Becoming creatures of habit: Among‐ and within‐individual variation in nesting behaviour shift with age

Abstract

The quantification of repeatability has enabled behavioural and evolutionary ecologists to assess the heritable potential of traits. For behavioural traits that vary across life, age‐related variation should be accounted for to prevent biasing the microevolutionary estimate of interest. Moreover, to gain a mechanistic understanding of ontogenetic variation in behaviour, among‐ and within‐individual variance should be quantified across life. We leveraged a 30‐year study of painted turtles (Chrysemys picta) to assess how age contributes to variation in the repeatability of nesting behaviours. We found that four components of nesting behaviour were repeatable and that accounting for age increased the repeatability estimate for maternal choice of canopy cover over nests. We detected canalization (diminished within‐individual variance with age) of canopy cover choice in a reduced data set despite no shift in repeatability. Additionally, random regression analysis revealed that females became more divergent from each other in their choice of canopy cover with age. Thus, properly modelling age‐related variance should more precisely estimate heritable potential, and assessing among‐ and within‐individual variance components in addition to repeatability will offer a more mechanistic understanding of behavioural variation across age.

Abstract

The nesting behavior of painted turtles becomes more consistent within individuals and more divergent among individuals as females age. Also, accounting for age‐related variation when measuring trait repeatability more precisely estimates heritable potential.

Open in new tabDownload slide

INTRODUCTION

Most microevolutionary studies estimate phenotypic selection, assuming traits are heritable, yet heritability is often low (Dochtermann, Schwab, Berdal, Dalos, & Royauté, 2019; Mousseau & Roff, 1987; Price & Schluter, 1991) or nonexistent (Becker et al., 2015; Kruuk, Slate, Pemberton, & Clutton‐Brock, 2003). Assessing the repeatability of a trait—the amount of phenotypic variation within a population explained by individual identity (Nakagawa & Schielzeth, 2010; Wolak, Fairbairn, & Paulsen, 2012)—is an important first step to determine evolutionary potential, as repeatability typically sets an upper bound on heritability (Boake, 1989; Falconer & Mackay, 1996; but see Dohm, 2002). However, factors other than individual identity may explain variation in phenotype expression (e.g. social context, Jolles, Taylor, & Manica, 2016; time between observations, Biro & Stamps, 2015; Gangloff, Sparkman, & Bronikowski, 2018). In such cases, accounting for confounding sources of variation provides an ‘adjusted’ repeatability that more accurately estimates the variance due to individual identity (Dingemanse & Dochtermann, 2014; Nakagawa & Schielzeth, 2010).

Understanding how behaviours and their consistency change with ontogeny is an important step in understanding the evolution of behaviour (Bell, Hankison, & Laskowski, 2009; Dochtermann et al., 2019; Wilson, 2018). Adding age as a covariate in repeatability models accounts for population‐level variation due to age and accounts for variation resulting from individuals being sampled at different life stages (e.g. Debeffe et al., 2015). However, longitudinal comparisons of behavioural consistency across age are limited in studies of wild animals because repeatability is often quantified from only a few observations per individual (Bell et al., 2009). Nevertheless, behavioural consistency generally increases with age in humans (reviewed in Roberts & DelVecchio, 2000) and usually increases or does not vary between age classes of other animals (Bell et al., 2009; Kok et al., 2019). Increasing repeatability with age could be caused by two nonmutually exclusive mechanisms: 1) the among‐individual phenotypic variance could increase as developmental trajectories progress, or 2) the within‐individual phenotypic variance could decline with age as traits canalize (Dochtermann & Royauté, 2019; Kok et al., 2019). To gain a mechanistic understanding of ontogenetic changes in repeatability, researchers must quantify how each variance component shifts across lifespan (e.g. Araya‐Ajoy & Dingemanse, 2017; Araya‐Ajoy, Mathot, & Dingemanse, 2015; Charmantier, Brommer, & Nussey, 2014; Class, Brommer, & van Oers, 2019; Fisher, David, Tregenza, & Rodríguez‐Muñoz, 2015).

