Does predation result in adult sex ratio skew in a sexually dimorphic insect genus?


Priscilla Wehi, Institute of Natural Resources, Massey University, Private Bag 11-222, Palmerston North 4442, New Zealand.
Tel.: +64 06 3569099; fax: +64 06 3505701; e-mail:


Theory proposes that sexually dimorphic, polygynous species are at particularly high risk of sex-biased predation, because conspicuous males are more often preyed upon compared to females. We tested the effects of predation on population sex ratio in a highly sexually dimorphic insect genus (Hemideina). In addition, introduction of a suite of novel mammalian predators to New Zealand during the last 800 years is likely to have modified selection pressures on native tree weta. We predicted that the balance between natural and sexual selection would be disrupted by the new predator species. We expected to see a sex ratio skew resulting from higher mortality in males with expensive secondary sexual weaponry; combat occurs outside refuge cavities between male tree weta. We took a meta-analytic approach using generalized linear mixed models to compare sex ratio variation in 58 populations for six of the seven species in Hemideina. We investigated adult sex ratio across these populations to determine how much variation in sex ratio can be attributed to sex-biased predation in populations with either low or high number of invasive mammalian predators. Surprisingly, we did not detect any significant deviation from 1 : 1 parity for adult sex ratio and found little difference between populations or species. We conclude that there is little evidence of sex-biased predation by either native or mammalian predators and observed sex ratio skew in individual populations of tree weta is probably an artefact of sampling error. We argue that sex-biased predation may be less prevalent in sexually dimorphic species than previously suspected and emphasize the usefulness of a meta-analytic approach to robustly analyse disparate and heterogeneous data.


Fisher’s prediction that primary 1 : 1 sex ratios are evolutionarily stable (Fisher, 1930) is not affected by the occurrence of polygamy or differential mortality of the sexes (Hamilton, 1967). This is despite fluctuating response to factors such as mate attractiveness, parental age and condition, and parental investment in offspring (Trivers & Willard, 1973; Clutton-Brock & Iason, 1986; West & Sheldon, 2002; Sheldon & West, 2004). In contrast, adult sex ratio bias is evident in many taxa (e.g. birds (Donald, 2007), ungulates (Clutton-Brock et al., 1997), frogs (Lodéet al., 2004), guppies (McKellar et al., 2009) and butterflies (Dyson & Hurst, 2004)) and can persist despite apparently negative consequences (e.g. Trewick, 1997; Dyson & Hurst, 2004). Adult sex ratio can affect mate choice and mate competition because it can influence the rate at which individuals encounter competitors or potential mates, and therefore the intensity of competition for mates or resources (Emlen & Oring, 1977; Magnhagen, 1991; Kokko & Rankin, 2006; Klug et al., 2010). For example, a male-biased adult sex ratio in crickets led to increased male–male competition and female choosiness (Gwynne, 1984). Differential predation on male frogs, resulting in female-biased sex ratio, affected male sexual behaviour and decreased the occurrence of polyandry (Lodéet al., 2004). Fitness consequences for both sexes can follow (Kasumovic et al., 2008) and in turn alter the opportunity for, and the intensity of, sexual selection (e.g. Dreiss et al., 2010). Thus, Le Galliard et al. (2005) demonstrated experimentally in lizards that an excess of adult males resulted in increased aggression towards adult females, decreasing their survival and fecundity and amplifying the existing male bias with an increased risk of population decline or extinction.

Differential mortality of the sexes that leads to adult sex ratio bias arises from extrinsic factors such as predation and intrinsic factors such as trade-offs in resource allocation within the body that lead to differential longevity, maturation or ageing (Zajitschek et al., 2009). Quantifying the role of predation as a factor of extrinsic mortality is notoriously difficult in natural populations (Christe et al., 2006), but sex ratio bias has nonetheless been reported in prey remains from a number of species including lizards (Costantini et al., 2007) and mammals (Christe et al., 2006). Sex-specific risk of predation is influenced by phenotypic variation (Magnhagen, 1991; Lodéet al., 2004; Christe et al., 2006). Strong sex-specific selection on traits common to both sexes typically results in sexual dimorphism, including the development of secondary sexual characteristics such as male weaponry. However, natural selection is thought to oppose sexual selection favouring exaggerated male weaponry and other sexually selected signals (Promislow et al., 1992; Andersson & Iwasa, 1996; Zuk & Kolluru, 1998; Bonduriansky et al., 2008), such that the evolution of sexual size dimorphism is associated with a pattern of male-biased mortality (Promislow et al., 1992; Moore & Wilson, 2002). High male mortality might arise through heavy parasitism (Moore & Wilson, 2002), inefficient movement (Oufiero & Garland, 2007), male combat (Kelly, 2006) or difficulty finding suitable refugia (Lappin, 2006), all of which may increase predation risk. In addition, male weaponry might compromise feeding (Emlen, 2008). Activity associated with mate competition can attract predators and lead to sex ratio bias (e.g. Gwynne, 1987; Acharya, 1995), particularly in polygynous species where male–male combat for females can also lead to increased injury and death (Bonduriansky et al., 2008). Adult sex ratio bias thus appears to be most frequent in polygynous, sexually dimorphic species with exaggerated male weaponry or visual or vocal displays (Promislow et al., 1992; Kelly, 2008). Nonetheless, sex ratio bias might also result from recent human-induced changes to the environment (such as the introduction of novel predators) that disrupt existing equilibria between selection for and against such male weapons and displays.

