Power and potential bias in field studies of natural selection

The advent of multiple regression analyses of natural selection has facilitated estimates of both the direct and indirect effects of selection on many traits in numerous organisms. However, low power in selection studies has possibly led to a bias in our assessment of the levels of selection shaping natural populations. Using calculations and simulations based on the statistical properties of selection coefficients, we find that power to detect total selection (the selection differential) depends on sample size and the strength of selection relative to the opportunity of selection. The power of detecting direct selection (selection gradients) is more complicated and depends on the relationship between the correlation of each trait and fitness and the pattern of correlation among traits. In a review of 298 previously published selection differentials, we find that most studies have had insufficient power to detect reported levels of selection acting on traits and that, in general, the power of detecting weak levels of selection is low given current study designs. We also find that potential publication bias could explain the trend that reported levels of direct selection tend to decrease as study sizes increase, suggesting that current views of the strength of selection may be inaccurate and biased upward. We suggest that studies should be designed so that selection is analyzed on at least several hundred individuals, the total opportunity of selection be considered along with the pattern of selection on individual traits, and nonsignificant results be actively reported combined with an estimate of power.     

Testing for environmentally induced bias in phenotypic estimates of natural selection: theory and practice

Measuring natural selection has been a fundamental goal of evolutionary biology for more than a century, and techniques developed in the last 20 yr have provided relatively simple means for biologists to do so. Many of these techniques, however, share a common limitation: when applied to phenotypic data, environmentally induced covariances between traits and fitness can lead to biased estimates of selection and misleading predictions about evolutionary change. Utilizing estimates of breeding values instead of phenotypic data with these methods can eliminate environmentally induced bias, although this approach is more difficult to implement. Despite this potential limitation to phenotypic methods and the availability of a potential solution, little empirical evidence exists on the extent of environmentally induced bias in phenotypic estimates of selection. In this article, we present a method for detecting bias in phenotypic estimates of selection and demonstrate its use with three independent data sets. Nearly 25% of the phenotypic selection gradients estimated from our data are biased by environmental covariances. We find that bias caused by environmental covariances appears mainly to affect quantitative estimates of the strength of selection based on phenotypic data and that the magnitude of these biases is large. As our estimates of selection are based on data from spatially replicated field experiments, we suggest that our findings on the prevalence of bias caused by environmental covariances are likely to be conservative.     

COMPARING STRENGTHS OF DIRECTIONAL SELECTION: HOW STRONG IS STRONG?

The fundamental equation in evolutionary quantitative genetics, the Lande equation, describes the response to directional selection as a product of the additive genetic variance and the selection gradient of trait value on relative fitness. Comparisons of both genetic variances and selection gradients across traits or populations require standardization, as both are scale dependent. The Lande equation can be standardized in two ways. Standardizing by the variance of the selected trait yields the response in units of standard deviation as the product of the heritability and the variance-standardized selection gradient. This standardization conflates selection and variation because the phenotypic variance is a function of the genetic variance. Alternatively, one can standardize the Lande equation using the trait mean, yielding the proportional response to selection as the product of the squared coefficient of additive genetic variance and the mean-standardized selection gradient. Mean-standardized selection gradients are particularly useful for summarizing the strength of selection because the mean-standardized gradient for fitness itself is one, a convenient benchmark for strong selection. We review published estimates of directional selection in natural populations using mean-standardized selection gradients. Only 38 published studies provided all the necessary information for calculation of mean-standardized gradients. The median absolute value of multivariate mean-standardized gradients shows that selection is on average 54% as strong as selection on fitness. Correcting for the upward bias introduced by taking absolute values lowers the median to 31%, still very strong selection. Such large estimates clearly cannot be representative of selection on all traits. Some possible sources of overestimation of the strength of selection include confounding environmental and genotypic effects on fitness, the use of fitness components as proxies for fitness, and biases in publication or choice of traits to study.     