Nest‐site choice is an important maternal behaviour that influences offspring phenotype and survival (Bernardo, 1996; Moore, Whiteman, & Martin, 2019; Refsnider & Janzen, 2010). Previous work across taxa has shown that nest‐site choice is under selection (e.g. fish, Tyler, 1995; amphibians, Rieger, Binckley, & Resetarits, 2004; reptiles, Brown & Shine, 2004; Mitchell, Warner, & Janzen, 2013; and birds, Martin, 1998; Latif, Heath, & Rotenberry, 2012; reviewed in Mousseau & Fox, 1998) and can be both repeatable and heritable (reviewed in Janzen, Delaney, Mitchell, & Warner, 2019). The microevolutionary potential of nest‐site choice has been particularly well studied in turtles because of their unique sex‐determining mechanism and the theoretical role of maternal nesting behaviour in sex ratio evolution (Bulmer & Bull, 1982; Butka & Freedberg, 2019; Mitchell, Maciel, & Janzen, 2013; Morjan, 2003a). Specifically, thermal conditions experienced by embryos determine gonadal sex in most turtle species (i.e. temperature‐dependent sex determination, TSD; Bull & Vogt, 1979; Valenzuela & Lance, 2004). Moreover, thermal conditions are strongly influenced by female choice of nest location, including overstory canopy cover (Janzen, 1994a; Mitchell, Warner, et al., 2013; Refsnider, Warner, & Janzen, 2013; Weisrock & Janzen, 1999), nest depth (Riley, Freedberg, & Litzgus, 2014; Telemeco et al., 2016; but see Refsnider, Bodensteiner, Reneker, & Janzen, 2013) and soil composition (Hays et al., 2001; Mitchell & Janzen, 2019; Tornabene, Bramblett, Zale, & Leathe, 2018).

Our interest in sex ratio evolution yielded a 30‐year study of the nesting ecology of painted turtles (Chrysemys picta) that enabled us to address the following questions: 1) Are indicators of nesting behaviours (i.e. canopy cover, distance to water, nest depth and soil type) repeatable? 2) To what extent is the repeatability of nest‐site choice conditional on maternal age (a la Wilson, 2018)? 3) Specifically, does among‐ or within‐individual variation in nest‐site choice shift with age?

METHODS

We collected data on the nesting ecology of C. picta from 1989 to 2018 at the Thomson Causeway Recreation Area along the Mississippi River in north‐west Illinois (Refsnider & Janzen, 2016; Warner, Miller, Bronikowski, & Janzen, 2016). We visually located turtles by transecting a nesting beach hourly from 0500 to 2000 hr each day from May through June. This protocol enabled us to sample nearly all nesting events occurring at this site for 30 years. After locating a nesting turtle, we observed from a distance, captured her after oviposition, measured carapace and plastron lengths and widths, uniquely and permanently notched marginal scutes for future identification and estimated age based on growth rings on the pectoral scute of the plastron and on previous encounter records (Hoekstra, Weber, Bronikowski, & Janzen, 2018). Female C. picta often mature at 5–7 years of age (Bowden, Harms, Paitz, & Janzen, 2004; Schwanz, Spencer, Bowden, & Janzen, 2010) and have a median lifespan of 10.6 years (Hoekstra et al., unpublished manuscript) at our site. Females reproduced every 1.5 (±1.3 SD) years and nested 1.4 (±0.5 SD) times per season in our data set. Although females may nest outside of our study area, most females that nest within our site do not nest elsewhere (Valenzuela & Janzen, 2001). Overall, we sampled 5,538 nesting events from 830 different females. Of these observations, we identified the age of 294 females which nested 1,969 times.

After processing a female, we excavated the nest, counted and weighed eggs, measured nest depth and then reconstructed the nest. We quantified overstory canopy cover with a spherical densiometer from 1995 to 2003 and 2006 to 2010 (e.g. Janzen, 1994a), and with hemispherical photography (processed with Gap Light Analyzer software (Cary Institute of Ecosystem Studies, Millbrook, New York)) from 2004 to 2005 and 2010 to 2018 (e.g. Mitchell, Maciel, et al., 2013). To estimate repeatability of canopy cover for all years simultaneously, we converted densiometer readings to % canopy cover by deriving an equation from nest sites for which we used both methods (R² = 0.63; sensu Schwanz et al., 2010; Refsnider, Milne‐Zelman, Warner, & Janzen, 2014):

Thus, two data sets existed for canopy cover: a reduced data set that only quantified canopy cover from hemispherical photographs and a full data set with additional observations converted from densiometer readings. The reduced data set has a lower potential for methodological bias because it only contains measurements from a single method, but the full data set has a much greater sample size and longer capture histories for individual females. For a subset of nesting events, we also recorded the substrate of the nest (entirely loam versus containing gravel) and how far females oviposited from water. We calculated distance to water using Cartesian coordinates and the programs INTERPNT (Boose, Boose, & Lezberg, 1998; as in Kolbe & Janzen, 2002) and ArcView (ESRI Inc. 1998) from 1988 to 2010 and with a laser rangefinder thereafter (Nikon Aculon).

