1. Grouping provides antipredatory benefits, and therefore aggregation tendencies increase under heightened predation risk. In socially breeding groups, however, conflicts over reproductive shares or safety tend to disintegrate groups. Group formation thereby involves a balance between the antipredatory benefits of aggregation and the destabilizing effect of reproductive conflict.
2. We study the grouping behaviour of a facultatively social precocial sea duck with uniparental female care, the eider (Somateria mollissima Linnaeus). Females tend their young solitarily or in groups of 2–5 females. Here, we focus on the effect predation on adults has on group-formation decisions of brood-caring females.
3. By modifying an existing bidding game over care, we model the effects of predation risk on the width of the window of selfishness, which delimits the reproductive sharing allowing cooperation within brood-rearing coalitions, and generate predictions about the relative frequencies of solitary versus cooperative parental care modes. Furthermore, we model the dilution effect as a function of female group size and predation risk.
4. The window of selfishness widens with increasing predation risk, and the dilution of predation risk increases with both female group size and increasing predation risk, yielding the following predictions: (i) cooperative brood care becomes more prevalent and, conversely, solitary brood care less prevalent under heightened predation risk and (ii) group sizes increase concomitantly.
5. We tested these predictions using 13 years of data on female grouping decisions and annual predation rates, while controlling for the potentially confounding effect of female body condition.
6. Our data supported both predictions, where heightened predation risk of nesting females, but not changes in their condition, increased the relative frequency of cooperative brood care. Increased female nesting mortality also resulted in larger groups of cooperative females.
7. The predation risk of incubating females has long-term implications for later parental care decisions. We discuss the potential consequences of predation-induced shifts in group size on per capita fitness and population-wide productivity.
Aggregating into groups provides antipredatory benefits such as increased probability of predator detection, predator confusion, dilution of individual predation risk given that predator-encounter probabilities do not increase linearly with group size, shared information and division of labour (Krause & Ruxton 2002). Selfish avoidance of predators by means of positioning other individuals between one self and the potential predator leads to aggregation (Hamilton 1971), and aggregation tendencies tend to increase under heightened risk of predation (e.g. Beecham & Fransworth 1999; Banks 2001).
Group assets, such as reproductive shares or safety, must be divided among its members (Hamilton 2000; Buston & Zink 2009). This division is all but straightforward and is in itself a source of conflict when some members of the group aspire for a larger portion of the assets than others (e.g. Hamilton 2000; Buston & Zink 2009). This reproductive conflict counteracts the unrestrained increase in group size predicted by selfish herding (Hamilton 1971) and is especially pronounced in socially breeding animals, where conflicts may arise over the division of reproduction (e.g. Beekman, Komdeur & Ratnieks 2003; Öst et al. 2007a). Group dominants may need to concede some reproductive benefits in return for keeping the group together (Vehrencamp 1979; Emlen 1982a). A dominant individual should hand out as few concessions as possible to enhance own fitness, but if group living is preferable, enough to keep the subordinate(s) in the group (Reeve 2000). These two extremes delimit ‘the window of selfishness’ (sensu Reeve 2000), within which individuals may be as selfish as possible, but exceeding the limits will dissolve the group. We note here that transactional ‘concessions’ models, where dominants have full control over reproduction and concede shares of reproduction to subordinates to keep the group together, and tug-of-war models, involving expensive fights over reproductive shares, may be unified under the same theoretical framework by viewing them as submodels that apply in different ecological settings (Shen & Reeve 2010). This vastly reduces the diversity of, and previous controversy between, different skew models (e.g. Johnstone 2000; Reeve & Shen 2006; Nonacs 2007). Here, we employ a transactional model of skew, which approximately corresponds to the alpha-pay submodel of the mutual-pay, bordered tug-of-war model of Shen & Reeve (2010).