Weta are a prominent and ecologically diverse group of ensiferan crickets (Anostostomatidae) that comprise many endemic, large and flightless species in New Zealand (Gibbs, 2001; Trewick & Morgan-Richards, 2009). Among these species, tree weta (Hemideina spp.) are a long-lived (average adult life span around 9 months, but up to 3 years (Leisnham et al., 2003)), nocturnal group that avoids diurnal predators through cavity living (Gibbs, 1998). They are sexually dimorphic (Hudson, 1920; Field & Deans, 2001) and polygynous (Gwynne & Jamieson, 1998), with males displaying megacephaly and highly exaggerated mandibles (Kelly, 2005a). These enlarged mandibles are used as weapons in male–male contests for access to females (Sandlant, 1981; Kelly, 2006), so that large mandibles might have fitness benefits, especially where male–male competition is intense (Klug et al., 2010). However, contests outside the safety of tree cavities (where female harems reside) might also expose male tree weta to elevated mortality risk from nocturnal predators such as the native owl (Ninox novaeseelandiae) (Lindsay & Ordish, 1964; Saint Girons et al., 1986; Clark, 1992; Haw & Clout, 1999; Haw et al., 2001), weka (Gallirallus australis) (Field & Glasgow, 2001) or gecko (Haplodactylus pacificus) (McIvor, 1972). The cumbersome mandibular morphology might also have other disadvantages, such as restricting the ability to quickly find refuge from predators where large refuge cavities are limited.

The opportunity to detect sex-specific effects in relation to a novel suite of predators arises with the recent arrival of mammalian predators in New Zealand and enables a natural predation experiment where there has been little time for evolutionary response. New Zealand lacks native terrestrial mammalian predators. Novel mammalian introductions began with the Polynesians around 800 years ago and accelerated with European colonization in the mid-1800s (Anderson, 1991; Atkinson & Cameron, 1993). These introductions have led to a significant decline in many native New Zealand species (Gibbs, 2009). Subtle impacts resulting from introduced mammalian predators include sex-specific predation in birds such as female kaka parrots (Nestor meridionalis) incubating young on nests (Wilson et al., 1998; Moorhouse et al., 2003). Thus, if there is a greater predation risk to female kaka because they spend more time than males at the nest where they are vulnerable, the sex ratio will be most male-biased at sites with high predator density. Tree weta form a large proportion of rat (Rattus rattus, R. exulans and R. norvegicus) diet in New Zealand (Daniel, 1973; Innes, 1979; King & Moody, 1982; Miller & Miller, 1995). They are also consumed by other introduced mammals that hunt at night including cats (Felis catus) (Richards, 1973), stoats (Mustela erminea) (Fitzgerald, 1964; King & Moody, 1982) and hedgehogs (Jones et al., 2005). Overlapping generations in tree weta ensure year-round availability for these predators, but sex-specific predation effects have not been conclusively demonstrated (Rufaut & Gibbs, 2003; Watts et al., 2011). If predation on conspicuous males is an important evolutionary force shaping tree weta ecology and behaviour, we might expect this effect to be exacerbated in regions with high numbers of novel predators. Predators such as rats now occur at high density throughout tree weta ranges with the exception of alpine regions but have been cleared or substantially reduced on some island and mainland sites. This allows direct comparisons between sites with and without mammalian predators.