Analysis

We conducted all analyses in R (version 3.6.0). All mixed models contained female identity as a random effect. We first assessed if age explained a significant amount of variation for the dependent variables canopy cover (full and reduced data sets), distance to water and soil type with linear mixed models. We used plastron length (instead of age) as a covariate for the analysis of nest depth because body size limits how deep females can construct nests (Morjan, 2003b; Refsnider et al., 2014). We also plotted covariates and dependent variables to verify linear relationships. Soil type had a slightly cubic distribution across age (Figure 1d), so we tested the fit of MCMCglmm models (discussed below) with a quadratic (DIC = 521.86) and cubic relationship (DIC = 521.41); however, the linear model (DIC = 521.53) fit similarly so we modelled age linearly in all subsequent models. We estimated repeatability by modelling female identity as a random effect in mixed models (Nakagawa & Schielzeth, 2010).

To assess whether the among‐ and within‐individual variance of nesting behaviours change with age, we first quantified repeatability and variance components for early‐ and late‐life nesting events using the rptR package (Stoffel, Nakagawa, & Schielzeth, 2017). For these analyses, we restricted the data set to known‐age females with ≥10 nesting events. Then, for each female, we binned her nesting events as either happening during early (5–10 years old) or late (>10 years old) life and further restricted the data set to females with ≥3 nesting events in each age bin. We chose this age for separation because doing so resulted in the most balanced sample sizes in age bins, the median age of females at our site is 10.6 years old, and females older than 10 were assured to be in the asymptotic growth phase (Hoekstra et al., 2018). We then estimated the repeatability and variances of each continuous nest‐site choice trait for both age bins. We report 95% confidence intervals for parameter estimates derived from rptR. We interpret estimates as significantly different from each other if their 1‐armed confidence intervals overlap by less than 50% (i.e. p < .05, Cumming, 2009; Kok et al., 2019).

Next, we evaluated whether among‐individual variation in nesting behaviour shifts continuously across age with hierarchical random regression analyses in the package MCMCglmm (Dingemanse, Kazem, Réale, & Wright, 2010; Hadfield, 2010). All covariates were mean‐centred and expressed in units of variance. We constructed a basic model of:

Next, we built a model by adding female identity as a random effect, which tested whether females differ in mean nesting behaviour. We then constructed a model that allowed the slope of each female to vary (i.e. idh(age):female ID), which tested whether females differ in how their behaviour shifts with age. Finally, we constructed a model that allowed slope–intercept covariation (i.e. us(age):female ID), which tested whether an individual's mean behaviour is related to the slope of that behaviour across age. Because age was mean‐centred in our analyses, a positive slope–intercept covariation indicates individuals become more divergent in the behaviour with age, whereas a negative covariation indicates that individuals become more similar with age. We used the model priors suggested for each model structure in the MCMCglmm course notes and evaluated model convergence by examining plots of the posterior distributions (Hadfield, 2010, 2015); all models passed this assessment. We report the posterior distribution mode and their 95% credible intervals for each variance component derived from MCMCglmm. We evaluated the significance of each additional term by comparing model DICs and evaluating credible intervals. We also constructed models with no covariate (i.e. intercept models) to examine how accounting for age‐related variation affected variance estimates.

RESULTS

All nesting behaviours varied with female age. Specifically, canopy cover and distance to water increased with age, whereas the frequency of loam nests decreased with age (Table 1; Figure 1). Nest depth increased with plastron length (Table 1).

All repeatability data sets were robust and ranged from an average of 3.9 nests for 115 individuals (soil type data set) up to an average of 6.8 nests for 291 individuals (full canopy cover data set; Table 2). Some variation in data coverage exists because not all variables were recorded for every nest over the duration of the study. All nesting behaviours were repeatable (R) in the intercept models and ranged from R = 0.149 for nest depth up to R = 0.607 (95% credible interval = 0.346–0.768) for soil type (Table 2; Figure S1).

Effects of age on repeatability

Accounting for variation due to age increased the estimates of repeatability for canopy cover from R = 0.291 (0.237–0.341) to R = 0.381 (0.330–0.448) for the full data set (Table 2; Figure S1). Although not statistically significant, the trend was similar for canopy cover in the reduced data set. The estimates of repeatability for distance to water and soil type did not change after accounting for variation due to age, whereas the repeatability estimate for nest depth dropped from R = 0.149 (0.119–0.187) to R = 0.091 (0.070–0.125) after accounting for variation explained by plastron length.

