Bayesian Inference in Large-scale Problems

Many modern applications fall into the category of "large-scale" statistical problems, in which both the number of observations n and the number of features or parameters p may be large. Many existing methods focus on point estimation, despite the continued relevance of uncertainty quantification in the sciences, where the number of parameters to estimate often exceeds the sample size, despite huge increases in the value of n typically seen in many fields. Thus, the tendency in some areas of industry to dispense with traditional statistical analysis on the basis that "n=all" is of little relevance outside of certain narrow applications. The main result of the Big Data revolution in most fields has instead been to make computation much harder without reducing the importance of uncertainty quantification. Bayesian methods excel at uncertainty quantification, but often scale poorly relative to alternatives. This conflict between the statistical advantages of Bayesian procedures and their substantial computational disadvantages is perhaps the greatest challenge facing modern Bayesian statistics, and is the primary motivation for the work presented here.

Two general strategies for scaling Bayesian inference are considered. The first is the development of methods that lend themselves to faster computation, and the second is design and characterization of computational algorithms that scale better in n or p. In the first instance, the focus is on joint inference outside of the standard problem of multivariate continuous data that has been a major focus of previous theoretical work in this area. In the second area, we pursue strategies for improving the speed of Markov chain Monte Carlo algorithms, and characterizing their performance in large-scale settings. Throughout, the focus is on rigorous theoretical evaluation combined with empirical demonstrations of performance and concordance with the theory.

One topic we consider is modeling the joint distribution of multivariate categorical data, often summarized in a contingency table. Contingency table analysis routinely relies on log-linear models, with latent structure analysis providing a common alternative. Latent structure models lead to a reduced rank tensor factorization of the probability mass function for multivariate categorical data, while log-linear models achieve dimensionality reduction through sparsity. Little is known about the relationship between these notions of dimensionality reduction in the two paradigms. In Chapter 2, we derive several results relating the support of a log-linear model to nonnegative ranks of the associated probability tensor. Motivated by these findings, we propose a new collapsed Tucker class of tensor decompositions, which bridge existing PARAFAC and Tucker decompositions, providing a more flexible framework for parsimoniously characterizing multivariate categorical data. Taking a Bayesian approach to inference, we illustrate empirical advantages of the new decompositions.

Latent class models for the joint distribution of multivariate categorical, such as the PARAFAC decomposition, data play an important role in the analysis of population structure. In this context, the number of latent classes is interpreted as the number of genetically distinct subpopulations of an organism, an important factor in the analysis of evolutionary processes and conservation status. Existing methods focus on point estimates of the number of subpopulations, and lack robust uncertainty quantification. Moreover, whether the number of latent classes in these models is even an identified parameter is an open question. In Chapter 3, we show that when the model is properly specified, the correct number of subpopulations can be recovered almost surely. We then propose an alternative method for estimating the number of latent subpopulations that provides good quantification of uncertainty, and provide a simple procedure for verifying that the proposed method is consistent for the number of subpopulations. The performance of the model in estimating the number of subpopulations and other common population structure inference problems is assessed in simulations and a real data application.

In contingency table analysis, sparse data is frequently encountered for even modest numbers of variables, resulting in non-existence of maximum likelihood estimates. A common solution is to obtain regularized estimates of the parameters of a log-linear model. Bayesian methods provide a coherent approach to regularization, but are often computationally intensive. Conjugate priors ease computational demands, but the conjugate Diaconis--Ylvisaker priors for the parameters of log-linear models do not give rise to closed form credible regions, complicating posterior inference. In Chapter 4 we derive the optimal Gaussian approximation to the posterior for log-linear models with Diaconis--Ylvisaker priors, and provide convergence rate and finite-sample bounds for the Kullback-Leibler divergence between the exact posterior and the optimal Gaussian approximation. We demonstrate empirically in simulations and a real data application that the approximation is highly accurate, even in relatively small samples. The proposed approximation provides a computationally scalable and principled approach to regularized estimation and approximate Bayesian inference for log-linear models.

