Distribution Of Random Effects Research Articles

Pair copula construction for longitudinal data with zero-inflated power series marginal distributions

ABSTRACT Assessing the temporal dependency among outcomes under investigation is critical in many fields. One complication in the modeling process of the discrete longitudinal data is the presence of excess zeros. We propose a framework for modeling count repeated measurements using members of power series family of distributions. The framework accommodates count outcomes having extra zeros. The longitudinal observations of response variable is modeled using pair copula constructions with a D-vine structure. The maximum likelihood estimates of parameters are obtained using a two-stage approach. Some simulation studies are performed for illustration of the proposed methods, for comparing its performance with that of a generalized linear mixed effects (GLME) model and for assessing the robustness of D-vine and GLME models with respect to the distribution of random effects. In the empirical analysis, the proposed method is applied for analysing a real data set of a kidney allograft rejection study.

A bias-corrected meta-analysis model for combining, studies of different types and quality.

Public health researchers may have to decide whether to perform a meta-analysis including only high-quality randomized clinical trials (RCTs) or whether to include a mixture of all the available evidence, namely RCTs of varying quality and observational studies (OS). The main hurdle when combining disparate evidence in a meta-analysis is that we are not only combining results of interest but we are also combining multiple biases. Therefore, commonly applied meta-analysis methods may lead to misleadingconclusions. In this paper, we present a new Bayesian hierarchical model, called the bias-corrected (BC) meta-analysis model, to combine different study types in meta-analysis. This model is based on a mixture of two random effects distributions, where the first component corresponds to the model of interest and the second component to the hidden bias structure. In this way, the resulting model of interest is adjusted by the internal validity bias of the studies included in a systematicreview. We illustrate the BC model with two meta-analyses: The first one combines RCTs and OS to assess effectiveness of vaccination to prevent invasive pneumococcal disease. The second one investigates the effectiveness of stem cell treatment in heart disease patients. Our results show that ignoring internal validity bias in a meta-analysis may lead to misleading conclusions. However, if a meta-analysis model contemplates a bias adjustment, then RCTs results may increase their precision by including OS in the analysis. The BC model has been implemented in JAGS and R, which facilitate its application inpractice.

High dimensional nuisance parameters: an example from parametric survival analysis

Parametric statistical problems involving both large amounts of data and models with many parameters raise issues that are explicitly or implicitly differential geometric. When the number of nuisance parameters is comparable to the sample size, alternative approaches to inference on interest parameters treat the nuisance parameters either as random variables or as arbitrary constants. The two approaches are compared in the context of parametric survival analysis, with emphasis on the effects of misspecification of the random effects distribution. Notably, we derive a detailed expression for the precision of the maximum likelihood estimator of an interest parameter when the assumed random effects model is erroneous, recovering simply derived results based on the Fisher information in the correctly specified situation but otherwise illustrating complex dependence on other aspects. Methods of assessing model adequacy are given. The results are both directly applicable and illustrate general principles of inference when there is a high-dimensional nuisance parameter. Open problems with an information geometrical bearing are outlined.

A graphical approach to assess the goodness-of-fit of random-effects linear models when the goal is to measure individual benefits of medical treatments in severely ill patients

