Social selection is density dependent but makes little contribution to total selection in New Zealand giraffe weevils

Abstract

Social selection occurs when traits of interaction partners influence an individual's fitness and can alter total selection strength. However, we have little idea of what factors influence social selection's strength. Further, social selection only contributes to overall selection when there is phenotypic assortment, but simultaneous estimates of social selection and phenotypic assortment are rare. Here, we estimated social selection on body size in a wild population of New Zealand giraffe weevils (Lasiorhynchus barbicornis). We measured phenotypic assortment by body size and tested whether social selection varied with sex ratio, density and interacted with the body size of the focal individual. Social selection was limited and unaffected by sex ratio or the size of the focal individual. However, at high densities social selection was negative for both sexes, consistent with size-based competitive interactions for access to mates. Phenotypic assortment was always close to zero, indicating negative social selection at high densities will not impede the evolution of larger body sizes. Despite its predicted importance, social selection may only influence evolutionary change in specific contexts, leaving direct selection to drive evolutionary change.

1. Introduction

Selection is an important concept in evolutionary biology, describing the link between traits and fitness. Typically, selection is characterized as a selection gradient or differential representing the relationship between the trait of a focal individual (e.g. body size) and a measure of fitness (e.g. its lifetime number of adult offspring; [1]). This ‘direct’ selection helps us understand the functional value of traits and predict how they might evolve. Further, direct selection is known to vary across space [2], time [3] and with ecological conditions [4,5], helping to generate the biodiversity of the natural world. Other forms of selection are possible, however. For instance, when organisms interact with others, such as by competing for access to resources or cooperating to raise young, they can influence each other's fitness. The link between a partner's traits or the traits of group mates and a focal individual's fitness is known as ‘social’ selection [6]. Estimating direct and social selection gradients is formally equivalent to estimating within- and among-group selection using contextual analysis, giving deep links between social selection theory and multilevel selection theory [6–8].

Social selection may not align with direct selection (Table 3 of [9]), which can alter the direction and tempo of trait evolution [8], possibly leading to maladaptation [10]. For instance, ‘selfish’ traits may increase the fitness of an individual that bears them but may be costly when displayed by those they interact with. Conversely, ‘altruistic’ traits may be costly for the individual that displays them but may be beneficial when expressed by other individuals that interact with them. Social selection is one of the ways social interactions can alter evolutionary change, alongside individuals having heritable effects on each other's phenotypes (indirect genetic effects [11]), and individuals choosing who they interact with through social niche construction and therefore manipulating their environment and the selection pressures they are under ([12], see [13] for a review). Given how fundamentally social interactions can alter evolutionary change, it is, therefore, vital we estimate parameters such as social selection in natural populations in a range of contexts [6,14,15].

Social selection alone cannot alter evolution, however. For social selection to contribute to total selection, and therefore evolutionary change, there must be covariation between the traits of an individual and the traits of those it interacts with (non-zero phenotypic assortment; [6]). The overall selection differential is calculated as (following [16], see also [6]):

$$({\beta }_D+\hspace{0.17em}{\beta }_S{\beta }_{z^{\mathrm{\prime }}z})\mathrm{var}(z).$$

Where β_D is the direct selection gradient, β_S is the social selection gradient, ${\beta }_{z^{\mathrm{\prime }}z}$ is the coefficient from the regression of the mean partner phenotype on the focal individual's phenotype, also known as the assortment coefficient, and var(z) is the phenotypic variance across focal individuals. Positive assortment indicates that individuals with similar traits interact e.g. aggressive individuals interact with other aggressive individuals. Negative assortment on the other hand indicates that individuals with dissimilar traits interact, e.g. resource-producing individuals interact with resource-consuming individuals. Assortment has been documented in groups of animals, and typically found to be positive (in male great tits, Parus major, [17], Chacma baboons, Papio ursinus, [18]; guppies, Poecilia reticulata, [19]). However, not all measures of assortment are equal, and only the interactant covariance (the covariance between an individual's traits and the mean trait value of those it interacts with) is correct for use in models of total selection [16]. Unfortunately, estimates of this parameter in natural populations are rare, especially alongside estimates of social selection (but see [20,21]). Therefore, despite its predicted importance, we have very little knowledge of how social selection contributes to total selection in natural populations.