Another challenging and somewhat non-standard joint modeling problem is inference on tail dependence in stochastic processes. In applications where extreme dependence is of interest, data are almost always time-indexed. Existing methods for inference and modeling in this setting often cluster extreme events or choose window sizes with the goal of preserving temporal information. In Chapter 5, we propose an alternative paradigm for inference on tail dependence in stochastic processes with arbitrary temporal dependence structure in the extremes, based on the idea that the information on strength of tail dependence and the temporal structure in this dependence are both encoded in waiting times between exceedances of high thresholds. We construct a class of time-indexed stochastic processes with tail dependence obtained by endowing the support points in de Haan's spectral representation of max-stable processes with velocities and lifetimes. We extend Smith's model to these max-stable velocity processes and obtain the distribution of waiting times between extreme events at multiple locations. Motivated by this result, a new definition of tail dependence is proposed that is a function of the distribution of waiting times between threshold exceedances, and an inferential framework is constructed for estimating the strength of extremal dependence and quantifying uncertainty in this paradigm. The method is applied to climatological, financial, and electrophysiology data.

The remainder of this thesis focuses on posterior computation by Markov chain Monte Carlo. The Markov Chain Monte Carlo method is the dominant paradigm for posterior computation in Bayesian analysis. It has long been common to control computation time by making approximations to the Markov transition kernel. Comparatively little attention has been paid to convergence and estimation error in these approximating Markov Chains. In Chapter 6, we propose a framework for assessing when to use approximations in MCMC algorithms, and how much error in the transition kernel should be tolerated to obtain optimal estimation performance with respect to a specified loss function and computational budget. The results require only ergodicity of the exact kernel and control of the kernel approximation accuracy. The theoretical framework is applied to approximations based on random subsets of data, low-rank approximations of Gaussian processes, and a novel approximating Markov chain for discrete mixture models.

Data augmentation Gibbs samplers are arguably the most popular class of algorithm for approximately sampling from the posterior distribution for the parameters of generalized linear models. The truncated Normal and Polya-Gamma data augmentation samplers are standard examples for probit and logit links, respectively. Motivated by an important problem in quantitative advertising, in Chapter 7 we consider the application of these algorithms to modeling rare events. We show that when the sample size is large but the observed number of successes is small, these data augmentation samplers mix very slowly, with a spectral gap that converges to zero at a rate at least proportional to the reciprocal of the square root of the sample size up to a log factor. In simulation studies, moderate sample sizes result in high autocorrelations and small effective sample sizes. Similar empirical results are observed for related data augmentation samplers for multinomial logit and probit models. When applied to a real quantitative advertising dataset, the data augmentation samplers mix very poorly. Conversely, Hamiltonian Monte Carlo and a type of independence chain Metropolis algorithm show good mixing on the same dataset.

Outlier effects on robust joint modelling of longitudinal and survival data

Robust joint modelling is an emerging field of research. Through the advancements in electronic patient healthcare records, the popularly of joint modelling approaches has grown rapidly in recent years providing simultaneous analysis of longitudinal and survival data. This research advances previous work through the development of a novel robust joint modelling methodology for one of the most common types of standard joint models, that which links a linear mixed model with a Cox proportional hazards model. Through t-distributional assumptions, longitudinal outliers are accommodated with their detrimental impact being down weighed and thus providing more efficient and reliable estimates. The robust joint modelling technique and its major benefits are showcased through the analysis of Northern Irish end stage renal disease patients. With an ageing population and growing prevalence of chronic kidney disease within the United Kingdom, there is a pressing demand to investigate the detrimental relationship between the changing haemoglobin levels of haemodialysis patients and their survival. As outliers within the NI renal data were found to have significantly worse survival, identification of outlying individuals through robust joint modelling may aid nephrologists to improve patient's survival. A simulation study was also undertaken to explore the difference between robust and standard joint models in the presence of increasing proportions and extremity of longitudinal outliers. More efficient and reliable estimates were obtained by robust joint models with increasing contrast between the robust and standard joint models when a greater proportion of more extreme outliers are present. Through illustration of the gains in efficiency and reliability of parameters when outliers exist, the potential of robust joint modelling is evident. The research presented in this thesis highlights the benefits and stresses the need to utilise a more robust approach to joint modelling in the presence of longitudinal outliers.

