Longitudinal data are common in biological research. However, real data sets vary considerably in terms of their structure and complexity and present many challenges for statistical modelling. This thesis proposes a series of methods using random coefficients for modelling two broad types of longitudinal response: normally distributed measurements and binary recapture data. Biased inference can occur in linear mixed-effects modelling if subjects are drawn from a number of unknown sub-populations, or if the residual covariance is poorly specified. To address some of the shortcomings of previous approaches in terms of model selection and flexibility, this thesis presents methods for: (i) determining the presence of latent grouping structures using a two-step approach, involving regression splines for modelling functional random effects and mixture modelling of the fitted random effects; and (ii) flexible of modelling of the residual covariance matrix using regression splines to specify smooth and potentially non-monotonic variance and correlation functions. Spatially explicit capture-recapture methods for estimating the density of animal populations have shown a rapid increase in popularity over recent years. However, further refinements to existing theory and fitting software are required to apply these methods in many situations. This thesis presents: (i) an analysis of recapture data from an acoustic survey of gibbons using supplementary data in the form of estimated angles to detections, (ii) the development of a multi-occasion likelihood including a model for stochastic availability using a partially observed random effect (interpreted in terms of calling behaviour in the case of gibbons), and (iii) an analysis of recapture data from a population of radio-tagged skates using a conditional likelihood that allows the density of animal activity centres to be modelled as functions of time, space and animal-level covariates.

Quando um conjunto de dados possui uma estrutura hierárquica, uma possível abordagem são os modelos de regressão multiníveis, que se justifica pelo fato de haver uma porção significativa da variabilidade dos dados que pode ser explicada por níveis macro. Neste trabalho, desenvolvemos a seleção de modelos de regressão multinível aplicados a dados educacionais. Esta análise divide-se em duas partes: seleção de variáveis e seleção de modelos. Esta última subdivide-se em dois casos: modelagem clássica e modelagem bayesiana. Buscamos através de critérios como o Lasso, AIC, BIC, WAIC entre outros, encontrar quais são os fatores que influenciam no desempenho em matemática dos alunos do nono ano do ensino fundamental do estado de São Paulo. Também investigamos o funcionamento de cada um dos critérios de seleção de variáveis e de modelos. Foi possível concluir que, sob a abordagem frequentista, o critério de seleção de modelos BIC é o mais eficiente, já na abordagem bayesiana, o critérioWAIC apresentou melhores resultados. Utilizando o critério de seleção de variáveis Lasso para abordagem clássica, houve uma diminuição de 34% dos preditores do modelo. Por fim, identificamos que o desempenho em matemática dos estudantes do nono ano do ensino fundamental do estado de São Paulo é influenciado pelas seguintes covariáveis: grau de instrução da mãe, frequência de leitura de livros, tempo gasto com recreação em dia de aula, o fato de gostar de matemática, o desempenho em matemática global da escola, desempenho em língua portuguesa do aluno, dependência administrativa da escola, sexo, grau de instrução do pai, reprovações e distorção idade-série. / When a dataset contains a hierarchical data structure, a possible approach is the multilevel regression modelling, which is justified by the significative amout of the data variability that can be explained by macro level processes. In this work, a selection of multilevel regression models for educational data is developed. This analysis is divided into two parts: variable selection and model selection. The latter is subdivided into two categories: classical and Bayesian modeling. Traditional criteria for model selection such as Lasso, AIC, BIC, and WAIC, among others are used in this study as an attempt to identify the factors influencing ninth grade students performance in Mathematics of elementary education in the State of São Paulo. Likewise, an investigation was conducted to evaluate the performance of each variable selection criteria and model selection methods applied to fitted models that will be mentioned throughout this work. It was possible to conclude that, under the frequentist approach, BIC is the most efficient, whereas under the bayesian approach, WAIC presented better results. Using Lasso under the frequentist approach, a decrease of 34% on the number of predictors was observed. Finally, we identified that the performance in Mathematics of students in the ninth year of elementary school in the state of São Paulo is most influenced by the following covariates: mothers educational level, frequency of book reading, time spent with recreation in classroom, the fact of liking Math, school global performance in Mathematics, performance in Portuguese, school administrative dependence, gender, fathers educational degree, failures and age-grade distortion.