We also have little data on the contexts where social selection is strongest. A few studies to date indicate that social selection and within- or among-group selection may vary with sex [20,22], local density [21–23] or season [20], but we lack substantial evidence to make any strong conclusions. Direct selection is known to vary based on demographic parameters such as population density [24] and sex ratio [25], as well as depending on other individual characteristics (correlational selection; [26]). A lack of knowledge of the conditions where social selection is strongest hampers our ability to predict how it may shape different populations, and therefore generate diversity. We aimed to address this gap in a wild population of New Zealand giraffe weevils (Lasiorhynchus barbicornis; Coleoptera: Brentidae). Both sexes are extremely variable in size [27–29], males bear an elongated rostrum used as a weapon during contests for mates [30], and body length is under positive linear selection in males and females (using a number of mates as the fitness component; [31]). As giraffe weevils form aggregations on trees and compete for access to mates, we predicted that social selection for body size would be negative, where the presence of larger rivals reduces a focal individual's fitness (following [21]). We included both sexes in all analyses as we wished to test whether females experienced social selection, as there is often the implicit assumption that females do not compete for access to mates, but whether females do in fact suffer a cost from competing with large rivals is rarely tested (but see [20]). We predicted that females would experience weaker social selection than males as females do not actively guard their mates, unlike males, and so large females should not be able to directly prevent others from mating.

We had a series of predictions for factors that could influence the strength of social selection. As individuals mate more often but for shorter periods of time at higher densities (C. Painting 2012, unpublished data), we predicted that social selection would be more negative at high densities and when the individual was of the more common sex (i.e. a male in a male-biased population), as these are conditions when they might be competing most fiercely for mates. We also predicted that smaller males would be less affected by the body size of rivals, as they can readily switch between fighting with similar-sized rivals and ‘sneaker’ tactics that allow them to gain copulations without direct competition [32]. Finally, following previous work which found positive assortment in various taxa (see above), we predicted that there would be positive assortment for body size in the individuals present on trees in both sexes, causing social selection to reduce the overall strength of selection on body size.

2. Methods

(a) . Data collection

The giraffe weevil population we studied resides in Matuku Reserve (36°51.92′ S, 174°28.32′ E), a coastal broadleaf forest west of Auckland, New Zealand. We located aggregations of adult giraffe weevils on karaka trees (Corynocarpus laevigatus), which were subsequently used for behavioural observations. The observations and data collection used in the current study are described in full in a previous study [31,33] but we briefly outline them again here.

To determine variation in mating success among males and females of different sizes we conducted daily observations for 1 h at three different trees (visited in a random order) that housed giraffe weevil aggregations. Aggregations typically covered only around 1 m² of the tree trunk, making the complete observation of all individuals in the aggregation relatively straightforward. Observations took place over two periods between 22 November and 22 December 2013 (31 days, N = 120 females, 132 males), and 22 January to 23 February 2014 (33 days, N = 301 females, 366 males). For the analysis, we excluded individuals only seen once, and those who were first seen in the last week of each observation period (following [31]). This left a dataset of 1234 records of 155 different females and 236 different males. At least 2 h prior to observations each day, we removed all previously unrecorded giraffe weevils from the tree for measurement and marking. We measured total body length using digital callipers to the nearest 0.01 mm; body length can vary between 15–90 mm in males and 13–50 mm in females [34]. We also measured rostrum length (the weapon used by males during combat), but this and all morphological traits are highly correlated with body length [28]. We used body length in our selection analyses, as body length includes the length of the rostrum and is likely under fecundity selection in females. We then painted individuals on the pronotum and elytra with a unique colour combination using queen bee marking paints (Lega, Italy) for identification before being released to the point of capture on the tree [35]. We observed all giraffe weevils present on each of the three trees for 1 h on each day of the observation period between 08.00 and 18.00 h. We stood at least one metre from the tree and used close-range binoculars (Pentax, Papilio) to avoid disturbing the weevils. During each observation, we recorded the identity of all giraffe weevils present on the tree that day as well as the occurrence of all matings using scan sampling with continuous recording. After observations, we thoroughly searched the tree to check for any individuals that had been inactive or hiding in cracks or under leaves, and we gave these individuals a mating frequency of zero. Within the hour it is very unlikely that individuals were able to copulate in these hiding places and be missed by us. These individuals were included in the calculation of the social environment (see below) as they could have interacted with the other weevils in the aggregation but chose not to. We conducted no observations on days of heavy rainfall because giraffe weevils are inactive during this time, resulting in two non-consecutive days being missed in the first observation period, and three non-consecutive days during the second.