In this research, we apply a meta-analytic approach to investigate sex ratio variations using a large data set collected for six Hemideina tree weta species. Differences in life history and selection pressure on secondary sexual characters that may result from sex-specific predation first require a test that adult sex ratio bias exists. We consider the following questions. First, what is the overall relationship between adult sex ratio and predation across studies in tree weta, after accounting for species and population variation? Second, is the relationship between sex ratio and predation dependent on the type of measure of sex ratio employed? We hypothesize that males (being the conspicuous dimorphic sex) are more vulnerable to predation as predicted by sexual selection theory. We consider whether tree weta sex ratios have been affected by the influx of a suite of predators quite distinct from any that have existed in New Zealand since the Pliocene (Field & Glasgow, 2001). We expect that if sex-biased predation occurs, it might be most marked in tree weta populations subject to both native and novel predators. A meta-analytic approach, using a large data set collected from throughout New Zealand, allows us to obtain a more reliable and powerful assessment of both evolutionary and novel predation impacts in relation to adult sex ratios in a sexually dimorphic, polygynous species.


Data collection

We conducted a literature search current to September 2010 using ISI Web of Science ( and Google Scholar ( to identify all published sources of sex ratio in Hemideina, using the topic word ‘weta’. We identified and excluded data that were reported more than once to avoid multiple uses of the data in our analysis. We also contacted all department of conservation offices throughout New Zealand to locate any unpublished, but available, data on tree weta sex ratios. This procedure proved very productive, with a large number of unpublished data sets located for this project (Table S1), thereby overcoming biases commonly associated with publication (Kotiaho & Tomkins, 2002; Jennions et al., 2004). In total, data were collected for six different Hemideina species (H. crassidens Blanchard 1851, H. femorata Hutton 1898, H. maori Pictet and Saussure 1891, H. ricta Hutton 1898, H. thoracica White 1842, and H. trewicki Morgan-Richards 1995) and 58 different populations. These populations were distributed throughout New Zealand, including some on offshore islands. No data were available for one species currently assigned to Hemideina (Hemideina broughi Buller 1896), but as this species is morphologically and phylogenetically aberrant (Morgan-Richards, 2001; Trewick & Morgan-Richards, 2004) it is, in any case, appropriate to exclude the species. All data sources used in the analyses are listed in the online supplement.

The numbers of males and females were obtained from all the collected sources. We recorded categorical variables of predation level relevant to the sample population and count method. Predation level was recorded as 1 (low) or 2 (high) in the first instance. Category 1 locations included ‘mainland island’ sites surrounded by a predator-proof fence to exclude mammalian predators (e.g. Boundary Stream, Maungatautari), offshore islands that have no exotic predators such as rats or cats recorded and sites with extensive trapping of mammalian predators and especially rats. Sites with some intermittent trapping, or the full suite of mammalian predators present without control, were recorded as category 2. Sites for which we had inadequate information were removed from the analysis. Three main count methods were recorded: observation in artificial cavities (most commonly used by conservation managers), night-time observations, and manual extraction from natural holes (frequently used to collect tree weta from the wild).

Statistical analyses

All statistical models were implemented using R (version 2.12.2; R Development Core Team, 2010). To analyse our proportional data (i.e. the numbers of males and females or sex ratios), which included nonindependent data points (i.e. replicates within populations and species), we applied a generalized linear mixed model (GLMM) with the quasi-binomial error structure (with a logit link function). GLMMs deal with both non-Gaussian data and pseudo-replication by explicitly modelling grouping structures of data as random factors (Bolker et al., 2009). We fitted GLMMs using the function glmmPQL in the MASS package (Venables & Ripley, 2002), which uses penalized quasi-maximum likelihood with a multiplicative overdispersion (Nakagawa & Schielzeth, 2010). If the overdispersion parameter (ω) equals 1, then the variation in the response variable (i.e. sex ratios) is fully explained by the variance of the specified distribution (e.g. the binomial distribution). Thus, no further predictors are required to explain the variation in data if ω = 1. We ran two GLMMs. In our two models, we specified two variables, 58 populations and six Hemideina species as nested random factors to account for nonindependence and also to calculate how much variation in the response can be attributed to these factors. In the first model, we only fitted the intercept as a fixed factor to obtain an overall sex ratio, as well as estimating ω and variance components for populations and species. In the second model, we added the predation levels and counting methods described above as fixed factors to identify the effects of these two variables associated with the main aims of this study. This model had fewer data points than the former because of missing information regarding the predation levels and counting methods.

It has been noted that, with some data structures, the function glmmPQL could give biased parameter estimations (Bolker et al., 2009). Therefore, we validated our results with two other R functions capable of GLMMs, the function lmer in the lme4 package (Bates & Maechler, 2010) and the glmmMCMC function in the glmmMCMC package (Hadfield, 2010). These functions provided near-identical results from glmmPQL, and thus, we deemed our results reliable.