Effect of a Mediterranean Diet Intervention on Dietary Glycemic Load and Dietary Glycemic Index: The PREDIMED Study

[EN]To compare the one year effect of two dietary interventions with MeDiet on GL and GI in the PREDIMED trial. Methods. Participants were older subjects at high risk for cardiovascular disease. This analysis included 2866 nondiabetic subjects. Diet was assessed with a validated 137-item food frequency questionnaire (FFQ). The GI of each FFQ item was assigned by a 5-step methodology using the International Tables of GI and GL Values. Generalized linear models were fitted to assess the relationship between the intervention group and dietary GL and GI at one year of follow-up, using control group as reference.

Increasing the power of genome wide association studies in natural populations using repeated measures : evaluation and implementation

1. Genomewide association studies (GWAS) enable detailed dissections of the genetic basis for organisms' ability to adapt to a changing environment. In long-term studies of natural populations, individuals are often marked at one point in their life and then repeatedly recaptured. It is therefore essential that a method for GWAS includes the process of repeated sampling. In a GWAS, the effects of thousands of single-nucleotide polymorphisms (SNPs) need to be fitted and any model development is constrained by the computational requirements. A method is therefore required that can fit a highly hierarchical model and at the same time is computationally fast enough to be useful. 2. Our method fits fixed SNP effects in a linear mixed model that can include both random polygenic effects and permanent environmental effects. In this way, the model can correct for population structure and model repeated measures. The covariance structure of the linear mixed model is first estimated and subsequently used in a generalized least squares setting to fit the SNP effects. The method was evaluated in a simulation study based on observed genotypes from a long-term study of collared flycatchers in Sweden. 3. The method we present here was successful in estimating permanent environmental effects from simulated repeated measures data. Additionally, we found that especially for variable phenotypes having large variation between years, the repeated measurements model has a substantial increase in power compared to a model using average phenotypes as a response. 4. The method is available in the R package RepeatABEL. It increases the power in GWAS having repeated measures, especially for long-term studies of natural populations, and the R implementation is expected to facilitate modelling of longitudinal data for studies of both animal and human populations.

Dynamic fator Models for bivariate Count Data: an application to fire activity

The study of forest re activity, in its several aspects, is essencial to understand the phenomenon and to prevent environmental public catastrophes. In this context the analysis of monthly number of res along several years is one aspect to have into account in order to better comprehend this tematic. The goal of this work is to analyze the monthly number of forest res in the neighboring districts of Aveiro and Coimbra, Portugal, through dynamic factor models for bivariate count series. We use a bayesian approach, through MCMC methods, to estimate the model parameters as well as to estimate the common latent factor to both series.

