Search

The area under the true ROC curve (AUC) is routinely used to determine how strongly a given model discriminates between the levels of a binary outcome. Standard inference with the AUC requires that outcomes be independent of each other. To overcome this limitation, a method was developed for the estimation of the variance of the AUC in the setting of two-level hierarchical data using probit-transformed prediction scores generated from generalized estimating equation models, thereby allowing for the application of inferential methods. This manuscript presents an extension of this approach so that inference for the AUC may be performed in a three-level hierarchical data setting (e.g., eyes nested within persons and persons nested within families). A method that accounts for the effect of tied prediction scores on inference is also described. The performance of 95% confidence intervals around the AUC was assessed through the simulation of three-level clustered data in multiple settings, including ones with tied data and variable cluster sizes. Across all settings, the actual 95% confidence interval coverage varied from 0.943 to 0.958, and the ratio of the theoretical variance to the empirical variance of the AUC varied from 0.920 to 1.013. The results are better than those from existing methods. Two examples of applying the proposed methodology are presented.

Exoplanet detection by direct imaging is a difficult task: the faint signals from the objects of interest are buried under a spatially structured nuisance component induced by the host star. The exoplanet signals can only be identified when combining several observations with dedicated detection algorithms. In contrast to most of existing methods, we propose to learn a model of the spatial, temporal and spectral characteristics of the nuisance, directly from the observations. In a pre-processing step, a statistical model of their correlations is built locally, and the data are centered and whitened to improve both their stationarity and signal-to-noise ratio (SNR). A convolutional neural network (CNN) is then trained in a supervised fashion to detect the residual signature of synthetic sources in the pre-processed images. Our method leads to a better trade-off between precision and recall than standard approaches in the field. It also outperforms a state-of-the-art algorithm based solely on a statistical framework. Besides, the exploitation of the spectral diversity improves the performance compared to a similar model built solely from spatio-temporal data., accepted to EUSIPCO 2023

In studies following selective sampling protocols for secondary outcomes, conventional analyses regarding their appearance could provide misguided information. In the large type 1 diabetes prevention and prediction (DIPP) cohort study monitoring type 1 diabetes-associated autoantibodies, we propose to model their appearance via a multivariate frailty model, which incorporates a correlation component that is important for unbiased estimation of the baseline hazards under the selective sampling mechanism. As further advantages, the frailty model allows for systematic evaluation of the association and the differences in regression parameters among the autoantibodies. We demonstrate the properties of the model by a simulation study and the analysis of the autoantibodies and their association with background factors in the DIPP study, in which we found that high genetic risk is associated with the appearance of all the autoantibodies, whereas the association with sex and urban municipality was evident for IA-2A and IAA autoantibodies. publishedVersion

Background There is high rate of under-five mortality in West Africa with little effort made to study determinants that significantly increase or decrease its risk across the West African sub-region. This is important since it will help in the design of effective intervention programs for each country or the entire region. The overall objective of this research evaluates the determinants of under-five mortality prior to the end of the 2015 Millennium Development Goals, to guide West African countries implement strategies that will aid them achieve the Sustainable Development Goal 3 by 2030. Method This study used the Demographic and Health Survey (DHS) data from twelve (12) out of the eighteen West African countries; Ghana, Benin, Cote d’ Ivoire, Guinea, Liberia, Mali, Niger, Nigeria, Sierra Leone, Burkina Faso, Gambia and Togo. Data were extracted from the children and women of reproductive age files as provided in the DHS report. The response or outcome variable of interest is under-five mortality rate. A Bayesian exponential, Weibull and Gompertz regression models via a gamma shared frailty model were used for the analysis. The deviance information criteria and Bayes factors were used to discriminate between models. These analyses were carried out using Stata version 15 software. Results The study recorded 101 (95% CI: 98.6–103.5) deaths per 1000 live births occurring among the twelve countries. Burkina Faso (124.4), Cote D’lvoire (110.1), Guinea (116.4), Nigeria (120.6) and Niger (118.3) recorded the highest child under-5 mortality rate. Gambia (48.1), Ghana (60.1) and Benin (70.4) recorded the least unde-5 mortality rate per 1000 livebirths. Multiple birth children were about two times more likely to die compared to singleton birth, in all except Gambia, Nigeria and Sierra Leone. We observed significantly higher hazard rates for male compared to female children in the combined data analysis (HR: 1.14, 95% CI: [1.10–1.18]). The country specific analysis in Benin, Cote D’lvoire, Guinea, Liberia, Mali and Nigeria showed higher under-5 mortality hazard rates among male children compared to female children whilst Niger was the only country to report significantly lower hazard rate of males compared to females. Conclusion There is still quite a substantial amount of work to be done in order to meet the Sustainable Development Goal 3 in 2030 in West Africa. There exist variant differences among some of the countries with respect to mortality rates and determinants which require different interventions and policy decisions.

Background As the barriers to incorporating RNA sequencing (RNA-Seq) into biomedical studies continue to decrease, the complexity and size of RNA-Seq experiments are rapidly growing. Paired, longitudinal, and other correlated designs are becoming commonplace, and these studies offer immense potential for understanding how transcriptional changes within an individual over time differ depending on treatment or environmental conditions. While several methods have been proposed for dealing with repeated measures within RNA-Seq analyses, they are either restricted to handling only paired measurements, can only test for differences between two groups, and/or have issues with maintaining nominal false positive and false discovery rates. In this work, we propose a Bayesian hierarchical negative binomial generalized linear mixed model framework that can flexibly model RNA-Seq counts from studies with arbitrarily many repeated observations, can include covariates, and also maintains nominal false positive and false discovery rates in its posterior inference. Results In simulation studies, we showed that our proposed method (MCMSeq) best combines high statistical power (i.e. sensitivity or recall) with maintenance of nominal false positive and false discovery rates compared the other available strategies, especially at the smaller sample sizes investigated. This behavior was then replicated in an application to real RNA-Seq data where MCMSeq was able to find previously reported genes associated with tuberculosis infection in a cohort with longitudinal measurements. Conclusions Failing to account for repeated measurements when analyzing RNA-Seq experiments can result in significantly inflated false positive and false discovery rates. Of the methods we investigated, whether they model RNA-Seq counts directly or worked on transformed values, the Bayesian hierarchical model implemented in the mcmseq R package (available at https://github.com/stop-pre16/mcmseq) best combined sensitivity and nominal error rate control.

Background Elderly population’s health is a major concern for most industrial nations. National health surveys provide a measure of the state of elderly health. One such survey is the Chinese Longitudinal Healthy Longevity Survey. It collects data on risk factors and outcomes on the elderly. We examine these longitudinal survey data to determine the changes in health and to identify risk factors as they impact health outcomes including the elderly’s ability to do a physical check. Methods We use a Partitioned GMM logistic regression model to identify risk factors. The model also accounts for the correlation between lagged time-dependent covariates and the outcomes. It addresses present and past measures of time-dependent covariates on simultaneous outcomes. The relation produces additional regression coefficients as byproduct of the Partitioned model, identifying the immediate, delayed effects (lag − 1), further delayed (lag-2), etc. Therefore, the model presents the opportunity for decision makers to monitor the covariate over time. This technique is particularly useful in healthcare and health related research. We use the Chinese Longitudinal Health Longevity Survey data to identify those risk factors and to display the utility of the model. Results We found that one’s ability to make own decisions, frequently consuming vegetables, exercise frequently, one’s ability to transfer without assistance, having visual difficulties and being able to pick book from floor while standing had varying effects of significance on one’s health and ability to complete physical checks as they get older. Conclusions The partitioning of the covariates as immediate effect, delayed effect or further delayed effect are important measures in a declining population.