In our first model (n(observation) = 258), where we had six Hemideina species and 58 populations as random factors, the overall proportion of males was close to 0.5 and no deviation from parity was detected (GLMM: sex ratio estimate = 48.28% [male %; note that the results were converted to percentage from the logit scale for interpretability], 95% CI = 42.73 to 53.86%, t200 = −0.607, = 0.544; Fig. 1a). The overdispersion parameter was close to 1 (ω = 1.253). This result demonstrated that most of the observed variance in sex ratios can be attributed to error in the binomial distribution. Our model also showed that population differences accounted for only 3% (variance component = 0.125; see Fig 1b) of the variance and that species differences accounted for < 1% (variance component = 0.032). Notably, the estimated male percentages for the six species from this model were as follows: H. crassidens (48.84%), H. femorata (48.86%), H. maori (48.35%), H. ricta (48.36%), H. thoracica (43.37%) and H. trewicki (47.47%).

Figure 1.

 Funnel plots of the proportion of male Hemideina spp. (sex ratio) and the sample sizes: (a) the entire data set with the meta-analytic mean for each sample (= 258), (b) data points at the population level with meta-analytic means (= 58), (c) a partial data set with predation levels indicated (= 213) and (d) the same partial data set with survey methods indicated (= 213). Method 1: night-time observation, Method 2: manual extraction from natural cavities and Method 3: observation in artificial cavities.

In the second model, we included the count method and predation parameters to specifically identify their effects on sex ratio estimates (n(observation) = 213, n(species) = 6, n(population) = 47). No statistical differences in adult sex ratio were detected between populations with high and low predation sites (GLMM: the difference between two predation levels = 5.15% [male detection per cent difference], 95% CI = −2.80 to 13.26%, t163 = 1.26, = 0.208). Similarly, there were no statistical differences in adult sex ratio detected between different count methods (GLMM: the difference between count methods 1 and 2 = −4.45%, the difference between count methods 1 and 3 = −3.43%; 95% CI = −7.12 to 16.4%, 95% CI = −13.9 to 8.22%, t163 = 0.738 and t163 = −0.594, = 0.462 and = 0.554, respectively; Fig. 1c,d). These results are consistent with the lack of difference between populations and species described earlier. The overdispersion parameter estimate for this model was lower than that of the first model (ω = 0.369), and the variance components of species and population differences were similar to those of the first models (variance component(population) 0.215, variance component(species) < 0.0001, respectively). When ω < 1, it is deemed as underdispersion, meaning that the fixed factors (i.e. predation levels and count methods) were redundant and should be removed from the model.


The meta-analysis presented here does not support our prediction of sex-biased predation in an insect genus with strong sexual selection on male weaponry. It has been argued that extrinsic mortality may be high in sexually dimorphic and polygynous species (Promislow et al., 1992; Moore & Wilson, 2002), where the effects of strong sexual selection and male combat may result in female-biased adult sex ratios. Here, we infer that proximate mechanisms such as predation are not drivers of adult sex ratio bias in tree weta. We find no support for any of the main hypothesized effects we examined and show that species and population differences account for only a small proportion of the variation in weta adult sex ratio. It is possible that predation is altogether unimportant in weta biology, and thus has little influence on adult sex ratio, but numerous dietary studies show that weta form a major component of native and introduced predator diets throughout the year. For example, in one study tree weta occurred in all rat stomachs necropsied in summer and autumn, as well as 65% of spring and 80% of winter rat stomachs (Miller & Miller, 1995). Experimental trials where mammalian and avian predators were introduced to H. crassidens (n = 33) have demonstrated that tree weta defence behaviours (‘freezing’, raising their spiny hind legs, gaping enlarged mandibles) are largely ineffective in deterring predator attacks (76% mortality overall; Field & Glasgow, 2001). There is little evidence for deviation from the 1 : 1 primary sex ratio of tree weta derived from a XO/XX sex-determining mechanism (Morgan-Richards, 1997), despite predictions that adult sex ratio in sexually dimorphic species may be female-biased if conspicuous males are preyed upon more often, and despite indications of apparent sex biases in some data sets (Townsend, 1995; Kelly, 2008; Watts et al., 2011; P. Wehi, M. Jorgensen and M. Morgan-Richards, unpublished data).

Populations with intense mammalian predation have provided us with a common garden experiment, because adaptation to these olfactory, tree-climbing hunters is unlikely to have occurred since their introduction by humans. Our finding of sex ratio parity indicates that sex-specific predation is absent or negligible in populations with a suite of native and exotic predators, exactly where we expected such bias to be most evident. This finding contrasts with the sex-specific vulnerability to mammals that is evident in some New Zealand species such as kaka parrots (Wilson et al., 1998; Moorhouse et al., 2003) and suggests that the common assumption that elaborate secondary sexual traits have a fitness cost or trade-off should be tested further in insects.

