Analysis of a Generalized Linear Mixed Model: A Case Study and Simulation Results

A class of generalized linear mixed models can be obtained by introducing random effects in the linear predictor of a generalized linear model, e.g. a split plot model for binary data or count data. Maximum likelihood estimation, for normally distributed random effects, involves high-dimensional numerical integration, with severe limitations on the number and structure of the additional random effects. An alternative estimation procedure based on an extension of the iterative re-weighted least squares procedure for generalized linear models will be illustrated on a practical data set involving carcass classification of cattle. The data is analysed as overdispersed binomial proportions with fixed and random effects and associated components of variance on the logit scale. Estimates are obtained with standard software for normal data mixed models. Numerical restrictions pertain to the size of matrices to be inverted. This can be dealt with by absorption techniques familiar from e.g. mixed models in animal breeding. The final model fitted to the classification data includes four components of variance and a multiplicative overdispersion factor. Basically the estimation procedure is a combination of iterated least squares procedures and no full distributional assumptions are needed. A simulation study based on the classification data is presented. This includes a study of procedures for constructing confidence intervals and significance tests for fixed effects and components of variance. The simulation results increase confidence in the usefulness of the estimation procedure.     

On Testing an Unspecified Function Through a Linear Mixed Effects Model with Multiple Variance Components

Summary We examine a generalized F ‐test of a nonparametric function through penalized splines and a linear mixed effects model representation. With a mixed effects model representation of penalized splines, we imbed the test of an unspecified function into a test of some fixed effects and a variance component in a linear mixed effects model with nuisance variance components under the null. The procedure can be used to test a nonparametric function or varying‐coefficient with clustered data, compare two spline functions, test the significance of an unspecified function in an additive model with multiple components, and test a row or a column effect in a two‐way analysis of variance model. Through a spectral decomposition of the residual sum of squares, we provide a fast algorithm for computing the null distribution of the test, which significantly improves the computational efficiency over bootstrap. The spectral representation reveals a connection between the likelihood ratio test (LRT) in a multiple variance components model and a single component model. We examine our methods through simulations, where we show that the power of the generalized F ‐test may be higher than the LRT, depending on the hypothesis of interest and the true model under the alternative. We apply these methods to compute the genome‐wide critical value and p ‐value of a genetic association test in a genome‐wide association study (GWAS), where the usual bootstrap is computationally intensive (up to 10⁸ simulations) and asymptotic approximation may be unreliable and conservative.     

Linear Mixed Model Selection for False Discovery Rate Control in Microarray Data Analysis

Summary In a microarray experiment, one experimental design is used to obtain expression measures for all genes. One popular analysis method involves fitting the same linear mixed model for each gene, obtaining gene‐specific p‐values for tests of interest involving fixed effects, and then choosing a threshold for significance that is intended to control false discovery rate (FDR) at a desired level. When one or more random factors have zero variance components for some genes, the standard practice of fitting the same full linear mixed model for all genes can result in failure to control FDR. We propose a new method that combines results from the fit of full and selected linear mixed models to identify differentially expressed genes and provide FDR control at target levels when the true underlying random effects structure varies across genes.     

混合家系遗传参数估计方法研究

针对混合家系遗传参数估计,本文在假定公畜方差组分和母畜方差组分相等这一理论基础上,通过对方差分析的期望均方组成分析,提出了新的遗传力估计方法,以及某些特殊情况下的近似估计方法。通过一个估测实例比较了几种遗传力估计方法,结果表明,本文方法与全同胞组分估计最为接近,而且遗传力标准误最小,本文近似估计方法的效果也较好。对各种方法而言,资料越不平衡其差异越大。本文方法可以在一定程度上弥补全同胞分析时,因实际资料的公母畜方差组分差异过大的缺陷,具有实际可行性。此外,由于本文方法是用单因方差分析解决二因方差分析问题,计算更为简便,并可免于计算混合家系平均亲缘相关系数。     

具有亲缘关系矩阵的混合模型的方差组分迭代估计——应用于计算BLUP的方差比值和遗传参数

论述的是来自非均街资料的混合模型中具有亲缘关系矩阵时利用迭代法估计方差组分问题。这篇文章表明计算程序是可行的,只要能够按照混合模型中固定效应的结构矩阵和Henderson方法3的固定效应的假设条件正确地计算二次型约化平方和,就可获得较为精确的方差组分估计值;而且表明方差初始比值k偏高或偏低,不影响迭代求解的最后结果,这是因为在迭代过程中可以通过结构矩阵x'x和x'x的控制而自行调整。这些方差组分不仅可应用于选种种畜用的BLUP计算,还可用来估计遗传参数。     