Dissertação de mestrado em Bioinformática, Os modelos de regressão logística procuram identificar a influência de diferentes variáveis/fatores numa variável resposta de interesse. Estes são normalmente utilizados na àrea da medicina pois permite verificar quais os fatores que influênciam a presença de determinadas patologias. Os estudos longitudinais, onde se verifica uma repetição de observações ou medições registadas ao longo do tempo, são os mais utilizados para aplicação dos modelos de regressão logística, no entanto grande parte deste modelos não consideram a correlação entre as variáveis em estudo. De modo a ultrapassar este problema foram desenvolvidos os modelos GEE (Generalized Estimating Equations) que consideram a correlação existente nos dados, possibilitando assim uma análise mais rigorosa da influência de diferentes factores. Esta dissertação procura identificar e explorar diferentes packages em R para aplicação dos modelos GEE através da sua aplicação a um caso de estudo. Ao longo desta dissertação será desenvolvida uma nova aplicação web designada SAGA, desenvolvida utilizando o package Shiny na linguagem R. Esta aplicação encontra-se disponível no seguinte link: https: //geemodelapp2022.shinyapps.io/Shiny_App/ . A aplicação SAGA tem como principal objetivo a análise de modelos GEE utilizando um conjunto de dados selecionado pelo utilizador, sendo possível identificar as diferentes variáveis de interesses que serão descritas ao longo da dissertação, bem como validar os modelos desenvolvidos através da validação por curvas ROC. Para além dos resultados dos modelos GEE, demonstrados na aplicação, também estão representadas as curvas ROC de cada modelo desenvolvido., Logistic regression models seek to identify the influence of different variables/factors on a response variable of interest. These are normally used in the field of medicine as it allows to verify which factors influence the presence of certain pathologies. Longitudinal studies, where there is a repetition of observations or measurements recorded over time, are the most used to apply logistic regression models, however most of these models do not consider the correlation between the variables under study. In order to overcome this problem, GEE (Generalized Estimating Equations) models were developed, which consider the existing correlation in the data, resulting in a more rigorous analysis of the influence of different factors. This dissertation, in a first phase, explores the different existing R packages for the application of GEE models, in order to identify differences or similarities between them using a case study for the application of GEE models. Throughout this dissertation, a new web application called SAGA will be developed, the application will be developed using the Shiny package in R language. This application is available at the following link: https: //geemodelapp2022.shinyapps.io/Shiny_App/. The main purpose of the SAGA application is to develop and analyse GEE models using a dataset selected by the user, where it will be possible to describe all the variables of interest in the development of the model as will be described in the course of this dissertation, as well as to validate the models developed through validation by ROC analysis. In addition to the results of the GEE models, shown in the application, the ROC curves of each model developed are also represented.

In agronomy, and more specifically in genetic breeding, high throughput genotyping has been widely used for more than 20 years to access increasingly rich and abundant genetic information. This has allowed the identification of positions along the genome involved in the variability of traits of interest. More recently, high throughput phenotyping methods have been developed. They give access to the monitoring of the evolution of several phenotypic traits over time. These longitudinal data allow a fine study of the dynamics of these traits, while identifying the environmental factors that influence their variability according to developmental stages.However, the analysis of such data raises several statistical challenges. This thesis proposes methodological developments in order to take into account the dependencies between observations and between variables, to select relevant genetic or environmental variables, or to estimate effects that evolve over time. The Bayesian framework is an elegant statistical formalism to address these different issues, especially through the construction of priors. We study and compare different priors to simultaneously infer and select fixed and/or random effects when they can be numerous. We consider different statistical modeling frameworks classically used for longitudinal data analysis. In particular, we focus on linear mixed models, varying coefficient models or signal regression.This work was motivated by various practical applications concerning the QTL detection, the temporal evolution of genetic architecture or the impact of climatic variations on phenotypic variability. Three datasets, from various agronomical contexts, are used to illustrate these new approaches.; En agronomie et plus spécifiquement en amélioration génétique, le génotypage haut débit a été largement mis à profit, depuis maintenant plus de 20 ans, pour accéder à une information génétique toujours plus riche et abondante. Celle-ci a permis d’identifier les positions le long du génome impliquées dans la variabilité de caractères d’intérêt. Plus récemment, les méthodes de phénotypage haut débit ont fait leur apparition. Elles donnent accès au suivi de l’évolution de plusieurs caractères phénotypiques au cours du temps. Ces données, longitudinales, permettent d’étudier finement la dynamique évolutive de ces caractères tout en identifiant les facteurs environnementaux qui influencent leur variabilité selon les stades de développement.Cependant, l’analyse de telles données soulève plusieurs défis statistiques. Cette thèse propose des développements méthodologiques afin de prendre en compte les dépendances entre observations et entre variables, de sélectionner les variables génétiques ou environnementales pertinentes, ou encore d’estimer des effets qui évoluent au cours du temps. Le cadre bayésien est un formalisme statistique élégant pour répondre à ces différentes problématiques notamment au travers de la construction de lois a priori. Nous étudions et comparons différentes lois a priori pour simultanément inférer et sélectionner les effets fixes et/ou aléatoires quand ceux-ci peuvent être nombreux. Nous considérons différents cadres de modélisation statistique classiquement utilisés pour l’analyse de données longitudinales. En particulier, nous nous focaliserons sur les modèles à coefficients variants, les modèles linéaires mixtes ou encore la régression sur signal.Ce travail a été motivé par différentes applications pratiques portant sur l’évolution temporelle de l’architecture génétique, la détection de QTL ou l’impact des variations climatiques sur la variabilité phénotypique. Trois jeux de données, issus de contextes agronomiques variés, sont utilisés pour illustrer ces nouvelles approches.

