The purpose of the study was to examine multilevel regression models in the context of multilevel structural equation modeling (SEM) in terms of accuracy of parameter estimates, standard errors, and fit indices in normal and nonnormal data under various sample sizes and differing estimators (maximum likelihood, generalized least squares, and weighted least squares). The finding revealed that the regression coefficients were estimated with little to no bias among the study design conditions investigated. However, the number of clusters (group level) appeared to have the greatest impact on bias among the parameter estimate standard errors at both level-1 and level-2. In small sample sizes (i.e., 300 and 500) the standard errors were negatively biased. When the number of clusters was 30 and cluster size was held at 10, the level-1 standard errors were biased downward by approximately 20% for the maximum likelihood and generalized least squares estimators, while the weighted least squares estimator produced level-1 standard errors that were negatively biased by 25%. Regarding the level-2 standard errors, the level-2 standard errors were biased downward by approximately 24% in nonnormal data, especially when the correlation among variables was fixed at .5 and kurtosis was held constant at 7. In this same setting (30 clusters with cluster size fixed at 10), when kurtosis was fixed at 4 and the correlation among variables was held at .7, both the maximum likelihood and generalized least squares estimators resulted in standard errors that were biased downward by approximately 11%. Regarding fit statistics, negative bias was noted among each of the fit indices investigated when the number of clusters ranged from 30 to 50 and cluster size was fixed at 10. The least amount of bias was associated with the maximum likelihood estimator in each of the data normality conditions examined. As sample size increased, bias decreased to near zero when the sample size was equal to or greater than 1,500 with similar results reported across estimation methods. Recommendations for the substantive researcher are presented and areas of future research are presented.

O uso de modelos multiníveis é uma alternativa interessante para analisar dados que estão estruturados de forma hierárquica, pois permite a obtenção de diferentes estimativas de parâmetros relativos a grupos distintos e, ao mesmo tempo, leva em consideração a dependência entre as observações em um mesmo grupo. Neste trabalho, desenvolvemos e aplicamos modelos de regressão multiníveis simétricos, a fim de fornecer alternativas ao modelo usual, sob normalidade. Além disso, apresentamos uma breve análise de diagnóstico e estudo de simulação. Como motivação, consideramos dados educacionais, a fim de avaliar se o número de reprovações no histórico escolar do aluno e a infraestrutura da escola são variáveis relevantes que afetam o baixo desempenho dos alunos do ensino básico na disciplina de Matemática / The use of multilevel models is an interesting alternative to analyze data that is structured in a hierarchical manner, since it allows the obtention of different parameters estimates for distinct groups and, at the same time, it takes into account the dependence of observations in the same group. In this dissertation, we develop and apply symmetrical multilevel regression models, for the purpose of providing alternatives to the usual model, under normality. Furthermore we present a brief diagnostics analysis and a simulation study. As motivation, we consider educational data in order to assess whether the number of failures in school history of students and the school infrastructure are important variables that affect the low performance of elementary school students in Mathematics

Housing deprivation is an important dimension of poverty. It is thus a key challenge of policy makers to secure decent housing. The purpose of this paper is to analyze the link between housing tenure types and housing deprivation in 24 European countries. Empirical analyses are based on EU-SILC 2007, enabling comparisons of deprivation across a large set of countries. A multilevel framework is employed. Two competing hypothesis are evaluated. First, whether a rental sector targeted towards low-income households, known as social housing, is successful in achieving adequate housing standards. Second, if a unified rental system covering broader income groups lowers the risk of housing deprivation. Housing deprivation is measured in terms of experiencing overcrowding and while also exhibiting any of the following deficits: a leaking roof; no bath/shower; no indoor toilet; or a dwelling considered too dark. Findings indicate a negative association between the size of the rental sector and the prevalence of housing deprivation. The organization of the rental sector appears most crucial and only the strategy of a rental sector encompassing broader parts of the population significantly reduces the prevalence of housing deprivation and its latent components. The association is robust in terms of confounding factors at the individual level and central country level contextual variables.