THE EFFECT OF ENVIRONMENTAL VARIABILITY ON THE HERITABILITIES OF TRAITS OF A FIELD CRICKET

The presence of heritable variation in traits is a prerequisite for evolution. The great majority of heritability (h²) estimates are performed under laboratory conditions that are characterized by low levels of environmental variability. Very little is known about the effect of environmental variability on the estimation of components of quantitative variation, although theoretical extrapolations from lab studies have been attempted. Here we investigate the effects of environmental heterogeneity on variance component estimation using full-sib families of Gryllus pennsylvanicus split between a homogeneous laboratory environment and a more variable field environment. Although large standard errors prevent demonstration of statistically significant differences among h² of traits measured in the two environments for all but one trait, the values of h² are, on average, lower in the variable field environment, with a mean reduction of 19%. Developmental time is an exception, exhibiting high levels of additive variance in the field, leading to a higher value of h² in the variable environment. Underlying the lower field h² estimates are greater components of environmental variance as expected, as well as lower components of genetic variance. In this study, there is no evidence that the increase in the environmental component of variance in the field is any more important in the reduction of h² than is the decrease in the additive genetic component. The implications of the relative changes in the two components of variance are discussed.     

MEASURING THE EFFECTS OF PAIRING SUCCESS,EXTRA-PAIR COPULATIONS AND MATE QUALITY ON THE OPPORTUNITY FOR SEXUAL SELECTION

Sexual selection can act through variation in the number of social mates obtained, variation in mate quality, or variation in success at obtaining extra-pair fertilizations. Because within-pair fertilizations (WPF) and extra-pair fertilizations (EPF) are alternate routes of reproduction, they are additive, rather than multiplicative, components of fitness. We present a method for partitioning total variance in reproductive success (a measure of the opportunity for selection) when fitness components are both additive and multiplicative and use it to partition the variance into components that correspond to each mechanism of sexual selection. Computer simulations show that extra-pair fertilizations can either increase or decrease total variance, depending on the covariance between within-pair and extra-pair success. Simulations also suggest that for socially monogamous species, extra-pair fertilizations have a greater effect than variation in mate quality or pairing status on the opportunity for selection. Application of our model to data gathered for a population of red-winged blackbirds (Agelaius phoeniceus) indicates that most of the variance in male reproductive success was attributable to within-pair sources of variance. Nevertheless, extra-pair copulations increased the opportunity for selection because males varied both in the proportion of their social young that they sired and in the number of extra-pair mates that they obtained. Furthermore, large and positive covariances existed between the number of extra-pair mates a male obtained and both social pairing success and within-pair paternity, indicating that, in this population, males preferred as social mates also were preferred as extra-pair mates.     

Fixed‐effect variance and the estimation of repeatabilities and heritabilities: issues and solutions

Linear mixed‐effects models are frequently used for estimating quantitative genetic parameters, including the heritability, as well as the repeatability, of traits. Heritability acts as a filter that determines how efficiently phenotypic selection translates into evolutionary change, whereas repeatability informs us about the individual consistency of phenotypic traits. As quantities of biological interest, it is important that the denominator, the phenotypic variance in both cases, reflects the amount of phenotypic variance in the relevant ecological setting. The current practice of quantifying heritabilities and repeatabilities from mixed‐effects models frequently deprives their denominator of variance explained by fixed effects (often leading to upward bias of heritabilities and repeatabilities), and it has been suggested to omit fixed effects when estimating heritabilities in particular. We advocate an alternative option of fitting models incorporating all relevant effects, while including the variance explained by fixed effects into the estimation of the phenotypic variance. The approach is easily implemented and allows optimizing the estimation of phenotypic variance, for example by the exclusion of variance arising from experimental design effects while still including all biologically relevant sources of variation. We address the estimation and interpretation of heritabilities in situations in which potential covariates are themselves heritable traits of the organism. Furthermore, we discuss complications that arise in generalized and nonlinear mixed models with fixed effects. In these cases, the variance parameters on the data scale depend on the location of the intercept and hence on the scaling of the fixed effects. Integration over the biologically relevant range of fixed effects offers a preferred solution in those situations.     

EVOLUTIONARY PREDICTABILITY IN NATURAL POPULATIONS: DO MATING SYSTEM AND NONADDITIVE GENETIC VARIANCE INTERACT TO AFFECT HERITABILITIES IN PLANTAGO LANCEOLATA?