Increasing ecological constraints on solitary reproduction will favour social reproduction over solitary breeding (Reeve 2000). If surviving, obtaining food or reproducing is more difficult for individuals than for groups, we may expect the benefits of grouping to exceed those of solitary living, i.e. the window of selfishness may be expected to widen (Reeve 2000). There are various ecological factors that may constrain solitary reproduction (Emlen 1982b; Hatchwell & Komdeur 2000), such as limited availability of breeding territories (Pasinelli & Walters 2002), predation risk (Mappes, Kaitala & Alatalo 1995), lacking experience (Komdeur 1996) and low bodily reserves (Öst et al. 2003a,b). Interestingly, predation may have a twofold effect on social breeding groups; first, by enhancing selfish herding (Hamilton 1971) and secondly by often placing a stronger constraint on solitarily as opposed to gregariously reproducing individuals (Mappes, Kaitala & Alatalo 1995).
Although grouping behaviour is a thoroughly studied field (Krause & Ruxton 2002 and references therein), some unresolved issues of importance remain. One such largely ignored question pertains to whether exposure to predators prior to the group-formation phase may have carry-over effects on future grouping behaviour. Also, it is yet unclear whether predation risk is expected to modify only the relative frequency of gregarious reproductive modes in the population or whether group sizes in these brood-rearing coalitions are also subject to predation-driven changes.
In this paper, we study the tendency for cooperative action (width of the window of selfishness) among brood-rearing eider females (Somateria mollissima Linnaeus) in relation to predation risk directed at themselves. In this monogamous species with uniparental female care, females may either pool their broods and share brood-rearing duties in coalitions of usually two to five females and their broods or they may care for the young solitarily (e.g. Öst et al. 2003a). The social brood care strategy mitigates the individual costs of caring for females having endured the 1-month incubation fast by allowing more time for feeding and self-maintenance, while the group as a whole benefits from increased collective predator detection (Öst, Mantila & Kilpi 2002). However, predation on ducklings is edge-biased (e.g. Swennen 1989), and as dominant females occupy more central positions (Öst, Jaatinen & Steele 2007b), the reproduction within coalitions is unequally divided. While solitary brood care thus imposes a greater strain on females, the survival prospects of own offspring are not compromised by female competition for the safest locations. The cost/benefit ratio of cooperating versus caring for young alone may change with increased ecological constraints such as increased predation. Indeed, the predator ensemble in the study area has changed over the past two decades, as one of the main predator of adults, the white-tailed eagle (Haliaeetus albicilla Linnaeus) has recently recovered from the effects of environmental toxins and persecution and become more abundant (see Discussion). The abundances of other predators of adults, mainly American mink (Mustela vison Schreber) and eagle owl (Bubo bubo Linnaeus), also fluctuate on a yearly basis.
Female brood-rearing coalitions are formed within a week of hatching of the eggs (Öst et al. 2003a). During this first week, females engage actively in social interactions and aggressive encounters with others, perhaps to evaluate the quality of prospective coalition partners. In our study population, coalitions are only formed by successful breeders and are not joined by failed nesters (Öst et al. 2003b). Once stabilized, brood-rearing coalitions usually persist for the full brood-rearing period (c. 40 days) (Öst et al. 2003a). Females are plastic in their decision of entering a coalition and commonly switch brood care strategies between years (Kilpi et al. 2001). This plastic, facultatively social brood care system, which is not based on kinship between coalition partners (Öst et al. 2005), is therefore well suited for studies on decisions regarding social versus solitary behaviour.
Two recent models portraying the coalition-forming process describe the effect of female quality asymmetry, assumed to stem mainly from differences in body condition, on parental efforts and reproductive skew within the coalition (Öst et al. 2003a, 2007a). This modelling shows that females of relatively good body condition are less likely to form coalitions because they require a disproportionally large reproductive share (Öst et al. 2003a). A high-condition female and a low-condition female, on the other hand, are more likely to form a coalition, because the lower-condition subordinate is willing to accept a lower share of reproduction as she will likely achieve even less reproductive output on her own. The dominant female benefits from the presence of the subordinate and her young through shared vigilance and diluted predation risk, and she should therefore give the subordinate the share in reproduction that keeps the coalition together (Öst et al. 2003a, 2007a).