(b) . Data analysis

To assess the strength of social selection, we fit a series of generalized linear mixed-effect models using the R package ‘glmmTMB’ [36]. For all models, we mean centred each continuous predictor variable and divided by its standard deviation to improve model fit and interpretability (see [37]). For quadratic terms we first mean centred and scaled the variable, then squared it and then divided by two (see [38]). Each model had the number of different individuals a focal weevil copulated with on that day (the component of absolute fitness) as the response variable, time of day as both linear and quadratic terms as fixed effects, tree identity, weevil identity and an Ornstein–Uhlenbeck temporal autocorrelation term (as dates were not evenly spaced), modelling both variation among dates and the similarity of measurements made closer in time on the same individual, as random effects, and a Poisson error distribution with a log-link. This approach gives fixed effect coefficients that can be readily transformed into selection gradients (see [38,39]) and means we do not scale fitness by the mean of the group, instead effectively looking at the strength of selection at the population level [40], matching the scale at which we calculate assortment (see below). We use number of different mating partners as a component of fitness in females (i.e. the rate of polyandry) as this was used by LeGrice et al. [31] when estimating selection in both males and females in this species (note that number of mating partners and both survival and recapture probability increase with body size). While laboratory studies suggest female fitness plateaus at higher mating rates [41], in studies on wild insects the lifetime rate of polyandry can predict the number of offspring produced that survive to adulthood (e.g. the field cricket Gryllus campestris; [42]). As such we think this is an appropriate measure of a component of fitness.

To estimate social selection, we must define which other weevils an individual is interacting with and calculate their mean body size, specifically excluding the focal individual from the calculation of the mean [7,9,16]. Estimating social selection is related to ‘contextual analysis’, where the within-group and among-group selection gradients are estimated [7,43], and one can transform the estimates of direct and social selection into estimates of within- and among-group selection (see [8], although note that phenotypic assortment is not required for among-group selection to contribute to overall selection). We used social selection rather than contextual analysis as we felt the parameters of direct and social selection were more intuitive when applied to the biology of our study species. We defined an individual's interaction partners as all same sex-individuals found in the same aggregation as a focal weevil on the same day, including those that achieved no mating success and were hiding in cracks (hereafter ‘rival mean body size’). The rival mean body size accurately represents the social environment as the aggregations we observed covered only a small area, meaning that all weevils, even those in cracks, could easily interact with others in the aggregation if they chose to. Further, daily survival probabilities based on mark–recapture data are high (around 0.9–0.95; [31]) and we observed only 27 out of 391 individuals at more than one tree, suggesting high site fidelity (although there is still variation within a tree in density and mean size of rivals across days, see electronic supplementary material, figures S1 and S2).

To estimate direct and social selection, we fit individual body size and the rival mean body size as predictors (model 1). We also included quadratic versions of both direct and social selection to determine whether either form of selection was nonlinear (terms were mean centred and scaled to unit variance, then squared, then halved). We included sex as a fixed effect, and the interactions between sex and both focal and rival body size for both linear and quadratic terms to test whether males and females experienced different selection, although we removed these interactions to calculate linear and quadratic selection gradients. Males were set as the default sex and so the interaction was modelled as the difference between females and males. We evaluated the ‘clarity’ (replacing the word ‘significant’ with ‘clarity’ avoids implying that finding a lack of a difference from zero is not important and emphasizes that effects not different from zero can still be positive or negative, but not ‘clearly’ so; see [44]) of the effect of all fixed effects using Wald χ²-tests with type III sum of squares using the Anova function within the R package ‘cars’ [45]. The degrees of freedom were 1 for all tests unless stated otherwise.