The strength of phenotypic selection in natural populations

How strong is phenotypic selection on quantitative traits in the wild? We reviewed the literature from 1984 through 1997 for studies that estimated the strength of linear and quadratic selection in terms of standardized selection gradients or differentials on natural variation in quantitative traits for field populations. We tabulated 63 published studies of 62 species that reported over 2,500 estimates of linear or quadratic selection. More than 80% of the estimates were for morphological traits; there is very little data for behavioral or physiological traits. Most published selection studies were unreplicated and had sample sizes below 135 individuals, resulting in low statistical power to detect selection of the magnitude typically reported for natural populations. The absolute values of linear selection gradients |beta| were exponentially distributed with an overall median of 0.16, suggesting that strong directional selection was uncommon. The values of |beta| for selection on morphological and on life-history/phenological traits were significantly different: on average, selection on morphology was stronger than selection on phenology/life history. Similarly, the values of |beta| for selection via aspects of survival, fecundity, and mating success were significantly different: on average, selection on mating success was stronger than on survival. Comparisons of estimated linear selection gradients and differentials suggest that indirect components of phenotypic selection were usually modest relative to direct components. The absolute values of quadratic selection gradients |gamma| were exponentially distributed with an overall median of only 0.10, suggesting that quadratic selection is typically quite weak. The distribution of gamma values was symmetric about 0, providing no evidence that stabilizing selection is stronger or more common than disruptive selection in nature.     

Synthetic analyses of phenotypic selection in natural populations: lessons, limitations and future directions

There are now thousands of estimates of phenotypic selection in natural populations, resulting in multiple synthetic reviews of these data. Here we consider several major lessons and limitations emerging from these syntheses, and how they may guide future studies of selection in the wild. First, we review past analyses of the patterns of directional selection. We present new meta-analyses that confirm differences in the direction and magnitude of selection for different types of traits and fitness components. Second, we describe patterns of temporal and spatial variation in directional selection, and their implications for cumulative selection and directional evolution. Meta-analyses suggest that sampling error contributes importantly to observed temporal variation in selection, and indicate that evidence for frequent temporal changes in the direction of selection in natural populations is limited. Third, we review the apparent lack of evidence for widespread stabilizing selection, and discuss biological and methodological explanations for this pattern. Finally, we describe how sampling error, statistical biases, choice of traits, fitness measures and selection metrics, environmental covariance and other factors may limit the inferences we can draw from analyses of selection coefficients. Current standardized selection metrics based on simple parametric statistical models may be inadequate for understanding patterns of non-linear selection and complex fitness surfaces. We highlight three promising areas for expanding our understanding of selection in the wild: (1) field studies of stabilizing selection, selection on physiological and behavioral traits, and the ecological causes of selection; (2) new statistical models and methods that connect phenotypic variation to population demography and selection; and (3) availability of the underlying individual-level data sets from past and future selection studies, which will allow comprehensive modeling of selection and fitness variation within and across systems, rather than meta-analyses of standardized selection metrics.     

Reducing environmental bias when measuring natural selection

Abstract.— Crucial to understanding the process of natural selection is characterizing phenotypic selection. Measures of phenotypic selection can be biased by environmental variation among individuals that causes a spurious correlation between a trait and fitness. One solution is analyzing genotypic data, rather than phenotypic data. Genotypic data, however, are difficult to gather, can be gathered from few species, and typically have low statistical power. Environmental correlations may act through traits other than through fitness itself. A path analytic framework, which includes measures of such traits, may reduce environmental bias in estimates of selection coefficients. We tested the efficacy of path analysis to reduce bias by re-analyzing three experiments where both phenotypic and genotypic data were available. All three consisted of plant species (Impatiens capensis, Arabidopsis thaliana , and Raphanus sativus) grown in experimental plots or the greenhouse. We found that selection coefficients estimated by path analysis using phenotypic data were highly correlated with those based on genotypic data with little systematic bias in estimating the strength of selection. Although not a panacea, using path analysis can substantially reduce environmental biases in estimates of selection coefficients. Such confidence in phenotypic selection estimates is critical for progress in the study of natural selection.     

