Estadística inferencial. 6. Ejemplo 2

Estadística inferencial. 6. Ejemplo 3

This PhD thesis is focused on disclosing features that might cause the increase in the use of analytical tools for better decision making. The theoretical part of this research is developed in two phases. At first, an exhaustive literature review was conducted with the purpose of identifying the main features in companies that impact positively the adoption of new analytical tools. This review brought our attention in four key drivers which were the foundation of the theoretical model: management support on data analysis, data-based competitive advantage, systemic thinking and communication outside the company. Secondly, a scale was proposed for classifying companies according with how its analytical capabilities are developed. The theoretical model and scale required to be validated with data from the real world. Four constructs derived from the model were operationalized in 17 items. An extensive statistical research related with the agreement, convergence, test-retest reliability and factor structure of the dimensions was conducted. Results allowed us to ascertain that our instrument is reliable and valid. Then, the questionnaire was sent to companies located in Barcelona area. The central part of the research analyzes data obtained from the companies. At first, the statistical engineering, which is interpreted as the link between the statistical thinking (the strategic management) and methods (the day-to-day operations), was adapted as guideline. A set of seven statistical tools were wisely assembled in a sequential order and relevant conclusions were obtained. Later, it was necessary to validate our preliminary conclusions with additional research and make them more robust. A second approach was utilized with this purpose. The evidential reasoning, which is a type of multi criteria decision analysis method, was implemented. Two different approaches lead us to similar results. At this phase of the thesis unstructured and soft features about the analytical practices were still missing. A complementary approach was needed to include aspects as personal values, beliefs and motivations and identify how they influence on analytical practices of the companies. The laddering methodology was utilized for these purposes. It is defined as a type of in-depth interview that is applied to understand how individuals transform attributes of any given concept into meaningful associations with respect to themselves. Consider this analogy; the data from questionnaires gave us "the picture of forest", then in-depth interviews yielded "the picture of the three". The last part of the thesis is reserved to provide guidelines to companies interested on increasing their analytical capabilities. Here it is offered a road map composed of five stages. The proposed order is: A company receive its diagnostic and is given a stage in the road map, later guidelines are provided to move the company upwards into the scale. The sequence of diagnostic-guidelines-diagnostic should be repeated until the company reach the highest level in the scale: analytics as competitive advantage. At the end of the thesis are presented two sets of values and attributes which were found decisive for increasing the adoption of analytical tools. In the first set, three values: honesty, serving the society and leadership impact the statistical thinking (the strategic level) in the company, whereas three attributes: the goal setting, creativity and information from outside are acting on the statistical methods (the operational level). The statistical engineering (the tactical level) establish a link between strategic and operational levels. All the tools and methods developed in this thesis, including the questionnaire, the scale for ranking the companies, the script for in-depth interviews, the road map for moving upward to higher levels in the scale and its related guidelines, represent an original and helpful toolkit for improving the analytical capabilities in companies.

