Árboles de clasificación y regresión: modelos Cart / Classification and regression trees: cart models

En muchas áreas de aplicación de la Estadística los individuos (unidades de observación) están caracterizados por información sobre múltiples características cuantitativas y/o cualitativas que conforman el PERFIL del caso: por otra parte tenemos una RESPUESTA (resultado, output) individual, en algunos casos cuantitativa, en otros cualitativa. Suele ocurrir que los valores más interesantes de esa respuesta se observan en individuos, en algún sentido, peculiares; identificar los perfiles o vectores que describen a esos individuos es de interés central, pero la búsqueda para hallarlos no es simplemente lineal sino más bien es comparable a una trayectoria de divisiones sucesivas: una estructura arbórea. Los modelos subyacentes no son tan simples como un modelo de regresión lineal múltiple. Esta metodología es intensiva en el uso de recurso computacional y requiere software ad-hoc. En este artículo describimos la metodología CART ilustrándola con tres aplicaciones relativas a problemas de salud de niños, jóvenes y trabajadores chilenos.

In many application areas of Statistics, individuals (observational units) are portrayed by multidimensional quantitative and / or qualitative information; on the other hand, we have an individual response (output) sometimes quantitative, in other qualitative. Often the most interesting response values are associated to individuals that, in some sense, are special; to identify profiles that describe such individuals is of central interest but the quest to find them is not merely linear but rather is comparable to a tree structure given by successive divisions. The underlying models are not as simple as a multiple linear regression model. This methodology is intensive in the use of computational resources demanding software ad-hoc. In this article we describe the methodology CART illustrated with three applications related to real chilean health problems of children, adolescent and workers.

Modelos de ecuaciones estructurales ¿qué es eso? / Structural equations models: what’s that?

Se presenta una secuencia de modelos matemáticos-estadísticos útiles para describir las relaciones entre varias variables (respuestas y descriptores) que permiten analizar un fenómeno de interés, teniendo en cuenta la estructura de covarianzas existente entre ellas. Las relaciones entre las variables descriptoras hacen difícil un modelamiento tipo regresión lineal múltiple. Adicionalmente, algunas de las variables consideradas como posibles descriptoras de la respuesta de interés pueden ser reflejo de un factor subyacente o variable latente. Esta complejidad, que pasa de un sistema de ecuaciones lineales a un análisis de trayectorias y de allí a un modelo de ecuaciones estructurales (Structural Equations Models) requiere, para tener una traducción en un modelo parsimonioso, de un apoyo computacional específico. Como en cualquier proceso de modelamiento estadístico estadístico es esencial dar al conocimiento de la teoría del área de aplicación la máxima importancia, ya que de ese conocimiento surgirán el planteamiento y la validación de hipótesis realistas.

This paper presents a sequence of mathematical-statistical models useful for describing several variables (answers and descriptors) which allows the analysis of a phenomenon of interest, taking into account the structure of covariances existing among them. The relationships among descriptive variables make multiple linear regression type modeling difficult. Additionally, some of the variables considered as possible descriptors of the answers of interest may reflect an underlying factor or latent variable. This complexity, which goes from a linear equations system to a trajectory analysis and from there to a structural equation model requires, for having a translation into a parsimonious model, a specific computational support. As with any process of statistical modeling, it is essential to give maximum importance to the knowledge of the theory of application area, since from this knowledge the putting forward and validation of realistic hypotheses will arise.

Attitudes towards genomic: research in four Latin American countries

The present reflection refers to data obtained about the social representations of genomic research and its applications through interviews with legislators and lawyers, biomedical researchers and civilians and the review of scientific and legal literature in four Latin American countries: Argentine, Chile, Mexico and Peru. Several issues are addressed: little access to prevention and therapeutic methods, lack of equity in health benefits, commercialization of gene sequences through patents which leads to commercial exploitation of underdeveloped countries, the possibility of physical or psychological damage or genetic discrimination, the possibility of genetic modifications or abortion for eugenic reasons, the necessity of safeguarding confidentiality, risks and benefits of the use of transgenics and cloning, the necessity of legal regulation to prevent the pathway towards genetic enhancement or reproductive human cloning and of regulating access to genetic information. Using the method of content analysis of verbal behaviour to evaluate the degree of anxiety and hostility of subjects in relation to the Human Genome Project (HGP) developed by Gottschalk and Gleser, an inverse relation between levels of anxiety and level of knowledge was observed which highlights the importance of educating the population. Differences in the level of hostility towards the HGP were also found among the groups.