (Forbes, 1990; Forbes and Lamey,
Address correspondence to L.D. Clifford. E-mail: [email protected]
Received 27 March 2000; revised 16 August 2000; accepted 22 September 2000.
2001 International Society for Behavioral Ecology
1996), and several field studies indicate that surplus offspring
can provide insurance against failure of core offspring in
some obligately siblicidal species, increasing parental reproductive success as a result (Anderson, 1990a; Gargett, 1977;
Kepler, 1969). These field studies were correlational, not experimental, so potentially confounding variables were not
controlled. For example, parents in better overall condition
might lay more eggs and take better care of offspring than do
parents in poor condition, leading to positive correlation of
clutch size and final reproductive success. Cash and Evans
(1986) conducted the lone experimental test of the Insurance
Egg Hypothesis (IEH) in an obligately siblicidal species, using
the American white pelican (Pelecanus erythrorhynchos). They
tested two predictions of the IEH and found support for both.
If overproduction of offspring increases the reproductive
success of individuals within a species, then all individuals are
expected to employ the strategy. However, some individuals
within obligately siblicidal species produce only a single offspring. These individuals may experience constraints that do
not allow them to employ the overproduction strategy. Alternatively, overproduction of offspring may not be the best strategy for all individuals in a population. In other words, different individuals may have different optimal clutch sizes (the
Individual Optimization Hypothesis; Högstedt, 1980; Perrins
and Moss, 1975).
The Individual Optimization Hypothesis (IOH) was originally proposed in the context of parents’ varying abilities to
provision broods of different sizes. Previous tests of the IOH
(e.g., Barber and Evans, 1995; Nur, 1986; Pettifor, 1993; Pettifor et al., 1988) have focused on the possibility that clutch
size is adjusted according to parents’ ability to raise hatched
offspring. Because obligate siblicide typically occurs shortly after the second chick hatches, brood size does not vary for the
majority of the nestling period. However, costs may accompany the incubation of the second egg, and the presence of
the chick for a short time before siblicide occurs, and there-
Clifford and Anderson • Insurance eggs in boobies
Table 1
Predictions and assumptions of the Insurance Egg Hypothesis (IEH)
and Individual Optimization Hypothesis (IOH) with respect to
experimental treatment groups, with inequalities indicating which
treatment groups should have a higher probability of producing at
least one hatchling and a higher probability of producing a fledgling
in pairwise comparisons
Insurance Egg Hypothesis
Individual Optimization
Reduced ⬍ control 2
Enlarged ⬎ control 1
Reduced ⬍ control 2
Enlarged ⬍ control 1
Reduced ⫽ control 1
Reduced ⬍ enlarged
Enlarged ⫽ control 2
Control 2 ⬎ control 1
Reduced ⬎ control 1
Reduced ⬎ enlarged
Enlarged ⬍ control 2
Control 2 ⬎ control 1
The critical prediction distinguishing the two hypotheses appears in
bold type.
fore the IOH can be extended to the incubation and laying
period. In addition, if individuals produce eggs of varying
quality as suggested by Simmons (1997), then two optimal
clutch sizes may exist, with high quality one-egg clutches producing highly hatchable eggs, and lower quality two-egg
clutches producing eggs of lower hatchability but a similar
probability of producing at least one hatchling.
Here we derive predictions from both the IEH and IOH
(Table 1) and report an experimental test of these hypotheses
using an obligately siblicidal seabird, the Nazca booby (Sula
granti). The Nazca boobies of the Galápagos have traditionally
been considered masked boobies (Sula dactylatra), but recent
analyses of morphological and breeding data (Pitman and
Jehl, 1998), and of mtDNA differentiation (Friesen et al.,
manuscript in review) support elevation of the Galápagos and
nearby populations to species status. In this article we follow
the nomenclature of Pitman and Jehl (1998) and refer to
these birds as Nazca boobies. We did not test all aspects of the
IOH and IEH, as we were not able to force Nazca boobies to
lay additional eggs or prevent them from laying, and thus
could not incorporate egg production and laying costs. Our
evaluation of these hypotheses is thus confined to the postlaying period.
Nazca boobies lay clutches of one or two eggs (Nelson,
1966); over one 3-year period at our study site, 44–66% of
clutches consisted of two eggs (Anderson, 1990a). Eggs are
incubated for approximately 43 days, and eggs hatch about 5
days apart (Anderson, 1993). If both eggs hatch in a two-egg
clutch, the first-hatched offspring pushes its sibling from the
nest scrape shortly after hatching, and it dies of exposure or
is scavenged by crabs or landbirds (Anderson, 1989; Nelson,
1978). Therefore, while clutch size varies among individuals,
final brood size does not. Chicks fledge at 113–120 days of
age (Nelson, 1978).
Nazca boobies provide a situation of unusual clarity for the
study of clutch size evolution. The obligately siblicidal nature
of their brood-size reduction results in surplus offspring having only replacement reproductive value (Mock and Parker,
1986) because it is only raised if the core offspring fails to
hatch or dies very early. The other component of reproductive
value, extra reproductive value, is that portion of the offspring’s survivorship which is not contingent on the fate of its
sibling; it equals zero for Nazca boobies (Mock and Parker,
1986). Because most hypotheses for adaptive clutch size evo-
lution focus on extra reproductive value, they are not relevant
in this case.