BackgroundTwo-dimensional personalized medicine (2-PM) models are tools for measuring individual benefits of medical treatments for chronic diseases which have potential applications in personalized medicine. These models assume normality for the distribution of random effects. It is necessary to examine the appropriateness of this assumption. Here, we propose a graphical approach to assessing the goodness-of-fit of 2-PM models with continuous responses.MethodsWe propose benefit quantile-quantile (BQQ) plots which compare the empirical quantiles of individual benefits from a patient sample predicted through an empirical Bayes (EB) approach versus the quantiles of the theoretical distribution of individual benefits derived from the assumption of normality for the random effects. We examine the performance of the approach by conducting a simulation study that compared 2-PM models with non-normal distributions for the random effects versus models with comparable normal distributions. Cramer-von Mises discrepancies were used to quantify the performance of the approach. The approach was illustrated with data from a clinical trial of imipramine for patients with depression.ResultsSimulations showed that BQQ plots were able to capture deviations from the normality assumption for the random effects and did not show any asymmetric deviations from the y = x line when the random effects were normally distributed. For the depression data, the points of the BQQ plot were scattered around closely to the y = x line, without presenting any asymmetric deviations. This implied the adequacy of the normality assumption for the random effects and the goodness-of-fit of the 2-PM model for the imipramine data.ConclusionBQQ plots are sensitive to violations of the normality assumption for the random effects, suggesting that the approach is a useful tool for examining the goodness-of-fit of random-effects linear models when the goal is to measure individual treatment benefits.

Robustness of linear mixed‐effects models to violations of distributional assumptions

Abstract Linear mixed‐effects models are powerful tools for analysing complex datasets with repeated or clustered observations, a common data structure in ecology and evolution. Mixed‐effects models involve complex fitting procedures and make several assumptions, in particular about the distribution of residual and random effects. Violations of these assumptions are common in real datasets, yet it is not always clear how much these violations matter to accurate and unbiased estimation. Here we address the consequences of violations in distributional assumptions and the impact of missing random effect components on model estimates. In particular, we evaluate the effects of skewed, bimodal and heteroscedastic random effect and residual variances, of missing random effect terms and of correlated fixed effect predictors. We focus on bias and prediction error on estimates of fixed and random effects. Model estimates were usually robust to violations of assumptions, with the exception of slight upward biases in estimates of random effect variance if the generating distribution was bimodal but was modelled by Gaussian error distributions. Further, estimates for (random effect) components that violated distributional assumptions became less precise but remained unbiased. However, this particular problem did not affect other parameters of the model. The same pattern was found for strongly correlated fixed effects, which led to imprecise, but unbiased estimates, with uncertainty estimates reflecting imprecision. Unmodelled sources of random effect variance had predictable effects on variance component estimates. The pattern is best viewed as a cascade of hierarchical grouping factors. Variances trickle down the hierarchy such that missing higher‐level random effect variances pool at lower levels and missing lower‐level and crossed random effect variances manifest as residual variance. Overall, our results show remarkable robustness of mixed‐effects models that should allow researchers to use mixed‐effects models even if the distributional assumptions are objectively violated. However, this does not free researchers from careful evaluation of the model. Estimates that are based on data that show clear violations of key assumptions should be treated with caution because individual datasets might give highly imprecise estimates, even if they will be unbiased on average across datasets.

Inference for variance components in linear mixed-effect models with flexible random effect and error distributions.

In many biomedical investigations, parameters of interest, such as the intraclass correlation coefficient, are functions of higher-order moments reflecting finer distributional characteristics. One popular method to make inference for such parameters is through postulating a parametric random effects model. We relax the standard normality assumptions for both the random effects and errors through the use of the Fleishman distribution, a flexible four-parameter distribution which accounts for the third and fourth cumulants. We propose a Fleishman bootstrap method to construct confidence intervals for correlated data and develop a normality test for the random effect and error distributions. Recognizing that the intraclass correlation coefficient may be heavily influenced by a few extreme observations, we propose a modified, quantile-normalized intraclass correlation coefficient. We evaluate our methods in simulation studies and apply these methods to the Childhood Adenotonsillectomy Trial sleep electroencephalogram data in quantifying wave-frequency correlation among different channels.

A Bayesian sensitivity analysis of the effect of different random effects distributions on growth curve models

Growth curve data consist of repeated measurements of a continuous growth process of human, animal, plant, microbial or bacterial genetic data over time in a population of individuals. A classical approach for analysing such data is the use of non-linear mixed effects models under normality assumption for the responses. But, sometimes the underlying population that the sample is extracted from is an abnormal population or includes some homogeneous sub-samples. So, detection of original properties of the population is an important scientific question of interest. In this paper, a sensitivity analysis of using different parametric and non-parametric distributions for the random effects on the results of applying non-linear mixed models is proposed for emphasizing the possible heterogeneity in the population. A Bayesian MCMC procedure is developed for parameter estimation and inference is performed via a hierarchical Bayesian framework. The methodology is illustrated using a real data set on study of influence of menarche on changes in body fat accretion.