Wind Farm Facilities in Germany Kill Noctule Bats from Near and Far

Over recent years, it became widely accepted that alternative, renewable energy may come at some risk for wildlife, for example, when wind turbines cause large numbers of bat fatalities. To better assess likely populations effects of wind turbine related wildlife fatalities, we studied the geographical origin of the most common bat species found dead below German wind turbines, the noctule bat (Nyctalus noctula). We measured stable isotope ratios of non-exchangeable hydrogen in fur keratin to separate migrants from local individuals, used a linear mixed-effects model to identify temporal, spatial and biological factors explaining the variance in measured stable isotope ratios and determined the geographical breeding provenance of killed migrants using isoscape origin models. We found that 72% of noctule bat casualties (n = 136) were of local origin, while 28% were long-distance migrants. These findings highlight that bat fatalities at German wind turbines may affect both local and distant populations. Our results indicated a sex and age-specific vulnerability of bats towards lethal accidents at turbines, i.e. a relatively high proportion of killed females were recorded among migratory individuals, whereas more juveniles than adults were recorded among killed bats of local origin. Migratory noctule bats were found to originate from distant populations in the Northeastern parts of Europe. The large catchment areas of German wind turbines and high vulnerability of female and juvenile noctule bats call for immediate action to reduce the negative cross-boundary effects of bat fatalities at wind turbines on local and distant populations. Further, our study highlights the importance of implementing effective mitigation measures and developing species and scale-specific conservation approaches on both national and international levels to protect source populations of bats. The efficacy of local compensatory measures appears doubtful, at least for migrant noctule bats, considering the large geographical catchment areas of German wind turbines for this species.

Influence of environmental variability and age on the body condition of small pelagic fish in the Gulf of Lions

Endogenous and environmental variables are fundamental in explaining variations in fish condition. Based on more than 20 yr of fish weight and length data, relative condition indices were computed for anchovy and sardine caught in the Gulf of Lions. Classification and regression trees (CART) were used to identify endogenous factors affecting fish condition, and to group years of similar condition. Both species showed a similar annual cycle with condition being minimal in February and maximal in July. CART identified 3 groups of years where the fish populations generally showed poor, average and good condition and within which condition differed between age classes but not according to sex. In particular, during the period of poor condition (mostly recent years), sardines older than 1 yr appeared to be more strongly affected than younger individuals. Time-series were analyzed using generalized linear models (GLMs) to examine the effects of oceanographic abiotic (temperature, Western Mediterranean Oscillation [WeMO] and Rhone outflow) and biotic (chlorophyll a and 6 plankton classes) factors on fish condition. The selected models explained 48 and 35% of the variance of anchovy and sardine condition, respectively. Sardine condition was negatively related to temperature but positively related to the WeMO and mesozooplankton and diatom concentrations. A positive effect of mesozooplankton and Rhone runoff on anchovy condition was detected. The importance of increasing temperatures and reduced water mixing in the NW Mediterranean Sea, affecting planktonic productivity and thus fish condition by bottom-up control processes, was highlighted by these results. Changes in plankton quality, quantity and phenology could lead to insufficient or inadequate food supply for both species.

Flexible regression models for functional neuroimaging

Current practice for analysing functional neuroimaging data is to average the brain signals recorded at multiple sensors or channels on the scalp over time across hundreds of trials or replicates to eliminate noise and enhance the underlying signal of interest. These studies recording brain signals non-invasively using functional neuroimaging techniques such as electroencephalography (EEG) and magnetoencephalography (MEG) generate complex, high dimensional and noisy data for many subjects at a number of replicates. Single replicate (or single trial) analysis of neuroimaging data have gained focus as they are advantageous to study the features of the signals at each replicate without averaging out important features in the data that the current methods employ. The research here is conducted to systematically develop flexible regression mixed models for single trial analysis of specific brain activities using examples from EEG and MEG to illustrate the models. This thesis follows three specific themes: i) artefact correction to estimate the `brain' signal which is of interest, ii) characterisation of the signals to reduce their dimensions, and iii) model fitting for single trials after accounting for variations between subjects and within subjects (between replicates). The models are developed to establish evidence of two specific neurological phenomena - entrainment of brain signals to an $\alpha$ band of frequencies (8-12Hz) and dipolar brain activation in the same $\alpha$ frequency band in an EEG experiment and a MEG study, respectively.