Our first aim is to explore the relative distribution of social and solitary brood care strategies under variable predation pressure. This aim is achieved by modifying the bidding game over care developed by Öst et al. (2007a) to generate predictions about the effects of predation risk on the width of the window of selfishness between prospective coalition partners. Second, we utilize the dilution effect function implicit in the model, to predict changes in group size as a response to changes in predation risk. To test our modelling predictions, we utilize an extensive 13-year data set on eiders, which allows us to explore the effects of predation on grouping decisions, while controlling for the effect of female body condition.
Materials and methods
The model is derived directly from the parental effort game of Öst et al. (2007a), allowing the total reproductive output to be divided between two partners (indexed by the subscript i), so that a proportion s goes to partner 1 and 1−s to partner 2. As in the model of Öst et al. (2007a), each female pays a cost Ki(ui), related to the effort, ui, she devotes to brood care. The cost is dependent on the condition, ki, of a female and decreases with increasing condition of the female. The overall brood-rearing capability of coalitions is assumed to exceed that of lone-tending females, based on dilution of predation risk, selfish herding and increased collective vigilance (Öst et al. 2003a, and reference therein). We compress all of this biology into a dilution parameter d,
where m is the mortality rate caused by predation and c a parameter describing the contribution of individuals to the joint brood. This parameter is held at 0·5, which for simplicity assumes that both partners contribute the same amount of individuals to the coalition (note that although both partners contribute equally to the dilution parameter, reproductive output is shared according to the skew parameter and need not be equal for both partners).
The effort individuals allocate depends on the effort of the partner. We calculate the effort of each partner using the stable effort function presented in Öst et al. (2007a), where the effort of partner 1 is
and the effort of partner 2 is
The bidding process assumes that each partner (i) submits a bid in advance, (ii) has knowledge of the other’s bid and (iii) adheres to the bid made. For more discussion on the bidding process, please consult Öst et al. (2007a) and McNamara et al. (2003).
The benefit resulting from allocated effort is
the payoffs of which diminish with increasing effort. The fitness of individuals in coalitions can then be calculated as
For lone-tending females, the fitness maximizing effort (sensu Öst et al. 2007a) is
and the fitness of lone tenders is calculated as
Coalitions form when it is advantageous for both partners (Öst et al. 2003a), which is true when both the inequalities F1 < C1 and F2 < C2 are fulfilled.
To derive predictions about the effect of predation mortality on the brood-rearing strategies of eider females, we vary predation mortality, m, from 0·01 to 0·99 by increments of 0·01, and for each increment of m we vary skew, s, from 0 to 1 with increments of 0·001. This allows us to record the levels of skew that are acceptable for both partners under different levels of predation mortality. The range of acceptable skew values is termed the window of selfishness, which is recorded for each level of predation mortality.
The partner effort model only applies to two individuals, yet eiders often tend their young in groups larger than two. Although the bidding model is best presented for two females for the sake of clarity, the dilution function presented in the model has the property of increasing the benefit of attending a group with a decreasing contribution (c) to the joint brood. Thus, the smaller proportion of the total group an individual (or its brood) represents, the larger the dilution effect is, and therefore the prospects of surviving are higher. To derive predictions about the effects of predation on female group size distributions in brood-rearing eiders, we use the following equation to calculate the magnitude of the dilution effect for all female group sizes between 1 and 7 [the range of stable group sizes in our population (Öst, Smith & Kilpi 2008a)]:
where m is the mortality rate and g is the group size. D thus represents the dilution effect as experienced by a female entering a group, where her contribution to the group is 1/g (the multiplayer equivalent to parameter c in eqn 1). We vary m from 0 to 1 by increments of 0·1.