To determine under which conditions social selection is strongest we then fitted a series of models. We used the same starting model as above, excluding quadratic terms as they had no clear effect (see Results). For model 2, we included an interaction between focal body size and rival mean body size to determine if smaller individuals experienced weaker social selection than larger individuals. Such interactive effects can be important in determining how social interactions contribute to evolution [46]. We also included the three-way interaction between sex, focal body size and rival mean body size, to see if males and females differed in the focal–rival relationship. As males of only smaller sizes (typically under 40 mm, see [32]) may engage in ‘sneaky’ copulations, we also fitted a model (model 3) where sex was a three-level categorical variable, either ‘female’, ‘male over 40 mm’, or ‘male 40 mm or under’, including the interactions between this new variable and both focal and rival mean body size. We then fitted two models to test whether demographic parameters influenced the social selection. Model 4 included weevil density (number of weevils observed on the tree on that day; median = 16, 25% quantile = 10, 75% quantile = 25) as a fixed effect and its interactions with both focal and rival mean body size, including the three-way interactions between density, sex, and either focal or rival mean body size. Model 5 was equivalent to the density model but included sex ratio (proportion of weevils on the tree on that day that were male; median = 0.64, 25% quantile = 0.60, 75% quantile = 0.71) instead of density. In models 4 and 5, the key terms are the interactions between density/sex ratio and the rival mean body size, as these terms indicate whether the impact of rival body size on focal individual fitness (i.e. the strength of social selection) increases or decreases with density/sex ratio (while the interaction between this term and sex indicates whether this effect differs between the sexes or not).

To estimate the overall phenotypic assortment within each sex we calculated the Pearson correlation between the body size of a focal individual and the rival mean body size, where the variables had been mean centred and divided by their standard deviation [16]. Note that small group sizes can lead to negative phenotypic assortment by chance [16], but our mean group size was 21 which is above the threshold where we expect negative assortment to occur by chance. Following our detection of density-dependent social selection (see Results), we then decided to test whether phenotypic assortment was density dependent (as has been found before, see [21], although note caveats above about negative assortment ‘appearing’ at low group sizes). We stress that we did this analysis after viewing our initial results and so should be interpreted appropriately. To do this we fitted a linear model with the rival mean body size as the response variable, the body size of a focal individual, the density of weevils on the tree, the focal individual's sex and all two- and three-way interactions between these variables as fixed effects. The random effect structure was the same as for the selection models, including the temporal autocorrelation term. The response variable and all continuous predictor variables were mean centred and divided by their own standard deviation. The key term here is the interaction between density and the body size of the focal individual, as this indicates whether the relationship between the focal individual and the rival mean body size changes with density (while the interaction between this term and sex indicates whether this effect differs between the sexes or not).

3. Results

Key model results are summarized in table 1, while full results from each model are reported in the supplementary materials (electronic supplementary material, tables S1–S5). There was no linear or quadratic social selection in either sex when not taking into account interactions with body size, density or sex ratio. We re-fitted a model without interactions between selection gradients and sex to estimate the direct (note not ‘directional’) linear and quadratic selection gradients as: β_D = 0.277 ± 0.181, γ_D = −0.005 ± 0.254 and the social linear and quadratic selection gradients as: β_S = −0.085, γ_S = −0.128 (standard errors for social selection gradients could not be calculated).

Table 1.

