A bayesian model for mixed responses and skew laten variable to measure HRQoL in children

The aim of the thesi is to formulate a suitable Item Response Theory (IRT) based model to measure HRQoL (as latent variable) using a mixed responses questionnaire and relaxing the hypothesis of normal distributed latent variable. The new model is a combination of two models already presented in literature, that is, a latent trait model for mixed responses and an IRT model for Skew Normal latent variable. It is developed in a Bayesian framework, a Markov chain Monte Carlo procedure is used to generate samples of the posterior distribution of the parameters of interest. The proposed model is test on a questionnaire composed by 5 discrete items and one continuous to measure HRQoL in children, the EQ-5D-Y questionnaire. A large sample of children collected in the schools was used. In comparison with a model for only discrete responses and a model for mixed responses and normal latent variable, the new model has better performances, in term of deviance information criterion (DIC), chain convergences times and precision of the estimates.

Comparison of phenotypes of childhood wheeze and cough in 2 independent cohorts

BACKGROUND Among children with wheeze and recurrent cough there is great variation in clinical presentation and time course of the disease. We previously distinguished 5 phenotypes of wheeze and cough in early childhood by applying latent class analysis to longitudinal data from a population-based cohort (original cohort). OBJECTIVE To validate previously identified phenotypes of childhood cough and wheeze in an independent cohort. METHODS We included 903 children reporting wheeze or recurrent cough from an independent population-based cohort (validation cohort). As in the original cohort, we used latent class analysis to identify phenotypes on the basis of symptoms of wheeze and cough at 2 time points (preschool and school age) and objective measurements of atopy, lung function, and airway responsiveness (school age). Prognostic outcomes (wheeze, bronchodilator use, cough apart from colds) 5 years later were compared across phenotypes. RESULTS When using a 5-phenotype model, the analysis distinguished 3 phenotypes of wheeze and 2 of cough as in the original cohort. Two phenotypes were closely similar in both cohorts: Atopic persistent wheeze (persistent multiple trigger wheeze and chronic cough, atopy and reduced lung function, poor prognosis) and transient viral wheeze (early-onset transient wheeze with viral triggers, favorable prognosis). The other phenotypes differed more between cohorts. These differences might be explained by differences in age at measurements. CONCLUSIONS Applying the same method to 2 different cohorts, we consistently identified 2 phenotypes of wheeze (atopic persistent wheeze, transient viral wheeze), suggesting that these represent distinct disease processes. Differences found in other phenotypes suggest that the age when features are assessed is critical and should be considered carefully when defining phenotypes.

Identifying Distinct Quitting Trajectories after an Unassisted Smoking Cessation Attempt: An Ecological Momentary Assessment Study

Objectives: This study aimed at identifying distinct quitting trajectories over 29 days after an unassisted smoking ces- sation attempt by ecological momentary assessment (EMA). In order to validate these trajectories we tested if they predict smoking frequency up to six months later. Methods: EMA via mobile phones was used to collect real time data on smoking (yes/no) after an unassisted quit attempt over 29 days. Smoking frequency one, three and six months after the quit attempt was assessed with online questionnaires. Latent class growth modeling was used to analyze the data of 230 self-quitters. Results: Four different quitting trajectories emerged: quitter (43.9%), late quitter (11.3%), returner (17%) and persistent smoker (27.8%). The quitting trajectories predicted smoking frequency one, three and six months after the quit attempt (all p < 0.001). Conclusions: Outcome after a smoking cessation attempt is better described by four distinct trajectories instead of a binary variable for abstinence or relapse. In line with the relapse model by Marlatt and Gordon, late quitter may have learned how to cope with lapses during one month after the quitting attempt. This group would have been allocated to the relapse group in traditional outcome studies.