La Tunisie a connu une forte hausse des poursuites d’études dans l’enseignement supérieur ces 20 dernières années. Les réformes imposées par l'augmentation des effectifs étudiants, ont conduit à une augmentation du nombre des établissements de l’enseignement supérieur. Un des enjeux de la Tunisie et de plusieurs pays africains est l’amélioration de l’efficacité du système éducatif afin de favoriser l’insertion professionnelle. Cette thèse se propose de s’interroger sur cette efficacité à partir d’une approche économique des rendements de l’éducation sur le marché du travail. Le premier chapitre propose une analyse des rendements privés de l’éducation dans le contexte des pays africains. Il souligne la forte hétérogénéité de ces rendements entre les pays, notamment lorsque l’on prend en compte le taux d’emploi. Le deuxième chapitre se centre sur l’effet établissement sur le rendement de l’enseignement supérieur tunisien. Les résultats suggèrent l’existence d’un effet de l’établissement sur le salaire des diplômés. La sélectivité des établissements et la qualification des enseignements semblent affecter la rémunération des diplômés. Le troisième chapitre se focalise sur l’évaluation d’une réforme des curricula de l’enseignement supérieur fournissant une éducation à l’entrepreneuriat dans les universités tunisiennes. Cette recherche, qui repose sur l’affectation aléatoire pour mesurer son impact sur les résultats sur le marché du travail ainsi que sur les compétences techniques et les compétences non cognitives des étudiants, souligne l’intérêt que peut avoir ce type de programme. / In Tunisia, enrollment rates in tertiary education had soared up over the past two decades. A significant increase of student annual flows imposed the implementation of reforms that led to an increase in the number of higher education institutions and universities. One of the challenges in Tunisia and many African countries is to improve the efficiency of education systems to promote employability and graduates employment. This thesis discuss the question of the efficiency of education from an economic approach based on returns to education in the labor market. The first chapter analyses private returns to education particularly higher education in African countries. Our findings highlight the large differences to RORE estimates across countries, particularly when it takes into account the employment rate. The second chapter focuses on the impact of university quality on labor market outcomes in Tunisia. The empirical results suggest that institutions selectivity and university professor’s qualifications seem to affect earnings of graduates. The third chapter focuses on the evaluation of reform of university curriculum providing entrepreneurship education to Tunisian university students. This research based on randomized assignment to the entrepreneurship track measure its impacts on labor market outcomes as well as on intermediary outcomes such as business skills and behavioral skills. This chapter underlines the role of entrepreneurship program.

Multilevel datasets are commonly used and increasingly popular in research in the organizational and other social sciences. These models are complex and have many elements beyond those found in more traditional linear models. However, research on how multilevel models perform is lacking. The current paper examined the impact of common factors (average cluster size, cluster size distribution, average number of clusters, strength of the intraclass correlation coefficient, and effect sizes of individual and cluster level variables, and their interaction) in multilevel datasets. Monte Carlo data simulation was used across 6,144 factor-combination conditions. The results of study factors on observed intraclass correlation coefficients, calculated design effect, and empirical design effect are discussed. The results of this study have implications for both researchers in both academic and applied fields. The scale of the simulation variables allow it to be germane to datasets from across the social sciences. However, the nature of data simulation and analysis is such that there are still many elements that can and should be accounted for in future research.

This study was a population-based retrospective cohort of all singleton births in British Columbia for the years 2001 to 2006. The purpose of this dissertation is to examine how social and physical environment factors influence the risk of adverse pregnancy outcomes and whether they interact with each other or with maternal characteristics to modify disease risk. The main environmental factors examined include ambient particulate air pollution (PM2.5), neighbourhood socioeconomic status (SES), neighbourhood immigrant density, neighbourhood level of post-secondary education level and the urban-rural context. Census dissemination areas (DAs) were used as the neighbourhood spatial unit. The data (N=242,472) was extracted from the BC Perinatal Data Registry (BCPDR) from Perinatal Services BC (PSBC). The main perinatal outcomes investigated include birth weight and indicators of fetal growth restriction such as small-for-gestational age (SGA), term low birth weight (tLBW), and intrauterine growth restriction (IUGR), preterm birth (PTB) and gestational age, gestational diabetes mellitus (GDM) and gestational hypertension (GH). Collectively, this dissertation contributes to the perinatal epidemiological literature linking particulate air pollution and neighbourhood SES context to adverse pregnancy outcomes. Assumptions about the linear effect of PM2.5 and smoking on birth weight are challenged showing that the effects are most pronounced between low and average exposures and that the magnitude of their effect is modified by neighbourhood and individual-level characteristics. These results suggest that focusing exclusively on individual risk factors may have limited success in improving outcomes without addressing the contextual influences at the neighbourhood-level. This dissertation therefore also contributes to the public health, sociological and community-urban development literature demonstrating that context and place matters. / Graduate / 0766 / 0573 / 0768 / anderse@uvic.ca