En aquesta tesi s’apliquen i desenvolupen models estadístics amb variables latents per a l’anàlisi conjunt (multivariant) de trastorns mentals, com a instrument de mesura en estudis epidemiològics. En l’anàlisi multivariant dels trastorns mentals no només es mesura la presència dels trastorns, sinó que també es té en compte les seves associacions, ajustant directament la comorbiditat mental, i permetent quantificar estats de salut psíquica individuals que no es poden observar directament. L’estructura de comorbiditat es construeix a partir d’estudis psiquiàtrics conceptuals previs. S’utilitzen models de tipus IRT (Item Response Theory) per construir les mesures dimensionals de salut mental que no es poden observar directament (les variables latents) a partir de la informació observada, categòrica, sobre els trastorns mentals. Els models de variables latents modelitzen simultàniament les variables observades i els individus. Al llarg d’aquesta tesi es demostra que els trastorns mentals definits categòricament segons el manual DSM-IV responen a fenòmens dimensionals. Els trastorns mentals més rellevants poden agrupar-se en un nombre petit de dimensions, i dins de cada dimensió existeix una gradació dels nivells de salut. Aquests models poden utilitzar-se per descriure nivells de salut mental d’una població en un moment determinat (utilitzant informació sobre trastorns patits en aquell moment) o per a estimar la predisposició a patir trastorns mentals en algun moment de la vida (trastorns vida). En aquesta tesi es realitzen estudis epidemiològics de la salut mental, enfocats en considerar la comorbiditat mental i l’estimació de les dimensions subjacents als trastorns observats, i modelant l’existència d’una classe sense trastorn (‘sana’) i ‘malalta’ dins la població. Les dades consisteixen en una mostra representativa de la població europea (ESEMeD). S’analitza l’estat ‘actual’ de la població per mitjà dels trastorns patits en l’últim any i es desenvolupa una metodologia per analitzar els trastorns vida tenint en compte que es disposa de mesures retrospectives, en les quals la informació sobre aparició de trastorns es troba censurada per la dreta (la gent sense trastorn observat fins al moment de la recollida de dades pot estar encara a risc de patir-lo en el futur). També es formulen i apliquen models on la dimensionalitat latent es mesura de manera discreta-ordinal, en comptes de l’assumpció contínua utilitzada habitualment. Cadascun d’aquests models descriuen diferents aspectes del la salut mental en la població europea. / Latent variable statistical models are developed and applied to the joint (multivariate) analyses of mental disorders, as measurement instruments for epidemiological studies. Multivariate analyses of mental disorders does not only take into account their presence, but also their associations, straightly modelling the mental comorbidity and allowing to (quantitatively) estimate underlying mental health states, which are not directly observable (non-manifested). Comorbidity structures were built upon previous conceptual psychiatric studies. Item Response Theory (IRT) models were used to estimate the non-manifested psychiatric dimensions (latent variables), based on the categorical manifested (observed) mental disorder information. Latent variable models allow for analysing the observed variables and the individuals, jointly. Along this research work has been shown that the categorical DSM-IV definitions of mental disorders follow dimensional constructs. Most relevant mental disorders can be grouped in small number of dimensions; within each dimension there is a gradation of mental health states. The presented latent variable statistical models can describe population mental health states at some specific time-point (e.g. current or 12-month observed disorders) or the individual predisposition to suffer from mental disorders at any moment in life (lifetime disorders). Mental health epidemiological studies were conducted, based on the dimensional structures of mental comorbidity that underlies to the manifested disorders. Given that only a minor proportion of the general population suffers from mental disorders, it was included in the statistical model the distinction (also latent) between individuals with and without mental disorders (‘ill’ and ‘healthy’, respectively). Data come from an European population representative sample (ESEMeD). First, the population’s current mental health state was analysed, based on the disorders suffered in their last year. Second, a new methodology has been developed in order to analyse the lifetime disorders, taking into account that only retrospective information was available, in which disorder’s onset information is right-censored (individuals without a mental disorder at the interview time-point may be still at risk to suffer it in the future). Moreover, it has been formulated and applied latent variable statistical models in which the dimensionality is modelled in a discrete-ordinal way, instead of the common continuous assumption. Each of these models describes different aspects of the European mental health states.

Els motors de cerca d’Internet recullen i emmagatzemen informació sobre els seus usuaris per tal d’oferir-los millors serveis. A canvi de rebre un servei personalitzat, els usuaris perden el control de les seves pròpies dades. Els registres de cerca poden revelar informació sensible de l’usuari, o fins i tot revelar la seva identitat. En aquesta tesis tractem com limitar aquests problemes de privadesa mentre mantenim suficient informació a les dades. La primera part d’aquesta tesis tracta els mètodes per prevenir la recollida d’informació per part dels motores de cerca. Ja que aquesta informació es requerida per oferir un servei precís, l’objectiu es proporcionar registres de cerca que siguin adequats per proporcionar personalització. Amb aquesta finalitat, proposem un protocol que empra una xarxa social per tal d’ofuscar els perfils dels usuaris. La segona part tracta la disseminació de registres de cerca. Proposem tècniques que la permeten, proporcionant k-anonimat i minimitzant la pèrdua d’informació. / Web Search Engines collects and stores information about their users in order to tailor their services better to their users' needs. Nevertheless, while receiving a personalized attention, the users lose the control over their own data. Search logs can disclose sensitive information and the identities of the users, creating risks of privacy breaches. In this thesis we discuss the problem of limiting the disclosure risks while minimizing the information loss. The first part of this thesis focuses on the methods to prevent the gathering of information by WSEs. Since search logs are needed in order to receive an accurate service, the aim is to provide logs that are still suitable to provide personalization. We propose a protocol which uses a social network to obfuscate users' profiles. The second part deals with the dissemination of search logs. We propose microaggregation techniques which allow the publication of search logs, providing $k$-anonymity while minimizing the information loss.