Quantitative genetics has been an immensely powerful tool in manipulating the phenotypes of domesticated plants and animals. Much of the predictive power of quantitative genetics depends on the breeder's control over the context in which phenotype and mating are being expressed. In the natural world, these contexts are often difficult to describe, let alone control. We are left, therefore, with a poor understanding of the limits of quantitative genetics in natural populations. One of the crucial contextual elements for assessing breeding value is the genetic background in which an individual's genes are being assessed. When interacting genes are polymorphic within a population, the degree of mating among relatives can influence the correlations among mates and the predictions of a response to selection. Population structure can strongly influence the degree to which dominance and epistasis influences additive genetic variance and heritability. The extent of inbreeding can also influence heritabilities through its effect on the environmental component of phenotypic variance. The applicability of standard quantitative genetic breeding designs to the measurement of heritabilities in natural populations therefore depends in part on: (1) the mating system of the population; and (2) the importance of gene interactions in determining phenotypic variation. We tested for an effect of mating structure on the partitioning of phenotypic variance and heritability by comparing two breeding designs in a common environment. Both breeding designs used 139 pollen parents taken from mapped locations in a population of Plantago lanceolata L., and crossed to 280 seed parents from the same population. One design was random-mating, the second was biased toward near-neighbor matings to an extent determined by field measure of pollen-mediated gene flow distances. The offspring were grown randomly mixed in a common garden. Nine traits were measured: central corm diameter, number of leaves, area of the most recently fully expanded leaf, density of hairs (cm^-2) on the leaves, dry weight per unit leaf area, photosynthetic capacity, transpiration rates, water use efficiency, and reproductive dry weight. Heritabilities and variance components from the two designs were compared using randomization tests. None of the variance components or the heritabilities differed significantly between breeding designs at the 0.05 level. The test could distinguish differences between the heritabilities measured in the two breeding designs as small as 0.11, on average. Thus, for the degree of inbreeding normally exhibited in P. lanceolata there is insufficient gene interaction present within populations to influence the partitioning of variance between additive and nonadditive components or to influence heritability estimates to a meaningful extent. We suggest that for Plantago other sources of variation in heritability estimates, such as maternal effects and genotype × environment interactions, are more important influences than the interaction between inbreeding and gene interactions, and standard heritability estimate based on random breeding is as accurate as one taking the natural mating structure into account.     

Genetic basis of between‐individual and within‐individual variance of docility

Between‐individual variation in phenotypes within a population is the basis of evolution. However, evolutionary and behavioural ecologists have mainly focused on estimating between‐individual variance in mean trait and neglected variation in within‐individual variance, or predictability of a trait. In fact, an important assumption of mixed‐effects models used to estimate between‐individual variance in mean traits is that within‐individual residual variance (predictability) is identical across individuals. Individual heterogeneity in the predictability of behaviours is a potentially important effect but rarely estimated and accounted for. We used 11 389 measures of docility behaviour from 1576 yellow‐bellied marmots (Marmota flaviventris) to estimate between‐individual variation in both mean docility and its predictability. We then implemented a double hierarchical animal model to decompose the variances of both mean trait and predictability into their environmental and genetic components. We found that individuals differed both in their docility and in their predictability of docility with a negative phenotypic covariance. We also found significant genetic variance for both mean docility and its predictability but no genetic covariance between the two. This analysis is one of the first to estimate the genetic basis of both mean trait and within‐individual variance in a wild population. Our results indicate that equal within‐individual variance should not be assumed. We demonstrate the evolutionary importance of the variation in the predictability of docility and illustrate potential bias in models ignoring variation in predictability. We conclude that the variability in the predictability of a trait should not be ignored, and present a coherent approach for its quantification.     

Estimation of Covariance Matrices in Unbalanced Random and Mixed Multivariate Models

Three HENDERSON'S Methods of estimating the variance components are generalized from one to p variables using a compact matrix notation. These results are obtained using a generalized Kronecker product of matrices, generalized trace of order p and a generalized quadratic form.     

Least Squares and Minimum MSE Estimators of Variance Components in Mixed Linear Models

The paper deals with the quadratic invariant estimators of the linear functions of variance components in mixed linear model. The estimator with locally minimal mean square error with respect to a parameter ? is derived. Under the condition of normality of the vector Y the theoretical values of MSE of several types of estimators are compared in two different mixed models; under a different types of distributions a simulation study is carried out for the behaviour of derived estimators.     