Bayesian analysis of Turkish Income and Living Conditions data, using clustered longitudinal ordinal modelling with Bridge distributed random effects

This article is motivated by the panel surveys, called Statistics on Income and Living Conditions (SILC), conducted annually on (randomly selected) country representative households to monitor EU 2020 aims on poverty reduction. We particularly consider the surveys conducted in Turkey within the scope of integration to the EU. Our main interests are on health aspects of economic and living conditions. The outcome is self-reported health that is clustered longitudinal ordinal, since repeated measures of it are nested within individuals and individuals are nested within families. Economic and living conditions have been measured through a number of individual- and family-level explanatory variables. The questions of interest are on the marginal relationships between the outcome and covariates that we address using a polytomous logistic regression with Bridge distributed random effects. This choice of distribution allows us to directly obtain marginal inferences in the presence of random effects. Widely used Normal distribution is also considered as the random effects distribution. Samples from the joint posterior densities of parameters and random effects are drawn using Markov Chain Monte Carlo. Interesting findings from the public health point of view are that differences were found between the subgroups of employment status, income level and panel year in terms of odds of reporting better health.

Vertex Exchange Method for non-parametric estimation of mixing distributions in logistic mixed models

The conventional normality assumption for the random effects distribution in logistic mixed models can be too restrictive in some applications. In our data example of a longitudinal study modelling employment participation of Australian women, the random effects exhibit non-normality due to a potential mover–stayer scenario. In such a scenario, the women observed to remain in the same initial response state over the study period may consist of two subgroups: latent stayers—those with extremely small probability of transitioning response states—and latent movers, those with a probability of transitioning response states. The similarities between estimating the random effects using non-parametric approaches and mover–stayer models have previously been highlighted. We explore non-parametric approaches to model univariate and bivariate random effects in a potential mover–stayer scenario. As there are limited approaches available to fit the non-parametric maximum likelihood estimate for bivariate random effects in logistic mixed models, we implement the Vertex Exchange Method (VEM) to estimate the random effects in logistic mixed models. The approximation of the non-parametric maximum likelihood estimate derived by the VEM algorithm induces more flexibility of the random effects, identifying regions corresponding to potential latent stayers in the non-employment category in our data example.

A multinomial quadrivariate D-vine copula mixed model for meta-analysis of diagnostic studies in the presence of non-evaluable subjects.

Diagnostic test accuracy studies observe the result of a gold standard procedure that defines the presence or absence of a disease and the result of a diagnostic test. They typically report the number of true positives, false positives, true negatives and false negatives. However, diagnostic test outcomes can also be either non-evaluable positives or non-evaluable negatives. We propose a novel model for the meta-analysis of diagnostic studies in the presence of non-evaluable outcomes, which assumes independent multinomial distributions for the true and non-evaluable positives, and, the true and non-evaluable negatives, conditional on the latent sensitivity, specificity, probability of non-evaluable positives and probability of non-evaluable negatives in each study. For the random effects distribution of the latent proportions, we employ a drawable vine copula that can successively model the dependence in the joint tails. Our methodology is demonstrated with an extensive simulation study and applied to data from diagnostic accuracy studies of coronary computed tomography angiography for the detection of coronary artery disease. The comparison of our method with the existing approaches yields findings in the real data application that change the current conclusions.