The field data used to test our predictions were gathered at Tvärminne (59°50′N, 23°15′E), western Gulf of Finland, during the years 1997–2009. Female eiders were captured during nesting by using hand nets. Upon capture, the females were weighed, measured for structural size (length of the radius-ulna), ringed with a standard metal ring as well as colour ringed for individual recognition at distance. Females were also equipped with a wing flag to ease recognition during brood care. The hatch date was estimated using an egg flotation test (Kilpi & Lindström 1997) or observed upon visiting the nest to capture the ducklings for other purposes than this study. Hatch dates estimated using egg floatation have been found to give 0·32 days later estimates of hatching than observed hatch dates, which is a non-significant difference (Kilpi & Lindström 1997). The female will escort the ducklings to sea within 24 h of hatching. During the study, we observed a total of 4417 breeding attempts (also including destroyed or abandoned nests) involving 2116 captures of 1238 individual females.
The yearly distributions of parental care strategies were obtained by observing the brood care behaviour of colour-ringed individuals throughout the brood-rearing period. The parental care strategy of females is in most cases easily determined as they use the same strategy all season. For individuals who switched strategy during the season (also see Discussion), we defined the strategy used as the one that the female used for the majority of the season. Females observed only once during the season were excluded from the data set unless the observation was made more than five days after hatching of the focal female’s brood. This will minimize bias in strategy distributions resulting from the first few days’ active social interactions during which females join and leave groups frequently (Öst et al. 2003a).
To address the question regarding the effect of female mortality on the selection of social vs. solitary parental care strategies and group size, we needed an estimate of female mortality. To obtain a measure of this, we estimated female nesting mortality as the number of killed females found per censused nest (Lehikoinen et al. 2008). This measure combines all forms of predation and the main predators, as deduced from the way the females had been killed and devoured, were American mink [23% of all kills (n = 26)], white-tailed sea eagle [15% of all kills (n = 17)] and eagle owl [6% of all kills (n = 7)]. In a total of 61 cases (55%), the predator remained unknown because of non-typical signs of attack or because of almost complete devouring of the carcass, leaving too few clues by which to determine the killer. As an independent check that our female nesting mortality variable accurately portrayed the actual predation risk experienced by nesting females, we used data on white-tailed sea eagle abundance from the Hanko Bird Observatory situated next to the Tvärminne archipelago (59°49′N, 22°54′E) (see Discussion). We calculated annual eagle abundance indices by dividing the sum total of daily numbers of resident white-tailed sea eagles observed during 1 April–15 June in 1997–2009 with the number of annual observation days during the same period.
As a measure of group size, we used ‘typical group size’ (sensu Jarman 1974), which better than the average group size describes the group size distribution as experienced by the group members (Reiczigel et al. 2008). Eider coalitions do vary slightly in size over the season because of females that are transient in broods, mainly in the social interaction period spanning the first 2 weeks after hatch (Öst et al. 2003a). In the first few days after ducklings have hatched, broods using overlapping feeding areas may also temporarily merge. However, the females truly participating in the coalition can, based on our detailed observations typically lasting up to several hours per brood, almost invariably be indentified based on their brood care behaviour and social interactions (Öst et al. 2003b). We used the mode number of females present in coalitions over the entire brood-rearing season as the size of individual groups. The following formula was implemented on these data to obtain yearly typical group size indices:
where g is the actual group size and G the typical group size. Only brood-rearing coalitions with at least two females were included, so as to separate the analysis of typical group size from our analysis of the relative frequency of cooperative brood care, which included the lone tenders.