Interaction between mercury (Hg), arsenic (As) and selenium (Se) affects the activity of glutathione S-transferase in breast milk: possible relationship with fish and shellfish intake

Breast milk is regarded as an ideal source of nutrients for the growth and development of neonates, but it can also be a potential source of pollutants. Mothers can be exposed to different contaminants as a result of their lifestyle and environmental pollution. Mercury (Hg) and arsenic (As) could adversely affect the development of fetal and neonatal nervous system. Some fish and shellfish are rich in selenium (Se), an essential trace element that forms part of several enzymes related to the detoxification process, including glutathione S-transferase (GST). The goal of this study was to determine the interaction between Hg, As and Se and analyze its effect on the activity of GST in breast milk. Milk samples were collected from women between day 7 and 10 postpartum. The GST activity was determined spectrophotometrically; total Hg, As and Se concentrations were measured by atomic absorption spectrometry. To explain the possible association of Hg, As and Se concentrations with GST activity in breast milk, generalized linear models were constructed. The model explained 44% of the GST activity measured in breast milk. The GLM suggests that GST activity was positively correlated with Hg, As and Se concentrations. The activity of the enzyme was also explained by the frequency of consumption of marine fish and shellfish in the diet of the breastfeeding women.

Investigating bias in recreational fishing surveys: Fishers listed in public telephone directories fish similarly to their unlisted counterparts

Several recent offsite recreational fishing surveys have used public landline telephone directories as a sampling frame. Sampling biases inherent in this method are recognised, but are assumed to be corrected through demographic data expansion. However, the rising prevalence of mobile-only households has potentially increased these biases by skewing raw samples towards households that maintain relatively high levels of coverage in telephone directories. For biases to be corrected through demographic expansion, both the fishing participation rate and fishing activity must be similar among listed and unlisted fishers within each demographic group. In this study, we tested for a difference in the fishing activity of listed and unlisted fishers within demographic groups by comparing their avidity (number of fishing trips per year), as well as the platform used (boat or shore) and species targeted on their most recent fishing trip. 3062 recreational fishers were interviewed at 34 tackle stores across 12 residential regions of Queensland, Australia. For each fisher, data collected included their fishing avidity, the platform used and species targeted on their most recent trip, their gender, age, residential region, and whether their household had a listed telephone number. Although the most avid fishers were younger and less likely to have a listed phone number, cumulative link models revealed that avidity was not affected by an interaction of phone listing status, age group and residential region (p > 0.05). Likewise, binomial generalized linear models revealed that there was no interaction between phone listing, age group and avidity acting on platform (p > 0.05), and platform was not affected by an interaction of phone listing status, age group, and residential region (p > 0.05). Ordination of target species using Bray-Curtis dissimilarity indices found a significant but irrelevant difference (i.e. small effect size) between listed and unlisted fishers (ANOSIM R < 0.05, p < 0.05). These results suggest that, at this time, the fishing activity of listed and unlisted fishers in Queensland is similar within demographic groups. Future research seeking to validate the assumptions of recreational fishing telephone surveys should investigate fishing participation rates of listed and unlisted fishers within demographic groups.

Epidemiology and genetic architecture of blood pressure: a family based study of Generation Scotland