We conducted this experiment at the large breeding colony
at Punta Cevallos, Isla Española, in the Galápagos Islands
(1⬚20⬘ S, 89⬚40⬘ W; see Anderson and Ricklefs, 1987) during
the 1995–1996 breeding season (October–June). On the first
day of the experiment we used a pencil to mark all eggs with
a small ‘‘X,’’ so that on subsequent days we could identify
newly-laid unmarked eggs. We checked for new clutches daily.
When a new clutch was found, we monitored it for 7–10 days
to determine the natural clutch size (laying intervals are 4–9
days; see Anderson, 1993), before assigning it to a treatment
group. After this period, clutches were randomly assigned (see
below) to treatment groups and some clutch sizes were adjusted. ‘‘Reduced’’ clutches had the second-laid ‘‘B-egg’’ removed from natural two-egg clutches; ‘‘enlarged’’ clutches
had a B-egg added to natural one-egg clutches; and
‘‘switched’’ clutches consisted of natural one-egg clutches that
were removed and replaced with a single age-matched egg
from another natural one-egg nest. The switched treatment
was designed to control for introduction of a foreign egg. In
addition, two control groups consisted of natural one-egg
clutches (C1) and natural two-egg clutches (C2).
We assigned clutches to treatment groups with the objective
of maintaining equal mean clutch initiation dates across treatments. However, one-egg clutches were less available (33.5%
of clutches) and our experimental protocol required twice as
many one-egg clutches as two-egg clutches. As a result, clutch
initiation dates differed between some treatments. The median clutch initiation dates for the treatment groups were as
follows: reduced, November 20 (n ⫽ 65); enlarged, November
20 (n ⫽ 57); C2, November 27 (n ⫽ 167); C1, November 27
(n ⫽ 36); switched one-egg clutches, December 6 (n ⫽ 60;
Kruskall-Wallis H ⫽ 66.2, df ⫽ 4, p ⬍ .001). We were able to
control statistically for differences in clutch initiation date
(see Statistical Analyses). Natural two-egg clutches were assigned to the reduced group when a natural one-egg clutch
was available to receive a donated egg and the donor’s B-egg
was 5–7 days younger than the recipient’s A-egg, thereby mimicking natural laying asynchrony. Otherwise, two-egg clutches
were assigned to the unmanipulated control group. One-egg
clutches were assigned to the switched group when two natural one-egg clutches were laid within 2 days of each other.
One-egg control clutches were treated identically, except that
no eggs were exchanged.
We monitored nests daily to determine the fates of eggs and
chicks. Because chicks from two-egg clutches hatch an average
of 5 days apart, a distinct size difference between A- and Bchicks allowed us to distinguish between them. We weighed
offspring every 10 days beginning on day of hatching (d 0)
and continued until fledging, and when they reached an easily
recognizable late developmental stage at which all but 1% of
the down coverage had been replaced by pennaceous feathers.
Test of the Insurance Egg Hypothesis
The IEH assumes that B-eggs can provide an offspring when
the A-egg fails to hatch, or the A-offspring dies before siblicide
occurs. Given the B-egg’s potential to act as insurance, the
IEH predicts that two-egg Nazca booby clutches should always
yield higher reproductive success than do one-egg clutches
(Table 1). We first examined the assumption that B-eggs provide a nestling when A-eggs fail or A-chicks die. We tested the
prediction of the IEH by comparing each group’s probability
of producing at least one hatchling and probability of pro-
Behavioral Ecology Vol. 12 No. 3
ducing a fledgling (fledging success). Fledging success may
not accurately reflect reproductive success if fledglings differ
in quality. We therefore compared offspring mass at the 1%
down stage (see above) between groups, as well as offspring
growth rate.
Test of the Individual Optimization Hypothesis
The IOH assumes that individuals vary in quality. Clutch size
may reflect parental quality, with high quality parents producing larger clutches than low quality parents do. This assumption is supported if parents that produce two eggs have higher
reproductive success than do parents that produce a single
egg. Given differences in parental quality, the IOH makes two
directional predictions in relation to our clutch size manipulations (Table 1). Clutches which have been adjusted by adding or removing an egg should have lower reproductive success than control clutches do, because individuals should do
best with the clutch size that they laid. We tested the assumption and predictions of the IOH by comparing the probability
of producing at least one hatchling and the probability of
fledging between treatment groups.
Statistical tests
The first dependent variable of interest, whether or not a
clutch produces at least one hatchling, is a binary variable.
Since Nazca boobies fledge a maximum of one offspring per
reproductive attempt, fledging success is also a binary variable. Logistic regression is designed for use with dichotomous
dependent variables such as these, and allows the inclusion of
a covariate. By including clutch initiation date as a covariate,
we were able remove variation between experimental groups
due to heterogenous clutch initiation dates, and thus to examine variation due to experimental treatment alone. We initially included clutch initiation date in the model as a continuous variable. However, in checking the logistic regression assumptions, we found that the logit transform did not linearize
the probability of producing at least one hatchling in relationship to clutch initiation date. When clutches were divided
into quartiles based on their clutch initiation dates, and the
mean probability of producing at least one hatchling for each
quartile was plotted against the clutch initiation date midpoint
for each group, we found that the probability of producing a
hatchling remained constant over the first three quartiles and
dropped in the fourth quartile. The same pattern existed for
fledging success. This suggested treatment of clutch initiation
date as a dichotomous variable also (Hosmer and Lemeshow,
1989): ‘‘early’’ if the clutch is initiated during the first 75%
of clutch initiation dates and ‘‘late’’ if during the last 25%.