Key model results from models 1 to 5. Given are the effect estimate, associated standard error, the χ²-value (with a note if the degrees of freedom were other than 1) and the p-value. For the difference between large males and both females and small males a single χ² statistic and p-value are given for the clarity of the three-level categorical variable (sex type factor), we used to test this effect. We also note which model the term stems from, see Data analysis, and the variance attributed to each random effect in model 1. Full results from each model are given in electronic supplementary material, tables S1–S5.

	estimate	standard error	χ2	p-value	model
intercept	−0.287	0.211	1.857	0.173	1
sex (female contrast)	0.470	0.167	7.888	0.005	1
direct selection—linear	0.239	0.205	1.352	0.245	1
social selection—linear	0.105	0.277	0.144	0.704	1
direct selection—quadratic	−0.301	0.353	0.724	0.395	1
social selection—quadratic	−0.126	0.260	0.233	0.630	1
time of day—linear	−0.105	0.044	5.719	0.017	1
time of day—quadratic	−0.005	0.086	0.003	0.953	1
sex: direct selection—linear	−0.026	0.207	0.016	0.900	1
sex: social selection—linear	0.029	0.281	0.011	0.917	1
sex: direct selection—quadratic	−0.219	0.353	0.387	0.534	1
sex: social selection—quadratic	0.266	0.255	1.083	0.298	1
focal body length—social selection	−0.089	0.181	0.242	0.623	2
sex: focal body length—social selection	0.010	0.183	0.003	0.956	2
female: direct selection	0.253	0.122	sex type factor	(d.f. = 2)	3
small male: direct selection	−0.083	0.101	4.337	0.114	3
female: social selection	0.116	0.125	sex type factor	(d.f. = 2)	3
small male: social selection	−0.143	0.112	1.663	0.435	3
density: direct selection	0.017	0.078	0.050	0.822	4
density: social selection	−0.441	0.161	7.526	0.006	4
sex: density—direct selection	−0.063	0.078	0.659	0.417	4
sex: density—social selection	−0.150	0.171	0.773	0.379	4
sex ratio: direct selection	0.033	0.089	0.139	0.709	5
sex ratio: social selection	−0.059	0.081	0.540	0.462	5
sex: sex ratio—direct selection	−0.042	0.089	0.221	0.639	5
sex: sex ratio—social selection	−0.062	0.081	0.587	0.443	5

			random effect	variance
			weevil	0.000	1
			tree	0.101	1
			date	0.181	1

The strength of social selection did not depend on the size of the focal individual for either sex, and it was not different among size classes of male. Direct selection also did not differ between large males and small males. When controlling for density, social selection was clearly negative for both sexes, and became stronger at higher densities (figure 1). Direct selection was clearly positive for both sexes and stronger for females in model 4, but not dependent on density for either sex. Sex ratio did not influence social selection or direct selection in either sex, while females had higher mating success at more male-biased sex ratios. Mating success tended to decline linearly throughout the day, with no quadratic effect. In all models, there was essentially no variance among weevils beyond the temporal autocorrelation term which indicated there were positive correlations between measurements on the same individual on close dates (electronic supplementary material, figure S3). There was some variance among trees and among dates.

Figure 1.

There was negative social selection when controlling for density, and the strength of social selection tended to become more negative at high densities (more than 40 weevils per tree, light blue) compared to low densities (40 or fewer weevils per tree, black). This was true for both females (left panel) and males (right panel). Note that we analysed density as a continuous variable, but we have used a categorical representation when plotting for ease of viewing. The grey areas around the trend lines indicate the 95% confidence intervals. (Online version in colour.)

The phenotypic assortment overall was near zero for both females (r_females = 0.066, t = 1.336, d.f. = 406, p = 0.182) and males (r_males = 0.053, t = 1.521, d.f. = 824, p = 0.129). Our subsequent test of whether phenotypic assortment was density dependent indicated that it was not for either sex (focal body size = −0.015, s.e. = 0.036, χ² = 0.173, p = 0.677; focal body size and density interaction = −0.007, s.e. = 0.032, χ² = 0.042, p = 0.838; sex interaction = −0.009, s.e. = 0.032, χ² = 0.087, p = 0.768; figure 2). As phenotypic assortment was never clearly different from zero social selection does not ever alter the total selection differential, which is the phenotypic variance in body length (149.222) multiplied by the direct selection gradient (above), giving 41.358.