The average number of nests per female per life stage ranged from 5.5 to 8.4 for the repeatability comparisons between early‐ versus late‐life nesting events (Table 3; Figure 2). Although not statistically different from each other, the repeatability of canopy cover in the full data set was greater for females later in life (R = 0.408 (0.294–0.516)) than for females early in life (R = 0.293 (0.182–0.408), p > .05). This increase in repeatability was driven by a nonsignificant increase in the variance among individuals (53.2 (28.9–83.6) to 78.5 (45.5–118.64), p > .05) in combination with a nonsignificant decrease in within‐individual variance (128.4 (109.1–148.2) to 114.0 (98.9–129.4), p > .05). Repeatabilities of canopy cover in the reduced data set, distance to water and nest depth did not vary between age classes (Table 3). Despite no change in repeatability, the within‐individual variance of canopy cover decreased with age in the reduced data set (Table 3).

The random regression results first indicate that female ID accounts for significant variation in all dependent variables (Table 4). Model support for random slopes suggests females also differed in how their behaviours changed with age for all dependent variables, but to a lesser extent for soil type (Table 4; Figure 3). Slope–intercept covariation was only strongly supported in the full canopy cover data set and weakly so (≤ delta DIC of 2.2) for canopy cover in the reduced data set, distance to water and soil type. Moreover, only the 95% credible interval for slope–intercept covariation in the full canopy cover data set did not overlap zero (Table 2). Specifically, the positive slope–intercept covariation for canopy cover indicates that females became more divergent in their choice of canopy cover with age (i.e. among‐individual variation increased with age).

DISCUSSION

The quantification of repeatability is a powerful tool that has enabled researchers to assess the evolutionary potential of behavioural traits. Yet, several questions have been limited in their assessment because of the difficulty in amassing more than a few observations per individual. Such is especially true for wild, long‐lived animals and for traits that may occur infrequently each season (e.g. reproductive behaviours). Here, we leveraged a long‐term study of the nesting ecology of Chrysemys picta to assess how the among‐ and within‐individual variance of nesting behaviours vary across life. We show that four components of nest‐site choice were repeatable and that accounting for age notably increased the repeatability estimate of canopy cover. Female choice of canopy cover canalized from early‐ to late‐life nesting events in the reduced data set despite no change in repeatability. Lastly, the random regressions show that individuals became more divergent in their choice of canopy cover with age.

Age explained a substantial amount of variation in each component of nest‐site choice, and plastron length covaried with nest depth. Specifically, females increased canopy cover, distance to water, nest depth and the frequency of gravel in nests with age and size. Such nest‐site variation with maternal age has been found in smaller subsets of data from our site (canopy cover—Warner et al., 2016; distance to water—Harms, Paitz, Bowden, & Janzen, 2005; Delaney & Janzen, 2020; nest depth—Morjan, 2003b). But why might nesting behaviour change predictably with age? Young females likely nest close to shore to reduce their own risk of predation by terrestrial predators (Delaney & Janzen, 2020; Harms et al., 2005; Tucker, Filoramo, & Janzen, 1999; but see Refsnider, Reedy, Warner, & Janzen, 2015). However, nests laid closer to shore tend to experience elevated predation rates because nest predators (such as raccoons) tend to forage along shorelines and other environmental edges (Kolbe & Janzen, 2002; Spencer, 2002; Strickland, Colbert, & Janzen, 2010). Thus, females increase the distance nests are laid from water to increase nest success as they age and the likelihood of future reproductive opportunities decreases (discussed further in Delaney & Janzen, 2019). Why canopy cover increases with maternal age is unclear. Variation in canopy cover exists across distance to water at our study site (Janzen & Morjan, 2001), so this finding is not likely explained by covariation with distance to water. In fact, choice of canopy cover is under selection because cover indirectly determines clutch sex ratio and neonatal survival in nests (via thermal conditions; Mitchell, Maciel, et al., 2013). Such increasing canopy cover with age should result in producing more males later in life in C. picta because the frequency of embryos that develop into males increases as incubation temperature decreases (Janzen, 1994a, 1994b; Refsnider & Janzen, 2016). Bigger (and older) females likely construct deeper nests because they have longer rear legs (Morjan, 2003b), and deeper nests should be buffered from environmental extremes more than shallower nests (e.g. thermal, Telemeco et al., 2016). Lastly, females increased the frequency of nests that contained gravel as they aged. Gravel does not occur within 15 m of shore at our site and only exists along anthropogenic structures. Thus, the frequency of gravel nests increases with age because the opportunity to nest in such substrates increases as older females nest farther from the shore. Whether female preference for substrate changes with age must be evaluated experimentally at our site. Nevertheless, gravel nests tend to be warmer and drier (Mitchell & Janzen, 2019) and experience lower predation rates compared to loam nests (Hoekstra et al., unpublished manuscript).