HIV prevalence and contextual risk-factors among injection drug users in Harris County, Texas

Injection drug use is the third most frequent risk factor for new HIV infections in the United States. A dual mode of exposure: unsafe drug using practices and risky sexual behaviors underlies injection drug users' (IDUs) risk for HIV infection. This research study aims to characterize patterns of drug use and sexual behaviors and to examine the social contexts associated with risk behaviors among a sample of injection drug users. ^ This cross-sectional study includes 523 eligible injection drug users from Houston, Texas, recruited into the 2009 National HIV Behavioral Surveillance project. Three separate set of analyses were carried out. First, using latent class analysis (LCA) and maximum likelihood we identified classes of behavior describing levels of HIV risk, from nine drug and sexual behaviors. Second, eight separate multivariable regression models were built to examine the odds of reporting a given risk behavior. We constructed the most parsimonious multivariable model using a manual backward stepwise process. Third, we examined whether HIV serostatus knowledge (self-reported positive, negative, or unknown serostatus) is associated with drug use and sexual HIV risk behaviors. ^ Participants were mostly male, older, and non-Hispanic Black. Forty-two percent of our sample had behaviors putting them at high risk, 25% at moderate risk, and 33% at low risk for HIV infection. Individuals in the High-risk group had the highest probability of risky behaviors, categorized as almost always sharing needles (0.93), seldom using condoms (0.10), reporting recent exchange sex partners (0.90), and practicing anal sex (0.34). We observed that unsafe injecting practices were associated with high risk sexual behaviors. IDUs who shared needles had higher odds of having anal sex (OR=2.89, 95%CI: 1.69-4.92) and unprotected sex (OR=2.66, 95%CI: 1.38-5.10) at last sex. Additionally, homelessness was associated with needle sharing (OR=2.24, 95% CI: 1.34-3.76) and cocaine use was associated with multiple sex partners (OR=1.82, 95% CI: 1.07-3.11). Furthermore, twenty-one percent of the sample was unaware of their HIV serostatus. The three groups were not different from each other in terms of drug-use behaviors: always using a new sterile needle, or in sharing needles or drug preparation equipment. However, IDUs unaware of their HIV serostatus were 33% more likely to report having more than three sexual partners in the past 12 months; 45% more likely to report to have unprotected sex and 85% more likely to use drug and or alcohol during or before at last sex compared to HIV-positive IDUs. ^ This analysis underscores the merit of LCA approach to empirically categorize injection drug users into distinct classes and identify their risk pattern using multiple indicators and our results show considerable overlap of high risk sexual and drug use behaviors among the high-risk class members. The observed clustering pattern of drug and sexual risk behavior among this population confirms that injection drug users do not represent a homogeneous population in terms of HIV risk. These findings will help develop tailored prevention programs.^

Mixture modeling for joint analysis of survival, discrete, and continuous data

Mixture modeling is commonly used to model categorical latent variables that represent subpopulations in which population membership is unknown but can be inferred from the data. In relatively recent years, the potential of finite mixture models has been applied in time-to-event data. However, the commonly used survival mixture model assumes that the effects of the covariates involved in failure times differ across latent classes, but the covariate distribution is homogeneous. The aim of this dissertation is to develop a method to examine time-to-event data in the presence of unobserved heterogeneity under a framework of mixture modeling. A joint model is developed to incorporate the latent survival trajectory along with the observed information for the joint analysis of a time-to-event variable, its discrete and continuous covariates, and a latent class variable. It is assumed that the effects of covariates on survival times and the distribution of covariates vary across different latent classes. The unobservable survival trajectories are identified through estimating the probability that a subject belongs to a particular class based on observed information. We applied this method to a Hodgkin lymphoma study with long-term follow-up and observed four distinct latent classes in terms of long-term survival and distributions of prognostic factors. Our results from simulation studies and from the Hodgkin lymphoma study demonstrated the superiority of our joint model compared with the conventional survival model. This flexible inference method provides more accurate estimation and accommodates unobservable heterogeneity among individuals while taking involved interactions between covariates into consideration.^

Modelling the Impact of Built Environment, Geographical Scales and Latent Constructs On Individual Travel Behaviour