Le Sentiment d'Efficacité Personnelle (SEP) des enseignants fait référence aux croyances que se font ces derniers de leur capacité à accomplir avec succès les tâches liées à leur mission d'enseignement. Le premier objectif de notre recherche vise à répondre aux critiques qui déplorent le manque de validation rigoureuse dont les échelles de SEP ont généralement fait l'objet. Pour ce faire, nous testons, sur base d’analyses factorielles confirmatoires, la validité de trois échelles couramment utilisées dans la littérature anglo-saxonne. La première échelle de mesure, intitulée "Teacher Efficacy Scale" (Gibson & Dembo, 1986) se compose de deux dimensions: le sentiment d’efficacité personnelle mesurant la croyance qu'un enseignant se fait de sa capacité à influencer les apprentissages des élèves, et le sentiment d’efficacité générale mesurant la croyance selon laquelle le corps enseignant est capable d'apporter des changements chez les élèves, en dépit des contraintes extérieures au milieu scolaire. La deuxième échelle intitulée "Teacher Sense of Efficacy Scale" (Tschannen-Moran & Woolfolk Hoy, 2001), mesure le SEP des enseignants selon trois dimensions: l’engagement des élèves, les stratégies d’enseignements et la gestion de la classe. Cette échelle se veut plus spécifique et davantage liée aux différentes tâches pédagogiques auxquelles sont confrontés les enseignants. Enfin, comme le SEP des enseignants n’est pas forcément uniforme à travers les différentes matières enseignées, la troisième échelle de mesure se focalise sur la perception de leur capacité à enseigner les mathématiques. Cette échelle, inspirée de la mesure de McGee et al. (2014) intitulée "Self-Efficacy for Teaching Mathematics Instrument", a été adaptée pour mieux refléter les compétences en mathématiques enseignées dans l’enseignement secondaire de la FW-B. L’échelle distingue trois groupes de compétences: les nombres, les grandeurs et le traitement de données.Une fois ces trois échelles validées, le second objectif de la thèse est d’évaluer dans quelle mesure le SEP des enseignants influence la réussite en mathématiques des élèves de 2e secondaire au CE1D. Nous examinerons l’impact potentiel des différentes dimensions du SEP et émettons l’hypothèse que le SEP des enseignants influence positivement les performances des élèves, mais que l’ampleur de la relation varie selon la dimension étudiée. Cette hypothèse est testée sur base d’une analyse statistique multiniveaux. L’intérêt de cette méthode est qu’elle permet de modéliser l’influence du SEP des enseignants tout en tenant compte des caractéristiques propres aux élèves et aux classes.Nos données sont issues d’une vaste enquête que nous avons organisée au cours de l’année scolaire 2014-2015 au sein des écoles secondaires de la FW-B. Un échantillon représentatif de 164 écoles secondaires a été sélectionné de manière aléatoire. Au sein de ces écoles, tous les élèves de 2e secondaire et leurs enseignants de mathématiques ont été invités à répondre à un questionnaire. Le questionnaire adressé aux élèves était constitué de questions sociodémographiques et d’une épreuve de mathématiques destinée à mesurer les acquis des élèves en début d’année scolaire. Le questionnaire enseignant nous a permis quant à lui de recueillir un ensemble d’information sur leur SEP, leurs attitudes et leurs pratiques pédagogiques. Enfin, grâce à une convention conclue avec l’Administration Générale de l'Enseignement et de la Recherche Scientifique de la FW-B, nous avons eu l’opportunité de coupler nos données aux résultats obtenus, par les élèves de notre échantillon, au CE1D. Notre échantillon final se constitue de 10395 élèves, 598 classes, 388 enseignants et 103 écoles secondaires. Les résultats de nos analyses factorielles confirmatoires remettent en question la validité de la "Teacher Efficacy Scale" développée par Gibson et Dembo en 1986. Ce manque de validité peut s'expliquer par le fait que, contrairement aux deux autres échelles, cette dernière ne reflète pas de manière assez précise la diversité et la complexité du métier d'enseignant. Elle ne permet donc pas de mesurer la concept de sentiment d'efficacité personnelle des enseignants tel que conceptualisé dans la théorie sociocognitive de Bandura (1997). Ces résultats confirment l'idée selon laquelle le sentiment d'efficacité personnelle des enseignants est un concept multidimensionnel qui ne peut pas se mesurer de manière globale. La mesure du sentiment d'efficacité personnelle des enseignants doit être spécifiquement associée à une tâche pédagogique ou à une matière d'enseignement. Cette conclusion va dans le sens des recommandations Bandura qui précise que les croyances d'efficacité doivent être mesurées en relation avec un domaine d’activités précis. Enfin, les résultats de nos analyses multiniveau ne confirment pas la relation direct entre le sentiment d'efficacité personnelle des enseignants et la réussite scolaire des élèves. Aucune des trois dimensions étudiées ne sont significativement liées aux résultats en mathématiques des élèves de 2e secondaire. / Doctorat en Sciences politiques et sociales / info:eu-repo/semantics/nonPublished

