| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.7 MB | Adobe PDF |
Resumo(s)
O presente trabalho centra-se no conceito de mediação, ou seja, como a relação entre duas variáveis pode ser explicada por uma ou várias variáveis, as quais se designam por variáveis mediadoras. Assim, depois de definir este conceito pormenorizadamente, é feita a distinção de outros conceitos que têm igualmente na sua base a introdução de uma ou várias “terceiras” variáveis, na relação entre duas variáveis. Aqui, apenas é desenvolvido o modelo de mediação simples, isto é, o modelo que envolve apenas uma variável mediadora, apesar da referência breve ao modelo de mediação múltipla. Os conceitos fundamentais de efeito directo, efeito indirecto ou de mediação e efeito total são também descritos. De seguida, é feito um levantamento das abordagens existentes referentes a este conceito central, havendo uma distinção entre as abordagens tradicionais - que se baseiam nas estimativas obtidas através de um conjunto de equações de regressão linear, o qual traduz o modelo de mediação simples – e a abordagem contrafactual - que se baseia no conceito de resultados potenciais. A abordagem contrafactual é mais fácil de ser aplicada a qualquer modelo estatístico. A aplicação deste assunto é realizada com base nos dados de um inquérito aplicado a três zonas com diferentes características, da cidade da Praia, Cabo Verde, tendo como objectivo analisar as desigualdades em saúde conforme cada zona. Algumas variáveis do questionário foram escolhidas para o estudo da mediação. Assim, são analisadas três situações que têm em comum a variável independente e dependente e variam quanto à variável mediadora. Todas estas variáveis foram tratadas como binárias. O objectivo é estimar, com base nos dados, os efeitos de interesse, para que se possa concluir quanto à presença ou ausência de mediação. Essa estimação é realizada com recurso ao pacote “mediation” do R, que usa a abordagem de resultados potenciais.
The present work focuses on the concept of mediation, that is, how the relationship between two variables can be explained by one or more variables, which are called mediating variables. Thus, after defining this concept in more detail, a distinction of other concepts that also have at their base the introduction of one or more "third" variables in that simple relationship between two variables, is made. Here, only the simple mediation model is developed, that is, the model that involves only one mediating variable, despite the brief reference to the multiple mediation model. The fundamental concepts of direct effect, indirect or mediation effect and total effect are also introduced. Next, a survey of existing approaches to this central concept is made. On the one hand, several traditional approaches are enumerated, all based on the estimates obtained through a set of regression equations, which translate the simple mediation model; on the other, the counterfactual approach is defined, which is based on the concept of potential results, which is easier to apply to any statistical model. The application of this subject is based on the data from a survey applied in three zones, with different characteristics, of the city of Praia, Cape Verde, in order to analyze health inequalities according to each zone. Based on the survey and in the variables withdrawn, some variables were chosen. Thus, three situations are analyzed, that has in common the independent variable and the dependent variable and varies in the mediating variable. All these variables were treated as binary. The objective is to estimate, based on the data, all the effects of interest, so that it can be concluded on the presence or absence of mediation. This estimation is done using the mediation package of R, which is based on the potential results approach.
The present work focuses on the concept of mediation, that is, how the relationship between two variables can be explained by one or more variables, which are called mediating variables. Thus, after defining this concept in more detail, a distinction of other concepts that also have at their base the introduction of one or more "third" variables in that simple relationship between two variables, is made. Here, only the simple mediation model is developed, that is, the model that involves only one mediating variable, despite the brief reference to the multiple mediation model. The fundamental concepts of direct effect, indirect or mediation effect and total effect are also introduced. Next, a survey of existing approaches to this central concept is made. On the one hand, several traditional approaches are enumerated, all based on the estimates obtained through a set of regression equations, which translate the simple mediation model; on the other, the counterfactual approach is defined, which is based on the concept of potential results, which is easier to apply to any statistical model. The application of this subject is based on the data from a survey applied in three zones, with different characteristics, of the city of Praia, Cape Verde, in order to analyze health inequalities according to each zone. Based on the survey and in the variables withdrawn, some variables were chosen. Thus, three situations are analyzed, that has in common the independent variable and the dependent variable and varies in the mediating variable. All these variables were treated as binary. The objective is to estimate, based on the data, all the effects of interest, so that it can be concluded on the presence or absence of mediation. This estimation is done using the mediation package of R, which is based on the potential results approach.
Descrição
Tese de mestrado, Estatística e Investigação Operacional (Estatística ) Universidade de Lisboa, Faculdade de Ciências, 2017
Palavras-chave
Mediação Causalidade Efeitos directos e indirectos Diferença de coeficientes e produtos de coeficientes Resultados contrafactuais Teses de mestrado - 2017