Measuring Natural Selection on Proportional Traits: Comparisons of Three Types of Selection Estimates for Resistance and Susceptibility to Herbivore Damage

To compare the strength of natural selection on different traits and in different species, evolutionary biologists typically estimate selection differentials and gradients in standardized units. Measuring selection differentials and gradients in standard deviation units or mean-standardized units facilitates such comparisons by converting estimates with potentially varied units to a common scale. In this note, I compare the performance of variance- and mean-standardized selection differentials and gradients for a unique and biologically important class of traits: proportional traits, that can only vary between zero and one, and their complements (1 minus the trait) using simple algebra and analysis of data from a field-study using morning glories. There is a systematic, mathematical relationship between unstandardized and variance-standardized selection gradients for proportional traits and their complements, but such a general relationship is lacking for mean-standardized gradients, potentially leading investigators to mistakenly conclude that a proportional change in a trait would have little effect on fitness. Despite this potential limitation, mean-standardized selection differentials and gradients represent a useful tool for studying natural selection on proportional traits, because by definition they measure how proportional changes in the mean of a trait lead to proportional changes in relative fitness.Co-ordinating editor: I. Olivieri     

THE MEASUREMENT OF SELECTION ON QUANTITATIVE TRAITS: BIASES DUE TO ENVIRONMENTAL COVARIANCES BETWEEN TRAITS AND FITNESS

The use of regression techniques for estimating the direction and magnitude of selection from measurements on phenotypes has become widespread in field studies. A potential problem with these techniques is that environmental correlations between fitness and the traits examined may produce biased estimates of selection gradients. This report demonstrates that the phenotypic covariance between fitness and a trait, used as an estimate of the selection differential in estimating selection gradients, has two components: a component induced by selection itself and a component due to the effect of environmental factors on fitness. The second component is shown to be responsible for biases in estimates of selection gradients. The use of regressions involving genotypic and breeding values instead of phenotypic values can yield estimates of selection gradients that are not biased by environmental covariances. Statistical methods for estimating the coefficients of such regressions, and for testing for biases in regressions involving phenotypic values, are described.     

It's about time: the temporal dynamics of phenotypic selection in the wild

Selection is a central process in nature. Although our understanding of the strength and form of selection has increased, a general understanding of the temporal dynamics of selection in nature is lacking. Here, we assembled a database of temporal replicates of selection from studies of wild populations to synthesize what we do (and do not) know about the temporal dynamics of selection. Our database contains 5519 estimates of selection from 89 studies, including estimates of both direct and indirect selection as well as linear and nonlinear selection. Morphological traits and studies focused on vertebrates were well-represented, with other traits and taxonomic groups less well-represented. Overall, three major features characterize the temporal dynamics of selection. First, the strength of selection often varies considerably from year to year, although random sampling error of selection coefficients may impose bias in estimates of the magnitude of such variation. Second, changes in the direction of selection are frequent. Third, changes in the form of selection are likely common, but harder to quantify. Although few studies have identified causal mechanisms underlying temporal variation in the strength, direction and form of selection, variation in environmental conditions driven by climatic fluctuations appear to be common and important.     

Below-threshold mortality: implications for studies in evolution,ecology and demography

Evolutionary biologists, ecologists and experimental gerontologists have increasingly used estimates of age-specific mortality as a critical component in studies of a range of important biological processes. However, the analysis of age-specific mortality rates is plagued by specific statistical challenges caused by sampling error. Here we discuss the nature of this ‘demographic sampling error’, and the way in which it can bias our estimates of (1) rates of ageing, (2) age at onset of senescence, (3) costs of reproduction and (4) demographic tests of evolutionary models of ageing. We conducted simulations which suggest that using standard statistical techniques, we would need sample sizes on the order of tens of thousands in most experiments to effectively remove any bias due to sampling error. We argue that biologists should use much larger sample sizes than have previously been used. However, we also present simple maximum likelihood models that effectively remove biases due to demographic sampling error even at relatively small sample sizes.     

Phenotypic selection in natural populations: what limits directional selection?

Studies of phenotypic selection document directional selection in many natural populations. What factors reduce total directional selection and the cumulative evolutionary responses to selection? We combine two data sets for phenotypic selection, representing more than 4,600 distinct estimates of selection from 143 studies, to evaluate the potential roles of fitness trade-offs, indirect (correlated) selection, temporally varying selection, and stabilizing selection for reducing net directional selection and cumulative responses to selection. We detected little evidence that trade-offs among different fitness components reduced total directional selection in most study systems. Comparisons of selection gradients and selection differentials suggest that correlated selection frequently reduced total selection on size but not on other types of traits. The direction of selection on a trait often changes over time in many temporally replicated studies, but these fluctuations have limited impact in reducing cumulative directional selection in most study systems. Analyses of quadratic selection gradients indicated stabilizing selection on body size in at least some studies but provided little evidence that stabilizing selection is more common than disruptive selection for most traits or study systems. Our analyses provide little evidence that fitness trade-offs, correlated selection, or stabilizing selection strongly constrains the directional selection reported for most quantitative traits.     