We therefore constructed our models with two independent
variables: treatment and lay period (early or late).
When we examined interactions between treatment group
and lay period for the dependent variable fledging success,
we found that the model including a treatment by lay period
interaction did not perform significantly better than a model
without the interaction (␹2 ⫽ 3.50, df ⫽ 3, p ⫽ .32). Therefore, we did not include a treatment by lay period interaction
term in our final model for fledging success. The model describing the probability of producing at least one hatchling
that included a treatment by lay period interaction performed
marginally better than the model without the interaction term
(␹2 ⫽ 5.67, df ⫽ 3, p ⫽ .13). This at least suggested that the
effect of treatment group did not remain constant across levels of lay period. We therefore analyzed the probability of producing at least one hatchling for early and late clutches separately. Parameter estimates and probability levels were calculated using the Statistica 4.5 nonlinear estimation module
(StatSoft Inc., 1993).
When testing for differences between treatment groups, we
employed the false discovery rate procedure (Benjamini and
Hochberg, 1995) to adjust p-values for multiple comparisons.
The procedure requires comparisons be ordered by decreasing p-values, and then compared to a critical significance level
beginning with the largest p-value. The critical significance
level for each comparison, di, is calculated by dividing the
specific comparison number by the total number of comparisons and then multiplying by the false discovery rate (the
probability of mistakenly rejecting a null hypothesis). For example, the fifth comparison (the comparison with the fifth
largest p-value) of six total comparisons, given a false discovery
rate of 0.05, has a di of 0.042 (⫽ 5/6 * 0.05). If the achieved
significance level is less than di for a given comparison, then
the null hypothesis is rejected for that comparison, and all
remaining comparisons (Benjamini and Hochberg, 1995; Curran-Everett, 2000).
To determine if offspring growth rates differed between
treatment, we performed repeated measures ANCOVA on offspring masses measured at 10-day intervals for offspring that
survived to fledging. We also analyzed offspring mass at fledging using ANCOVA.
Consequence of incubating a foreign egg
Enlarged clutches consisted of one egg that belonged to the
parents and one foreign egg. If parents recognized the introduction of the foreign egg, they might have altered their incubation or attendance pattern, possibly affecting the probability of producing a hatchling. To evaluate this possibility, we
compared the probability of producing at least one hatchling
for one-egg control clutches, in which parents incubated their
own egg, with switched clutches, in which parents incubated
a single foreign egg. We initially performed logistic regression
including lay period as a covariate, and found that the coefficient for lay period was not significant (t ⫽ 0.90, df ⫽ 93, p
⫽ .37), nor was the coefficient discriminating between the two
groups (t ⫽ 1.29, df ⫽ 93, p ⫽ .20). Since the regression
model indicated that lay period did not contribute significantly to the model, we performed a 2 ⫻ 2 contingency table
analysis to compare the proportions of clutches that successfully initiated broods in these two groups (switched 24/60 ⫽
0.40; C1 20/36 ⫽ 0.56) and found no significant difference
(Yates’ corrected ␹2 ⫽ 1.61, df ⫽ 1, p ⫽ .20), indicating that
parents incubating a foreign egg had the same probability of
producing at least one hatchling as did parents incubating
their own egg. In addition, an egg recognition experiment
demonstrated that parents do not discriminate their own 10
day old eggs from age-matched foreign eggs (Clifford LD and
Anderson DJ, unpublished data).
If parents did discriminate foreign eggs from their own and
acted in a way that lowered the probability of producing a
hatchling, then we would expect enlarged clutches (which
contained a foreign egg) to have a lower probability of producing at least one hatchling than C2 clutches did (which did
not contain a foreign egg). The proportion of enlarged
clutches that produced at least one hatchling (0.83, n ⫽ 57)
was not lower than the proportion in C2 clutches (0.77, n ⫽
167; Yates’s corrected ␹2 ⫽ 0.53, df ⫽ 1, p ⫽ .47), indicating
that the presence of a foreign egg in the nest did not alter
the parents’ behavior in a way that negatively affected the
probability of the foreign egg hatching. As a group, these analyses indicated that the presence of a foreign egg in enlarged
clutches did not confound comparisons with treatment groups
lacking foreign eggs. For the remaining analyses, we used data
from the C1 group, and not the switched group, in order to
be conservative in our comparisons.
Clifford and Anderson • Insurance eggs in boobies
Table 2
Predicted probabilities and significance tests from logistic regression models describing the probability of a clutch producing a hatchling and
the probability of a clutch producing a fledgling
Predicted probabilities
Early lay period
Late lay period
Early lay period
Late lay period
Across lay periods
Reduced versus C2
Enlarged versus C1
Enlarged versus reduced
C2 versus C1
Enlarged versus C2
Reduced versus C1
p-value, achieved significance level; di, critical significance level from false discovery rate procedure. After arranging comparisons in order of
decreasing p-values, if p-value ⬍ di then the null hypothesis for that comparison and for all remaining comparisons are rejected. Superscripts
on p-values indicate ordering. One marginally significant and all significant comparisons appear in bold.
