Skip to main content

Dataset for: Mating competition and adult sex ratio in wild Trinidadian guppies

Cite this dataset

Chuard, Pierre; Grant Brown, James; Brown, Grant (2022). Dataset for: Mating competition and adult sex ratio in wild Trinidadian guppies [Dataset]. Dryad.


Most experimental tests of mating systems theory have been conducted in the laboratory, using operational sex ratios (ratio of ready-to-mate male to ready-to-mate female) that are often not representative of natural conditions. Here, we first measured the range of adult sex ratio (proportion of adult males to adult females; ASR) in two populations of Trinidadian guppies (Poecilia reticulata) differing in ambient predation risk (high vs. low). We then explored, under semi-wild conditions, the effect of ASR (i.e. 0.17, 0.50, 0.83) on mating competition patterns in these populations. ASR in the wild was female-biased and did not significantly differ between the two populations. The range of ASR in our experiment was representative of natural ASRs. As expected, we observed an increase in intrasexual aggression rates in both sexes as the relative abundance of competitors increased. In support of the risky competition hypothesis, all measured behaviors had lower rates in a high vs. low predation-risk population, likely due to the costs of predation. In terms of mating tactics, a male-biased ASR did not lead males to favor forced mating over courtship, indicating that males did not compensate for the cost of competition by switching to a less costly alternative mating tacticOverall, this study highlights the need for field experiments using natural ranges of ASRs to test the validity of mating systems theory in a more complex, ecologically relevant context.



The Upper Aripo is a low-predation site (Croft et al. 2006; Botham et al. 2008; Deacon et al. 2018), with few predator species, at low densities, foraging primarily on newly emerged juvenile guppies. Predators include Hart’s rivulus, Anablepsoides hartii (Magurran 2005), and a small freshwater prawn, Macrobrachium crenulatum (personal observations GE Brown). By contrast, the Lower Aripo River is a high-predation risk site (Croft et al. 2006; Deacon et al. 2018), with a higher diversity and density of guppy predators including pike cichlids, Crenicichla sp., blue acara cichlids, Aequidens pulcher, black acara cichlids, Cichlasoma bimaculatum, wolf fish, Hoplias malabaricus, and two-spot sardines, Astyanax bimaculatus (Endler and Houde 1995; Croft et al. 2006; Brown et al. 2009; Walsh and Reznick 2009), all preying on adult guppies. Natural selection is thought to have led to locally-adapted populations to high vs. low ambient predation-risk (Fitzpatrick et al. 2015). The selective pressures are strong enough that these gene pools remain resilient to infrequent gene flow. These two populations are more genetically similar to each other than to other Trinidadian guppy populations differing in ambient predation risk (Carvalho et al. 1991), suggesting more recent geographical isolation of the two populations, or continual gene flow from the Upper to the Lower Aripo.

Compared to the Upper Aripo River, the guppy population in the Lower Aripo River experiences markedly different ecological conditions, including higher predation risk, lower guppy densities and higher stream productivity (Reznick et al. 2001). Hereafter, we will refer to these two populations as experiencing high and low ambient predation risk, respectively, as it is the most likely factor shaping behavioral patterns in Poeciliidae (Heinen et al. 2013; Elvidge et al. 2016; Chuard et al. 2016, 2018, 2020). However, we acknowledge that ecological factors other than predation risk may explain any population differences detected in this study (see Discussion). 

Experimental design

To estimate the range of ASRs within each population, calculated here as the proportion of adult males to adult females, we sampled individual pools using seine nets of 3-mm stretched-mesh (l x 4 m). We sampled seven pools in the Lower Aripo in 2014 from 1 to 5 times every 2.5 days on average, and sampled an additional eight different pools once in 2015 (Table 2). In the Upper Aripo, we sampled eight pools in 2014 from 2 to 6 times, and eight different pools once in 2015 (Table 2). The distance between a pool and its nearest neighboring pool was between 2 and 4 m (Pettersson et al. 2004); adjacent pools were separated by short riffle areas. We obtained a total of 20 ASR estimates from the Lower Aripo and 34 from the Upper Aripo in 2014. During both sampling seasons, we estimated ASR in each pool by sampling the entire pool in an upstream direction 2.5 times on average (range: 2-5) before obtaining an estimate of ASR for a given day. We fished until most individuals were sampled, as indicated by at least a 90% decrease in the number of individuals caught from one sampling attempt to the next. In 2014, the same pool was typically sampled every 2-3 days (Table 2).