La relación entre la estructura urbana y la movilidad ha sido estudiada desde hace más de 70 años. El entorno urbano incluye múltiples dimensiones como por ejemplo: la estructura urbana, los usos de suelo, la distribución de instalaciones diversas (comercios, escuelas y zonas de restauración, parking, etc.). Al realizar una revisión de la literatura existente en este contexto, se encuentran distintos análisis, metodologías, escalas geográficas y dimensiones, tanto de la movilidad como de la estructura urbana. En este sentido, se trata de una relación muy estudiada pero muy compleja, sobre la que no existe hasta el momento un consenso sobre qué dimensión del entorno urbano influye sobre qué dimensión de la movilidad, y cuál es la manera apropiada de representar esta relación. Con el propósito de contestar estas preguntas investigación, la presente tesis tiene los siguientes objetivos generales: (1) Contribuir al mejor entendimiento de la compleja relación estructura urbana y movilidad. y (2) Entender el rol de los atributos latentes en la relación entorno urbano y movilidad. El objetivo específico de la tesis es analizar la influencia del entorno urbano sobre dos dimensiones de la movilidad: número de viajes y tipo de tour. Vista la complejidad de la relación entorno urbano y movilidad, se pretende contribuir al mejor entendimiento de la relación a través de la utilización de 3 escalas geográficas de las variables y del análisis de la influencia de efectos inobservados en la movilidad. Para el análisis se utiliza una base de datos conformada por tres tipos de datos: (1) Una encuesta de movilidad realizada durante los años 2006 y 2007. Se obtuvo un total de 943 encuestas, en 3 barrios de Madrid: Chamberí, Pozuelo y Algete. (2) Información municipal del Instituto Nacional de Estadística: dicha información se encuentra enlazada con los orígenes y destinos de los viajes recogidos en la encuesta. Y (3) Información georeferenciada en Arc-GIS de los hogares participantes en la encuesta: la base de datos contiene información respecto a la estructura de las calles, localización de escuelas, parking, centros médicos y lugares de restauración. Se analizó la correlación entre e intra-grupos y se modelizaron 4 casos de atributos bajo la estructura ordinal logit. Posteriormente se evalúa la auto-selección a través de la estimación conjunta de las elecciones de tipo de barrio y número de viajes. La elección del tipo de barrio consta de 3 alternativas: CBD, Urban y Suburban, según la zona de residencia recogida en las encuestas. Mientras que la elección del número de viajes consta de 4 categorías ordinales: 0 viajes, 1-2 viajes, 3-4 viajes y 5 o más viajes. A partir de la mejor especificación del modelo ordinal logit. Se desarrolló un modelo joint mixed-ordinal conjunto. Los resultados indican que las variables exógenas requieren un análisis exhaustivo de correlaciones con el fin de evitar resultados sesgados. ha determinado que es importante medir los atributos del BE donde se realiza el viaje, pero también la información municipal es muy explicativa de la movilidad individual. Por tanto, la percepción de las zonas de destino a nivel municipal es considerada importante. En el contexto de la Auto-selección (self-selection) es importante modelizar conjuntamente las decisiones. La Auto-selección existe, puesto que los parámetros estimados conjuntamente son significativos. Sin embargo, sólo ciertos atributos del entorno urbano son igualmente importantes sobre la elección de la zona de residencia y frecuencia de viajes. Para analizar la Propensión al Viaje, se desarrolló un modelo híbrido, formado por: una variable latente, un indicador y un modelo de elección discreta. La variable latente se denomina “Propensión al Viaje”, cuyo indicador en ecuación de medida es el número de viajes; la elección discreta es el tipo de tour. El modelo de elección consiste en 5 alternativas, según la jerarquía de actividades establecida en la tesis: HOME, no realiza viajes durante el día de estudio, HWH tour cuya actividad principal es el trabajo o estudios, y no se realizan paradas intermedias; HWHs tour si el individuo reaiza paradas intermedias; HOH tour cuya actividad principal es distinta a trabajo y estudios, y no se realizan paradas intermedias; HOHs donde se realizan paradas intermedias. Para llegar a la mejor especificación del modelo, se realizó un trabajo importante considerando diferentes estructuras de modelos y tres tipos de estimaciones. De tal manera, se obtuvieron parámetros consistentes y eficientes. Los resultados muestran que la modelización de los tours, representa una ventaja sobre la modelización de los viajes, puesto que supera las limitaciones de espacio y tiempo, enlazando los viajes realizados por la misma persona en el día de estudio. La propensión al viaje (PT) existe y es específica para cada tipo de tour. Los parámetros estimados en el modelo híbrido resultaron significativos y distintos para cada alternativa de tipo de tour. Por último, en la tesis se verifica que los modelos híbridos representan una mejora sobre los modelos tradicionales de elección discreta, dando como resultado parámetros consistentes y más robustos. En cuanto a políticas de transporte, se ha demostrado que los atributos del entorno urbano son más importantes que los LOS (Level of Service) en la generación de tours multi-etapas. la presente tesis representa el primer análisis empírico de la relación entre los tipos de tours y la propensión al viaje. El concepto Propensity to Travel ha sido desarrollado exclusivamente para la tesis. Igualmente, el desarrollo de un modelo conjunto RC-Number of trips basado en tres escalas de medida representa innovación en cuanto a la comparación de las escalas geográficas, que no había sido hecha en la modelización de la self-selection. The relationship between built environment (BE) and travel behaviour (TB) has been studied in a number of cases, using several methods - aggregate and disaggregate approaches - and different focuses – trip frequency, automobile use, and vehicle miles travelled and so on. Definitely, travel is generated by the need to undertake