Test of assumptions of the IEH
Final brood size was one in every nest that hatched two chicks.
Most B-chicks (97/103 ⫽ 0.942) were apparently ejected from
the nest or died in the nest within 6 days of hatching; all
except one were dead by 15 days after hatching. In that case
the B-chick survived to 46 days. No parents produced two
The B-egg provided a hatchling when the A-egg failed to
hatch in 8.8% (5/57) of enlarged clutches, and when the Achick died in 3.5% (2/57) of enlarged clutches. The total
replacement rate for enlarged clutches was 12.3% (7/57). In
C2 clutches, the B-egg provided a hatchling when the A-egg
failed to hatch in 8.4% (14/167) of clutches, and when the
A-chick died in 1.2% (2/167) of clutches. The total replacement rate of B hatchlings for C2 clutches was 9.6% (16/167).
The B-egg produced the surviving fledgling in 13.9% (5/36)
of enlarged clutches that produced a fledgling, and in 10.5%
(12/114) of C2 clutches that produced a fledgling in this experiment. Thus B-eggs had insurance value in both experimental and control broods.
Probability of producing a hatchling
The addition of treatment to a logistic regression model for
early clutches that included only the intercept significantly
Figure 1
Proportion of clutches in each treatment group that produced at
least one hatchling, classified by lay period. Bars represent the 95%
confidence interval for proportions, and sample sizes are in
improved the model (␹2 ⫽ 15.05, df ⫽ 3, p ⬍ .01), indicating
that membership in a treatment group influenced a clutch’s
probability of producing at least one hatchling. For early
clutches, the logistic regression model describing the probability of producing a hatchling showed that C2 clutches had
a significantly higher probability of producing a hatchling
than did reduced or C1 clutches (Table 2). Enlarged clutches
also had a significantly higher probability of producing at least
one hatchling than did reduced or C1 clutches. Enlarged and
C2 clutches did not differ in their probability of producing a
hatchling, nor did reduced and C1 clutches. These results supported all six of the IEH’s predictions, but only one of two
predictions made by the IOH.
The addition of treatment to a logistic regression model for
late clutches that included only the intercept did not improve
the model (␹2 ⫽ 3.36, df ⫽ 3, p ⫽ .34), indicating that the
probability of producing at least one hatchling was not influenced by treatment in late clutches. The probability of successfully initiating broods did not differ for any of the treatments in late clutches (Table 2 and Figure 1). Because of the
small number of nests in the late lay period, power was less
than 80% to detect even a 50% difference between groups.
Fledging success
Treatment group also significantly influenced a clutch’s probability of producing a fledgling (␹2 ⫽ 17.62, df ⫽ 1, p ⬍ .001).
Controlling for lay period, enlarged clutches had a significantly higher probability of producing a fledgling than C1
clutches did (Table 2). C2 clutches had a marginally significantly higher probability of producing a fledgling than reduced clutches did. C2 clutches had a significantly higher
probability of producing a fledgling than C1 clutches did, and
reduced clutches had a marginally significantly higher probability of producing a fledgling than C1 clutches did. The
probability of producing a fledgling did not differ between
enlarged and reduced clutches, nor between enlarged and C2
clutches. Four of the six predictions of the IEH and one of
the two predictions of the IOH were supported by these data.
Chick growth rates and mass at fledging
The growth rates of offspring that fledged in different treatment groups did not differ from each other (F3, 164 ⫽ 0.872,
p ⫽ .46), and we detected no difference in offspring mass at
Figure 2
Proportion of clutches in each treatment group that produced a
fledgling, classified by lay period. Bars represent 95% confidence
intervals, and sample sizes are in parentheses.
the 1% down stage across treatment groups (ANCOVA, F3, 192
⫽ 0.83, p ⫽ 0.48).
The Insurance Egg Hypothesis
The logistic regression model describing the probability of
producing at least one hatchling for early clutches provided
support for all six of the IEH’s predictions. Nazca booby
clutches with two eggs were more likely to produce a hatchling than were single-egg clutches, regardless of the parents’
original clutch size (Figure 1). The C2, C1, and enlarged
groups maintained their relationships to each other (Figure
1) in both the early and late clutch initiation periods, and
they all had a lower probability of producing a hatchling late
in the season. In contrast, the reduced group had approximately the same probability of producing a hatchling early in
the season and late in the season (0.65 compared to 0.64).
While our data show an immediate effect of insurance eggs
on the probability of producing at least one hatchling, data
on fledging success better indicate whether the IEH is a sufficient ultimate explanation for variation in overall reproductive success among natural clutch sizes. Four of the six predictions of the IEH were supported by the fledging success
data. Enlarged clutches had higher fledging success than C1
clutches did, indicating that parents that produced only one
egg would have had higher reproductive success if they had
laid two eggs. C2 clutches had higher fledging success than
reduced clutches did, indicating that parents that produced
two eggs would have had lower reproductive success had they
produced only one egg. C2 clutches and enlarged clutches
did not have significantly different fledging success, and C2
clutches had significantly higher fledging success than C1
The logistic regression model for fledging success in Nazca
boobies showed a decrease in the probability of producing a
fledgling late in the season that paralleled the seasonal decline in the probability of producing a hatchling (Figures 1
and 2). A seasonal decline in reproductive success is a commonly observed pattern in birds and is associated with a concomitant decrease in clutch size (Crick et al., 1993; Klomp,
1970; Perrins, 1970). Our data showed such a seasonal decline
in reproductive success independent of clutch size, since we
saw a within-clutch size decrease in both the probability of
producing a hatchling and fledging success.