Submitted by Lucas Vasconcelos Vieira (vasconcelos.vieira@unesp.br) on 2021-07-26T13:03:59Z No. of bitstreams: 1 Lucas_Vieira_Tese.pdf: 1974487 bytes, checksum: 95b27607a38a25672497d68eda3b4586 (MD5) Approved for entry into archive by Ana Lucia de Grava Kempinas (ana.kempinas@unesp.br) on 2021-07-26T13:55:59Z (GMT) No. of bitstreams: 1 vieira_lv_dr_botfca.pdf: 1974487 bytes, checksum: 95b27607a38a25672497d68eda3b4586 (MD5) Made available in DSpace on 2021-07-26T13:55:59Z (GMT). No. of bitstreams: 1 vieira_lv_dr_botfca.pdf: 1974487 bytes, checksum: 95b27607a38a25672497d68eda3b4586 (MD5) Previous issue date: 2021-05-27 Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) Em experimentos com medidas repetidas, a premissa de simetria composta (esfericidade), isto é, de correlação constante entre observações tomadas na mesma unidade, nem sempre é observada. Não obstante, os dados são frequentemente analisados como se correlações entre as medidas repetidas fossem constantes, como no caso da ANOVA para medidas repetidas, ou utilizando métodos que apenas consideram todas as correlações distintas, como no caso da ANOVA multivariada, podendo impactar na taxa de rejeição da hipótese nula, e consequentemente afetar a taxa de erro do tipo I e o poder estatístico. Nesse contexto, o presente estudo propôs investigar a aplicação de modelos lineares mistos com diferentes pressupostos sobre a matriz de covariância em conjuntos de dados provenientes de experimentos com medidas repetidas simulados. Foram avaliados 84 cenários que variaram quanto o padrão da matriz de covariância (14 estruturas), número de medidas repetidas (4 e 8) e de repetições (4, 8 e 12). Foram simulados 10,000 conjuntos de dados para cada cenário baseado em uma distribuição normal multivariada e que foram posteriormente analisados utilizando modelos lineares mistos aliados a máxima verossimilhança restrita. A taxa de erro do tipo I e o poder estatístico para o teste de hipótese da interação entre tratamentos e medidas repetidas foram estimados como a proporção de valores p menores ou iguais a 0.01 ou 0.05 de um total de 10,000 testes para cada cenário. Os modelos também foram avaliados quanto a habilidade de ajuste aos dados a partir dos critérios de seleção BIC. Assim, a frequência com a qual as estruturas de covariância foram escolhidas pelos critérios de seleção foi computada. Os resultados indicam que o pressuposto escolhido com maior frequência pelos critérios de informação resultou da estrutura de covariância especificada que correspondeu à estrutura de covariância empírica dos conjuntos de dados analisados, particularmente para aqueles conjuntos com tamanho amostral maiores. Os resultados também indicam que o uso de modelos de covariância que não reconheça correlações heterogêneas entre as medidas repetidas pode inflar o erro do tipo I a níveis muito liberais ou reduzi-lo a níveis muito conservadores, podendo afetar a conclusão dos experimentos agrícolas. Para um nível de significância e 0.05, o enviesamento da taxa de erro do tipo I foi superior a 2α, enquanto para um nível de significância de 1%, o enviesamento foi superior a 4α, quando o pressuposto sobre a estrutura de covariância dos dados não considerou a existência de correlações heterogêneas, particularmente para aqueles conjuntos de dados com variâncias também heterogêneas, com um número maior de medidas repetidas e um menor tamanho amostral. A proporção de modelos de covariância escolhidos pelo critério de seleção foi superior para os modelos de covariância que correspondiam a estrutura empírica dos dados, particularmente para aquelas simulações com maior número de medidas repetidas e maior tamanho amostral. The premise in experiments with repeated measures is that observations taken in the same experimental unit are correlated and that correlations decrease proportionally to the increase in the distance between measurements in time or space. Nevertheless, these experiments are often analyzed as if the correlations between the repeated measures were constant or using methods that only consider correlations different, which may impact on the rejection rate of the null hypothesis, and ultimately type I error rate and statistical power. In this context, the present study investigated the application of mixed linear models with different assumptions about the covariance matrix in data sets from simulated experiments with repeated measures. 84 scenarios that varied in terms of the covariance matrix pattern (14 structures), number of repeated measurements (4 and 8) and sample size (4, 8 and 12) were evaluated. 10,000 datasets were simulated for each scenario based on a multivariate normal distribution and were subsequently analyzed using mixed linear models. Type I error rate and statistical power for the hypothesis test of the interaction between treatments and repeated measures were estimated as the proportion of p values less than or equal to 0.01 or 0.05 out of a total of 10,000 tests for each scenario. The models were also evaluated for their ability to fit the data using Bayesian Information Criteria (BIC). Thus, the frequency with which the covariance structures were chosen by the selection criteria was computed. Results indicate that the assumption chosen most frequently by the information criteria resulted from the specified covariance structure that corresponded to the empirical covariance structure of the analyzed data sets, particularly for those with larger number of repeated measures and sample sizes. Results also indicate that the use of covariance models that do not recognize heterogeneous correlations between repeated measures can inflate type I error or reduce it to very conservative levels, which may affect the conclusion of agricultural experiments. For a 5% significance level, type I error bias was greater than 2α, while for 1% significance level, bias was over 4α. In addition, the statistical power was reduced when the assumption about the covariance matrix of the data sets did not correspond to the empirical covariance structure, particularly for those datasets with a smaller sample size.

Statistical analyses always lead to results that include an imprecision to a degree. Combining information refers to quantitative methods that can synthesize information from multiple sources on the same topic to reach a more precise and general result. A challenging issue in combining information is that the individual results may be correlated to each other. For example, in ecology, outcomes from multiple species are dependent on each other due to their shared evolutionary history. This dissertation focuses on two fields in combining information, meta-analysis and combining p-values, and examines techniques that can incorporate the dependence among the correlated measurements into the method for combining information. The results show that, by doing so, it is possible to reduce the risk of incorrect results, such as false positives. Furthermore, this dissertation introduces an open-source software that implements methods for combining p-values and adjustment techniques to incorporate the correlations among them.

Generalised estimating equations with the sandwich standard-error estimator provide a promising method of analysis for stepped wedge cluster randomised trials. However, they have inflated type-one error when used with a small number of clusters, which is common for stepped wedge cluster randomised trials. We present a large simulation study of binary outcomes comparing bias-corrected standard errors from Fay and Graubard; Mancl and DeRouen; Kauermann and Carroll; Morel, Bokossa, and Neerchal; and Mackinnon and White with an independent and exchangeable working correlation matrix. We constructed 95% confidence intervals using a t-distribution with degrees of freedom including clusters minus parameters (DFC-P), cluster periods minus parameters, and estimators from Fay and Graubard (DFFG), and Pan and Wall. Fay and Graubard and an approximation to Kauermann and Carroll (with simpler matrix inversion) were unbiased in a wide range of scenarios with an independent working correlation matrix and more than 12 clusters. They gave confidence intervals with close to 95% coverage with DFFG with 12 or more clusters, and DFC-P with 18 or more clusters. Both standard errors were conservative with fewer clusters. With an exchangeable working correlation matrix, approximated Kauermann and Carroll and Fay and Graubard had a small degree of under-coverage.

Background Event-related potentials (ERP) data are widely used in brain studies that measure brain responses to specific stimuli using electroencephalogram (EEG) with multiple electrodes. Previous ERP data analyses haven’t accounted for the structured correlation among observations in ERP data from multiple electrodes, and therefore ignored the electrode-specific information and variation among the electrodes on the scalp. Our objective was to evaluate the impact of early adversity on brain connectivity by identifying risk factors and early-stage biomarkers associated with the ERP responses while properly accounting for structured correlation. Methods In this study, we extend a penalized generalized estimating equation (PGEE) method to accommodate structured correlation of ERPs that accounts for electrode-specific data and to enable group selection, such that grouped covariates can be evaluated together for their association with brain development in a birth cohort of urban-dwelling Bangladeshi children. The primary ERP responses of interest in our study are N290 amplitude and the difference in N290 amplitude. Results The selected early-stage biomarkers associated with the N290 responses are representatives of enteric inflammation (days of diarrhea, MIP1b, retinol binding protein (RBP), Zinc, myeloperoxidase (MPO), calprotectin, and neopterin), systemic inflammation (IL-5, IL-10, ferritin, C Reactive Protein (CRP)), socioeconomic status (household expenditure), maternal health (mother height) and sanitation (water treatment). Conclusions Our proposed group penalized GEE estimator with structured correlation matrix can properly model the complex ERP data and simultaneously identify informative biomarkers associated with such brain connectivity. The selected early-stage biomarkers offer a potential explanation for the adversity of neurocognitive development in low-income countries and facilitate early identification of infants at risk, as well as potential pathways for intervention. Trial registration The related clinical study was retrospectively registered with https://doi.org/ClinicalTrials.gov, identifier NCT01375647, on June 3, 2011.

Background A recently proposed method for estimating qPCR amplification efficiency E analyzes fluorescence intensity ratios from pairs of points deemed to lie in the exponential growth region on the amplification curves for all reactions in a dilution series. This method suffers from a serious problem: The resulting ratios are highly correlated, as they involve multiple use of the raw data, for example, yielding ~ 250 E estimates from ~ 25 intensity readings. The resulting statistics for such estimates are falsely optimistic in their assessment of the estimation precision. Results Monte Carlo simulations confirm that the correlated pairs method yields precision estimates that are better than actual by a factor of two or more. This result is further supported by estimating E by both pairwise and Cq calibration methods for the 16 replicate datasets from the critiqued work, and then comparing the ensemble statistics for these methods. Conclusion Contrary to assertions in the proposing work, the pairwise method does not yield E estimates a factor of 2 more precise than estimates from Cq calibration fitting (the standard curve method). On the other hand, the statistically correct direct fit of the data to the model behind the pairwise method can yield E estimates of comparable precision. Ways in which the approach might be improved are discussed briefly.