O uso de modelos multiníveis é uma alternativa interessante para analisar dados que estão estruturados de forma hierárquica, pois permite a obtenção de diferentes estimativas de parâmetros relativos a grupos distintos e, ao mesmo tempo, leva em consideração a dependência entre as observações em um mesmo grupo. Neste trabalho, desenvolvemos e aplicamos modelos de regressão multiníveis simétricos, a fim de fornecer alternativas ao modelo usual, sob normalidade. Além disso, apresentamos uma breve análise de diagnóstico e estudo de simulação. Como motivação, consideramos dados educacionais, a fim de avaliar se o número de reprovações no histórico escolar do aluno e a infraestrutura da escola são variáveis relevantes que afetam o baixo desempenho dos alunos do ensino básico na disciplina de Matemática / The use of multilevel models is an interesting alternative to analyze data that is structured in a hierarchical manner, since it allows the obtention of different parameters estimates for distinct groups and, at the same time, it takes into account the dependence of observations in the same group. In this dissertation, we develop and apply symmetrical multilevel regression models, for the purpose of providing alternatives to the usual model, under normality. Furthermore we present a brief diagnostics analysis and a simulation study. As motivation, we consider educational data in order to assess whether the number of failures in school history of students and the school infrastructure are important variables that affect the low performance of elementary school students in Mathematics

No description available.

No description available.

Thesis advisor: Henry Braun / This study evaluates the accuracy, precision, and stability of three different methods of cross-survey analysis in order to determine their suitability for estimating the proportions of low-incidence populations. Population parameters of size and demographic distribution are necessary for planning and policy development. The estimation of these parameters for low-incidence populations poses a number of methodological challenges. Cross-survey analysis methodologies offer an alternative to generate useful, low-incidence population estimates not readily available in today's census without conducting targeted, costly surveys to estimate group size directly. The cross-survey methods evaluated in the study are meta-analysis of complex surveys (MACS), pooled design-based cross-survey (PDCS), and Bayesian multilevel regression with post-stratification (BMRP). The accuracy and precision of these methods were assessed by comparing the estimates of the proportion of the adult Jewish population in Canada generated by each method with benchmark estimates. The stability of the estimates, in turn, was determined by cross-validating estimates obtained with data from two random stratified subsamples drawn from a large pool of US surveys. The findings of the study indicate that, under the right conditions, cross-survey methods have the potential to produce very accurate and precise estimates of low-incidence populations. The study did find that the level of accuracy and precision of these estimates varied depending on the cross-survey method used and on the conditions under which the estimates were produced. The estimates obtained with PDCS and BMRP methodologies were more accurate than the ones generated by the MACS approach. The BMRP approach generated the most accurate estimates. The pooled design-based cross-survey method generated relatively accurate estimates across all the scenarios included in the study. The precision of the estimates was found to be related to the number of surveys considered in the analyses. Overall, the findings clearly show that cross-survey analysis methods provide a useful alternative for estimation of low-incidence populations. More research is needed to fully understand the factors that affect the accuracy and precision of estimates generated by these cross-survey methods. / Thesis (PhD) — Boston College, 2016. / Submitted to: Boston College. Lynch School of Education. / Discipline: Educational Research, Measurement and Evaluation.

Due to the inherently hierarchical nature of many natural phenomena, data collected rests in nested entities. As an example, students are nested in schools, school are nested in districts, districts are nested in counties, and counties are nested within states. Multilevel models provide a statistical framework for investigating and drawing conclusions regarding the influence of factors at differing hierarchical levels of analysis. The work in this paper serves as an introduction to multilevel models and their comparison to Ordinary Least Squares (OLS) regression. We overview three basic model structures: variable intercept model, variable slope model, and hierarchical linear model and illustrate each model with an example of student data. Then, we contrast the three multilevel models with the OLS model and present a method for producing confidence intervals for the regression coefficients. / text