Behavioral Ecology Vol. 12 No. 3
The Individual Optimization Hypothesis
We found that parents laying two eggs were more successful
at raising a chick from hatching to fledging than were parents
laying one egg. This result supports the IOH’s assumption of
a positive correlation between clutch size and parental quality.
Reduced clutches had marginally significantly higher fledging
success than C1 clutches, and they fledged a similar proportion of young to that of enlarged clutches. In both of these
comparisons, it appears that an additional effect on reproductive success interacts with the insurance effect during the
nestling period. The additional factor appears to involve intrinsic parental quality. In the case of enlarged versus reduced
clutches, enlarged clutches had a significant advantage at
hatching but not at fledging. Parents of enlarged clutches
originally laid only a single egg, and parents of reduced
clutches originally laid two eggs. Parents of enlarged clutches
were thus apparently less capable of finding food during egg
formation, since food limitation during this period accounts
for most variation in clutch size (Clifford LD and Anderson
DJ, in press). This variation in intrinsic parental quality was
associated with a marginally significant advantage of reduced
parents over enlarged parents in raising a hatchling to fledging among early layers (t ⫽ 1.91, df ⫽ 189, p ⫽ .057; Clifford
LD and Anderson DJ, unpublished data).
Differences in parents’ abilities to produce a fledgling
could result from differences in abilities to absorb the costs
of incubating additional eggs. Some studies do suggest that
incubation costs are reflected in the parents’ abilities to raise
offspring (Heaney and Monaghan, 1996; Monaghan and Nager, 1997) and therefore affect fledging success and/or offspring condition.
That parents producing large clutches are higher quality
parents than those producing small clutches has been suggested by correlations between clutch size and offspring survival rate in other avian taxa (e.g., Kittiwake gull Rissa tridactyla; Coulson and Porter, 1985; Blue tit Parus caeruleus; Nur,
1986). In addition, experimental manipulations resulting in
parents with the same brood sizes showed that recruitment
rates were higher for great tits (Parus major) and blue tits
(Parus caeruleus) that originally laid larger clutches (Pettifor,
1993; Pettifor et al., 1988), further indicating a quality difference between parents of large and small clutches.
While our data did support the IOH’s assumption that large
clutches are produced by high quality parents, the critical test
of the IOH is whether parents with manipulated clutch sizes
have lower reproductive success than parents with unmanipulated clutch sizes. Experimentally enlarged clutches should
have lower reproductive success than C1 clutches. This prediction is opposite that of the IEH, which predicts that enlarged clutches should have higher reproductive success than
C1 clutches. Our data from Nazca boobies do not support this
prediction of the IOH, as enlarged clutches had a higher
probability of producing at least one hatchling as well as a
higher probability of producing a fledgling than C1 clutches.
Therefore, barring egg-laying costs and assuming that fledging success is an accurate estimate of parental fitness, Nazca
boobies that lay one-egg clutches do not appear to be laying
their optimal clutch size.
Parents with enlarged clutches were given ‘‘free’’ eggs without incurring the cost of producing and laying them. These
costs have been shown to reduce chick survival and female
condition in other species (Heaney and Monaghan, 1995;
Monaghan et al., 1995, 1998). In addition, these costs might
affect parents’ future survival or reproductive success (Charnov and Krebs, 1974; Williams, 1966). If we were able to incorporate these costs into the experiment, we might find that
the optimal clutch size for some Nazca boobies was indeed
Clifford and Anderson • Insurance eggs in boobies
Table 3
Observed hatching success (HS) for C2 and enlarged clutches, as well as the observed and expected
numbers and proportions of clutches that produced two, one, or zero hatchlings
Number of clutches (proportion) that produced
Two chicks
One chick
Zero chicks
HS, number of eggs that hatched/number of eggs laid. Expected proportions are calculated from
observed hatching success for each group (e.g., expected proportion of C2 clutches to produce two
chicks ⫽ (0.638)(0.638) ⫽ 0.407).
* ␹2 ⫽ 32.4, df ⫽ 2, p ⬍ .001.
Insurance value and parental quality
In the early clutch initiation period, two-egg clutches had a
clear advantage over one-egg clutches in terms of the probability of producing at least one hatchling. However, for clutches in the late clutch initiation period, the logistic regression
model failed to detect any significant differences. This failure
does not contradict the IEH for several reasons. First, small
sample sizes resulted in insufficient power to detect differences of the magnitude observed. Second, while acknowledging
low statistical power, differences between the C2, C1, and enlarged were in the direction predicted by the IEH. And last,
clutches initiated in our ‘‘late’’ lay period are of little significance to selection for clutch size because they represent a
small proportion of clutches (only 13.3% of 458 clutches laid
by a random sample of Nazca boobies in the Punta Cevallos
colony were laid during this same period) and late clutches
have lower reproductive success than early clutches do (Fernández P and Anderson DJ, unpublished data).