Hypertension is a major risk factor for cardiovascular disease and mortality, and a growing global public health concern, with up to one-third of the world’s population affected. Despite the vast amount of evidence for the benefits of blood pressure (BP) lowering accumulated to date, elevated BP is still the leading risk factor for disease and disability worldwide. It is well established that hypertension and BP are common complex traits, where multiple genetic and environmental factors contribute to BP variation. Furthermore, family and twin studies confirmed the genetic component of BP, with a heritability estimate in the range of 30-50%. Contemporary genomic tools enabling the genotyping of millions of genetic variants across the human genome in an efficient, reliable, and cost-effective manner, has transformed hypertension genetics research. This is accompanied by the presence of international consortia that have offered unprecedentedly large sample sizes for genome-wide association studies (GWASs). While GWAS for hypertension and BP have identified more than 60 loci, variants in these loci are associated with modest effects on BP and in aggregate can explain less than 3% of the variance in BP. The aims of this thesis are to study the genetic and environmental factors that influence BP and hypertension traits in the Scottish population, by performing several genetic epidemiological analyses. In the first part of this thesis, it aims to study the burden of hypertension in the Scottish population, along with assessing the familial aggregation and heritialbity of BP and hypertension traits. In the second part, it aims to validate the association of common SNPs reported in the large GWAS and to estimate the variance explained by these variants. In this thesis, comprehensive genetic epidemiology analyses were performed on Generation Scotland: Scottish Family Health Study (GS:SFHS), one of the largest population-based family design studies. The availability of clinical, biological samples, self-reported information, and medical records for study participants has allowed several assessments to be performed to evaluate factors that influence BP variation in the Scottish population. Of the 20,753 subjects genotyped in the study, a total of 18,470 individuals (grouped into 7,025 extended families) passed the stringent quality control (QC) criteria and were available for all subsequent analysis. Based on the BP-lowering treatment exposure sources, subjects were further classified into two groups. First, subjects with both a self-reported medications (SRMs) history and electronic-prescription records (EPRs; n =12,347); second, all the subjects with at least one medication history source (n =18,470). In the first group, the analysis showed a good concordance between SRMs and EPRs (kappa =71%), indicating that SRMs can be used as a surrogate to assess the exposure to BP-lowering medication in GS:SFHS participants. Although both sources suffer from some limitations, SRMs can be considered the best available source to estimate the drug exposure history in those without EPRs. The prevalence of hypertension was 40.8% with higher prevalence in men (46.3%) compared to women (35.8%). The prevalence of awareness, treatment and controlled hypertension as defined by the study definition were 25.3%, 31.2%, and 54.3%, respectively. These findings are lower than similar reported studies in other populations, with the exception of controlled hypertension prevalence, which can be considered better than other populations. Odds of hypertension were higher in men, obese or overweight individuals, people with a parental history of hypertension, and those living in the most deprived area of Scotland. On the other hand, deprivation was associated with higher odds of treatment, awareness and controlled hypertension, suggesting that people living in the most deprived area may have been receiving better quality of care, or have higher comorbidity levels requiring greater engagement with doctors. These findings highlight the need for further work to improve hypertension management in Scotland. The family design of GS:SFHS has allowed family-based analysis to be performed to assess the familial aggregation and heritability of BP and hypertension traits. The familial correlation of BP traits ranged from 0.07 to 0.20, and from 0.18 to 0.34 for parent-offspring pairs and sibling pairs, respectively. A higher correlation of BP traits was observed among first-degree relatives than other types of relative pairs. A variance-component model that was adjusted for sex, body mass index (BMI), age, and age-squared was used to estimate heritability of BP traits, which ranged from 24% to 32% with pulse pressure (PP) having the lowest estimates. The genetic correlation between BP traits showed a high correlation between systolic (SBP), diastolic (DBP) and mean arterial pressure (MAP) (G: 81% to 94%), but lower correlations with PP (G: 22% to 78%). The sibling recurrence risk ratio (λS) for hypertension and treatment were calculated as 1.60 and 2.04 respectively. These findings confirm the genetic components of BP traits in GS:SFHS, and justify further work to investigate genetic determinants of BP. Genetic variants reported in the recent large GWAS of BP traits were selected for genotyping in GS:SFHS using a custom designed TaqMan® OpenArray®. The genotyping plate included 44 single nucleotide polymorphisms (SNPs) that have been previously reported to be associated with BP or hypertension at genome-wide significance level. A linear mixed model that is adjusted for age, age-squared, sex, and BMI was used to test for the association between the genetic variants and BP traits. Of the 43 variants that passed the QC, 11 variants showed statistically significant association with at least one BP trait. The phenotypic variance explained by these variant for the four BP traits were 1.4%, 1.5%, 1.6%, and 0.8% for SBP, DBP, MAP, and PP, respectively. The association of genetic risk score (GRS) that were constructed from selected variants has showed a positive association with BP level and hypertension prevalence, with an average effect of one mmHg increase with each 0.80 unit increases in the GRS across the different BP traits. The impact of BP-lowering medication on the genetic association study for BP traits has been established, with typical practice of adding a fixed value (i.e. 15/10 mmHg) to the measured BP values to adjust for BP treatment. Using the subset of participants with the two treatment exposure sources (i.e. SRMs and EPRs), the influence of using either source to justify the addition of fixed values in SNP association signal was analysed. BP phenotypes derived from EPRs were considered the true phenotypes, and those derived from SRMs were considered less accurate, with some phenotypic noise. Comparing SNPs association signals between the four BP traits in the two model derived from the different adjustments showed that MAP was the least impacted by the phenotypic noise. This was suggested by identifying the same overlapped significant SNPs for the two models in the case of MAP, while other BP traits had some discrepancy between the two sources

