The context for this research is statistical inference, the process of making predictions or inferences about a population from observation and analyses of a sample. In this context, many researchers want to grasp what inferences can be made that are valid, in the sense of being able to uphold or justify by argument or evidence. Another pressing question among users of statistical methods is: how can spurious relationships be distinguished from genuine ones? Underlying both of these issues is the concept of evidence. In response to these (and similar) questions, two questions I work on in this essay are: (1) what is a genuine principle of evidence? and (2) do error probabilities have more than a long-run role? Concisely, I propose that felicitous genuine principles of evidence should provide concrete guidelines on precisely how to examine error probabilities, with respect to a test's aptitude for unmasking pertinent errors, which leads to establishing sound interpretations of results from statistical techniques. The starting point for my definition of genuine principles of evidence is Allan Birnbaum's confidence concept, an attempt to control misleading interpretations. However, Birnbaum's confidence concept is inadequate for interpreting statistical evidence, because using only pre-data error probabilities would not pick up on a test's ability to detect a discrepancy of interest (e.g., "even if the discrepancy exists" with respect to the actual outcome. Instead, I argue that Deborah Mayo's severity assessment is the most suitable characterization of evidence based on my definition of genuine principles of evidence. / Master of Arts

Dans les études géophysiques, l'assimilation de données a pour but d'estimer l'état d'un système ou les paramètres d'un modèle physique de façon optimale. Pour ce faire, l'assimilation de données a besoin de trois types d'informations : des observations, un modèle physique/numérique et une description statistique de l'incertitude associée aux paramètres du système. Dans ma thèse, de nouvelles méthodes d'assimilation de données sont utilisées pour l'étude de la physico-chimie de l'atmosphère: (i) On y utilise de manière conjointe la méthode 4D-Var avec un modèle sous-maille statistique pour tenir compte des erreurs de représentativité. (ii) Des échelles multiples sont prises en compte dans la méthode d'estimation BLUE. (iii) Enfin, la méthode du maximum de vraisemblance est appliquée pour estimer des hyper-paramètres qui paramètrisent les erreurs à priori. Ces trois approches sont appliquées de manière spécifique à des problèmes de modélisation inverse des sources de polluant atmosphérique. Dans une première partie, la modélisation inverse est utilisée afin d'estimer les émissions de monoxyde de carbone sur un domaine représentant la France. Les stations du réseau d'observation considérées sont impactées par les erreurs de représentativité. Un modèle statistique sous-maille est introduit. Il est couplé au système 4D-Var afin de réduire les erreurs de représentativité. En particulier, les résultats de la modélisation inverse montrent que la méthode 4D-Var seule n'est pas adaptée pour gérer le problème de représentativité. Le système d'assimilation des données couplé conduit à une meilleure représentation de la variabilité de la concentration de CO avec une amélioration très significative des indicateurs statistiques. Dans une deuxième partie, on évalue le potentiel du réseau IMS (International Monitoring System) du CTBTO pour l'inversion d'une source accidentelle de radionucléides. Pour évaluer la performance du réseau, une grille multi-échelle adaptative pour l'espace de contrôle est optimisée selon un critère basé sur les degrés de liberté du signal (DFS). Les résultats montrent que plusieurs régions restent sous-observées par le réseau IMS. Dans la troisième et dernière partie, sont estimés les émissions de Composés Organiques Volatils (COVs) sur l'Europe de l'ouest. Cette étude d'inversion est faite sur la base des observations de 14 COVs extraites du réseau EMEP. L'évaluation des incertitudes des valeurs des inventaires d'émission et des erreurs d'observation sont faites selon le principe du maximum de vraisemblance. La distribution des inventaires d'émission a été supposée tantôt gaussienne et tantôt semi-normale. Ces deux hypothèses sont appliquées pour inverser le champs des inventaires d'émission. Les résultats de ces deux approches sont comparés. Bien que la correction apportée sur les inventaires est plus forte avec l'hypothèse Gaussienne que semi-normale, les indicateurs statistiques montrent que l'hypothèse de la distribution semi-normale donne de meilleurs résultats de concentrations que celle Gaussienne. / Data assimilation in geophysical sciences aims at optimally estimating the state of the system or some parameters of the system's physical model. To do so, data assimilation needs three types of information: observations and background information, a physical/numerical model, and some statistical description that prescribes uncertainties to each componenent of the system.In my dissertation, new methodologies of data assimilation are used in atmospheric chemistry and physics: the joint use of a 4D-Var with a subgrid statistical model to consistently account for representativeness errors, accounting for multiple scale in the BLUE estimation principle, and a better estimation of prior errors using objective estimation of hyperparameters. These three approaches will be specifically applied to inverse modelling problems focussing on the emission fields of tracers or pollutants. First, in order to estimate the emission inventories of carbon monoxide over France, in-situ stations which are impacted by the representativeness errors are used. A subgrid model is introduced and coupled with a 4D-Var to reduce the representativeness error. Indeed, the results of inverse modelling showed that the 4D-Var routine was not fit to handle the representativeness issues. The coupled data assimilation system led to a much better representation of theCO concentration variability, with a significant improvement of statistical indicators, and more consistent estimation of the CO emission inventory. Second, the evaluation of the potential of the IMS (International Monitoring System) radionuclide network is performed for the inversion of an accidental source. In order to assess the performance of the global network, a multiscale adaptive grid is optimised using a criterion based on degrees of freedom for the signal (DFS). The results show that several specific regions remain poorly observed by the IMS network. Finally, the inversion of the surface fluxes of Volatile Organic Compounds (VOC) are carried out over Western Europe using EMEP stations. The uncertainties of the background values of the emissions, as well as the covariance matrix of the observation errors, are estimated according to the maximum likelihood principle. The prior probability density function of the control parameters is chosen to be Gaussian or semi-normal distributed. Grid-size emission inventories are inverted under these two statistical assumptions. The two kinds of approaches are compared. With the Gaussian assumption, the departure between the posterior and the prior emission inventories is higher than when using the semi-normal assumption, but that method does not provide better scores than the semi-normal in a forecast experiment.