Two of the predictions of the IEH were not supported by
the fledging success data. Enlarged clutches had an insurance
advantage at hatching (Table 2), but reduced clutches had a
countervailing advantage in parental quality thereafter, so the
two groups did not differ in overall fledging success (Table
2). We were able to experimentally decouple parental quality
from clutch size; however, our estimate of parental quality will
covary with clutch size in natural situations. Only two of the
six predictions regarding fledging success in Table 2 do not
confound clutch size and parental quality, and these constitute the essential test of the IEH, given our discovery of variation in parental quality. Both of these predictions (reduced
versus C2 and enlarged versus C1) were strongly supported.
Variation in probablity of producing a hatchling
In addition to parental quality differences between birds that
laid different clutch sizes, we also found an apparent within
clutch size quality difference reflected in hatching success
(the proportion of eggs laid that hatched). Enlarged clutches
produced two hatchlings, one hatchling and zero hatchlings
at the frequency expected from a binomial expansion based
on hatching success (Table 3). For example, the probability
of two eggs hatching in an enlarged clutch is roughly equal
to (0.596)(0.596) ⫽ 0.355, the equivalent of 20 clutches in
our sample. The actual number of enlarged clutches that produced two hatchlings was 21. But C2 clutches produced two
hatchlings and zero hatchlings more often than expected, and
one hatchling less often than expected from the observed
hatching success. This indicates that some C2 parents produce
two low quality eggs or lose both more often than enlarged
parents. This observation is consistent with Nazca booby data
reported by Anderson (1990a), which showed that the pro-
portion of two-egg clutches in which either both hatched or
both failed was higher than expected if hatching was independent of nest. While these data reflected only intrinsic
hatchability, ours do not distinguish intrinsic hatchability and
extrinsic causes of egg loss.
Other hypotheses for overproduction of offspring
Two other hypotheses have been proposed to explain overproduction of offspring. The Resource Tracking Hypothesis
(Lack, 1947, 1954; Temme and Charnov, 1987) proposes that
surplus offspring serve a bet-hedging function; in years of
high resources they are raised in addition to core offspring,
but in low resource years they are eliminated. In 14 years of
field work at the Punta Cevallos colony, we have never seen
parents raise two chicks to fledging (Anderson DJ, unpublished data). Therefore this hypothesis is unlikely to explain
two-egg clutches. The Offspring Facilitation Hypothesis proposes that surplus offspring aid core offspring to survive and/
or reproduce (Mock and Forbes, 1995), and is commonly associated with cannibalism. However, Nazca booby B-chicks are
not consumed by conspecifics, nor do they offer any obvious
aid to A-chicks.
The Insurance Egg Hypothesis appears to be the best explanation for overproduction of offspring in the Nazca booby.
We have shown that surplus B-eggs can provide insurance
against the failure of A-eggs. Experimentally enlarged clutches
produced more hatchlings and fledglings than control clutches did, and experimentally reduced clutches produced fewer.
Therefore, selection should favor the production of two-egg
clutches over one-egg clutches, and one-egg clutches probably
result from some proximate constraint experienced by the
parents. Evidence that females are limited by food availability
is provided by supplemental feeding experiments that increased clutch size in birds (see reviews in Arcese and Smith,
1988; Boutin, 1990; Martin, 1987; Meijer et al., 1990; see also
Aparicio, 1994; Nilsson, 1991; Soler and Soler, 1996), and litter size in mammals (review in Boutin, 1990). Age may also
act as a proximate constraint; clutch size is known to increase
with age in many species (see review in Fowler, 1995).
The problem of unused insurance value is not unique to
Nazca boobies; other obligate brood reducing species also lay
one-egg clutches, with varying frequencies. For example, rockhopper penguins (Eudyptes chrysocome) apparently always lay
insurance eggs (Williams, 1981; but see St. Clair and St. Clair,
1996), while several eagle species lay insurance eggs from 2–
87% of the time (Brown, 1966; Cramp and Simmons, 1980;
Gargett, 1977; Hustler and Howells, 1988; Simmons, 1997).
Future experimental work, such as supplemental feeding,
should attempt to identify the causal factors preventing indi-
viduals from taking advantage of insurance eggs in obligately
siblicidal taxa with significant proportions of one-egg clutches.
Insurance function has been most widely explored in obligately siblicidal birds with small clutch sizes, but is also applicable to species that experience less frequent offspring loss
(Forbes, 1990, 1991; Forbes et al., 1997; Lamey et al., 1996;
Mock and Parker, 1986; Mock et al., 1990). The insurance
reproductive value for last hatched nestlings in facultatively
siblicidal species can be equal to or greater than that found
in obligately siblicidal species (Wiebe, 1996). Brood size manipulations in the double-crested cormorant (Phalacrocorax
auritus), a facultatively siblicidal seabird that lays clutches of
three–four eggs, provided limited evidence that the insurance
value of eggs increased the reproductive success of parents
(Hunt and Evans, 1997). However, fledging success for these
broods was not ascertained. An experimental test of the IEH
in the lesser kestrel (Falco naumanni) provided support for
the IEH in a species that lays two–six eggs (Aparicio, 1997).