Because the brood-rearing strategies of eiders are influenced by body condition (e.g. Öst et al. 2003a), we investigated potential confounding trends in body condition over the study period and its possible collinearity with our measures of predation risk. This is important because high collinearity between explanatory variables may cause the parameter estimates to be extremely imprecise and unstable. We calculated global condition indices for all trapped females (n = 1950 observations of 1156 females) for the entire study period, provided that they had incubated eggs for >8 days (egg laying may otherwise still be in progress; Öst et al. 2008b). As a global condition index, we used the standardized residuals of a regression of log-transformed projected weight at hatching (response variable) on log-transformed radius-ulna length; indices were derived for the pooled data from 1997 to 2009. A female’s weight at hatching was estimated by subtracting an estimate of the weight she would be expected to lose during the remaining incubation time from her measured incubation weight. Each female was weighed once, but because females do not feed during incubation and we captured females at different times in their incubation, we can estimate average weight loss rate during incubation as the slope of the regression of log(body weight) (response variable) on log(incubation time) and projected hatching date (Öst et al. 2008b). The average weight loss per day was 19·05 g (±0·11 SE) (n = 1870; excluding 80 females with a hatched brood that had completed incubation), with annual means ranging from 8·0 to 24·0 g. The inter-relationship between global body condition indices (response variable) and time (explanatory variable) were analysed by GLMMs to account for pseudoreplication arising from repeated observations of the same females in different years (random term: female identity). A preliminary analysis revealed that body condition was strongly correlated with female nesting mortality (rp = −0·62, P = 0·02), and so body condition could not be included in any of the models described below because of its collinearity with female nesting mortality. Instead, the relative importance of body condition and female nesting mortality on our dependent variables (proportions of social and solitary brood care and typical group size) was assessed based on the performance of single-variable general linear models (GLMs) as judged by the Akaike’s information criterion (AIC) corrected for small sample size (AICC). Models <2 units from the best model with the lowest AICC value (ΔAICC < 2) were considered equally well supported by the data (Burnham & Anderson 2002).
We used three GLM models to explore the effect of female nesting mortality on (i) the relative frequencies of solitary and (ii) social brood care tactics and (iii) typical group size. The residuals of all of these models conformed to the assumption of normality.
Model results and predictions
We find that below a certain level of predation, there is no benefit of forming coalitions and that the window of selfishness widens as predation increases (Fig. 1). Thus, based on the bidding game between two partners, we predict that (i) the widening range of mutually acceptable levels of reproductive skew under elevated predation risk will facilitate coalition forming, and hence coalition forming will be more common under higher predation risk, and inversely, lone tending will decrease in prevalence under elevated predation risk. Finally, the magnitude of the dilution effect, D, increases with group size and predation mortality (Fig. 2). We therefore predict that not only does the prevalence of coalition formation increase with elevated predation risk but (ii) group sizes are expected to increase concurrently.
Testing the predictions
Female predation during incubation tended to increase slightly over the 13-year study period (GLM: F1,11 = 4·56, R2 = 0·29, b = 0·0022, P = 0·056), as did the coalition-forming strategy (GLM: F1,11 = 25·61, R2 = 0·70, b = 0·029, P < 0·001). Our model explaining the proportion of females attending brood-rearing coalitions was significant and explained a large proportion of the variation in this variable (GLM: F1,11 = 12·55, R2 = 0·53, P = 0·005). The coalition-forming strategy increased significantly with increasing female mortality (b = 6·14, t = 3·54, P = 0·005; Fig. 3). This result indicates that an increase in female nesting mortality of 0·01 (one additional dead female found per 100 nests censused) increases the proportion of the social brood care strategy by ca. 0·06. Likewise, the model explaining the relative frequency of lone tenders was significant, explaining a large proportion of the variation (F1,11 = 13·70, R2 = 0·55, P = 0·003). Lone tending was less prevalent with increasing female mortality (b = −2·38, t = −3·70, P = 0·003; Fig. 3). The proportion of females caring for their brood alone thus decreases by ca. 0·02 when nesting mortality increases by 0·01.
The model explaining typical group sizes was significant (GLM: F1,11 = 6·86, R2 = 0·38, P = 0·02), showing that typical group sizes were larger with increasing female mortality during incubation (b = 15·71, t = 2·62, P = 0·02; Fig. 4). Typical group size will thus increase by ca 0·16 females for each increase of 0·01 in nesting mortality.