Para analisar dados binários oriundos de uma estrutura hierárquica com dois níveis (por exemplo, aluno e escola), uma alternativa bastante utilizada é a suposição da distribuição binomial para as unidades experimentais do primeiro nível (aluno) condicionalmente a um efeito aleatório proveniente de uma distribuição normal para as unidades do segundo nível (escola). Neste trabalho, propõe-se a adição de um efeito aleatório normal no primeiro nível de um modelo linear generalizado hierárquico binomial para contemplar uma possível variabilidade extra-binomial decorrente da dependência entre os ensaios de Bernoulli de um mesmo indivíduo. Obtém-se o processo de estimação por máxima verossimilhança para este modelo a partir da verossimilhança marginal dos dados, após uma dupla aplicação do método de quadratura de Gauss-Hermite adaptativa como aproximação para as integrais dos efeitos aleatórios. Realiza-se um estudo de simulação para contrastar propriedades inferenciais do modelo aspirante com o modelo linear generalizado binomial, um modelo de quase-verossimilhança e o tradicional modelo linear generalizado hierárquico em dois níveis. / A common alternative when analyzing binary data originated from a two-level hierarchical structure (for instance, student and school) is to assume a binomial distribution for the experimental units of the first level (student) conditionally to a normal random effect for the second level units (school). In this work, we propose the inclusion of a second normal random effect in the first level to contemplate a possible extra-binomial variability due to the dependence among the Bernoulli trials in the same individual. We obtain the maximum likelihood estimation process for this hierarchical model starting from the marginal likelihood of the data, after a double application of the adaptive Gauss-Hermite quadrature as an approximation of the integrals of the random effects. We conduct a simulation study to compare the inferential properties of the advocated model with the generalized linear (binomial) model, a quasi-likelihood model and the usual two-level hierarchical generalized linear model.

A partire dagli anni ’80, la diffusione dei metodi statistici, abbinata ai progressi nelle capacità computazionali dei personal computers, ha progressivamente facilitato i ricercatori nel testare empiricamente le proprie teorie. Gli economisti sono diventati in grado di eseguire milioni di regressioni prima di pranzo senza abbandonare le proprie scrivanie. Purtroppo, ciò ha portato ad un accumulo di evidenze spesso eterogenee, quando non contradditorie se non esplicitamente in conflitto. Per affrontare il problema, questa tesi fornirà una panoramica dei metodi meta-analitici disponibili in economia. Nella prima parte verranno introdotte le intuizioni alla base dei modelli gerarchici a fattori fissi e casuali capaci di risolvere le problematicità derivanti dalla presenza di osservazioni non indipendenti. Verrà inoltre affrontato il tema dell’errore sistematico di pubblicazione in presenza di elevata eterogeneità tra gli studi. La metodologia verrà successivamente applicata, nella seconda e terza parte, a due diverse aree della letteratura economica: l’impatto del rapporto banca-impresa sulle prestazioni aziendali e il dibattito sulla relazione fra democrazia e crescita. Mentre nel primo caso la correlazione negativa non è influenzata da fattori specifici ai singoli paesi, il contrario è vero per spiegare l’impatto (statisticamente non significativo) delle istituzioni democratiche sullo sviluppo economico. Quali siano questi fattori è però meno chiaro; gli studiosi non hanno ancora individuato le co-variate – o la corretta misurazione di esse – capaci di spiegare questa discussa relazione. / Starting in the late 1980s, improved computing performances and spread knowledge of statistical methods allowed researchers to put their theories to test. Formerly constrained economists became able [to] run millions of regressions before lunch without leaving their desks. Unfortunately, this led to an accumulation of often conflicting evidences. To address such issue, this thesis will provide an overview of the meta-analysis methods available in economics. The first paper will explain the intuitions behind fixed and random effects models in such a framework. It will then detail how multilevel modelling can help overcome hierarchical dependence issues. Finally, it will address the problem of publication bias in presence of high between-studies heterogeneity. Such methods will be then applied, in the second and third papers, to two different areas of the economics literature: the effect of relationship banking on firm performances and the democracy and growth conundrum. Results are far-reaching. While in the first case the documented negative relation is not driven by country-specific characteristics the opposite is true for the (statistically insignificant) impact of democratic institutions on economic growth. What these characteristics are is, however, less clear. Scholars have not yet found the covariates - or their suitable proxies - that matter to explain such much-debated relationship.