activities and obtain services, and there is a general consensus that urban components affect TB. However researches are still needed to better understand which components of the travel behaviour are affected most and by which of the urban components. In order to fill the gap in the research, the present dissertation faced two main objectives: (1) To contribute to the better understanding of the relationship between travel demand and urban environment. And (2) To develop an econometric model for estimating travel demand with urban environment attributes. With this purpose, the present thesis faced an exhaustive research and computation of land-use variables in order to find the best representation of BE for modelling trip frequency. In particular two empirical analyses are carried out: 1. Estimation of three dimensions of travel demand using dimensions of urban environment. We compare different travel dimensions and geographical scales, and we measure self-selection contribution following the joint models. 2. Develop a hybrid model, integrated latent variable and discrete choice model. The implementation of hybrid models is new in the analysis of land-use and travel behaviour. BE and TB explicitly interact and allow richness information about a specific individual decision process For all empirical analysis is used a data-base from a survey conducted in 2006 and 2007 in Madrid. Spatial attributes describing neighbourhood environment are derived from different data sources: National Institute of Statistics-INE (Administrative: municipality and district) and GIS (circular units). INE provides raw data for such spatial units as: municipality and district. The construction of census units is trivial as the census bureau provides tables that readily define districts and municipalities. The construction of circular units requires us to determine the radius and associate the spatial information to our households. The first empirical part analyzes trip frequency by applying an ordered logit model. In this part is studied the effect of socio-economic, transport and land use characteristics on two travel dimensions: trip frequency and type of tour. In particular the land use is defined in terms of type of neighbourhoods and types of dwellers. Three neighbourhood representations are explored, and described three for constructing neighbourhood attributes. In particular administrative units are examined to represent neighbourhood and circular – unit representation. Ordered logit models are applied, while ordinal logit models are well-known, an intensive work for constructing a spatial attributes was carried out. On the other hand, the second empirical analysis consists of the development of an innovative econometric model that considers a latent variable called “propensity to travel”, and choice model is the choice of type of tour. The first two specifications of ordinal models help to estimate this latent variable. The latent variable is unobserved but the manifestation is called “indicators”, then the probability of choosing an alternative of tour is conditional to the probability of latent variable and type of tour. Since latent variable is unknown we fit the integral over its distribution. Four “sets of best variables” are specified, following the specification obtained from the correlation analysis. The results evidence that the relative importance of SE variables versus BE variables depends on how BE variables are measured. We found that each of these three spatial scales has its intangible qualities and drawbacks. Spatial scales play an important role on predicting travel demand due to the variability in measures at trip origin/destinations within the same administrative unit (municipality, district and so on). Larger units will produce less variation in data; but it does not affect certain variables, such as public transport supply, that are more significant at municipality level. By contrast, land-use measures are more efficient at district level. Self-selection in this context, is weak. Thus, the influence of BE attributes is true. The results of the hybrid model show that unobserved factors affect the choice of tour complexity. The latent variable used in this model is propensity to travel that is explained by socioeconomic aspects and neighbourhood attributes. The results show that neighbourhood attributes have indeed a significant impact on the choice of the type of tours either directly and through the propensity to travel. The propensity to travel has a different impact depending on the structure of each tour and increases the probability of choosing more complex tours, such as tours with many intermediate stops. The integration of choice and latent variable model shows that omitting important perception and attitudes leads to inconsistent estimates. The results also indicate that goodness of fit improves by adding the latent variable in both sequential and simultaneous estimation. There are significant differences in the sensitivity to the latent variable across alternatives. In general, as expected, the hybrid models show a major improvement into the goodness of fit of the model, compared to a classical discrete choice model that does not incorporate latent effects. The integrated model leads to a more detailed analysis of the behavioural process. Summarizing, the effect that built environment characteristics on trip frequency studied is deeply analyzed. In particular we tried to better understand how land use characteristics can be defined and measured and which of these measures do have really an impact on trip frequency. We also tried to test the superiority of HCM on this field. We can concluded that HCM shows a major improvement into the goodness of fit of the model, compared to classical discrete choice model that does not incorporate latent effects. And consequently, the application of HCM shows the importance of LV on the decision of tour complexity. People are more elastic to built environment attributes than level of services. Thus, policy implications must take place to develop more mixed areas, work-places in combination with commercial retails.