This paper proposes a new classification of correlated data types based upon the relative number of direct connections among observations, producing a family of correlated observations embracing seven categories, one whose empirical counterpart currently is unknown, and ranging from independent (i.e., no links) to approaching near-complete linkage (i.e., n(n &ndash, 1)/2 links). Analysis of specimen datasets from publicly available data sources furnishes empirical illustrations for these various categories. Their descriptions also include their historical context and calculation of their effective sample sizes (i.e., an equivalent number of independent observations). Concluding comments contain some state-of-the-art future research topics.

Large-scale surveys, such as national forest inventories and vegetation monitoring programs, usually have complex sampling designs that include geographical stratification and units organized in clusters. When models are developed using data from such programs, a key question is whether or not to utilize design information when analyzing the relationship between a response variable and a set of covariates. Standard statistical regression methods often fail to account for complex sampling designs, which may lead to severely biased estimators of model coefficients. Furthermore, ignoring that data are spatially correlated within clusters may underestimate the standard errors of regression coefficient estimates, with a risk for drawing wrong conclusions. We first review general approaches that account for complex sampling designs, e.g. methods using probability weighting, and stress the need to explore the effects of the sampling design when applying logistic regression models. We then use Monte Carlo simulation to compare the performance of the standard logistic regression model with two approaches to model correlated binary responses, i.e. cluster-specific and population-averaged logistic regression models. As an example, we analyze the occurrence of epiphytic hair lichens in the genus Bryoria; an indicator of forest ecosystem integrity. Based on data from the National Forest Inventory (NFI) for the period 1993–2014 we generated a data set on hair lichen occurrence on >100,000 Picea abies trees distributed throughout Sweden. The NFI data included ten covariates representing forest structure and climate variables potentially affecting lichen occurrence. Our analyses show the importance of taking complex sampling designs and correlated binary responses into account in logistic regression modeling to avoid the risk of obtaining notably biased parameter estimators and standard errors, and erroneous interpretations about factors affecting e.g. hair lichen occurrence. We recommend comparisons of unweighted and weighted logistic regression analyses as an essential step in development of models based on data from large-scale surveys.

High throughput technologies like microarrays, next generation sequencing and mass spectrometry enable the measurement of tens of thousands of biomarker candidates in pilot studies. Biological systems are often too complex to be based on simple single cause-effect associations and from the medical practice point of view, a single biomarker may not possess the desired sensitivity and/or specificity for disease classification and outcome prediction. Therefore, the efforts of researchers currently aims at combining biomarkers. The intention of biomarker pilot studies with small sample sizes is often to explore the possibility of finding good biomarker combinations and not to find and evaluate a final combination of biomarkers with high predictive value. The aim of the pilot study is to answer the question whether it is worthwhile to extend the study to a larger study and to obtain information about the required sample size. In this paper, we propose a method to judge the potential in a small biomarker pilot study without the need to explicitly identifying and confirming a specific subset of biomarkers. In addition, we provide a method for sample size estimation for an extended study when the results of the pilot study look promising. Abbreviations: ROC: receiver operating characteristic curve; AUC: Area Under the ROC curve; HAUCA: high AUC abundance; ER: Estrogen receptor; BCs: Biomarker candidates; w: with; wt: without

Fundação de Amparo à Pesquisa do Estado do Amazonas (FAPEAM) Parameter estimation in nonlinear mixed-effects models is often challenging. In this thesis, a comparison of estimation methods for these models is proposed under a frequentist approach. In the first study, a comparison of maximum likelihood estimates under an exact method via Monte Carlo expectation-maximization (MCEM) and an approximate method based on a Taylor expansion, frequently used in the literature, is provided. In a second study, a restricted maximum likelihood estimation method is proposed, aiming to decrease the bias for the variance components estimates, based on the integration of the likelihood function on the fixed-effects, also in an exact likelihood context. These estimates are compared to the maximum likelihood ones. For the latter comparison, stochastic approximation of expectation-maximization (SAEM) algorithms are considered. The random effects and errors are assumed to follow multivariate symmetric distributions, namely the scale mixture of normal distributions, which include the normal, t and slash distributions. Finally, a general nonlinear mixed-effects model is proposed, where no linear relation is assumed in the random effects structure. In all the proposals, real data sets and simulation studies are used to illustrate the estimates’ properties. A estimação de parâmetros em modelos não lineares com efeitos mistos é muitas vezes desafiadora. Neste trabalho, propomos a comparação de alguns de métodos de estimação nesses modelos sob o enfoque frequentista. Em um primeiro momento, propomos um estimador de máxima verossimilhança em um esquema de estimação exata contra o estimador de máxima versossimilhança em um modelo linearizado pela expansão de Taylor, o que é frequentemente utilizado na literatura. No primeiro cenário usamos o algoritmo MCEM. Em um segundo momento, visando diminuir o viés para estimativas das componentes de variância, propomos um estimador de máxima verossimilhança restrita também dentro de um esquema de estimação exata, baseada na integração da função de verossimilhança em relação aos efeitos fixos. Esse estimador é comparado com o de máxima verossimilhança. Neste caso, usamos o algoritmo SAEM, para os dois métodos de estimação. Assume-se para os erros e efeitos aleatórios algumas distribuições simétricas multivariadas de escala de misturas de distribuições normais, que compõem a classe de distribuições de caudas pesadas, a saber: normal, t e slash. Por último propomos um modelo não linear mais flexível, em que não é assumida uma forma linear para a inclusão dos efeitos aleatórios. Em todos os casos utilizamos dados reais e estudos de simulação para avaliar as propriedades dos estimadores.

We study the recovery of missing data from multiple smart grid datasets within a matrix completion framework. The datasets contain the electrical magnitudes required for monitoring and control of the electricity distribution system. Each dataset is described by a low rank matrix. Different datasets are correlated as a result of containing measurements of different physical magnitudes generated by the same distribution system. To assess the validity of matrix completion techniques in the recovery of missing data, we characterize the fundamental limits when two correlated datasets are jointly recovered. We then proceed to evaluate the performance of Singular Value Thresholding (SVT) and Bayesian SVT (BSVT) in this setting. We show that BSVT outperforms SVT by simulating the recovery for different correlated datasets. The performance of BSVT displays the tradeoff behaviour described by the fundamental limit, which suggests that BSVT exploits the correlation between the datasets in an efficient manner., Comment: Extended version of submission to IEEE Data Science Workshop 2019

Traditional cluster analysis methods used in ordinal data, for instance k-means and hierarchical clustering, are mostly heuristic and lack statistical inference tools to compare among competing models. To address this we propose a latent transitional model, a finite mixture model that includes both observed and latent covariates and apply it for the first time to the case of longitudinal ordinal data. This model-based clustering model is an extension of the proportional odds model and includes a first-order transitional term, occasion effects and interactions which provide flexible ways to capture different time patterns by cluster as well as time-heterogeneous transitions. We estimate model parameters within a Bayesian setting using a Markov chain Monte Carlo scheme and block-wise Metropolis–Hastings sampling. We illustrate the model using 2001–2011 self-reported health status (SRHS) from the Household, Income and Labour Dynamics in Australia survey. SRHS is recorded as an ordinal variable with five levels: poor, fair, good, very good and excellent. Using the Widely Applicable Information Criterion for model comparison, we find evidence for six latent groups. Transitions in the original data and the estimated groups are visualized using heatmaps.

Using a sample from a population to estimate the proportion of the population with a certain category label is a broadly important problem. In the context of microbiome studies, this problem arises when researchers wish to use a sample from a population of microbes to estimate the population proportion of a particular taxon, known as the taxon's relative abundance. In this paper, we propose a beta-binomial model for this task. Like existing models, our model allows for a taxon's relative abundance to be associated with covariates of interest. However, unlike existing models, our proposal also allows for the overdispersion in the taxon's counts to be associated with covariates of interest. We exploit this model in order to propose tests not only for differential relative abundance, but also for differential variability. The latter is particularly valuable in light of speculation that dysbiosis, the perturbation from a normal microbiome that can occur in certain disease conditions, may manifest as a loss of stability, or increase in variability, of the counts associated with each taxon. We demonstrate the performance of our proposed model using a simulation study and an application to soil microbial data.