Diversas teorias ao longo do tempo apresentam explicações sobre as estruturas de capital das organizações. As principais são a teoria Pecking Order, Teoria Trade Off e Teoria Free Cash-Flow, com base na teoria de Agência. Todas essas teorias apresentam relações teóricas entre alguns determinantes de estruturas de capital vinculadas a firma que poderiam interferir na decisão de financiamento. Uma segunda linha de estudos, vinculada a esta, apresenta que determinantes externas a firma também interferem nesta estrutura de capital, porém as variáveis de firma comportam-se de forma semelhante em diferentes cenários econômicos. (RAJAN e ZINGALES, 1995; BOOTH et. al., 2001; de JONG et. al., 2008; GURCHARAN, 2010; KAYO e KIMURA, 2011). Considerando as pesquisas anteriores, desenvolveu-se uma investigação para a confirmação desta hipótese, com o objetivo de identificar quais variáveis são mais importantes na tomada de decisão financeira e se há variabilidade em cenários temporais e ambientes econômicos distintos. Para tal foram analisadas 10.243 empresas sediadas em 61 países distintos no período de 2002-2011, totalizando o número de 58.423 observações firma ano, por meio de um modelo de regressão linear hierárquica de três níveis com medidas repetidas, verificando qual a importância das variáveis de firma e país no endividamento, se há variação das mesmas em países com diferentes contextos econômicos e em períodos de crescimento e retração econômica. Foram analisadas cinco determinantes clássicas de firma (lucratividade, tangibilidade, proteção fiscal não advinda da dívida, tamanho e oportunidades de crescimento), e onze variáveis de país que possuem relação com o endividamento (PIB, inflação, taxa de impostos, volume negociado em ações, liquidez de bolsa, capitalização das empresas listadas, índice risco país, taxa de juros, enforcement jurídico, nível de proteção ao investidor e nível de disclosure de negócios). A partir das análises realizadas, foi identificado que o endividamento está ligado em maior grau a características das firmas e ao tempo, e em menor grau, porém significante, às características do ambiente. Foi identificado que não há mudanças extremamente significativas no comportamento das variáveis de firma entre economias desenvolvidas e em desenvolvimento, assim como entre períodos pré e pós-crise financeira de 2008. Em relação as determinantes de país analisadas, observa-se que as mesmas apresentam comportamento adverso em função da crise de 2008, perdendo capacidade explicativa, e não apresentam comportamento de mudança de sinal dos coeficientes quando comparados países com desenvolvimento econômico distinto. Identifica-se que características do desenvolvimento econômico ficam mais evidentes no processo de financiamento, como acesso a recursos em economias com menor desenvolvimento. Os resultados apresentam convergência com os estudos anteriores como os de Moore (1986), Rajan e Zingales (1995), Booth et. al. (2001), Kayo e Kimura (2011), Bebzcuk e Galindo (2011), Akbar et. al (2012), entre outros. / Several theories over time present explanations of the capital structures of organizations. The main theories are the Pecking Order Theory, Trade Off and Free Cash-Flow Theory, based on the Agency Theory. All these theories have some theoretical relationships between determinants of capital structures linked to firm that could interfere in the financing decision. A second line of studies, linked to this, shows that determinants outside the firm also interfere in capital structure, but the firm variables behave similarly in different economic scenarios (RAJAN and ZINGALES, 1995; BOOTH et. al., 2001; de JONG et. al., 2008; GURCHARAN, 2010; KAYO and KIMURA, 2011). Considering previous researches, we developed an investigation to confirm this hypothesis, identifying which variables are the most important in financial decision-making and there is variability in temporal scenarios and different economic environments. To this end, we analyzed 10,243 companies based in 61 different countries in the period 2002-2011, a total number of 58,423 firm year observations, through a hierarchical linear regression model of three levels with repeated measures, checking the importance of the variables firm and country in debt, if there is variation in the same countries with different economic contexts and periods of growth and downturn. We analyzed five firm classical determinants ( profitability , tangibility, non-debt tax shield , size and growth opportunities) , and eleven variables that are related to country debt ( GDP , inflation, taxes , trading volume in shares , stock liquidity , capitalization of listed companies, country risk index , interest rate , law enforcement , level of investor protection and disclosure level) . From the analysis, it was identified the debt is linked to a greater degree the characteristics of firms and time, and on a lesser degree, but significant, with characteristics of the environment. It wasn\'t identified very significant changes in the behavior of firm variables between developed and developing countries, as well as between pre-and post- 2008 financial crisis. Regarding the determinants of country analyzed, it is observed that they present adverse behavior due to the 2008 crisis, losing explanatory power, and have no behavior change in sign of the coefficients when comparing countries with different economic development. Characteristics of economic development become more evident in the funding process, such as access to resources in less developed economies. The results show convergence with previous studies such as Moore (1986) , Rajan and Zingales (1995) Booth et al. al. (2001) Kayo and Kimura (2011) , Bebzcuk and Galindo (2011) , Akbar et al. al (2012 ), among others