Meta-analysis provides a robust analytical approach for handling disparate data sets. We conclude that sex ratio variation inferred in some of the individual studies contributing to our data set most likely represents type I errors. The results of our model reveal that apparent female bias under low predation conditions (Kelly, 2008; P. Wehi, M. Jorgensen and M. Morgan-Richards, unpublished data) and apparent male bias under high mammalian predation conditions (Watts et al., 2011) were adequately explained by the expected variance in the binomial distribution in our model. Any factor which introduces variation into the probability of male (or female) survivorship should cause greater variation in the data than the expected variance (i.e. overdispersion) in the model. In our analysis, the overdispersion parameter (ω) was close to or < 1, indicating that sample errors account for apparent deviations from parity (cf. Postma et al., 2011). Other indirect evidence that could potentially support a hypothesis of sex-specific predation on adult tree weta (for example, where 38 adult female and 47 adult male tree weta exoskeletons were found in kiore (Rattus exulans) husking stations (Rufaut & Gibbs, 2003)) would require substantial expansion of the data set to avoid similar type I error. In addition, we detected little difference in the variance of population sex ratios for tree weta among the three main methods used to count tree weta. Any of the methods examined should therefore produce comparable snapshot estimates of tree weta sex ratio.

Potential bias towards positive results in publication can be offset by the use of unpublished sources in meta-analysis. In addition, meta-analysis allows inference beyond individual studies, which are susceptible to sampling errors and frequently have low statistical power, thus reducing biases associated with ‘vote-counting’ procedures and in particular an increase in the probability that null hypotheses will be accepted erroneously, i.e. type II error (Jennions et al., 2001). In this study, we sourced much of our data from unpublished work to increase the population sample size. Our findings support an earlier estimate of adult sex ratio parity in tree weta (population sample sizes 11 < > 83, = 12 studies) (Stringer, 2001) although subsequent publications have indicated deviations from parity (Kelly, 2008; Watts et al., 2011). Interpretation of results from individual studies, however, is not always straightforward. One large population sample in our data set was significantly male-biased (= 604, Fig. 1a; Spurr & Berben, 2004). The authors attributed this bias to male tree weta colonization after a poison drop targeting invasive mammals, with a concomitant 10-fold increase in population density. Those findings emphasize the importance of considering time scale and sufficient long-term monitoring to determine demographic parameters such as sex ratio at the individual population level. However, at the population level, this data point (Fig. 1b) does not deviate markedly from the expected variation (which was confirmed by the results from our GLMMs), emphasizing the importance of meta-analysis.

Emlen (2008) has pointed out the lack of a theoretical framework for weapon evolution. Hemideina have evolved sexually dimorphic male weaponry in stark contrast to their sister lineage, the giant weta (DeinacridaTrewick & Morgan-Richards, 2004), which shows no such dimorphism. The reason for this is unclear, and interpretations that large male mandibles used in male–male competition have a natural selection cost (Jennions et al., 2001) require identification of these costs or constraints in tree weta. Substantial injury and death can result from male combat in Hemideina crassidens (Kelly, 2006); high-intensity interactions (wrestling and locking jaws) were recorded in 7% of laboratory observations of male–male combat, but the incidence of injury in the wild is unknown. Hypotheses that tree weta males may have elevated parasite loads or decreased immunocompetence associated with the costs of dimorphism are not well supported (Robb et al., 2003; Kelly, 2005b; Kelly & Jennions, 2009). The results of our models suggest that much remains to be understood about the evolutionary ecology of weta, and more generally the costs of sexual selection, specifically exaggerated male weaponry, across different taxa.


Gareth Boyt, Brendan Christensen, Chris Green, Ralph Powlesland, Cam Robertson and staff from the Department of Conservation and the Boundary Stream Mainland Island generously shared their unpublished data, as did Kirsti Bennett, Niki Minards, Melissa Jacobson and Melissa Griffin. William Wehi, Niwa Wehi and many other enthusiastic assistants helped in the field. Mark Davis of Punga Direct collected weta from Kinleith Forest. Corinne Watts kindly made her paper available prior to publication. Maungatautari Ecological Island Trust, Hamilton City Council and the Department of Conservation provided permission for the authors to collect tree weta at a range of sites. Michael Jennions and three anonymous referees made constructive comments on the paper. This research was funded by FRST Postdoctoral fellowship MAUX0905 to PMW.