Experimental designs such as matched-pair or longitudinal studies yield mRNA sequencing (mRNA-Seq) counts that are correlated across samples. Most of the approaches for the analysis of correlated mRNA-Seq data are restricted to a specific design and/or balanced data only (with the same number of samples in each group). We propose a model that is applicable to the analysis of correlated mRNA-Seq data of different types: paired, clustered, longitudinal, or others. Any combination of explanatory variables, as well as unbalanced data, can be processed within the proposed modeling framework. The model assumes that exon counts of a particular gene of an individual sample jointly follow a multivariate negative-binomial distribution. Additional correlation between exon counts obtained for, for example, individual samples within the same pair or cluster, is taken into account by including into the model a cluster-level normally distributed random effect. An interesting feature of the model is that it provides explicit expression for marginal correlation between exon counts at different levels. The performance of the model is evaluated by using a simulation study and an analysis of two real-life data sets: a paired mRNA-Seq experiment for 24 patients with clear-cell renal-cell carcinoma and a longitudinal mRNA-Seq experiment for 29 patients with Lyme disease.

The vast and complex wealth of information available to researchers often leads to a systematic review, which involves a detailed and comprehensive plan and search strategy with the goal of identifying, appraising, and synthesizing all relevant studies on a particular topic. A meta–analysis, conducted ideally as part of a comprehensive systematic review, statistically synthesizes evidence from multiple independent studies to produce one overall conclusion. The increasingly widespread use of meta–analysis has led to growing interest in meta–analytic methods for rare events and sparse data. Conventional approaches tend to perform very poorly in such settings. Recent work in this area has provided options for sparse data, but these are still often hampered when heterogeneity across the available studies differs based on treatment group. Heterogeneity arises when participants in a study are more correlated than participants across studies, often stemming from differences in the administration of the treatment, study design, or measurement of the outcome. We propose several new exact methods that accommodate this common contingency, providing more reliable statistical tests when such patterns on heterogeneity are observed. First, we develop a permutation–based approach that can also be used as a basis for computing exact confidence intervals when estimating the effect size. Second, we extend the permutation–based approach to the network meta–analysis setting. Third, we develop a new exact confidence distribution approach for effect size estimation. We show these new methods perform markedly better than traditional methods when events are rare, and heterogeneity is present.

Many studies handle binary longitudinal data by model-based classifiers, although their assumptions are often unsatisfied with real data. In contrast, tree-based classifiers are free of distributional assumptions, and several of these classifiers use bootstrap samples to construct multiple trees and combine them, in order to reduce prediction error. For paired data, Adler et al. (2011a) compared two types of bagging, namely subject-based strategies and observation-based strategies (one, all). In this thesis, we extended these strategies to longitudinal data. To evaluate the performance of the strategies, we compared them twice. Subject-based-strategies are classified by bagging, and the observation-based strategies are classified either by bagged or a single tree. We found the random (bootstrap) to be the best strategy whether the covariates are time-fixed or time-varying. We illustrated the five strategies on a subset of a well-known dataset on mother’s stress and children’s morbidity.

© 2016, Copyright © Taylor & Francis Group, LLC. The combined model as introduced by Molenberghs et al. (2007, 2010) has been shown to be an appealing tool for modeling not only correlated or overdispersed data but also for data that exhibit both these features. Unlike techniques available in the literature prior to the combined model, which use a single random-effects vector to capture correlation and/or overdispersion, the combined model allows for the correlation and overdispersion features to be modeled by two sets of random effects. In the context of count data, for example, the combined model naturally reduces to the Poisson-normal model, an instance of the generalized linear mixed model in the absence of overdispersion and it also reduces to the negative-binomial model in the absence of correlation. Here, a Poisson model is specified as the parent distribution of the data conditional on a normally distributed random effect at the subject or cluster level and/or a gamma distribution at observation level. Importantly, the development of the combined model and surrounding derivations have relevance well beyond mere data analysis. It so happens that the combined model can also be used to simulate correlated data. If a researcher is interested in comparing marginal models via Monte Carlo simulations, a necessity to generate suitable correlated count data arises. One option is to induce correlation via random effects but calculation of such quantities as the bias is then not straightforward. Since overdispersion and correlation are simultaneous features of longitudinal count data, the combined model presents an appealing framework for generating data to evaluate statistical properties, through a pre-specification of the desired marginal mean (possibly in terms of the covariates and marginal parameters) and a marginal variance-covariance structure. By comparing the marginal mean and variance of the combined model to the desired or pre-specified marginal mean and variance, respectively, the implied hierarchical parameters and the variance-covariance matrices of the normal and Gamma random effects are then derived from which correlated Poisson data are generated. We explore data generation when a random intercept or random intercept and slope model is specified to induce correlation. The data generator, however, allows for any dimension of the random effects although an increase in the random-effects dimension increases the sensitivity of the derived random effects variance-covariance matrix to deviations from positive-definiteness. A simulation study is carried out for the random-intercept model and for the random intercept and slope model, with or without the normal and Gamma random effects. We also pay specific attention to the case of serial correlation. peerreview_statement: The publishing and review policy for this title is described in its Aims & Scope. aims_and_scope_url: http://www.tandfonline.com/action/journalInformation?show=aimsScope&journalCode=lssp20 ispartof: Communications in Statistics: Simulation and Computation vol:45 issue:7 pages:2491-2510 ispartof: location:PORTUGAL, Caparica status: published

Differential Privacy (DP) has received increasing attention as a rigorous privacy framework. Many existing studies employ traditional DP mechanisms (e.g., the Laplace mechanism) as primitives to continuously release private data for protecting privacy at each time point (i.e., event-level privacy), which assume that the data at different time points are independent, or that adversaries do not have knowledge of correlation between data. However, continuously generated data tend to be temporally correlated, and such correlations can be acquired by adversaries. In this paper, we investigate the potential privacy loss of a traditional DP mechanism under temporal correlations. First, we analyze the privacy leakage of a DP mechanism under temporal correlation that can be modeled using Markov Chain. Our analysis reveals that, the event-level privacy loss of a DP mechanism may \textit{increase over time}. We call the unexpected privacy loss \textit{temporal privacy leakage} (TPL). Although TPL may increase over time, we find that its supremum may exist in some cases. Second, we design efficient algorithms for calculating TPL. Third, we propose data releasing mechanisms that convert any existing DP mechanism into one against TPL. Experiments confirm that our approach is efficient and effective., Comment: accepted in TKDE special issue "Best of ICDE 2017". arXiv admin note: substantial text overlap with arXiv:1610.07543

Several statistical methods are nowadays available for the analysis of gene expression data recorded through microarray technology. In this article, we take a closer look at several Gaussian mixture models which have recently been proposed to model gene expression data. It can be shown that these are special cases of a more general model, called the mixture of structural equation models (mixture of SEMs), which has been developed in psychometrics. This model combines mixture modelling and SEMs by assuming that component-specific means and variances are subject to a SEM. The connection with SEM is useful for at least two reasons: (1) it shows the basic assumptions of existing methods more explicitly and (2) it helps in straightforward development of alternative mixture models for gene expression data with alternative mean/covariance structures. Different specifications of mixture of SEMs for clustering gene expression data are illustrated using two benchmark datasets.Keywords: biclustering, correlated data, microarray data, mixture of SEMs, simultaneous clustering and dimensional reduction