Table 2 Mean (range) of physical and demographic characteristics of the pools sampled in the Lower (high predation) and Upper (low predation) Aripo populations in 2014 and 2015


Lower Aripo       2014 (n=7)

Lower Aripo  2015 (n=8)

Upper Aripo 2014 (n=8)

Upper Aripo 2015 (n=8)

Area (m2)









Depth (m)









Velocity (m/s)









Guppies per pool


Samples per pool

















Days between sampling










If both sexes were present, we retained the sampled individuals for the mating competition experiments (see below) in 30-L buckets filled with 10 L of river water. In the rare cases where only one sex was present, we recorded the data and released individuals into the current downstream of their pool. We placed a maximum of 30-40 fish per bucket in mixed-sex groups and changed the water every 45 min to avoid changes in temperature and dissolved oxygen. We tested individuals on the day of sampling.

We performed mating competition experiments in the wild, as described in Chuard et al. (2016), at both the Upper and Lower Aripo River (i.e. low vs. high predation sites) using transparent rectangular Plexiglas enclosures (40 x 30 x 30 cm) with a bottom placed on the river substrate. The experiment took place during the day between April 19th and May 1st, 2014. We used three ASR (as a proxy for OSR, see Chuard et al. 2016) treatments each composed of six individuals (i.e. 0.17, 0.50, 0.83), to match the range of OSRs used by Jirotkul (1999). We replicated each treatment 30 times in each of the two populations. To be comparable to Chuard et al. (2016)’s laboratory experiment, the amount of water flowing through the enclosures was minimal. In any given trial, we used guppies from the same pool, so that individuals may have been familiar with one another (see Chuard et al. 2016). We gave individuals a 5-min period to acclimate to the test enclosure before the start of observations. We justified this relatively short acclimation period based the familiarity of individuals within a trial, and by the high natural dynamic of fission-fusion processes between pools in guppies (Croft et al. 2003), often driven by ASR, indicating that guppies are able to rapidly adjust their behavior to their local environment. We used Go ProTM cameras placed underwater outside the enclosure to obtain the same side view as the observer in Chuard et al. (2016). We used a large, 3-mm stretched-mesh net (6-m long x 0.5-m high) to form a rectangular barrier of 1.5 x 1.5 m around our Plexiglas enclosures to isolate guppies in our experiment from those swimming freely outside of the netted area in the stream. We dispersed a small pinch of flake food (TetraminTM) as described in Chuard et al. (2016) uniformly over the surface 5 min before a trial to limit aggressive interactions related to foraging (Robb and Grant 1998). Fish typically consumed all large flakes during the 5 min following feeding, leaving only small particles available during the observation. We distinguished males from females based on their conspicuous coloration. We measured the same behavioral traits using a continuous scanning method (except for the seconds it took to write down the behavior observed) as in Chuard et al. (2016) over 10 min: (1) intrasexual aggression rates for both males and females separately by summing the frequency of biting, chasing (Gorlick 1976), tail beating (Liley 1966) and pushing (Magurran and Seghers 1991) for each sex; and, male mating tactics quantified as the rates of (2) courtship (i.e. sigmoid display) and (3) forced mating attempts (i.e. sneaking). Male guppies court females by erecting their dorsal fin and moving up and down while bending their body in a sigmoid shape (Houde 1987). As an alternative mating tactic, male guppies use sneaking attempts where they try to insert their gonopodium inside a female’s gonopore without initial courting (Farr 1980). Immediately after a trial, we released individuals downstream from the location of capture and below some natural physical barriers (e.g. small waterfall, riffles) to avoid resampling. These displacements are unlikely to have affected future ASR sampling as guppies show highly dynamic fusion-fission processes and individuals would have recolonized the pool within minutes (Croft et al. 2003).

Statistical analysis

We used a linear mixed-model (LMM) for the ASR analysis since our data residuals met the normality assumptions, and general linear mixed-models (GLMM) for the behavioral analyses. We fitted our behavioral analyses count data to the negative binomial distribution and tested for over-dispersion. All models were not significantly over-dispersed (P>0.48). For the analysis of ASR in the wild, we performed an LMM on the ASR estimates. We used ASR as the dependent variable; population and year as categorical explanatory variables; and pool as a random factor. In addition, within each population, we estimated repeatability (i.e. 0-1, 1 being 100% repeatability) of ASR over time within the 4 and 8 pools for which we obtained multiple data points in 2014 in the Lower and Upper Aripo River, respectively. For each population, the data met the assumptions of normality, so under a Gaussian distribution. we used ASR as the dependent variable and pool as the grouping factor, and ran one thousand permutations to obtain a p-value for repeatability.