GGE biplot analysis of multi-environment yield trials of rice produced in a temperate climate

Low temperature is one of the main environmental constraints for rice ( Oryza sativa L.) grain production yield. It is known that multi-environment studies play a critical role in the sustainability of rice production across diverse environments. However, there are few studies based on multi-environment studies of rice in temperate climates. The aim was to study the performance of rice plants in cold environments. Four experimental lines and six cultivars were evaluated at three locations during three seasons. The grain yield data were analyzed with ANOVA, mixed models based on the best linear unbiased predictors (BLUPs), and genotype plus Genotype × Environment interaction (GGE) biplot. High genotype contribution (> 25%) was observed in grain yield and the interaction between genotype and locations was not very important. Results also showed that ‘Quila 241319’ was the best experimental line with the highest grain yield (11.3 t ha-1) and grain yield stability across the environments; commercial cultivars were classified as medium grain yield genotypes.

Quantification of air quality in space and time and its effects on health

The long-term adverse effects on health associated with air pollution exposure can be estimated using either cohort or spatio-temporal ecological designs. In a cohort study, the health status of a cohort of people are assessed periodically over a number of years, and then related to estimated ambient pollution concentrations in the cities in which they live. However, such cohort studies are expensive and time consuming to implement, due to the long-term follow up required for the cohort. Therefore, spatio-temporal ecological studies are also being used to estimate the long-term health effects of air pollution as they are easy to implement due to the routine availability of the required data. Spatio-temporal ecological studies estimate the health impact of air pollution by utilising geographical and temporal contrasts in air pollution and disease risk across $n$ contiguous small-areas, such as census tracts or electoral wards, for multiple time periods. The disease data are counts of the numbers of disease cases occurring in each areal unit and time period, and thus Poisson log-linear models are typically used for the analysis. The linear predictor includes pollutant concentrations and known confounders such as socio-economic deprivation. However, as the disease data typically contain residual spatial or spatio-temporal autocorrelation after the covariate effects have been accounted for, these known covariates are augmented by a set of random effects. One key problem in these studies is estimating spatially representative pollution concentrations in each areal which are typically estimated by applying Kriging to data from a sparse monitoring network, or by computing averages over modelled concentrations (grid level) from an atmospheric dispersion model. The aim of this thesis is to investigate the health effects of long-term exposure to Nitrogen Dioxide (NO2) and Particular matter (PM10) in mainland Scotland, UK. In order to have an initial impression about the air pollution health effects in mainland Scotland, chapter 3 presents a standard epidemiological study using a benchmark method. The remaining main chapters (4, 5, 6) cover the main methodological focus in this thesis which has been threefold: (i) how to better estimate pollution by developing a multivariate spatio-temporal fusion model that relates monitored and modelled pollution data over space, time and pollutant; (ii) how to simultaneously estimate the joint effects of multiple pollutants; and (iii) how to allow for the uncertainty in the estimated pollution concentrations when estimating their health effects. Specifically, chapters 4 and 5 are developed to achieve (i), while chapter 6 focuses on (ii) and (iii). In chapter 4, I propose an integrated model for estimating the long-term health effects of NO2, that fuses modelled and measured pollution data to provide improved predictions of areal level pollution concentrations and hence health effects. The air pollution fusion model proposed is a Bayesian space-time linear regression model for relating the measured concentrations to the modelled concentrations for a single pollutant, whilst allowing for additional covariate information such as site type (e.g. roadside, rural, etc) and temperature. However, it is known that some pollutants might be correlated because they may be generated by common processes or be driven by similar factors such as meteorology. The correlation between pollutants can help to predict one pollutant by borrowing strength from the others. Therefore, in chapter 5, I propose a multi-pollutant model which is a multivariate spatio-temporal fusion model that extends the single pollutant model in chapter 4, which relates monitored and modelled pollution data over space, time and pollutant to predict pollution across mainland Scotland. Considering that we are exposed to multiple pollutants simultaneously because the air we breathe contains a complex mixture of particle and gas phase pollutants, the health effects of exposure to multiple pollutants have been investigated in chapter 6. Therefore, this is a natural extension to the single pollutant health effects in chapter 4. Given NO2 and PM10 are highly correlated (multicollinearity issue) in my data, I first propose a temporally-varying linear model to regress one pollutant (e.g. NO2) against another (e.g. PM10) and then use the residuals in the disease model as well as PM10, thus investigating the health effects of exposure to both pollutants simultaneously. Another issue considered in chapter 6 is to allow for the uncertainty in the estimated pollution concentrations when estimating their health effects. There are in total four approaches being developed to adjust the exposure uncertainty. Finally, chapter 7 summarises the work contained within this thesis and discusses the implications for future research.