Clinical studies enable us to measure many longitudinales variables. When our goal is to find a link between a response and some covariates, one can use regularisation methods, such as LASSO which have been extended to Generalized Estimating Equations (GEE). They allow us to select a subgroup of variables of interest taking into account intra-patient correlations. Databases often have unfilled data and measurement problems resulting in inevitable missing data. The objective of this thesis is to integrate missing data for variable selection in the presence of longitudinal data. We use mutiple imputation and introduce a new imputation function for the specific case of variables under detection limit. We provide a new variable selection method for correlated data that integrate missing data : the Multiple Imputation Penalized Generalized Estimating Equations (MI-PGEE). Our operator applies the group-LASSO penalty on the group of estimated regression coefficients of the same variable across multiply-imputed datasets. Our method provides a consistent selection across multiply-imputed datasets, where the optimal shrinkage parameter is chosen by minimizing a BIC-like criteria. We then present an application on knee osteoarthritis aiming to select the subset of biomarkers that best explain the differences in joint space width over time., Les études cliniques permettent de mesurer de nombreuses variables répétées dans le temps. Lorsque l'objectif est de les relier à un critère clinique d'intérêt, les méthodes de régularisation de type LASSO, généralisées aux Generalized Estimating Equations (GEE) permettent de sélectionner un sous-groupe de variables en tenant compte des corrélations intra-patients. Les bases de données présentent souvent des données non renseignées et des problèmes de mesures ce qui entraîne des données manquantes inévitables. L'objectif de ce travail de thèse est d'intégrer ces données manquantes pour la sélection de variables en présence de données longitudinales. Nous utilisons la méthode d'imputation multiple et proposons une fonction d'imputation pour le cas spécifique des variables soumises à un seuil de détection. Nous proposons une nouvelle méthode de sélection de variables pour données corrélées qui intègre les données manquantes : le Multiple Imputation Penalized Generalized Estimating Equations (MI-PGEE). Notre opérateur utilise la pénalité group-LASSO en considérant l'ensemble des coefficients de régression estimés d'une même variable sur les échantillons imputés comme un groupe. Notre méthode permet une sélection consistante sur l'ensemble des imputations, et minimise un critère de type BIC pour le choix du paramètre de régularisation. Nous présentons une application sur l'arthrose du genoux où notre objectif est de sélectionner le sous-groupe de biomarqueurs qui expliquent le mieux les différences de largeur de l'espace articulaire au cours du temps.

This paper presents a Photovoltaic (PV) power conversion model and a forecasting approach which uses spatial dependency of variables along with their temporal information. The power produced by a PV plant is forecasted by a PV conversion model using the predictions of three weather variables, namely, irradiance on the tilted plane, ambient temperature, and wind speed. The predictions are accomplished using a spatio-temporal algorithm that exploits the sparsity of correlations between time series data of different meteorological stations in the same region. The performances of the forecasting algorithm as well as the PV conversion model are investigated using real data recorded at various locations in Italy. The comparisons with various benchmark methods show the effectiveness of the proposed approaches over short-term forecasts.

Analysis of longitudinal count data has, for long, been done using a generalized linear mixed model (GLMM), in its Poisson-normal version, to account for correlation by specifying normal random effects. Univariate counts are often handled with the negativebinomial (NEGBIN) model taking into account overdispersion by use of Gamma random effects. Inherently though, longitudinal count data commonly exhibit both features of correlation and overdispersion simultaneously, necessitating analysis methodology that can account for both. The introduction of the combined model (CM) by (Molenberghs, Verbeke, and Dem´etrio 2007) and (Molenberghs, Verbeke, Dem´etrio, and Vieira 2010) serves this purpose, not only for count data but for the general exponential family of distributions. Here, a Poisson model is specified as the parent distribution of the data with a normally distributed random effect at the subject or cluster level and/or a gamma distribution at observation level. The GLMM and NEGBIN are special cases. Data can be simulated from (1) the general CM, with random effects, or, (2) its marginal version directly. This paper discusses an implementation of (1) in SAS software (SAS Institute Inc. (2011)). One needs to reflect on the mean of both the combined (hierarchical) and marginal models in order to generate correlated and/or overdispersed counts. A pre-specification of the desired marginal mean (in terms of covariates and marginal parameters), a marginal variance-covariance structure and the hierarchical mean (in terms of covariates and regression parameters) is required. The implied hierarchical parameters, the variance-covariance matrix of the random effects, and the variance-covariance matrix of the overdispersion part are then derived from which correlated Poisson data are generated. Sample calls of the SAS macro are presented as well as output. The authors gratefully acknowledge support from IAP Research Network P7/06 of the Belgian Government (Belgian Science Policy).

Incomplete data often arise in various clinical trials such as crossover trials, equivalence trials, and pre and post-test comparative studies. Various methods have been developed to construct confidence interval (CI) of risk difference or risk ratio for incomplete paired binary data. But, there is little works done on incomplete continuous correlated data. To this end, this manuscript aims to develop several approaches to construct CI of the difference of two means for incomplete continuous correlated data. Large sample method, hybrid method, simple Bootstrap-resampling method based on the maximum likelihood estimates (B 1) and Ekbohm’s unbiased estimator (B 2), and percentile Bootstrap-resampling method based on the maximum likelihood estimates (B 3) and Ekbohm’s unbiased estimator (B 4) are presented to construct CI of the difference of two means for incomplete continuous correlated data. Simulation studies are conducted to evaluate the performance of the proposed CIs in terms of empirical coverage probability, expected interval width, and mesial and distal non-coverage probabilities. Empirical results show that the Bootstrap-resampling-based CIs B 1, B 2, B 4 behave satisfactorily for small to moderate sample sizes in the sense that their coverage probabilities could be well controlled around the pre-specified nominal confidence level and the ratio of their mesial non-coverage probabilities to the non-coverage probabilities could be well controlled in the interval [0.4, 0.6]. If one would like a CI with the shortest interval width, the Bootstrap-resampling-based CIs B 1 is the optimal choice.

Correlated data frequently arise in contexts such as, for example, repeated measures and meta-analysis. The amount of information in such data depends not only on the sample size, but also on the structure and strength of the correlations among observations from the same independent block. A general concept is discussed, the effective sample size, as a way of quantifying the amount of information in such data. It is defined as the sample size one would need in an independent sample to equal the amount of information in the actual correlated sample. This concept is widely applicable, for Gaussian data and beyond, and provides important insight. For example, it helps explain why fixed-effects and random-effects inferences of meta-analytic data can be so radically divergent. Further, we show that in some cases the amount of information is bounded, even when the number of measures per independent block approaches infinity. We use the method to devise a new denominator degrees-of-freedom method for fixed-effects testing. It is compared to the classical Satterthwaite and Kenward-Roger methods for performance and, more importantly, to enhance insight. A key feature of the proposed degrees-of-freedotn method is that it, unlike the others, can be used for non-Gaussian data, too. This article has supplementary material online.

The start-up phase data of a process are the spine of traditional SPC charting and testing methods and are usually assumed to be i.i.d. observations from the in-control distribution. In this work a new method is proposed to model normally distributed start-up phase data where we allow for serial dependence and randomly occurring unidirectional level shifts of the underlying parameter of interest. The theoretic development is based on a Bayesian sequentially updated EWMA model with normal mixture errors. The new approach makes use of available prior information and provides a framework for drawing decisions and making prediction on line, even with a single observation. Copyright © 2008 John Wiley & Sons, Ltd.