Random Effects Misspecification Can Have Severe Consequences for Random Effects Inference in Linear Mixed Models

SummaryThere has been considerable and controversial research over the past two decades into how successfully random effects misspecification in mixed models (i.e. assuming normality for the random effects when the true distribution is non‐normal) can be diagnosed and what its impacts are on estimation and inference. However, much of this research has focused on fixed effects inference in generalised linear mixed models. In this article, motivated by the increasing number of applications of mixed models where interest is on the variance components, we study the effects of random effects misspecification on random effects inference in linear mixed models, for which there is considerably less literature. Our findings are surprising and contrary to general belief: for point estimation, maximum likelihood estimation of the variance components under misspecification is consistent, although in finite samples, both the bias and mean squared error can be substantial. For inference, we show through theory and simulation that under misspecification, standard likelihood ratio tests of truly non‐zero variance components can suffer from severely inflated type I errors, and confidence intervals for the variance components can exhibit considerable under coverage. Furthermore, neither of these problems vanish asymptotically with increasing the number of clusters or cluster size. These results have major implications for random effects inference, especially if the true random effects distribution is heavier tailed than the normal. Fortunately, simple graphical and goodness‐of‐fit measures of the random effects predictions appear to have reasonable power at detecting misspecification. We apply linear mixed models to a survey of more than 4 000 high school students within 100 schools and analyse how mathematics achievement scores vary with student attributes and across different schools. The application demonstrates the sensitivity of mixed model inference to the true but unknown random effects distribution.

English

The current interest in meta-analysis of count data in which some studies have zero events (sparse data) has led to re-assessment of commonly used meta-analysis methods to establish their validity in such scenarios. The general consensus is that methods which exclude studies with zero events should be avoided. In the family of parametric methods, random effects models come out highly recommended. While acknowledging the strength of this approach, one of its aspects with potentially undesirable impact on the results, is often overlooked. The random effects approach accounts for the variation in the effect measure across studies by using models with random slopes. It has been shown that parameters associated with a random structure risk being estimated with biased unless the distribution of the random effects is correctly specified. In meta-analysis the parameter of interest, average effect measure, is associated with a random structure (random slope). Information on how the effect measure point and precision estimates are affected by misspecification of random effects distribution is still lacking. To fill in the information gap, we used a simulation study to investigate the impact of misspecification of distribution of random effects in this context and provide guidelines in using the random effects approach. Our results indicated that relative bias in the estimated effect measure could be as high as 30% and 95% confidence interval coverage as low as 0%. These results send a clear message that possible effects of misspecification of the distribution of random effects should not be ignored. In light of these findings, we have proposed a sensitivity analysis that also establishes whether a random slope model is necessary.

A semiparametric stochastic mixed effects model for bivariate cyclic longitudinal data

AbstractWe propose a flexible semiparametric stochastic mixed effects model for bivariate cyclic longitudinal data. The model can handle either single cycle or, more generally, multiple consecutive cycle data. The approach models the mean of responses by parametric fixed effects and a smooth nonparametric function for the underlying time effects, and the relationship across the bivariate responses by a bivariate Gaussian random field and a joint distribution of random effects. The proposed model not only can model complicated individual profiles, but also allows for more flexible within‐subject and between‐response correlations. The fixed effects regression coefficients and the nonparametric time functions are estimated using maximum penalized likelihood, where the resulting estimator for the nonparametric time function is a cubic smoothing spline. The smoothing parameters and variance components are estimated simultaneously using restricted maximum likelihood. Simulation results show that the parameter estimates are close to the true values. The fit of the proposed model on a real bivariate longitudinal dataset of pre‐menopausal women also performs well, both for a single cycle analysis and for a multiple consecutive cycle analysis. The Canadian Journal of Statistics 48: 471–498; 2020 © 2020 Statistical Society of Canada

MixWILD: A program for examining the effects of variance and slope of time-varying variables in intensive longitudinal data