Para analisar dados binários oriundos de uma estrutura hierárquica com dois níveis (por exemplo, aluno e escola), uma alternativa bastante utilizada é a suposição da distribuição binomial para as unidades experimentais do primeiro nível (aluno) condicionalmente a um efeito aleatório proveniente de uma distribuição normal para as unidades do segundo nível (escola). Neste trabalho, propõe-se a adição de um efeito aleatório normal no primeiro nível de um modelo linear generalizado hierárquico binomial para contemplar uma possível variabilidade extra-binomial decorrente da dependência entre os ensaios de Bernoulli de um mesmo indivíduo. Obtém-se o processo de estimação por máxima verossimilhança para este modelo a partir da verossimilhança marginal dos dados, após uma dupla aplicação do método de quadratura de Gauss-Hermite adaptativa como aproximação para as integrais dos efeitos aleatórios. Realiza-se um estudo de simulação para contrastar propriedades inferenciais do modelo aspirante com o modelo linear generalizado binomial, um modelo de quase-verossimilhança e o tradicional modelo linear generalizado hierárquico em dois níveis. / A common alternative when analyzing binary data originated from a two-level hierarchical structure (for instance, student and school) is to assume a binomial distribution for the experimental units of the first level (student) conditionally to a normal random effect for the second level units (school). In this work, we propose the inclusion of a second normal random effect in the first level to contemplate a possible extra-binomial variability due to the dependence among the Bernoulli trials in the same individual. We obtain the maximum likelihood estimation process for this hierarchical model starting from the marginal likelihood of the data, after a double application of the adaptive Gauss-Hermite quadrature as an approximation of the integrals of the random effects. We conduct a simulation study to compare the inferential properties of the advocated model with the generalized linear (binomial) model, a quasi-likelihood model and the usual two-level hierarchical generalized linear model.

This study examined the growth patterns of action teams over time. Cognitive and non-cognitive (i.e., motivational) team composition variables were hypothesized to differentially predict initial levels of and changes over time in team performance. In order to test the hypotheses 78 two-person teams flew three equivalent missions on a low-fidelity computer-based Apache helicopter simulator. Random Coefficient Modeling analyses indicated that, as expected, team composition of general cognitive ability positively predicted initial team performance, whereas team composition of motivational traits did not. However, none of the team composition variables predicted team performance change. Implications, limitations and directions for future research are discussed.

At present there is a reliance on behavioral interventions that have been limited in their effectiveness to reduce the public health burden of chronic disease, partly because the effects of social context on the initiation and maintenance of health behaviors is not incorporated into public health policy and practice. Yet current research indicates that there are macro-level structural and contextual influences on population health that cannot be reduced to individual or compositional effects. This study investigated the associations between social structural factors, community social context, individual characteristics, and self-reported correlates of disease. Distal influences included social structural inequalities such as income inequality and absolute deprivation or poverty. Pertinent mechanisms through which these influences might have operated on disease included social contextual factors, such as social capital. Both political economy and the ecosocial perspective were selected to inform this study and to provide the theoretical framework from which hypotheses were derived.The design was a multilevel, retrospective, nonexperimental study using secondary data. The study linked three data sources (2001 Behavioral Risk Factor Surveillance System, Social Capital Community Benchmark Study, and U.S. Census) by Federal Information Processing Standards codes in order for individuals to be placed in their community or state contexts. Results provided mixed evidence of the direct role of structural and contextual inequalities on self-rated health. Any direct effects of social structural inequalities on the health outcomes disappeared once individual factors were included in the models. Findings demonstrated that one dimension of social capital, organizational activism, retained its significant direct effect on general health status, once individual characteristics were considered. Conclusions suggested indirect associations whereby the negative influence of social structural inequalities on health was mediated by the erosion of social trust, which in turn was associated with engaging in risk behavior, thus increasing the odds of reporting hypertension, fair/poor general health, and mental distress. Although results were inconsistent, this study contributed to advancing Healthy People 2010 goals of increasing quality of life and reducing health disparities by advancing understanding of the multilevel nature of perceived health and the chronic diseases they predict.