There was no temporal trend in body condition during the 13-year study period (GLMM: n = 1950 observations of 1156 females, b (year) = −0·00072 (±0·0062 SE), t = −0·12, P = 0·91). The single-variable models for the proportion of lone tenders and coalition-forming females based on body condition fared poorly when compared to the corresponding models based on female nesting mortality (ΔAICC for lone tending and coalition forming 9·59 and 7·08, respectively). This suggests that the changes in the relative proportions of solitary and social brood care strategies are primarily driven by changes in female predation rather than changes in female body condition. However, single-variable models for typical group size based on body condition (AICc = 13·20) and female nesting mortality (AICC = 13·54; ΔAICC = 0·34) were equally well supported. This means that females attending larger brood-rearing groups were in poorer body condition (b = −0·93, t = −2·71, P = 0·02).
The field data supported our predictions, so that elevated predation risk on adult females during the incubation phase resulted not only in a higher frequency of brood-rearing coalitions (Fig. 3), but also the typical group sizes of females attending these coalitions increased (Fig. 4). The frequencies of lone tenders showed a corresponding decrease with increased female nesting mortality. Although all brood-rearing eider females have successfully hatched a brood, there are clear parallels between our results and those obtained in species conventionally recognized as cooperative breeders. Thus, our finding of a decrease in lone tending with increasing predation risk corresponds with the finding of helpers more frequently postponing independent reproduction when constraints on solitary reproduction are high (e.g. Ekman et al. 2004). It is noteworthy that the long-term changes in parental care strategies occurred concurrent with a decreasing trend in the size of our study population (Lehikoinen et al. 2008). This fact clearly refutes explanations based on non-adaptive, accidental brood mixing (e.g. Savard 1987), according to which cooperative brood-rearing tactics should be more common with increasing numbers of prospective cooperation partners, i.e. when the densities of broods and nesting parents are higher (e.g. Gosser & Conover 2000). Body condition was not a force driving the changes in the frequencies of lone-tending and coalition-forming tactics, indicating that the changes in strategy distributions can likely be accounted to changes in the reigning predation regime. Poor body condition was, however, linked with an increase in female typical group sizes.
We have clearly shown that predation risk directed at females during the incubation phase has far-reaching implications for grouping decisions some days or even weeks later. In the light of current empirical evidence, this type of ‘carry-over effect of danger’ seems to be rare with respect to causing changes in parental care strategies. So far, most studies that have investigated long-term effects of predation risk have found changes in subsequent clutch size (e.g. Doligez & Clobert 2003; Eggers et al. 2006), nest-site selection (e.g. Forstmeister & Weiss 2004; Eggers et al. 2006), or they have used whole colonies rather than individuals as their prime study target (e.g. Kenyon, Smith & Butler 2007). It is conceivable to envision several proximate mechanisms whereby perceived predation risk may spread in colonially nesting birds such as eiders. If females can observe conspecifics being attacked or killed within sight, this can have long-lasting effects, e.g., through elevated levels of stress hormones (e.g. Cockrem & Silverin 2002), thereby affecting grouping decisions after the brood has left the nest. We used actual female nesting mortality as a proxy for predation risk, but based on our experience with this colonial duck species, predation events seldom occur in isolation, but are likely to be observed by several other breeders nearby. It is also likely that actual predation and predation attempts are positively correlated, although we were unable to collect data on predation attempts.