The use of intensive sampling methods, such as ecological momentary assessment (EMA), is increasingly prominent in medical research. However, inferences from such data are often limited to the subject-specific mean of the outcome and between-subject variance (i.e., random intercept), despite the capability to examine within-subject variance (i.e., random scale) and associations between covariates and subject-specific mean (i.e., random slope). MixWILD (Mixed model analysis With Intensive Longitudinal Data) is statistical software that tests the effects of subject-level parameters (variance and slope) of time-varying variables, specifically in the context of studies using intensive sampling methods, such as ecological momentary assessment. MixWILD combines estimation of a stage 1 mixed-effects location-scale (MELS) model, including estimation of the subject-specific random effects, with a subsequent stage 2 linear or binary/ordinal logistic regression in which values sampled from each subject’s random effect distributions can be used as regressors (and then the results are aggregated across replications). Computations within MixWILD were written in FORTRAN and use maximum likelihood estimation, utilizing both the expectation-maximization (EM) algorithm and a Newton–Raphson solution. The mean and variance of each individual’s random effects used in the sampling are estimated using empirical Bayes equations. This manuscript details the underlying procedures and provides examples illustrating standalone usage and features of MixWILD and its GUI. MixWILD is generalizable to a variety of data collection strategies (i.e., EMA, sensors) as a robust and reproducible method to test predictors of variability in level 1 outcomes and the associations between subject-level parameters (variances and slopes) and level 2 outcomes.

Effect of misspecification of random effects distribution on the performance of parameters estimation methods in binary logistic mixed models

Nous avons compare, a l’aide de la simulation, une methode d’estimation Bayesienne (INLA) a trois methodes classiques d’estimation (PQL, HLM et AGHQ) sous six distributions des effets aleatoires dans le cadre des modeles logistiques binaires mixtes. Les resultats ont montre la superiorite de AGHQ et HLM sur les autres pour toutes les distributions considerees dans le cas des effets fixes. Pour les effets aleatoires, les methodes classiques ont montre les meilleures performances pour les distributions symetriques consideres. AGHQ, HLM et INLA ont donne de meilleurs resultats par rapport a PQL pour les distributions normale et uniforme pour tous les echantillons consideres.

A Class of Hierarchical Random Effect Models for Posterior Insurance Ratemaking

Poisson random effect panel data model is widely used in non-life insurance for posterior rate making. Recently, Lee et al. (2020) identify one shortcoming of the shared random effect assumption, which introduces both the marginal count distribution and the serial dependence between counts at different periods. Therefore, it is not possible to accommodate separately these two features. This shortcoming also exists for any other nonlinear (say, gamma, or negative binomial) random effect regression model. In this note we propose a family of hierarchical random effect model, allowing to separate its effect on the marginal and the serial dependence structure of the observable dependent variables. The model we propose has several advantages. First it is very tractable, with closed form likelihood function and posterior rate making formula. Second, it is very general and suitable for a large number of marginal distributions of the random effect and conditional distributions of the response variable, such as the gamma-Poisson, beta-negative binomial pair distributions, as well as some new non conjugate pairs. Finally, it is also applicable to the simultaneous pricing of a large number of risks encountered in fleet insurance.

Forecasting With Dynamic Panel Data Models

This paper considers the problem of forecasting a collection of short time series using cross‐sectional information in panel data. We construct point predictors using Tweedie's formula for the posterior mean of heterogeneous coefficients under a correlated random effects distribution. This formula utilizes cross‐sectional information to transform the unit‐specific (quasi) maximum likelihood estimator into an approximation of the posterior mean under a prior distribution that equals the population distribution of the random coefficients. We show that the risk of a predictor based on a nonparametric kernel estimate of the Tweedie correction is asymptotically equivalent to the risk of a predictor that treats the correlated random effects distribution as known (ratio optimality). Our empirical Bayes predictor performs well compared to various competitors in a Monte Carlo study. In an empirical application, we use the predictor to forecast revenues for a large panel of bank holding companies and compare forecasts that condition on actual and severely adverse macroeconomic conditions.