Figure 2.

Phenotypic assortment was not different from zero at any density. The graph shows the predicted phenotypic assortments at high densities (more than 40 weevils per tree, light blue) compared to low densities (40 or fewer weevils per tree, black), but note that we analysed density as a continuous variable; we have used a categorical representation when plotting for ease of viewing. The grey areas around the trend lines indicate the 95% confidence intervals. (Online version in colour.)

4. Discussion

We estimated the strength of social selection across a range of contexts for both male and female giraffe weevils. In contrast to our predictions, we found that social selection was typically absent, although negative social selection was apparent when controlling for density, and social selection increased in strength at high densities, a result in line with our predictions. An increase in the strength of negative social selection as densities increase is consistent with the idea that weevils are engaging in higher competition for access to mates at high densities. However, as phenotypic assortment was always close to zero, the social selection we observed will make limited contributions to overall selection. Therefore, it seems that direct selection will govern the evolution of body and weapon size in giraffe weevils.

Our results are consistent with several previous studies exploring how traits like body size can benefit focal individuals while being costly when expressed in interacting individuals. Formica et al. [21] found positive direct selection and negative social selection for body size (when using mating success as a proxy for fitness, but this is not true when using survival) in fungus beetles (Bolitotherus cornutus), matching our result when we accounted for variation in density. Similarly, Santostefano et al. [20] found positive direct selection and negative social selection for body size in male chipmunks (Tamias striatus) in the summer (but never for females). In an equivalent result using contextual analysis, jewelweed (Impatiens capensis) show positive within-group selection and negative among-group selection for size [47]. Meanwhile, several studies have used a ‘mixed’ method, which should not differ substantially from a social selection approach if groups are large enough (see [9]). These studies have found positive direct and negative social selection for size in the ant Pristomyrmex pungens [48] and sea rocket (Cakile edentula) at low densities [23]. Other studies, however, find positive direct selection and either positive or variable social selection for size (black-throated blue warblers, Setophaga caerulescens, [49]; Silene tatarica [50], note latter study used the ‘mixed’ method), or, using contextual analysis, find within- and among-group selection to be aligned (for growth rate in juvenile North American red squirrels, Tamiasciurus hudsonicus, [22]; for body mass in fledgling great tits, P. major, [51]). Therefore, while opposing direct and social selection for body size may be more common than any other situation, consistent with competition for limited resources (such as food or members of the limiting sex, [52]) where larger individuals gain the most and therefore cause their rivals to suffer relative losses, it is by no means the rule. Given we have clear theoretical reasons to expect social selection (and social interactions more broadly) to alter the speed and direction of evolutionary change, we need to gather more estimates of direct and social selection so that we can identify general rules for when social interactions do fundamentally alter evolution.

Interestingly, we found clear density-dependent social selection for both males and females, and so females experience reduced mating success at high densities when there are large females on the same tree. This finding challenges the implicit assumption that females do not compete for mating partners. The mechanism for this social selection in females remains to be explored, although there is evidence larger males prefer to mate with larger females [31], which might lead to fewer matings for smaller females. Another possibility is that at high densities males are spending more of their time fighting other males, leaving less time to copulate with females, resulting in choosier males to the detriment of small females sharing trees with large females.

Several contextual factors we had predicted would influence social selection did not. Sex ratio had no effect on social selection, which surprised us given we assumed social selection represents competition for mates, which should be stronger for the sex which is more common. The sex ratio varied from 0 to 1.0 so we do not think a lack of variation in our dataset prevented us from finding a pattern. Possibly, many males or females on a tree on any given day are not competing for mates, making our measure of sex ratio uninformative. We also found a focal individual's body size did not influence the impact of rivals on fitness. We had expected smaller males to be less severely affected by large rivals, as they are able to obtain matings by switching from a female-defence strategy to ‘sneaking’ copulations with females guarded by large males [32]. However, given we only detected any negative effect of larger rivals at high densities, we might require a dataset with more samples of trees with a high density of giraffe weevils to test this prediction. While correlational direct selection has received some attention [53], we possess very limited information about which traits of individuals influence the strength of social selection (but see [22]). Beyond body size, certain behavioural traits, such as how responsive an individual's behaviour is to the traits of others, or how susceptible its phenotype is to being modified by the actions of others, might modulate how strongly an individual is influenced by rivals and could alter evolutionary trajectories [46], but this remains to be tested.