The egg-laying and incubation phase is by far the most dangerous part of the life cycle of most ground-breeding birds (e.g. Caro 2005). Changes in perceived predation risk experienced by nesting individuals may thereby also affect subsequent parental care decisions. While our knowledge of the recent population fluctuations of American minks and eagle owls in our study area remains limited, the presence of one of the main predators of female eiders, the white-tailed sea eagle, has increased rapidly over the study period (Fig. 5). This trend implies that the changes in actual predation measured by us go hand in hand with changes in perceived predation risk, making the landscape more risky for breeding eiders. The day-active white-tailed sea eagle is a large and conspicuous predator, and hence the indirect effects of perceived predation risk on eider parental care and grouping decisions are likely substantial, despite the relatively low incidence of actual eagle predation itself.
Selection of brood-rearing strategies in eiders has so far been considered to be almost exclusively determined by body condition (Bustnes & Erikstad 1991; Kilpi et al. 2001; Öst et al. 2003a, 2007a). Our results partly support the importance of body condition, but also point to the unfortunate neglect of predation risk in determining brood care strategies. As the effects of body condition and predation risk both seem to be important for determining brood care strategies, their effects should be considered simultaneously. As an illustrative example, we found that group sizes increased when predation on females is more severe. However, this being the case, the females found in the largest groups are still the ones that are, on average, in poorest body condition (Öst et al. 2003b; and this study). The strong correlation between female predation during nesting and body condition is interesting in itself and would merit further research efforts. Circumstantial evidence from our study population suggests that females in poor body condition may have a shorter life expectancy (Öst & Steele 2010); however, this correlation still awaits justification by formal survival analysis and the causality underlying this correlation remains unknown.
We regularly observe attempts of group entry and subsequent rejection by group members (Öst et al. 2003b). The formation of a brood-rearing coalition is the result of a negotiation process, described as a bidding game by Öst et al. (2007a). This negotiation process adheres to the transactional model framework (Reeve, Emlen & Keller 1998), where females negotiating over coalition forming exchange benefits. The group members may deter entry of extra females when the group becomes saturated, so that an additional member does not contribute enough benefits for admittance into the group. Both our theoretical and empirical results suggest that increased predation-induced constraints on solitary reproduction have led to a widening of the window of selfishness between prospective coalition partners, allowing for a wider range of mutually acceptable brood compositions, with a concomitant increase in social brood-rearing tactics. The same process, leading females to accept smaller shares of the group’s reproductive output, also allows larger groups to form.
According to Öst, Smith & Kilpi (2008a), two to three females is the female group size that maximizes duckling survival in years with average duckling survival. Interestingly, a substantial proportion of the current brood-rearing coalitions exceed this number of tending females, and provided that the current increasing trend in group sizes continues, this may have important consequences for group productivity. Specifically, it could be an example of the tragedy of the commons (Hardin 1968; Rankin, Bargum & Kokko 2007), where individual-level selection may lead to suboptimal group sizes that reduce per capita reproductive output and, hence, lower population-wide productivity. Such population-level consequences may be of concern in the already declining study population.
Perhaps, the most important remaining gap in our knowledge relates to the individual fitness consequences of a predation-driven increase in sociality. We can see several equally plausible outcomes. First, the fitness of individual females may have decreased because of increasing constraints on solitary brood care. Alternatively, however, the per capita fitness of females may have stayed unchanged because of a complete compensation of the increased constraints by means of increased cooperation. Knowledge of individual lifetime reproductive success is needed to resolve the inter-relationships between cooperation, ecological constraints on solitary breeding and reproductive success. Another, perhaps less elusive goal of future research is to clarify the decision-making process involved in the switching between parental care modes, which is relatively frequent in the early social interaction period (Öst et al. 2003a). Focusing on the factors affecting these tactic switch points, particularly brood success (Pöysä, Virtanen & Milonoff 1997), body condition and their interactions, is likely to be especially rewarding.
We thank numerous field workers for their efforts. The observers at Hanko Bird observatory are acknowledged for their tenacious data gathering. We also thank two anonymous reviewers for detailed comments that improved the manuscript. Tvärminne Zoological Station provided facilities and the Academy of Finland (grant no. 128039 to K.J. and M.Ö.) and Maj and Tor Nessling Foundation (A.L.) funding.