The Poisson random effect model for experience ratemaking: Limitations and alternative solutions

Poisson random effect models with a shared random effect have been widely used in actuarial science for analyzing the number of claims. In particular, the random effect is a key factor in a posteriori risk classification. However, the necessity of the random effect may not be properly assessed due to the dual role of the random effect; it affects both the marginal distribution of the number of claims and the dependence among the numbers of claims obtained from an individual over time. We first show that the score test for the nullity of the variance of the shared random effect can falsely indicate significant dependence among the numbers of claims even though they are independent. To mitigate this problem, we propose to separate the dual role of the random effect by introducing additional random effects to capture the overdispersion part, which are called saturated random effects. In order to circumvent heavy computational issues by the saturated random effects, we choose a gamma distribution for the saturated random effects because it gives the closed form of marginal distribution. In fact, this choice leads to the negative binomial random effect model that has been widely used for the analysis of frequency data. We show that safer conclusions about the a posteriori risk classification can be made based on the negative binomial mixed model under various situations. We also derive the score test as a sufficient condition for the existence of the a posteriori risk classification based on the proposed model.

ESTIMATION OF INCOME CHARACTERISTICS FOR REGIONS IN POLAND USING SPATIO-TEMPORAL SMALL AREA MODELS

Abstract The paper presents the comparison of estimation results for spatial and spatiotemporal small area models. The study was carried out for income-related variables drawn from the Polish Household Budget Survey and explanatory variables from the Polish Local Data Bank for the years 2003-2013. The properties of EBLUPs (Empirical Best Linear Unbiased Predictors) based on spatiotemporal models, which utilize spatial correlation between neighbouring areas as well as historical data, were compared and contrasted with EBLUPs based on spatial models obtained separately for each year and with EBLUPs based on the Rao-Yu model. The computations were performed using sae, sae2 and spdep packages for R-project environment. In the case of sae package, the eblupFH, eblupSFH and the eblupSTFH functions were used for point estimation along with the mseFH, mseSFH and the pbmseSTFH functions for the MSE estimation, whereas the eblupRY function was applied for the purposes of sae2 package. The precision of direct estimators was guaranteed by the adoption of the Balanced Repeated Replication method. The results of the analysis demonstrate that a visible reduction of the estimation error was achieved for the implemented spatiotemporal small-area models, especially when significant spatial and time autocorrelations were observed. These results are even more valuable than those achieved by the means of the Rao-Yu model. In the computations, three author-defined functions were adopted, which not only enabled the author to perform the extract of random effects for spatial, spatiotemporal and Rao-Yu models, but also made it possible to obtain their decomposition with respect to spatial and temporal parts, thus creating a novel solution. The comparison was carried out using choropleth maps for spatial effects and distributions of temporal random effects for the considered years.

An EM algorithm to improve the estimation of the probability of clonal relatedness of pairs of tumors in cancer patients

BackgroundWe previously introduced a random-effects model to analyze a set of patients, each of which has two distinct tumors. The goal is to estimate the proportion of patients for which one of the tumors is a metastasis of the other, i.e. where the tumors are clonally related. Matches of mutations within a tumor pair provide the evidence for clonal relatedness. In this article, using simulations, we compare two estimation approaches that we considered for our model: use of a constrained quasi-Newton algorithm to maximize the likelihood conditional on the random effect, and an Expectation-Maximization algorithm where we further condition the random-effect distribution on the data.ResultsIn some specific settings, especially with sparse information, the estimation of the parameter of interest is at the boundary a non-negligible number of times using the first approach, while the EM algorithm gives more satisfactory estimates. This is of considerable importance for our application, since an estimate of either 0 or 1 for the proportion of cases that are clonal leads to individual probabilities being 0 or 1 in settings where the evidence is clearly not sufficient for such definitive probability estimates.ConclusionsThe EM algorithm is a preferable approach for our clonality random-effect model. It is now the method implemented in our R package Clonality, making available an easy and fast way to estimate this model on a range of applications.