All aspects of nest‐site choice were repeatable behaviours in our study. Previous work has shown that the choice of canopy cover is repeatable in C. picta. In fact, earlier analysis of smaller samples from our site revealed repeatability estimates of 0.18 (females with ≥ 2 nests, n = 79), 0.21 (females with ≥ 3 nests, n = 40; Janzen & Morjan, 2001) and 0.14 (n = 631, McGaugh, Schwanz, Bowden, Gonzalez, & Janzen, 2010) for canopy cover, which are similar to our intercept model estimates of 0.21 (reduced data set, n = 202) and 0.29 (full data set, n = 291). This result should not be surprising given that a variety of nest‐site choice components is repeatable for reptiles with TSD (canopy cover, geographic location, distance to water and vegetation, and nest temperature, reviewed in Janzen et al., 2019; Patricio et al., 2018), fish (nest size and composition, Rushbrook, Dingemanse, & Barber, 2008; nest size, Japoshvili, Lehtonen, Wong, & Lindstrom, 2012) and a bird (nest composition, Mennerat, Perret, & Lambrechts, 2009). However, we add that distance to water is also a repeatable behaviour in C. picta, and we show for the first time in any ground‐nesting animal that nest depth and substrate can be repeatable components of nest‐site choice. We also document that females differed in how their behaviours shifted across life, suggesting these age‐related reaction norms could be targets of selection.

Accounting for trait variation due to age increased estimates of repeatability in our study. Age has also been an important covariate when assessing the repeatability of traits in other systems (Araya‐Ajoy & Dingemanse, 2017; Debeffe et al., 2015; Douhard et al., 2017; Fisher et al., 2015; Snowder, Stellflug, & Van Vleck, 2002;). Ignoring the effects of age on trait variation could alter the repeatability estimate of interest (e.g. upper bound to heritability). For example, if females change their preference for nest sites as they age, among‐individual variance may be driven by sampling individuals of different ages rather than genetic differences in preference. Similarly, if individuals are sampled over a wide age range, within‐individual variance will be higher if age is not accounted for. A similar phenomenon occurs when the amount of time between observations is ignored (Biro & Stamps, 2015), with time generally decreasing consistency in individual behaviour and lowering repeatabilities (Bell et al., 2009). Accounting for age in our study reduced the within‐individual variance of canopy cover, whereas among‐individual variance remained similar, thus elevating repeatability. Interestingly, accounting for plastron length in the analysis of nest depth caused both within‐ and among‐individual variance to decrease, but the among‐individual decrease was sufficient to reduce the repeatability estimate. In particular, the among‐individual variance decreased because plastron length varies among individuals, especially for individuals that were sampled at different ranges of their lifespan. Nevertheless, accounting for age during the estimation of repeatability shifted the amount of variation explained within and among individuals enough to significantly alter trait repeatability.

To test how behaviour varies among and within individuals across age, researchers either split observations into discrete categories (e.g. juvenile versus adult) and test whether repeatability varies across life stages (Bell et al., 2009; Kok et al., 2019) or examine how among‐ and within‐individual variation shift across age using a random regression framework (Araya‐Ajoy & Dingemanse, 2017; Araya‐Ajoy et al., 2015; Charmantier et al., 2014; Class et al., 2019; Dingemanse et al., 2010; Fisher et al., 2015). We found that the repeatability of canopy cover increased 11.5% from early‐ to late‐life nesting events in the full data set and that individuals became more divergent in their choice of canopy cover with age (i.e. positive slope–intercept covariation). When repeatabilities vary across life stages in wild animals, they tend to increase with age, which could be due to trait canalization with age (reduction in within‐individual variance) or diverging developmental trajectories (increasing among‐individual variance; reviewed in Kok et al., 2019). Yet, the mechanism of variation in repeatabilities is rarely evaluated statistically. However, an increase in the repeatability of personality traits in mosquitofish (Gambusia holbrooki) was driven by a decline in within‐individual variance (Polverino, Cigliano, Nakayama, & Mehner, 2016). In addition, increased repeatability of gizzard mass in red knots (Calidris canutus) was driven by an increase in among‐individual variance and a reduction in within‐individual variance from juvenile to adult life stages (Kok et al., 2019), which may be expected for morphological traits. In our study, the increase in choice of canopy cover repeatability was driven by a nonsignificant increase in among‐individual variation and a nonsignificant decrease in within‐individual variation with age. Our random regression analyses, however, found that individuals significantly diverged in their choice of canopy cover with age. Moreover, although we found no change in the repeatability of canopy cover in the reduced data set, we detected a decrease in within‐individual variation. Thus, focusing on trait repeatability without also evaluating each variance component could cause researchers to miss interesting variation and leave potential mechanisms to speculation.