We found phenotypic assortment was never clearly different from zero for both sexes. Due to this near-zero assortment, social selection on giraffe weevil body size will not contribute to total selection. Note that, although taking a contextual analysis approach is equivalent to the social selection approach we used, in contextual analysis there is no requirement for phenotypic assortment to allow among-group selection to contribute to the response to selection (see [8]). Limited phenotypic assortment is consistent with individuals mostly randomly aggregating on trees without respect to the body size of other individuals on the tree. In giraffe weevils, assortment by body size has been observed in mating pairs [31], suggesting there is some capacity for partner choice once a weevil has arrived at an aggregation, but our current results show limited assortment occurs in the process of forming aggregations on trees. There may be environmental factors, such as the distance between suitable trees, that prevent strong assortment from occurring. If a giraffe weevil can only access a single or small number of suitable trees, and body sizes are initially evenly spread across the landscape, then no strong assortment could occur regardless of whether individuals have preferences for certain sized rivals or not.

Estimates of phenotypic assortment are accumulating in the literature and are often positive. Positive assortment can arise through environment by phenotype associations, through active selection for like individuals, or from individual's confirming to match their social partners [18]. If positive phenotypic assortment is indeed common, then social selection will often contribute to total selection, and if social selection is typically in the opposite direction to direct selection [9], will, therefore, tend to reduce overall selection. However, there are only a few estimates of the interactant covariance, the specific parameter used in social selection models [16]. Formica et al. [21] estimated the interactant covariance for body size in aggregations of forked fungus beetles (B. cornutus) and found a negative covariance. This would cause negative social selection for body size to increase the magnitude of the overall positive selection for body size. By contrast, while Santostefano et al. [20] found a negative interactant covariance among female chipmunks (T. striatus) for body mass, they found no covariance among males for body mass. Since social selection was only present in males, social selection would not contribute to overall selection in either sex. In summary, while we may expect social selection to weaken overall selection, evidence from systems where both social selection and the interactant covariance have been estimated suggests that it often does not contribute at all. Further, a lack of estimates of how phenotypic assortment changes with key demographic parameters such as density prevents us from understanding whether there are some contexts social selection does contribute to total selection. Each of direct selection, social selection, the phenotypic variance and phenotypic assortment can vary with conditions. Therefore, evolution may vary substantially in both direction and speed in different environments, and further could change differently across environments for traits that show a higher degree of social selection, or a greater tendency towards positive assortment, and so on. While we currently lack the data to assess this possibility, it represents an exciting future line of work.

Overall, we have contributed to our knowledge of how selection operates in wild animals. As predicted, the social selection was in the opposite direction to direct selection and was stronger at high densities. However, social selection was not clearly different from zero when not controlling for variation in density and did not vary with sex ratio or the size of the focal individual. Further, phenotypic assortment was never different from zero, indicating that social selection will have a limited contribution to overall selection even at high densities. Therefore, despite its predicted importance, social selection will only have a minor impact on the evolutionary change of body size in New Zealand giraffe weevils.

Data accessibility

The data and the R code used to manipulate the dataset, analyse the data and produce all figures are provided in the electronic supplementary material [54].

Authors' contributions

D.N.F.: conceptualization, formal analysis, visualization, writing-original draft, writing-review and editing; R.J.L.: data curation, investigation, methodology, writing-review and editing; C.J.P.: conceptualization, data curation, funding acquisition, methodology, project administration, resources, supervision, writing-original draft, writing-review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

D.N.F. was supported by the University of Aberdeen. R.L.G. was supported by a University of Auckland Masters Scholarship during data collection. C.J.P. was supported by a Rutherford Foundation Postdoctoral Fellowship during the writing of this manuscript.