INFERRING FITNESS LANDSCAPES

Since 1983, study of natural selection has relied heavily on multiple regression of fitness on the values for a set of traits via ordinary least squares (OLSs), as proposed by Lande and Arnold, to obtain an estimate of the quadratic relationship between fitness and the traits, the fitness surface. However, well‐known statistical problems with this approach can affect inferences about selection. One key concern is that measures of lifetime fitness do not conform to a normal or any other standard sampling distribution, as needed to justify the usual statistical tests. Another is that OLS may yield an estimate of the sign of the fitness function's curvature that is opposite to the truth. We here show that the recently developed aster modeling approach, which explicitly models the components of fitness as the basis for inferences about lifetime fitness, eliminates these problems. We illustrate selection analysis via aster using simulated datasets involving five fitness components expressed in each of four years. We demonstrate that aster analysis yields accurate estimates of the fitness function in cases in which OLS misleads, as well as accurate confidence regions for directional selection gradients. Further, to evaluate selection when many traits are under consideration, we recommend model selection by information criteria and frequentist model averaging.     

Impacts of environmental heterogeneity on natural selection in a wild bird population*

Natural selection has been studied for several decades, resulting in the computation of thousands of selection estimates. Although the importance of environmental conditions on selection has often been suggested, published estimates rarely take into account the effects of environmental heterogeneity on selection patterns. Here, we estimated linear and nonlinear viability selection gradients on morphological traits of 12-day old nestlings in a wild population of tree swallows (Tachycineta bicolor) across a large-scale heterogeneous study system in southern Québec, Canada. We assessed the environmental drivers of nestling survival and evaluated their effects on strength and direction of selection gradients. Separate analyses of environmental variables showed that high temperatures and heavy rainfall caused stronger positive linear selection on morphological traits. Weaker linear selection was also measured in more extensively cultivated areas. Both strength and shape of nonlinear quadratic and correlational components of selection were modified by environmental variables. Considering all environmental variables revealed that precipitation since hatching affected patterns of linear selection on traits, while temperatures since hatching shaped nonlinear selection patterns. Our study underlines the importance of quantifying linear and nonlinear natural selection under various environmental conditions and how the evolutionary response of traits may be affected by ongoing human-induced environmental changes.     

Estimating infection prevalence: Best practices and their theoretical underpinnings

Accurately estimating infection prevalence is fundamental to the study of population health, disease dynamics, and infection risk factors. Prevalence is estimated as the proportion of infected individuals (“individual‐based estimation”), but is also estimated as the proportion of samples in which evidence of infection is detected (“anonymous estimation”). The latter method is often used when researchers lack information on individual host identity, which can occur during noninvasive sampling of wild populations or when the individual that produced a fecal sample is unknown. The goal of this study was to investigate biases in individual‐based versus anonymous prevalence estimation theoretically and to test whether mathematically derived predictions are evident in a comparative dataset of gastrointestinal helminth infections in nonhuman primates. Using a mathematical model, we predict that anonymous estimates of prevalence will be lower than individual‐based estimates when (a) samples from infected individuals do not always contain evidence of infection and/or (b) when false negatives occur. The mathematical model further predicts that no difference in bias should exist between anonymous estimation and individual‐based estimation when one sample is collected from each individual. Using data on helminth parasites of primates, we find that anonymous estimates of prevalence are significantly and substantially (12.17%) lower than individual‐based estimates of prevalence. We also observed that individual‐based estimates of prevalence from studies employing single sampling are on average 6.4% higher than anonymous estimates, suggesting a bias toward sampling infected individuals. We recommend that researchers use individual‐based study designs with repeated sampling of individuals to obtain the most accurate estimate of infection prevalence. Moreover, to ensure accurate interpretation of their results and to allow for prevalence estimates to be compared among studies, it is essential that authors explicitly describe their sampling designs and prevalence calculations in publications.     