Individual differences in predicting occupational success: The effect of population heterogeneity

Using a sample of 339 university graduates from the University of Alicante (Spain) three years after completion of their studies, we studied the relationships between general intelligence (GI), personality traits, emotional intelligence (EI), academic performance, and occupational attainment and compared the results of conventional regression analysis with the results obtained from applying regression mixture models. The results reveal the influence of unobserved population heterogeneity (latent class) on the relationship between predictors and criteria and the improvement in the prediction obtained from applying regression mixture models compared to applying a conventional regression model.

Migraine With Aura and Migraine Without Aura Are Not Distinct Entities: Further Evidence From a Large Dutch Population Study

It is often debated whether migraine with aura (MA) and migraine without aura (MO) are etiologically distinct disorders. A previous study using latent class analysis (LCA) in Australian twins showed no evidence for separate subtypes of MO and MA. The aim of the present study was to replicate these results in a population of Dutch twins and their parents, siblings and partners (N = 10,144). Latent class analysis of International Headache Society (IHS)-based migraine symptoms resulted in the identification of 4 classes: a class of unaffected subjects (class 0), a mild form of nonmigrainous headache (class 1), a moderately severe type of migraine (class 2), typically without neurological symptoms or aura (8% reporting aura symptoms), and a severe type of migraine (class 3), typically with neurological symptoms, and aura symptoms in approximately half of the cases. Given the overlap of neurological symptoms and nonmutual exclusivity of aura symptoms, these results do not support the MO and MA subtypes as being etiologically distinct. The heritability in female twins of migraine based on LCA classification was estimated at .50 (95% confidence intervals [0CI} .27 -.59), similar to IHS-based migraine diagnosis (h(2) = .49, 95% Cl .19-.57). However, using a dichotomous classification (affected-unaffected) decreased heritability for the IHS-based classification (h(2) = .33, 95% Cl .00-.60), but not the LCA-based classification (h(2) = .51, 95% Cl. 23-.61). Importantly, use of the LCA-based classification increased the number of subjects classified as affected. The heritability of the screening question was similar to more detailed LCA and IHS classifications, suggesting that the screening procedure is an important determining factor in genetic studies of migraine.

Semi-supervised learning of hierarchical latent trait models for data visualisation

An interactive hierarchical Generative Topographic Mapping (HGTM) ¸iteH_GTM has been developed to visualise complex data sets. In this paper, we build a more general visualisation system by extending the HGTM visualisation system in 3 directions: bf (1) We generalize HGTM to noise models from the exponential family of distributions. The basic building block is the Latent Trait Model (LTM) developed in ¸iteKaban_pami. bf (2) We give the user a choice of initializing the child plots of the current plot in either em interactive, or em automatic mode. In the interactive mode the user interactively selects ``regions of interest'' as in ¸iteH_GTM, whereas in the automatic mode an unsupervised minimum message length (MML)-driven construction of a mixture of LTMs is employed. bf (3) We derive general formulas for magnification factors in latent trait models. Magnification factors are a useful tool to improve our understanding of the visualisation plots, since they can highlight the boundaries between data clusters. The unsupervised construction is particularly useful when high-level plots are covered with dense clusters of highly overlapping data projections, making it difficult to use the interactive mode. Such a situation often arises when visualizing large data sets. We illustrate our approach on a toy example and apply our system to three more complex real data sets.