Other studies have suggested that insurance is a contributing
factor to clutch size evolution in non-obligately siblicidal species with large clutch sizes (Beissinger and Waltman, 1991;
Krebs, 1999). Future considerations of clutch size evolution
should not neglect the potential insurance component of eggs
and offspring.
We thank the Galápagos National Park Service for permission to work
in the park, and the Charles Darwin Research Station and TAME
airline for logistical support. We thank B. Shelton, J. L. Norris, and
D. Case for statistical assistance, M. R. Silman and L. S. Forbes for
critical discussion of this topic, and S. R. Beissinger, K. P. Huyvaert,
D. W. Mock, D. F. Westneat, and two anonymous reviewers for comments on an earlier version. National Science Foundation grant
DEB93–04579 and DEB96–29539 and Wake Forest University provided financial support.
Anderson DJ, 1989. The role of hatching asynchrony in siblicidal
brood reduction of two booby species. Behav Ecol Sociobiol 25:363–
Anderson DJ, 1990a. Evolution of obligate siblicide in boobies. 1. A
test of the insurance-egg hypothesis. Am Nat 135:334–350.
Anderson DJ, 1990b. On the evolution of human brood size. Evolution 44:438–440.
Anderson DJ, 1993. Masked booby (Sula dactylatra). In: The Birds of
North America, No. 73 (Poole A, Gill FB, eds). Philadelphia: The
Academy of Natural Sciences. Washington, DC: The American Ornithologists’ Union; 1–16.
Anderson DJ, Ricklefs RE, 1987. Radio-tracking masked and blue-footed boobies (Sula spp.) in the Galápagos Islands. Nat Geogr Res 3:
Aparicio JM, 1994. The seasonal decline in clutch size: an experiment
with supplementary food in the kestrel, Falco tinnunculus. Oikos
Aparicio JM, 1997. Costs and benefits of surplus offspring in the lesser
kestrel (Falco naumanni). Behav Ecol Sociobiol 41:129–137.
Arcese P, Smith JNM, 1988. Effects of population density and supplemental food on reproduction in song sparrows. J Anim Ecol 57:
Barber CA, Evans RM, 1995. Clutch-size manipulations in the yellowheaded blackbird: a test of the individual optimization hypothesis.
Condor 97:352–360.
Bartlett J, 1987. Filial cannibalism in burying beetles. Behav Ecol Sociobiol 21:179–183.
Beissinger SR, Waltman JR, 1991. Extraordinary clutch size and hatching asynchrony of a neotropical parrot. Auk 108:863–871.
Benjamini Y, Hochberg Y, 1995. Controlling the false discovery rate:
a practical and powerful approach to multiple testing. J R Statist
Soc B 57:289–300.
Boutin S, 1990. Food supplementation experiments with terrestrial
vertebrates: patterns, problems, and the future. Can J Zool 68:203–
Behavioral Ecology Vol. 12 No. 3
Brown LH, 1966. Observations on some Kenya eagles. Ibis 108:531–
Cash KJ, Evans RM, 1986. Brood reduction in the American white
pelican (Pelecanus erythrorhyncus). Behav Ecol Sociobiol 18:413–
Charnov EL, Krebs JR, 1974. On clutch size and fitness. Ibis 116:217–
Clifford LD, Anderson DJ, in press. Food limitation explains most
clutch size variations in the Nazca booby. J Anim Ecol.
Coulson JC, Porter JM, 1985. Reproductive success of the kittiwake
Rissa tridactlya: the roles of clutch size, chick growth rates and parental quality. Ibis 127:450–466.
Cramp S, Simmons KEL, 1980. Birds of the Western Palearctic, vol.
11. Hawks to bustards. Oxford: Oxford University Press.
Crick HQP, Gibbons DW, Magrath RD, 1993. Seasonal changes in
clutch size in British birds. J Anim Ecol 62:263–273.
Curran-Everett D, 2000. Multiple comparisons: philosophies and illustrations. Am J Physiol Regul Integr Comp Physiol 279:R1–R8.
Dorward DR, 1962. Comparative biology of the white booby and the
brown booby Sula spp. at Ascension. Ibis 103b:174–200.
Ehrlén J, 1991. Why do plants produce surplus flowers? A reserveovary model. Am Nat 138:918–933.
Forbes LS, 1990. Insurance offspring and the evolution of avian clutch
size. J Theor Biol 147:345–359.
Forbes LS, 1991. Insurance offspring and brood reduction in a variable environment: the costs and benefits of pessimism. Oikos 62:
Forbes LS, Lamey TC, 1996. Insurance, developmental accidents, and
the risks of putting all your eggs in one basket. J Theor Biol 180:
Forbes LS, Thornton S, Glassey B, Forbes M, Buckley NJ, 1997. Why
do parent birds play favourites? Nature 390:351–352.
Fowler GS, 1995. Stages of age-related reproductive success in birds:
simultaneous effects of age, pair-bond duration and reproductive
experience. Amer Zool 35:318–328.
Gargett V, 1977. A 13-year population study of the black eagles in the
Matopos, Rhodesia, 1964–1976. Ostrich 48:17–27.
Graves J, Whiten A, Henzi P, 1984. Why does the herring gull lay three
eggs? Anim Behav 32:798–805.
Heaney V, Monaghan P, 1995. A within-clutch trade-off between egg
production and rearing in birds. Proc R Soc Lond B 261:361–365.