A quantitative review of heterozygosity–fitness correlations in animal populations

The ease of obtaining genotypic data from wild populations has renewed interest in the relationship between individual genetic diversity and fitness-related traits (heterozygosity–fitness correlations, or HFC). Here we present a comprehensive meta-analysis of HFC studies using powerful multivariate techniques which account for nonindependence of data. We compare these findings with those from univariate techniques, and test the influence of a range of factors hypothesized to influence the strength of HFCs. We found small but significantly positive effect sizes for life-history, morphological, and physiological traits; while theory predicts higher mean effect sizes for life-history traits, effect size did not differ consistently with trait type. Newly proposed measures of variation were no more powerful at detecting relationships than multilocus heterozygosity, and populations predicted to have elevated inbreeding variance did not exhibit higher mean effect sizes. Finally, we found evidence for publication bias, with studies reporting weak, nonsignificant effects being under-represented in the literature. In general, our review shows that HFC studies do not generally reveal patterns predicted by population genetic theory, and are of small effect (less than 1% of the variance in phenotypic characters explained). Future studies should use more genetic marker data and utilize sampling designs that shed more light on the biological mechanisms that may modulate the strength of association, for example by contrasting the strength of HFCs in mainland and island populations of the same species, investigating the role of environmental stress, or by considering how selection has shaped the traits under investigation.     

Taxon sampling, correlated evolution, and independent contrasts

Independent contrasts are widely used to incorporate phylogenetic information into studies of continuous traits, particularly analyses of evolutionary trait correlations, but the effects of taxon sampling on these analyses have received little attention. In this paper, simulations were used to investigate the effects of taxon sampling patterns and alternative branch length assignments on the statistical performance of correlation coefficients and sign tests; "full-tree" analyses based on contrasts at all nodes and "paired-comparisons" based only on contrasts of terminal taxon pairs were also compared. The simulations showed that random samples, with respect to the traits under consideration, provide statistically robust estimates of trait correlations. However, exact significance tests are highly dependent on appropriate branch length information; equal branch lengths maintain lower Type I error than alternative topological approaches, and adjusted critical values of the independent contrast correlation coefficient are provided for use with equal branch lengths. Nonrandom samples, with respect to univariate or bivariate trait distributions, introduce discrepancies between interspecific and phylogenetically structured analyses and bias estimates of underlying evolutionary correlations. Examples of nonrandom sampling processes may include community assembly processes, convergent evolution under local adaptive pressures, selection of a nonrandom sample of species from a habitat or life-history group, or investigator bias. Correlation analyses based on species pairs comparisons, while ignoring deeper relationships, entail significant loss of statistical power and as a result provide a conservative test of trait associations. Paired comparisons in which species differ by a large amount in one trait, a method introduced in comparative plant ecology, have appropriate Type I error rates and high statistical power, but do not correctly estimate the magnitude of trait correlations. Sign tests, based on full-tree or paired-comparison approaches, are highly reliable across a wide range of sampling scenarios, in terms of Type I error rates, but have very low power. These results provide guidance for selecting species and applying comparative methods to optimize the performance of statistical tests of trait associations.     

Accounting for Sampling Error When Inferring Population Synchrony from Time-Series Data: A Bayesian State-Space Modelling Approach with Applications

Background

Data collected to inform time variations in natural population size are tainted by sampling error. Ignoring sampling error in population dynamics models induces bias in parameter estimators, e.g., density-dependence. In particular, when sampling errors are independent among populations, the classical estimator of the synchrony strength (zero-lag correlation) is biased downward. However, this bias is rarely taken into account in synchrony studies although it may lead to overemphasizing the role of intrinsic factors (e.g., dispersal) with respect to extrinsic factors (the Moran effect) in generating population synchrony as well as to underestimating the extinction risk of a metapopulation.

Methodology/Principal findings

The aim of this paper was first to illustrate the extent of the bias that can be encountered in empirical studies when sampling error is neglected. Second, we presented a space-state modelling approach that explicitly accounts for sampling error when quantifying population synchrony. Third, we exemplify our approach with datasets for which sampling variance (i) has been previously estimated, and (ii) has to be jointly estimated with population synchrony. Finally, we compared our results to those of a standard approach neglecting sampling variance. We showed that ignoring sampling variance can mask a synchrony pattern whatever its true value and that the common practice of averaging few replicates of population size estimates poorly performed at decreasing the bias of the classical estimator of the synchrony strength.