Natural selection on the genetical component of variance in body condition in a wild bird population

Although there is substantial evidence that skeletal measures of body size are heritable in wild animal populations, it is frequently assumed that the nonskeletal component of body weight (or ‘condition’) is determined primarily by environmental factors, in particular nutritional state. We tested this assumption by quantifying the genetic and environmental components of variance in fledgling body condition index (=relative body weight) in a natural population of collared flycatchers (Ficedula albicollis), and compared the strength of natural selection on individual breeding values with that on phenotypic values. A mixed model analysis of the components of variance, based on an ‘animal model’ and using 18 years of data on 17 717 nestlings, revealed a significant additive genetic component of variance in body condition, which corresponded to a narrow sense heritability (h²) of 0.30 (SE=0.03). Nongenetic contributions to variation in body condition were large, but there was no evidence of dominance variance nor of contributions from early maternal or common environment effects (pre‐manipulation environment) in condition at fledging. Comparison of pre‐ and post‐selection samples revealed virtually identical h² of body condition index, despite the fact that there was a significant decrease (35%) in the levels of additive genetic variance from fledging to breeding. The similar h² in the two samples occurred because the environmental component of variance was also reduced by selection, suggesting that natural selection was acting on both genotypic and environmental variation. The effects of selection on genetic variance were confirmed by calculation of the selection differentials for both phenotypic values and best linear unbiased predictor (BLUP) estimates of breeding values: there was positive directional selection on condition index both at the phenotypic and the genotypic level. The significant h² of body condition index is consistent with data from human and rodent populations showing significant additive genetic variance in relative body mass and adiposity, but contrasts with the common assumption in ecology that body condition reflects an individual’s nongenetic nutritional state. Furthermore, the substantial reduction in the additive genetic component of variance in body condition index suggests that selection on environmental deviations cannot alone explain the maintenance of additive genetic variation in heritable traits, but that other mechanisms are needed to explain the moderate to high heritabilities of traits under consistent and strong directional selection.     

Quantifying natural seasonal variation in mutation parameters with mutation accumulation lines

Mutations create novel genetic variants, but their contribution to variation in fitness and other phenotypes may depend on environmental conditions. Furthermore, natural environments may be highly heterogeneous. We assessed phenotypes associated with survival and reproductive success in over 30,000 plants representing 100 mutation accumulation lines of Arabidopsis thaliana across four temporal environments at a single field site. In each of the four assays, environmental variance was substantially larger than mutational variance. For some traits, whether mutational variance was significantly varied between seasons. The founder genotype had mean trait values near the mean of the distribution of the mutation accumulation lines in all field experiments. New mutations also contributed more phenotypic variation than would be predicted, given phenotypic and sequence‐level divergence among natural populations of A. thaliana. The combination of large environmental variance with a mean effect of mutation near zero suggests that mutations could contribute substantially to standing genetic variation.     

Does natural selection alter genetic architecture? An evaluation of quantitative genetic variation among populations of Allonemobiussocius and A. fasciatus

To make long-term predictions using present quantitative genetic theory it is necessary to assume that the genetic variance–covariance matrix ( G ) remains constant or at least changes by a constant fraction. In this paper we examine the stability of the genetic architecture of two traits known to be subject to natural selection; femur length and ovipositor length in two species of the cricket Allonemobius. Previous studies have shown that in A. fasciatus and A. socius natural selection favours an increased body size southwards but a decreased ovipositor length. Such countergradient selection should tend to favour a change in G . In the total sample of eight populations of A. socius and one of A. fasciatus we show that there is significant variation in all genetic covariance components, i.e. V_A for body size, V_A for ovipositor length, and Cov_A. This variation results entirely from an increase in the covariances of A. fasciatus. However, although larger, these components are approximately proportionally increased, thereby leading to no statistically significant change in the genetic correlation. A proportional increase in the covariance components is consistent with changes resulting from genetic drift. On the other hand, the genetic covariance components are significantly correlated with the length of the growing season suggesting that the change in the genetic architecture is the result of selection and drift.     

Can dominance genetic variance be ignored in evolutionary quantitative genetic analyses of wild populations?

Accurately estimating genetic variance components is important for studying evolution in the wild. Empirical work on domesticated and wild outbred populations suggests that dominance genetic variance represents a substantial part of genetic variance, and theoretical work predicts that ignoring dominance can inflate estimates of additive genetic variance. Whether this issue is pervasive in natural systems is unknown, because we lack estimates of dominance variance in wild populations obtained in situ. Here, we estimate dominance and additive genetic variance, maternal variance, and other sources of nongenetic variance in eight traits measured in over 9000 wild nestlings linked through a genetically resolved pedigree. We find that dominance variance, when estimable, does not statistically differ from zero and represents a modest amount (2-36%) of genetic variance. Simulations show that (1) inferences of all variance components for an average trait are unbiased; (2) the power to detect dominance variance is low; (3) ignoring dominance can mildly inflate additive genetic variance and heritability estimates but such inflation becomes substantial when maternal effects are also ignored. These findings hence suggest that dominance is a small source of phenotypic variance in the wild and highlight the importance of proper model construction for accurately estimating evolutionary potential.     