Living organisms can be viewed as complex biological machines. In order to function, they must regulate their internal mechanism to do the right thing, at the right time, and in the right amount. Part of this regulation is encoded in gene regulatory networks. These are built up of genes which produce special proteins (transcription factors, tf) that regulate other tf-producing genes. Thus a network is formed with genes (nodes) linked together by their mutual regulation (edges).By constructing simplified models, we investigate such gene networks. The models allow us to probe general principles behind what shapes these networks (paper II), as well as specific networks such as that which endows the plant Arabidopsis thaliana with the ability to predict dawn and dusk (paper III). We also present a model for dynamically generating transcriptional networks which encode function from a single variable-length binary representation of dna (string of ones and zeroes). This gives a natural way for the network to evolve by mutations. However, performing a meaningful and efficient crossover operation on two dna strings of different length becomes a challenge. We address this by introducing a heuristic algorithm, which we compare against existing methods (paper IV).Additionally, we present a correct error estimation for the popular least squares method that is valid also for nonlinear functions applied to highly correlated data (paper I). For model fitting to correlated data, one has previously been constrained to use either a maximum likelihood approach, which leads to strong bias in the estimated parameters, or a least squares approach, which gives an incorrect error estimate. We also derive the first order contribution of the bias for both the maximum likelihood and the least squares method, and introduce a minimum variance function fitting method suited for Brownian motion.

This paper presents a solar power forecasting scheme, which uses spatial and temporal time series data along with a photovoltaic (PV) power conversion model. The PV conversion model uses the forecast of three different variables, namely, irradiance on the tilted plane, ambient temperature, and wind speed, in order to estimate the power produced by a PV plant at the grid connection terminals. The forecast values are obtained using a spatio-Temporal method that uses the data recorded from a target meteorological station as well as data of its surrounding stations. The proposed forecasting method exploits the sparsity of correlations between time series data in a collection of stations. The performance of both the PV conversion model and the spatio-Temporal algorithm is evaluated using high-resolution real data recorded in various locations in Italy. Comparison with other benchmark methods illustrates that the proposed method significantly improves the solar power forecasts, particularly over short-Term horizons.

The objective of this study was to evaluate the efficiency of spatial statistical analysis in the selection of genotypes in a plant breeding program and, particularly, to demonstrate the benefits of the approach when experimental observations are not spatially independent. The basic material of this study was a yield trial of soybean lines, with five check varieties (of fixed effect) and 110 test lines (of random effects), in an augmented block design. The spatial analysis used a random field linear model (RFML), with a covariance function estimated from the residuals of the analysis considering independent errors. Results showed a residual autocorrelation of significant magnitude and extension (range), which allowed a better discrimination among genotypes (increase of the power of statistical tests, reduction in the standard errors of estimates and predictors, and a greater amplitude of predictor values) when the spatial analysis was applied. Furthermore, the spatial analysis led to a different ranking of the genetic materials, in comparison with the non-spatial analysis, and a selection less influenced by local variation effects was obtained. O objetivo deste trabalho foi avaliar a eficiência da análise estatística espacial na seleção de genótipos de plantas num programa de melhoramento. Buscou-se demonstrar os benefícios potenciais dessa abordagem quando as observações experimentais não são espacialmente independentes. O material consistiu de um ensaio de competição de linhagens de soja, com cinco cultivares testemunhas (de efeitos fixos) e 110 novos genótipos (de efeitos aleatórios), delineado em blocos aumentados. O ajuste espacial foi feito pelo modelo linear de campo aleatório (RFLM), com função de autocovariância estimada a partir dos resíduos da análise sob erros independentes. Os resultados apontaram uma autocorrelação residual de magnitude e alcance significativos, o que garantiu à abordagem espacial uma melhoria considerável na discriminação dos tratamentos genéticos - aumento do poder dos testes estatísticos, redução nos erros padrão de estimativas e de preditores e alargamento na amplitude das predições genotípicas. A análise espacial levou a um diferente ordenamento das linhagens em relação à análise não espacial e, finalmente, a uma seleção menos influenciada por efeitos da variação local.

The objective of this study was to evaluate the efficiency of spatial statistical analysis in the selection of genotypes in a plant breeding program and, particularly, to demonstrate the benefits of the approach when experimental observations are not spatially independent. The basic material of this study was a yield trial of soybean lines, with five check varieties (of fixed effect) and 110 test lines (of random effects), in an augmented block design. The spatial analysis used a random field linear model (RFML), with a covariance function estimated from the residuals of the analysis considering independent errors. Results showed a residual autocorrelation of significant magnitude and extension (range), which allowed a better discrimination among genotypes (increase of the power of statistical tests, reduction in the standard errors of estimates and predictors, and a greater amplitude of predictor values) when the spatial analysis was applied. Furthermore, the spatial analysis led to a different ranking of the genetic materials, in comparison with the non-spatial analysis, and a selection less influenced by local variation effects was obtained.O objetivo deste trabalho foi avaliar a eficiência da análise estatística espacial na seleção de genótipos de plantas num programa de melhoramento. Buscou-se demonstrar os benefícios potenciais dessa abordagem quando as observações experimentais não são espacialmente independentes. O material consistiu de um ensaio de competição de linhagens de soja, com cinco cultivares testemunhas (de efeitos fixos) e 110 novos genótipos (de efeitos aleatórios), delineado em blocos aumentados. O ajuste espacial foi feito pelo modelo linear de campo aleatório (RFLM), com função de autocovariância estimada a partir dos resíduos da análise sob erros independentes. Os resultados apontaram uma autocorrelação residual de magnitude e alcance significativos, o que garantiu à abordagem espacial uma melhoria considerável na discriminação dos tratamentos genéticos - aumento do poder dos testes estatísticos, redução nos erros padrão de estimativas e de preditores e alargamento na amplitude das predições genotípicas. A análise espacial levou a um diferente ordenamento das linhagens em relação à análise não espacial e, finalmente, a uma seleção menos influenciada por efeitos da variação local.

The paper describes the analysis of data originating from the German Deep Drill Program. The amount of ‘cataclastic rocks’ is modelled with data resulting from a series of measurements taken from deep drill samples ranging from 1000 up to 5000 m depth. The measurements thereby describe the amount of strongly deformed rock particles and serve as indicator for the occurrence of cataclastic shear zones, which are areas of severely‘ground’ stones due to movements of different layers in the earth crust. The data represent a ‘depth series’ as analogue to a ‘time series’, with mean, dispersion and correlation structure varying in depth. The general smooth structure is thereby disturbed by peaks and outliers so that robust procedures have to be applied for estimation. In terms of statistical modelling technology three different peculiarities of the data have to be tackled simultaneously, that is estimation of the correlation structure, local bandwidth selection and robust smoothing. To do so, existing routines are adapted and combined in new ‘two-stage’ estimation procedures.

PURPOSE: This study evaluated an Adolescent Community Reinforcement Approach (A-CRA) and Assertive Continuing Care (ACC) program targeting Hispanic adolescents at risk for substance abuse. METHOD: The Clinic for Education, Treatment, and Prevention of Addiction (CETPA, Inc.), a behavioral health provider offering culturally appropriate substance use and mental health services, carried out the intervention. We examined longitudinal substance use data in relation to time spent in the program and possible confounders. RESULTS: We analyzed data from 72 adolescent clients collected between 2010 and 2012. Self-reported data were evaluated to determine if time spent in the program was associated with substance use reduction. The data were correlated, zero-inflated, and overdispersed; consequently, we employed a mixed-effects zero-inflated negative-binomial model. Time spent in CETPA’s program was significantly associated with reductions in the number of days of substance use (p= .039), but not with the likelihood of fully abstaining from use (p= .290). For non-abstinent participants who spend a year in the program, our models revealed an average decline of 46% in reported days of substance use. CONCLUSIONS: A culturally tailored and age-appropriate substance abuse program for Hispanic adolescents resulted in a significant reduction of the numbers of days using alcohol, drugs, or other illicit substances. The A-CRA/ACC approach can yield successful results in culturally diverse settings.