O problema de Inferência sobre uma proporção, amplamente divulgado na literatura estatística, ocupa papel central no desenvolvimento das várias teorias de Inferência Estatística e, invariavelmente, é objeto de investigação e discussão em estudos comparativos entre as diferentes escolas de Inferência. Ademais, a estimação de proporções, bem como teste de hipóteses para proporções, é de grande importância para as diversas áreas do conhecimento, constituindo um método quantitativo simples e universal. Nesse trabalho, é feito um estudo comparativo entre as abordagens clássica e bayesiana do problema de testar as hipóteses de ocorrência ou não de 2º turno em um cenário típico de eleição majoritária (maioria absoluta) em dois turnos no Brasil. / The problem of inference about a proportion, widely explored in the statistical literature, plays a key role in the development of several theories of statistical inference and, invariably, is the object of investigation and discussion in comparative studies among different schools of inference. In addition, the estimation of proportions, as well as test of hypothesis for proportions, is very important in many areas of knowledge as it constitutes a simple and universal quantitative method. In this work a comparative study between the Classical and Bayesian approaches to the problem of testing the hypothesis of occurrence of second round (or not) in a typical scenario of a majoritarian election (absolute majority) in two rounds in Brazil is developed.

Data assimilation in geophysical sciences aims at optimally estimating the state of the system or some parameters of the system's physical model. To do so, data assimilation needs three types of information: observations and background information, a physical/numerical model, and some statistical description that prescribes uncertainties to each componenent of the system.In my dissertation, new methodologies of data assimilation are used in atmospheric chemistry and physics: the joint use of a 4D-Var with a subgrid statistical model to consistently account for representativeness errors, accounting for multiple scale in the BLUE estimation principle, and a better estimation of prior errors using objective estimation of hyperparameters. These three approaches will be specifically applied to inverse modelling problems focussing on the emission fields of tracers or pollutants. First, in order to estimate the emission inventories of carbon monoxide over France, in-situ stations which are impacted by the representativeness errors are used. A subgrid model is introduced and coupled with a 4D-Var to reduce the representativeness error. Indeed, the results of inverse modelling showed that the 4D-Var routine was not fit to handle the representativeness issues. The coupled data assimilation system led to a much better representation of theCO concentration variability, with a significant improvement of statistical indicators, and more consistent estimation of the CO emission inventory. Second, the evaluation of the potential of the IMS (International Monitoring System) radionuclide network is performed for the inversion of an accidental source. In order to assess the performance of the global network, a multiscale adaptive grid is optimised using a criterion based on degrees of freedom for the signal (DFS). The results show that several specific regions remain poorly observed by the IMS network. Finally, the inversion of the surface fluxes of Volatile Organic Compounds (VOC) are carried out over Western Europe using EMEP stations. The uncertainties of the background values of the emissions, as well as the covariance matrix of the observation errors, are estimated according to the maximum likelihood principle. The prior probability density function of the control parameters is chosen to be Gaussian or semi-normal distributed. Grid-size emission inventories are inverted under these two statistical assumptions. The two kinds of approaches are compared. With the Gaussian assumption, the departure between the posterior and the prior emission inventories is higher than when using the semi-normal assumption, but that method does not provide better scores than the semi-normal in a forecast experiment.