Conclusion/Significance

The state-space model used in this study provides a flexible way of accurately quantifying the strength of synchrony patterns from most population size data encountered in field studies, including over-dispersed count data. We provided a user-friendly R-program and a tutorial example to encourage further studies aiming at quantifying the strength of population synchrony to account for uncertainty in population size estimates.     

Interspecific patterns of phenotypic selection do not predict intraspecific patterns

We estimated linear (β) and nonlinear (γ) selection gradients to quantify host plant‐mediated selection on the trait gall size in each of 22 unequally sampled subpopulations of the cynipid gall wasp Belonocnema treatae. We characterized the relationship between variation in subpopulation sample size and the magnitude of and the variance among selection gradients. We then tested the hypothesis that the intraspecific patterns we observed would follow two patterns that have emerged from published estimates of linear and nonlinear selection gradients compiled across species, namely that the average magnitude of β and γ and the variance among estimated β and γ decrease with increasing sample size. For both β and γ, intraspecific patterns of phenotypic selection in relation to sample size were not predicted by interspecific patterns. Thus, our results suggest that when selection is heterogeneous among subpopulations, variation in the biological basis for selection is more important in influencing estimates of selection than is variation in study size. Our study highlights the value of inspecting selection in relation to sampling effort at the level at which understanding the sources of variation in selection is most important, among populations within species.     

ESTIMATING AND VISUALIZING FITNESS SURFACES USING MARK–RECAPTURE DATA

Understanding how selection operates on a set of phenotypic traits is central to evolutionary biology. Often, it requires estimating survival (or other fitness‐related life‐history traits) which can be difficult to obtain for natural populations because individuals cannot be exhaustively followed. To cope with this issue of imperfect detection, we advocate the use of mark‐recapture data and we provide a general framework for both the estimation of linear and nonlinear selection gradients and the visualization of fitness surfaces. To quantify the strength of selection, the standard second‐order polynomial regression method is integrated in mark‐recapture models. To visualize the form of selection, we use splines to display selection acting on multivariate phenotypes in the most flexible way. We employ Markov chain Monte Carlo sampling in a Bayesian framework to estimate model parameters, assessing traits relevance and calculating the optimal amount of smoothing. We illustrate our approach using data from a wild population of Common blackbirds (Turdus merula) to investigate survival in relation to morphological traits, and provide evidence for correlational selection using the new methodology. Overall, the framework we propose will help in exploring the full potential of mark‐recapture data to study natural selection.     

Effects of sample size on estimates of population growth rates calculated with matrix models

Background

Matrix models are widely used to study the dynamics and demography of populations. An important but overlooked issue is how the number of individuals sampled influences estimates of the population growth rate (λ) calculated with matrix models. Even unbiased estimates of vital rates do not ensure unbiased estimates of λ–Jensen''s Inequality implies that even when the estimates of the vital rates are accurate, small sample sizes lead to biased estimates of λ due to increased sampling variance. We investigated if sampling variability and the distribution of sampling effort among size classes lead to biases in estimates of λ.

Methodology/Principal Findings

Using data from a long-term field study of plant demography, we simulated the effects of sampling variance by drawing vital rates and calculating λ for increasingly larger populations drawn from a total population of 3842 plants. We then compared these estimates of λ with those based on the entire population and calculated the resulting bias. Finally, we conducted a review of the literature to determine the sample sizes typically used when parameterizing matrix models used to study plant demography.

Conclusions/Significance

We found significant bias at small sample sizes when survival was low (survival = 0.5), and that sampling with a more-realistic inverse J-shaped population structure exacerbated this bias. However our simulations also demonstrate that these biases rapidly become negligible with increasing sample sizes or as survival increases. For many of the sample sizes used in demographic studies, matrix models are probably robust to the biases resulting from sampling variance of vital rates. However, this conclusion may depend on the structure of populations or the distribution of sampling effort in ways that are unexplored. We suggest more intensive sampling of populations when individual survival is low and greater sampling of stages with high elasticities.