Novel visualization methods for protein data

Visualization of high-dimensional data has always been a challenging task. Here we discuss and propose variants of non-linear data projection methods (Generative Topographic Mapping (GTM) and GTM with simultaneous feature saliency (GTM-FS)) that are adapted to be effective on very high-dimensional data. The adaptations use log space values at certain steps of the Expectation Maximization (EM) algorithm and during the visualization process. We have tested the proposed algorithms by visualizing electrostatic potential data for Major Histocompatibility Complex (MHC) class-I proteins. The experiments show that the variation in the original version of GTM and GTM-FS worked successfully with data of more than 2000 dimensions and we compare the results with other linear/nonlinear projection methods: Principal Component Analysis (PCA), Neuroscale (NSC) and Gaussian Process Latent Variable Model (GPLVM).

Semisupervised learning of hierarchical latent trait models for data visualization

Recently, we have developed the hierarchical Generative Topographic Mapping (HGTM), an interactive method for visualization of large high-dimensional real-valued data sets. In this paper, we propose a more general visualization system by extending HGTM in three ways, which allows the user to visualize a wider range of data sets and better support the model development process. 1) We integrate HGTM with noise models from the exponential family of distributions. The basic building block is the Latent Trait Model (LTM). This enables us to visualize data of inherently discrete nature, e.g., collections of documents, in a hierarchical manner. 2) We give the user a choice of initializing the child plots of the current plot in either interactive, or automatic mode. In the interactive mode, the user selects "regions of interest," whereas in the automatic mode, an unsupervised minimum message length (MML)-inspired construction of a mixture of LTMs is employed. The unsupervised construction is particularly useful when high-level plots are covered with dense clusters of highly overlapping data projections, making it difficult to use the interactive mode. Such a situation often arises when visualizing large data sets. 3) We derive general formulas for magnification factors in latent trait models. Magnification factors are a useful tool to improve our understanding of the visualization plots, since they can highlight the boundaries between data clusters. We illustrate our approach on a toy example and evaluate it on three more complex real data sets. © 2005 IEEE.

Visualisation of bioinformatics datasets

Analysing the molecular polymorphism and interactions of DNA, RNA and proteins is of fundamental importance in biology. Predicting functions of polymorphic molecules is important in order to design more effective medicines. Analysing major histocompatibility complex (MHC) polymorphism is important for mate choice, epitope-based vaccine design and transplantation rejection etc. Most of the existing exploratory approaches cannot analyse these datasets because of the large number of molecules with a high number of descriptors per molecule. This thesis develops novel methods for data projection in order to explore high dimensional biological dataset by visualising them in a low-dimensional space. With increasing dimensionality, some existing data visualisation methods such as generative topographic mapping (GTM) become computationally intractable. We propose variants of these methods, where we use log-transformations at certain steps of expectation maximisation (EM) based parameter learning process, to make them tractable for high-dimensional datasets. We demonstrate these proposed variants both for synthetic and electrostatic potential dataset of MHC class-I. We also propose to extend a latent trait model (LTM), suitable for visualising high dimensional discrete data, to simultaneously estimate feature saliency as an integrated part of the parameter learning process of a visualisation model. This LTM variant not only gives better visualisation by modifying the project map based on feature relevance, but also helps users to assess the significance of each feature. Another problem which is not addressed much in the literature is the visualisation of mixed-type data. We propose to combine GTM and LTM in a principled way where appropriate noise models are used for each type of data in order to visualise mixed-type data in a single plot. We call this model a generalised GTM (GGTM). We also propose to extend GGTM model to estimate feature saliencies while training a visualisation model and this is called GGTM with feature saliency (GGTM-FS). We demonstrate effectiveness of these proposed models both for synthetic and real datasets. We evaluate visualisation quality using quality metrics such as distance distortion measure and rank based measures: trustworthiness, continuity, mean relative rank errors with respect to data space and latent space. In cases where the labels are known we also use quality metrics of KL divergence and nearest neighbour classifications error in order to determine the separation between classes. We demonstrate the efficacy of these proposed models both for synthetic and real biological datasets with a main focus on the MHC class-I dataset.