O problema de Inferência sobre uma proporção, amplamente divulgado na literatura estatística, ocupa papel central no desenvolvimento das várias teorias de Inferência Estatística e, invariavelmente, é objeto de investigação e discussão em estudos comparativos entre as diferentes escolas de Inferência. Ademais, a estimação de proporções, bem como teste de hipóteses para proporções, é de grande importância para as diversas áreas do conhecimento, constituindo um método quantitativo simples e universal. Nesse trabalho, é feito um estudo comparativo entre as abordagens clássica e bayesiana do problema de testar as hipóteses de ocorrência ou não de 2º turno em um cenário típico de eleição majoritária (maioria absoluta) em dois turnos no Brasil. / The problem of inference about a proportion, widely explored in the statistical literature, plays a key role in the development of several theories of statistical inference and, invariably, is the object of investigation and discussion in comparative studies among different schools of inference. In addition, the estimation of proportions, as well as test of hypothesis for proportions, is very important in many areas of knowledge as it constitutes a simple and universal quantitative method. In this work a comparative study between the Classical and Bayesian approaches to the problem of testing the hypothesis of occurrence of second round (or not) in a typical scenario of a majoritarian election (absolute majority) in two rounds in Brazil is developed.

[pt] Em redes sem fio, o desvanecimento do sinal causado pela propagação por caminhos múltiplos pode ser mitigado através do uso de diversidade cooperativa [1–3]. Neste contexto, esquemas de seleção de repetidores são essenciais por causa de seu alto desempenho [4–6]. Esta tese é focada no desenvolvimento de técnicas de seleção de repetidores, que utilizam buffers. Como primeira contribuição, apresentamos uma estrutura de chaveamento para sistemas de repetidores MIMO em que um nó de origem pode transmitir diretamente para um nó de destino ou auxiliado por repetidores. Em particular, apresentamos uma nova técnica de seleção de repetidores baseada no chaveamento e seleção do melhor canal, denominada Switched Max-Link, que faz uso do critério de seleção Maximum Minimum Distance (MMD). Como segunda contribuição, apresentamos uma estratégia de seleção de repetidores para sistemas cooperativos de múltiplas antenas que são auxiliados por um nó processador central, em que um cluster formado por dois usuários é selecionado para transmitir simultaneamente um ao outro com a ajuda de repetidores. Em particular, apresentamos uma nova estratégia de seleção de repetidores Multi-Way com base na seleção do melhor link, explorando o uso de buffers e codificação de rede em camada física (PLNC), denominada Multi-Way Buffer-Aided Max-Link (MW-Max-Link). Como terceira contribuição, apresentamos uma estrutura de uplink dirigida por nuvem para sistemas de repetidores Multi-Way de múltiplas antenas, que ajuda na detecção conjunta de símbolos na nuvem, onde os usuários são selecionados para transmitir simultaneamente uns aos outros auxiliados por repetidores. Em particular, desenvolvemos um novo protocolo de seleção de repetidores Multi-Way com base na seleção do melhor link, explorando o uso de buffers em nuvem e PLNC, denominado Multi-Way Cloud-Driven Best-User-Link (MWC-Best-User-Link). É realizada uma análise das técnicas propostas e existentes em termos de custo computacional, probabilidade de erro de pareamento, soma das taxas e atraso médio e simulações são empregadas para avaliar o desempenho dessas técnicas. / [en] In wireless networks, signal fading caused by multipath propagation can be mitigated through the use of cooperative diversity [1–3]. In this context, relay selection schemes are key because of their high performance [4–6]. Thus, this thesis is focused on developing relay selection techniques, that uses buffers. In this work, as a first contribution, we present a switched relaying framework for multiple-input multiple-output (MIMO) relay systems where a source node may transmit directly to a destination node or aided by relays equipped with buffers. In particular, we develop a novel relay selection protocol based on switching and the selection of the best link, denoted as Switched Max-Link, that uses the novel Maximum Minimum Distance (MMD) relay selection criterion. After that, as a second contribution, we present a relay-selection strategy for multi-way cooperative multi-antenna systems that are aided by a central processor node, where a cluster formed by two users is selected to simultaneously transmit to each other with the help of relays. In particular, we present a novel multi-way relay selection strategy based on the selection of the best link, exploiting the use of buffers and physical-layer network coding (PLNC), that is called Multi-Way Buffer-Aided Max-Link (MW-Max-Link). Moreover, as a third contribution, we present a cloud-driven uplink framework for multi-way multiple-antenna relay systems which aids joint symbol detection in the cloud and where users are selected to simultaneously transmit to each other aided by relays. In particular, we develop a novel multi-way relay selection protocol based on the selection of the best link, exploiting the use of cloud buffers and PLNC, denoted as Multi-Way Cloud- Driven Best-User-Link (MWC-Best-User-Link). An analysis of the proposed and existing techniques in terms of computational cost, pairwise error probability, sum-rate and average delay is carried out. Simulations are then employed to evaluate the performance of these techniques.