For all behavioral analyses, we used GLMMs, with ASR as a factor as in Chuard et al. (2016), but with 3 ASR treatments (i.e. 0.17, 0.50, 0.83) instead of 5, under a negative binomial distribution. As ASR consisted of more than two levels for the analysis of mating tactic rates (i.e. no 0.17 treatment data for intrasexual aggression as the competing sex had no competitor), we treated ASR as planned contrasts to explore the predicted linear relationships (i.e. linear contrasts). Planned contrasts allow for the ordering of factor levels (i.e. 0.17 first, 0.50 second, 0.83 third) necessary for testing linear relationships of equidistant levels and also test for quadratic relationships, but we only displayed the quadratic results if significant. Chuard et al. (2016) found no significant dome-shaped (e.g. quadratic) relationship between ASR and aggression rates/mating tactic propensities (see Weir et al. 2011; Schacht et al. 2017). 

To facilitate comparisons between the sexes, for the analysis of intrasexual aggression rates, we defined ASR for female intrasexual aggression rates as 1 - ASR. For instance, for an ASR of 0.83 for male intrasexual aggression rates, the corresponding ASR for female intrasexual aggression rates is 0.17. We used sex-specific aggression rate (counts per 10-min observation period) as the dependent variable; ASR treatment (i.e. ASR for male intrasexual aggression rates, 1 - ASR for female intrasexual aggression rates), population, and sex as categorical explanatory variables; observer and trial ID as random factors; and the number of same-sex competitors as an offset term. Offset terms were treated as proportions for all statistical analyses. However, we refer to these proportions as ratios throughout the manuscript to remain consistent with the previous literature on adult sex ratios. Since we measured aggression rates in both sexes simultaneously, we added trial ID as a random factor for the repeated measure in the model. In addition, we also included observer as a random factor to account for the 3 different observers who analyzed the video recordings blindly.

We used a similar approach for the analysis of reproductive behavior. For the separate analyses of courtship and sneaking rate as dependent variables, we used both the planned contrast for the ASR treatments and population as explanatory variables; observer as a random factor; and the number of males as an offset term. Because our enclosures were small and lacked any visual barriers, we considered that all fish had encountered all others at all times during our trials (de Jong et al. 2012). Hence, to analyze mating propensity for each mating tactic, we used the number of possible male-female dyads as an offset term in the previous model (i.e. 5 for the 0.17 and 0.83 ASR treatments, 9 for the 0.5 ASR treatment). To test our eighth prediction on the effect of ambient predation risk on mating tactic rates overall, we ran the same model as the one for courtship and sneaking rates but with the two mating tactics as dependent variables within the same multivariate model.

We displayed the statistical outcomes of our LMM as: (Fdf1, df2, probability); and the statistical results of our GLMMs as: (β, [95% CI], z score, probability) where β represents the regression coefficient and CI the confidence intervals. We included all explanatory variable interactions in our original models, however, to increase the statistical power of our overparametrized behavioral GLMMs, we removed non-significant interactions when applicable as suggested by Engqvist (2005) and Burnham et al. (2011). 

For all analyses, we only reported unpredicted interactions (including quadratic contrasts) if significant after the non-significant interactions were removed from the model if necessary. We did not perform statistical corrections (see Moran 2003) as our models were based on a priori predictions and the different data subsets were not multiply tested. For graphical purposes, we divided our behavioral data by the number of competitors to visualize the outcome of the statistical analyses. We performed all statistical tests in R (3.1.2; R Development Core Team 2015) using the lmer()function () of the lme4 package (Bates and Sarkar 2007) for the LMM, followed by the anova() function (Type III) of the lmerTest package (Kuznetsova et al. 2017) to obtain probabilities. Regarding GLMM analyses, we used the contrast()function (Chambers and Hastie 1992) for planned linear and quadratic contrasts, and the glmmadmb() function of the glmmADMB package (with log link function; Fournier et al. 2012) for GLMMs. We calculated repeatability using rpt()function from the rptR package (Stoffel et al. 2017). 


Natural Sciences and Engineering Research Council