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.

Combining Information from Multiple Sources in Bayesian Modeling

Surveys can collect important data that inform policy decisions and drive social science research. Large government surveys collect information from the U.S. population on a wide range of topics, including demographics, education, employment, and lifestyle. Analysis of survey data presents unique challenges. In particular, one needs to account for missing data, for complex sampling designs, and for measurement error. Conceptually, a survey organization could spend lots of resources getting high-quality responses from a simple random sample, resulting in survey data that are easy to analyze. However, this scenario often is not realistic. To address these practical issues, survey organizations can leverage the information available from other sources of data. For example, in longitudinal studies that suffer from attrition, they can use the information from refreshment samples to correct for potential attrition bias. They can use information from known marginal distributions or survey design to improve inferences. They can use information from gold standard sources to correct for measurement error.

This thesis presents novel approaches to combining information from multiple sources that address the three problems described above.

The first method addresses nonignorable unit nonresponse and attrition in a panel survey with a refreshment sample. Panel surveys typically suffer from attrition, which can lead to biased inference when basing analysis only on cases that complete all waves of the panel. Unfortunately, the panel data alone cannot inform the extent of the bias due to attrition, so analysts must make strong and untestable assumptions about the missing data mechanism. Many panel studies also include refreshment samples, which are data collected from a random sample of new

individuals during some later wave of the panel. Refreshment samples offer information that can be utilized to correct for biases induced by nonignorable attrition while reducing reliance on strong assumptions about the attrition process. To date, these bias correction methods have not dealt with two key practical issues in panel studies: unit nonresponse in the initial wave of the panel and in the

refreshment sample itself. As we illustrate, nonignorable unit nonresponse

can significantly compromise the analyst's ability to use the refreshment samples for attrition bias correction. Thus, it is crucial for analysts to assess how sensitive their inferences---corrected for panel attrition---are to different assumptions about the nature of the unit nonresponse. We present an approach that facilitates such sensitivity analyses, both for suspected nonignorable unit nonresponse

in the initial wave and in the refreshment sample. We illustrate the approach using simulation studies and an analysis of data from the 2007-2008 Associated Press/Yahoo News election panel study.

The second method incorporates informative prior beliefs about

marginal probabilities into Bayesian latent class models for categorical data.

The basic idea is to append synthetic observations to the original data such that

(i) the empirical distributions of the desired margins match those of the prior beliefs, and (ii) the values of the remaining variables are left missing. The degree of prior uncertainty is controlled by the number of augmented records. Posterior inferences can be obtained via typical MCMC algorithms for latent class models, tailored to deal efficiently with the missing values in the concatenated data.

We illustrate the approach using a variety of simulations based on data from the American Community Survey, including an example of how augmented records can be used to fit latent class models to data from stratified samples.

The third method leverages the information from a gold standard survey to model reporting error. Survey data are subject to reporting error when respondents misunderstand the question or accidentally select the wrong response. Sometimes survey respondents knowingly select the wrong response, for example, by reporting a higher level of education than they actually have attained. We present an approach that allows an analyst to model reporting error by incorporating information from a gold standard survey. The analyst can specify various reporting error models and assess how sensitive their conclusions are to different assumptions about the reporting error process. We illustrate the approach using simulations based on data from the 1993 National Survey of College Graduates. We use the method to impute error-corrected educational attainments in the 2010 American Community Survey using the 2010 National Survey of College Graduates as the gold standard survey.

Oil Market Modelling: A Comparative Analysis of Fundamental and Latent Factor Approaches

We formally compare fundamental factor and latent factor approaches to oil price modelling. Fundamental modelling has a long history in seeking to understand oil price movements, while latent factor modelling has a more recent and limited history, but has gained popularity in other financial markets. The two approaches, though competing, have not formally been compared as to effectiveness. For a range of short- medium- and long-dated WTI oil futures we test a recently proposed five-factor fundamental model and a Principal Component Analysis latent factor model. Our findings demonstrate that there is no discernible difference between the two techniques in a dynamic setting. We conclude that this infers some advantages in adopting the latent factor approach due to the difficulty in determining a well specified fundamental model.