Analysis of asymmetries in the African fruit bats Eidolon helvum and Rousettus egyptiacus (Mammalia: Megachiroptera) from the islands of the Gulf of Guinea. I. Variance and size components of bilateral variation

A set of cranial characters was examined in the fruit bats Rousettus egyptiacus and Eidolon helvum to compare trends and relative importance of major components of bilateral morphometric variation, and their relationship with character size. Using two‐way, sides‐by‐individuals ANOVA , four components of variation were estimated for each bilateral variable: individual variation (I), directional asymmetry (DA), non‐directional asymmetry (NDA) and measurement error (E). Both species exhibit similar major trends of variation in asymmetry across characters, as shown by principal component analysis, using variance components as variables. Degree of interspecific congruence among characters was confirmed by a two‐way ANOVA with species and variance components as fixed factors. Congruence of asymmetry patterns between species suggests that the concept of population asymmetry parameter (PAP) could be extended to higher hierarchies. PAPs above the species level may result from common mechanisms or similar developmental constraints acting on species’ buffering capacities and morphological integration processes.     

Partitioning of population genetic variance under multiplicative-epistatic gene action

Determining the way in which different QTLs interact (epistasis) in their effects on the phenotype is crucial to many areas in population genetics and evolutionary biology. For example, in the founder event, a separated population readapts to a new environment through the release of cryptic gene-gene interactions. In hybrid zones, hybrid speciation must be subjected to natural selection for epistasis resulting from genomic recombinations between different species. However, there is a severe shortage of relevant methodologies to estimate epistatic genetic effects and variances. A statistical model has recently been proposed to estimate the number of QTLs, their genetic effects and allelic frequencies in segregating populations. This model is based on multiplicative gene action and derived from a two-level intra- and interspecific mating design. In this paper, we formulate a statistical procedure for partitioning the genetic variance into additive, dominant and various kinds of epistatic components in an intra- or mixed intra- and interspecific hybrid population. The procedure can be used to study the genetic architecture of fragmented populations and hybrid zones, thus allowing for a better recognition of the role of epistasis in evolution and hybrid speciation. A real example for two Populus species, P. tremuloides and P. tremula, is provided to illustrate the procedure. In this example, we found that considerable new genetic variation is formed through genomic recombination between two aspen species. Received: 1 May 1999 / Accepted: 27 July 1999     

Selection,structure and the heritability of behaviour

Characters which are closely linked to fitness often have low heritabilities (V_A/V_P). Low heritabilities could be because of low additive genetic variation (V_A), that had been depleted by directional selection. Alternatively, low heritabilities may be caused by large residual variation (V_R=V_P – V_A) compounded at a disproportionately higher rate than V_A across integrated characters. Both hypotheses assume that each component of quantitative variation has an independent effect on heritability. However, V_A and V_R may also covary, in which case differences in heritability cannot be fully explained by the independent effects of elimination‐selection or compounded residual variation. We compared the central tendency of published behavioural heritabilities (mean=0.31, median=0.23) with morphological and life history data collected by 26 ). Average behavioural heritability was not significantly different from average life history heritability, but both were smaller than average morphological heritability. We cross‐classified behavioural traits to test whether variation in heritability was related to selection (dominance, domestic/wild) or variance compounding (integration level). There was a significant three‐way interaction between indices of selection and variance compounding, related to the absence of either effect at the highest integration level. At lower integration levels, high dominance variance indicated effects of selection. It was also indicated by the low CV_A of domestic species. At the same time CV_R increased disproportionately faster than CV_A across integration levels, demonstrating variance compounding. However, neither CV_R nor CV_A had a predominant effect on heritability. The partial regression coefficients of CV_R and CV_A on heritability were similar and a path analysis indicated that their (positive) correlation was also necessary to explain variation in heritability. These results suggest that relationships between additive genetic and residual components of quantitative genetic variation can constrain their independent direct effects on behavioural heritability.     

Efficiency of the Sample Variance for the Lognormal Distribution

Two unbiased estimators T and ?? of the variance ? = var (Y) of a lognormal distribution are considered. Here T is the sample variance and ?? is the minimum variance unbiased estimator of ?. The values of the ratio E = 100 var (??)/var (T) are tabulated for some values of the sample size n and of the coefficient of variation δ.