Heaney V, Monaghan P, 1996. Optimal allocation of effort between
reproductive phases: the trade-off between egg production and
rearing in birds. Proc R Soc Lond B 263:1719–1724.
Högstedt G, 1980. Evolution of clutch size in birds: Adaptive variation
in relation to territory quality. Science 210:1148–1150.
Hosmer DW, Lemeshow S, 1989. Appied logistic regression. New York:
John Wiley & Sons.
Hunt JD, Evans RM, 1997. Brood reduction and the insurance-egg
hypothesis in double crested cormorants. Colonial Waterbirds 20:
Hustler K, Howells WW, 1988. The effect of primary production on
breeding success and habitat selection in the African hawk-eagle.
Condor 90:583–587.
Kepler CB, 1969. Breeding biology of the blue-faced booby Sula dactylatra personata on Green Island, Kure Atoll. Publ. 8. Cambridge,
Massachusetts: Nuttall Ornithological Club.
Klomp H, 1970. The determination of clutch size in birds. Ardea 58:
Krebs EA, 1999. Last but not least: nestling growth and survival in
asynchronously hatching crimson rosellas. J Anim Ecol 68:266–281.
Lack D, 1947. The significance of clutch size. Ibis 89:302–352.
Lack D, 1954. The natural regulation of animal numbers. Oxford:
Clarendon Press.
Lamey TC, Evans RM, Hunt JD, 1996. Insurance reproductive value
and facultative brood reduction. Oikos 77:285–290.
Martin TE, 1987. Food as a limit on breeding birds: a life-history
perspective. Ann Rev Ecol Syst 18:453–87.
Meijer T, Daan S, Hall M, 1990. Family planning in the kestrel (Falco
tinnunculus): the proximate control of covariation of laying date
and clutch size. Behav 114:117–136.
Millar JS, 1973. Evolution of litter-size in the pika, Ochotona princeps
(Richardson). Evolution 27:134–143.
Mock DW, Drummond H, Stinson CH, 1990. Avian siblicide. Am Sci
Clifford and Anderson • Insurance eggs in boobies
Mock DW, Forbes LS, 1995. The evolution of parental optimism.
Trends Ecol Evol 10:130–134.
Mock DW, Parker GA, 1986. Advantages and disadvantages of egret
and heron brood reduction. Evolution 40:459–470.
Mock DW, Parker GA, 1997. The evolution of sibling rivalry. Oxford:
Oxford University Press.
Monaghan P, Bolton M, Houston DC, 1995. Egg production constraints and the evolution of avian clutch size. Proc R Soc Lond B
Monaghan P, Nager RG, 1997. Why don’t birds lay more eggs? Trends
Ecol Evol 12:270–274.
Monaghan P, Nager RG, Houston DC, 1998. The price of eggs: increased investment in egg production reduces the offspring rearing
capacity of parents. Proc R Soc Lond B 265:1731–1735.
Nelson JB, 1966. Clutch size in the sulidae. Nature 210:435–436.
Nelson JB, 1978. The sulidae. Oxford: Oxford University Press.
Nilsson J.-Å, 1991. Clutch size determination in the marsh tit (Parus
palustris). Ecology 72:1757–1762.
Nur N, 1986. Is clutch size variation in the blue tit (Parus caeruleus)
adaptive? An experimental study. J Anim Ecol 55:983–999.
Perrins CM, 1970. The timing of birds’ breeding seasons. Ibis 112:
Perrins CM, Moss D, 1975. Reproductive rates in the great tit. J Anim
Ecol 44:659–706.
Pettifor RA, 1993. Brood-manipulation experiments. I. The Number
of offspring surviving per nest in blue tits (Parus caeruleus). J Anim
Ecol 62:131–144.
Pettifor RA, Perrins CM, McCleery RH, 1988. Individual optimization
of clutch size in great tits. Nature 336:160–162.
Pitman RL, Jehl JR, Jr., 1998. Geographic variation and reassessment
of species limits in the ‘‘masked’’ boobies of the eastern Pacific
Ocean. Wilson Bull 110:155–170.
Rosenheim JA, Hongkham D, 1996. Clutch size in an obligately siblicidal parasitoid wasp. Anim Behav 51:841–852.
Simmons R, 1997. Why don’t all siblicidal eagles lay insurance eggs?
The egg quality hypothesis. Behav Ecol 8:544–550.
Soler M, Soler JJ, 1996. Effects of experimental food provisioning on
reproduction in the Jackdaw Corvus monedula, a semi-colonial species. Ibis 138:377–383.
StatSoft, Inc., 1993. Statistica. Tulsa, Oklahoma: StatSoft.
St. Clair CC, St. Clair RC, 1996. Causes and consequences of egg loss
in the rockhopper penguins, Eudyptes chrysocome. Oikos 77:459–
Temme DH, Charnov EL, 1987. Brood size adjustments in birds: Economical tracking in a temporally varying environment. J Theor Biol
Wiebe KL, 1996. The insurance egg-hypothesis and extra reproductive
value of last-laid eggs in clutches of American kestrels. Auk 113:
Williams AJ, 1981. The clutch size of macaroni and rockhopper penguins. Emu 81:87–90.
Williams GC, 1966. Natural selection, the costs of reproduction, and
a refinement of Lack’s principle. Am Nat 100:687–690.