A few important considerations of microevolutionary potential exist when traits and their consistency vary across age. First, when age is used as a covariate, the estimation of repeatability is of the intercept of the age‐related change in the trait rather than simply of the trait. Second, the heritability of a trait, for which repeatability sets an upper bound, may vary across ontogeny (Wilson & Réale, 2006). For example, heritability can shift if the environment affects phenotypic expression differentially across life (Réale, Festa‐Bianchet, & Jorgenson, 1999; Wilson, Kruuk, & Coltman, 2005). In addition, a trait can be influenced by different genes or differential gene expression throughout ontogeny (Vinuela, Snoek, Riksen, & Kammenga, 2010; Yao et al., 2014). Thus, the complexity of trait expression across life should be considered when interpreting the repeatability and heritability of traits recorded over lifespans. Lastly, selection acts on the phenotypes expressed at each age, but aggregate evolutionary responses to selection depend on lifetime reproductive success. Thus, age‐related traits may provide inefficient targets for selection because individuals that die early will not express late‐life phenotypes (Abrams, 1993).

Our findings also have substantial implications for how species with TSD and other at‐risk species may respond to rapid environmental change. Heritability estimates for TSD‐related traits tend to be low (Janzen et al., 2019), but we show that accounting for age, just like accounting for winter warmth (McGaugh et al., 2010), increased the repeatability of canopy cover in C. picta. Thus, the heritable potential to respond to climate change may be greater than anticipated for at least this one species with TSD. Additionally, the repeatability of canopy cover increased with age in C. picta. Choices of canopy cover expressed by older turtles therefore provide more efficient targets for selection, although turtles must live to older ages to express late‐life phenotypes.

The quantification of repeatability has been broadly implemented by researchers to assess the heritable potential of traits (Bell et al., 2009; Falconer & Mackay, 1996; Wilson, 2018). We show that four components of nesting behaviour are repeatable in C. picta, and accounting for age‐related variation in nest‐site choice altered repeatability in two components of behaviour. We found that female choice of canopy cover canalized from early to late life in the reduced data set despite no change to repeatability. Lastly, our random regression analysis shows that individuals became more divergent in their choice of canopy cover across ontogeny. Future work in evolutionary ecology that examines each variance component, rather than simply repeatability (ratio of the variances) (sensu Dochtermann & Royauté, 2019; Jenkins, 2011; Niemelä, Niehoff, Gasparini, Dingemanse, & Tuni, 2019; Tüzün, Müller, Kock, & Stoks, 2017), offers a mechanistic understanding of variation in repeatability and can uncover trait variation even in the absence of repeatability shifts.

ACKNOWLEDGMENTS

We thank the many students and researchers that collected field data at Turtle Camp over the past three decades. We thank two anonymous reviewers for comments that improved the manuscript. We are grateful to Illinois Department of Natural Resources, United States Army Corps of Engineers and United States Fish and Wildlife Service for permits and access to the field site. NSF grants BSR‐8914686, DEB‐9629529, DEB‐0089680, DEB‐0640932 and DEB‐1242510 to FJJ funded this research. The fieldwork was approved by the Iowa State Institutional Animal Care and Use Committee, most recently protocol #12‐03‐5570‐J. This is Kellogg Biological Station contribution number 2160.

Peer Review

The peer review history for this article is available at https://publons.com/publon/10.1111/jeb.13701.

DATA AVAILABILITY STATEMENT

Data available on Dryad at https://doi.org/10.5061/dryad.70rxwdbvq

REFERENCES