Blind deformulation is an important stake for several industries. This work was motivated by the identification and quantification of contaminants originated from food packaging systems. Many substances originating from plastic materials are indeed suspected to be endocrine disruptors but remain chiefly difficult to separate with spectroscopic techniques. We propose a tailored two-scale pursuit methodology to identify and quantify an arbitrary number of substances from the H-1 NMR spectrum of the mixture. Identified substances are included within a library of spectra and can be combined with undocumented ones. To preserve the initial resolution of NMR spectra, peak lines are spanned onto Gaussian kernels so that they can be identified, even when the positions and shapes of multiplets in the mixture are modified within tolerance ranges or when multiplets are overlapping. The deconvolution procedure starts with a crude pairwise search to build a list of likely substances, which is subsequently expanded as nested scenarios. Scenarios are built according to the risk of confusing similar substances. Quantification is carried out on a preference list of substances selected as in a voting system. Using a primary library of 52 substances (corresponding to 279 multiplets and 5620 lines), the reliability and robustness of the method were tested extensively in numerical experiments and by performing the brute-force deformulation of five processed common thermoplastics.

Assessing between-study variability in the context of conventional random-effects meta-analysis is notoriously difficult when incorporating data from only a small number of historical studies. In order to borrow strength, historical and current data are often assumed to be fully homogeneous, but this can have drastic consequences for power and Type I error if the historical information is biased. In this paper, we propose empirical and fully Bayesian modifications of the commensurate prior model (Hobbs et al., 2011) extending Pocock (1976), and evaluate their frequentist and Bayesian properties for incorporating patient-level historical data using general and generalized linear mixed regression models. Our proposed commensurate prior models lead to preposterior admissible estimators that facilitate alternative bias-variance trade-offs than those offered by pre-existing methodologies for incorporating historical data from a small number of historical studies. We also provide a sample analysis of a colon cancer trial comparing time-to-disease progression using a Weibull regression model.

In survival analysis, researchers often encounter multivariate survival time data, in which failure times are correlated even in the presence of model covariates. It is argued that because observations are clustered by unobserved heterogeneity, the application of standard survival models can result in biased parameter estimates and erroneous model-based predictions. In this article, the author describes and compares four methods handling unobserved heterogeneity in survival analysis: the Andersen-Gill approach, the robust sandwich variance estimator, the hazard model with individual frailty, and the retransformation method. An empirical analysis provides strong evidence that in the presence of strong unobserved heterogeneity, the application of a standard survival model can yield equally robust parameter estimates and the likelihood ratio statistic as does a corresponding model adding an additional parameter for random effects. When predicting the survival function, however, a standard model on multivariate survival time data can result in serious prediction bias. The retransformation method is effective to derive an adjustment factor for correctly predicting the survival function.

We consider parametric accelerated failure time models with random effects to predict the probability of possibly correlated failures occurring in oil wells. In this context, we first consider empirical Bayes predictors (EBP) based on aWeibull distribution for the failure times and on a Gaussian distribution for the random effects.We also obtain empirical best linear unbiased predictors (EBLUP) using a linear mixed model for which the form of the distribution of the random effects is not specified. We compare both approaches using data obtained from an oil-drilling company and suggest how the results may be employed in designing a preventive maintenance program.

International audience; Let $Y\in\R^n$ be a random vector with mean $s$ and covariance matrix $\sigma^2P_n\tra{P_n}$ where $P_n$ is some known $n\times n$-matrix. We construct a statistical procedure to estimate $s$ as well as under moment condition on $Y$ or Gaussian hypothesis. Both cases are developed for known or unknown $\sigma^2$. Our approach is free from any prior assumption on $s$ and is based on non-asymptotic model selection methods. Given some linear spaces collection $\{S_m,\ m\in\M\}$, we consider, for any $m\in\M$, the least-squares estimator $\hat{s}_m$ of $s$ in $S_m$. Considering a penalty function that is not linear to the dimensions of the $S_m$'s, we select some $\hat{m}\in\M$ in order to get an estimator $\hat{s}_{\hat{m}}$ with a quadratic risk as close as possible to the minimal one among the risks of the $\hat{s}_m$'s. Non-asymptotic oracle-type inequalities and minimax convergence rates are proved for $\hat{s}_{\hat{m}}$. A special attention is given to the estimation of a non-parametric component in additive models. Finally, we carry out a simulation study in order to illustrate the performances of our estimators in practice.

Statistical models of marine ecosystems use a variety of data sources to estimate parameters using composite or weighted likelihood functions with associated weighting issues and questions on how to obtain variance estimates. Regardless of the method used to obtain point estimates, a method is required for variance estimation. A bootstrap technique is introduced for the evaluation of uncertainty in such models, taking into account inherent spatial and temporal correlations in the datasets, which are commonly transferred as assumptions from a likelihood estimation procedure into Hessian-based variance estimation procedures. The technique is demonstrated on a real dataset and the effects of the number of bootstrap samples on estimation bias and variance estimates are studied. Although the modelling framework and bootstrap method can be applied to multispecies and multiarea models, for clarity the case study described is of a single-species and single-area model.Keywords: bootstrapping, correlated data, fish population dynamics, non-linear modelsAfrican Journal of Marine Science 2014, 36(1): 99–110

The generalized linear models (GLM) lately received a lot of attention in modelling the insurance data. However, the violation of assumptions about the independence of underlying data set often causes problems and misinterpretation of achieved results. The need for more exible instruments has been spoken out and consequently various proposals have been made. This thesis deals with GLM based techniques enabling to handle correlated data sets. The usage have been made of generalized linear mixed models (GLMM) and generalized estimating equations (GEE). The main aim of this thesis is to provide a solid statistical background and perform a practical application to demonstrate and compare features of various models. Powered by TCPDF (www.tcpdf.org)

Correlated data with complex association structures arise from longitudinal studies and cluster randomized trials. However, some methodological challenges in the design and analysis of such studies or trials have not been overcome. In this thesis, we address three of the challenges: 1) Power analysis for population based longitudinal study investigating gene-environment interaction effects on chronic disease: For longitudinal studies with interest in investigating the gene-environment interaction in disease susceptibility and progression, rigorous statistical power estimation is crucial to ensure that such studies are scientifically useful and cost-effective since human genome epidemiology is expensive. However conventional sample size calculations for longitudinal study can seriously overestimate the statistical power due to overlooking the measurement error, unmeasured etiological determinants, and competing events that can impede the occurrence of the event of interest. 2) Comparing the performance of different multiple imputation strategies for missing binary outcomes in cluster randomized trials: Though researchers have proposed various strategies to handle missing binary outcome in cluster randomized trials (CRTs), comprehensive guidelines on the selection of the most appropriate or optimal strategy are not available in the literature. 3) Comparison of population-averaged and cluster-specific models for the analysis of cluster randomized trials with missing binary outcome: Both population-averaged and cluster-specific models are commonly used for analyzing binary outcomes in CRTs. However, little attention has been paid to their accuracy and efficiency when analyzing data with missing outcomes. The objective of this thesis is to provide researchers recommendations and guidance for future research in handling the above issues. Doctor of Philosophy (PhD)

© 2014 Taylor & Francis Group, LLC. Generalized Estimating Equations (GEE) are a widespread tool for modeling correlated data, based on properly formulating a marginal regression function, combined with working assumptions about the correlation function. Should interest be placed in addition on the correlation function, then, apart from second-order GEE, pseudo-likelihood (PL) also provides an attractive alternative, especially in its pairwise form, where the covariance between each pair of the response vector is modeled as well. An elegant PL approach is formulated in this paper, based on a flexible bivariate Poisson model. The performance of the PL-method is studied, relative to GEE, using simulations. Data on repeated counts of epileptic seizures in a two-arm clinical trial are analyzed. A macro has been developed by the authors and made available on their web pages. peerreview_statement: The publishing and review policy for this title is described in its Aims & Scope. aims_and_scope_url: http://www.tandfonline.com/action/journalInformation?show=aimsScope&journalCode=lsta20 ispartof: Communications in Statistics: Simulation and Computation vol:43 issue:22 pages:4790-4805 status: published