Modelação do risco de perda dentária por doença periodontal

O prognóstico da perda dentária é um dos principais problemas na prática clínica de medicina dentária. Um dos principais fatores prognósticos é a quantidade de suporte ósseo do dente, definido pela área da superfície radicular dentária intraóssea. A estimação desta grandeza tem sido realizada por diferentes metodologias de investigação com resultados heterogéneos. Neste trabalho utilizamos o método da planimetria com microtomografia para calcular a área da superfície radicular (ASR) de uma amostra de cinco dentes segundos pré-molares inferiores obtida da população portuguesa, com o objetivo final de criar um modelo estatístico para estimar a área de superfície radicular intraóssea a partir de indicadores clínicos da perda óssea. Por fim propomos um método para aplicar os resultados na prática. Os dados referentes à área da superfície radicular, comprimento total do dente (CT) e dimensão mésio-distal máxima da coroa (MDeq) serviram para estabelecer as relações estatísticas entre variáveis e definir uma distribuição normal multivariada. Por fim foi criada uma amostra de 37 observações simuladas a partir da distribuição normal multivariada definida e estatisticamente idênticas aos dados da amostra de cinco dentes. Foram ajustados cinco modelos lineares generalizados aos dados simulados. O modelo estatístico foi selecionado segundo os critérios de ajustamento, preditibilidade, potência estatística, acurácia dos parâmetros e da perda de informação, e validado pela análise gráfica de resíduos. Apoiados nos resultados propomos um método em três fases para estimação área de superfície radicular perdida/remanescente. Na primeira fase usamos o modelo estatístico para estimar a área de superfície radicular, na segunda estimamos a proporção (decis) de raiz intraóssea usando uma régua de Schei adaptada e na terceira multiplicamos o valor obtido na primeira fase por um coeficiente que representa a proporção de raiz perdida (ASRp) ou da raiz remanescente (ASRr) para o decil estimado na segunda fase. O ponto forte deste estudo foi a aplicação de metodologia estatística validada para operacionalizar dados clínicos na estimação de suporte ósseo perdido. Como pontos fracos consideramos a aplicação destes resultados apenas aos segundos pré-molares mandibulares e a falta de validação clínica.

Procedimento para estudo da escolha modal em viagens realizadas por estudantes de instituições de ensino médio, mediado pela psicologia social

Tese (doutorado)—Universidade de Brasília, Faculdade de Tecnologia, Departamento de Engenharia Civil e Ambiental, 2016.