Tesi per compendi de publicacions. / This thesis is about statistics' contributions to industry. It is an article compendium comprising four articles divided in two blocks: (i) two contributions for a water supply company, and (ii) significance of the effects in Design of Experiments. In the first block, great emphasis is placed on how the research design and statistics can be applied to various real problems that a water company raises and it aims to convince water management companies that statistics can be very useful to improve their services. The article "A methodology to model water demand based on the identification of homogeneous client segments. Application to the city of Barcelona", makes a comprehensive review of all the steps carried out for developing a mathematical model to forecast future water demand. It pays attention on how to know more about the influence of socioeconomic factors on customer's consumption in order to detect segments of customers with homogenous habits to objectively explain the behavior of the demand. The second article -related to water demand management, "An Approach to disaggregating total household water consumption into major end-uses" describes the procedure to assign water consumption to microcomponents (taps, showers, cisterns, washer machines and dishwashers) on the basis of the readings of water consumption of the water meter. The main idea to accomplish this is, to determine which of the devices has caused the consumption, to treat the consumption of each device as a stochastic process. In the second block of the thesis, a better way to judge the significance of effects in unreplicated factorial experiments is described. The article "Proposal of a Single Critical Value for the Lenth Method" analyzes the many analytical procedures that have been proposed for identifying significant effects in not replicated two level factorial designs. Many of them are based on the original "Lenth Method and explain and try to overcome the problems that it presents". The article proposes a new strategy to choose the critical values to better differentiate the inert from the active factors. The last article "Analysing DOE with Statistical Software Packages: Controversies and Proposals" review the most important and commonly used in industry statistical software with DOE capabilities: JMP, Minitab, SigmaXL, StatGraphics and Statistica and evaluates how well they resolve the problem of analyzing the significance of effects in unreplicated factorial designs

Se describe someramente un trabajo en el que se examina con cierto detalle un conjunto de métodos propuestos en Little, R. J.A. y Smith, P.J. {1987}, "Editing and Imputation for Quantitative Survey Data", Journal of the American Statistical Association, Vol. 82, pags. 58 -68. El referido trabajo es la tesis de Mestrado em Estadística presentada, y aprobada, en IMPA por Pedro Luis do Nascimento Silva.

La cuantificación del valor estadístico de la vida (VEV) se ha vuelto importante para la evaluación de proyectos y políticas públicas, esta valoración se hace a través de distintos métodos que miden el intercambio entre riesgo y salario, uno de los métodos más usados es el modelo de salarios hedónicos. Bajo este esquema conceptual, el presente trabajo realizó una estimación del VEV para el Perú en el año 2014, a partir de una base de datos del mercado de trabajo peruano extraído de la ENAHO (2014) y de la construcción de índices de riesgo laboral de cada actividad económica para el período de la estimación. Los principales resultados de esta investigación comprobaron la correlación entre salario y riesgo mortal en el trabajo, así también se obtuvo un VEV de S/.8.61 millones para el mercado laboral peruano a precios del 2014. Este valor podrá ser usado en los análisis costo beneficio de los proyectos de inversión que incidan en la vida de la población estudiada y en políticas públicas que busquen reducir el riesgo de muerte prematuro en la sociedad. / Tesis

Se hace un esbozo de la justificación estadística y matemática propia de la Psicometría para el problema de medición psicológica, con una introducción a la axiomática y principales relaciones de la teoría clásica de los tests. Nuestro abordaje se basa en una revisión de la literatura pertinente, pero, adicionalmente, contiene resultados y derivaciones propias. Un valor adicional está constituido por el tratamiento como modelo y su conexión desde la teoría matemática de medición y la justificación de la aleatoriedad de los puntajes observados, así como incorporar la propuesta de Zimmerman sobre la definición de validez. El presente estudio pretende ofrecer una presentación del modelo psicométrico de la Teoría clásica de los tests, el cual contiene parte del modelo lineal clásico para una medición en una población de sujetos.

Este trabajo se compone de dos partes que se enmarcan en el tarea del procesamiento estadístico de imágenes. La primera, corresponde al modelado estadístico de una imagen y la segunda a la medición de la calidad de una imagen en el contexto de fusión. Cada una de las partes de este trabajo es autocontenida. En el Capítulo 1 de la primer parte, se describe el marco teórico concerniente al modelado estocástico de una imagen de textura mediante los Campos Aleatorios de Gibbs-Markov. Además, se establecen las condiciones su ficientes para obtener la descomposición de un potencial de Gibbs en configuraciones basadas en el esquema de codificación de Besag. En el Capítulo 2, se propone un nuevo método de estimación basado en el esquema de codificación de Besag, denominado estimador de Mínimo Cuadrado Condicional Coding (MCCC), implementado mediante un algoritmo paralelo. Para evaluar su performance, se llevó a cabo un estudio de simulación Monte Carlo en situaciones específicas de un modelo de textura Gibbs-Markov, el Modelo Autobinomial. El estudio consistió en un análisis descriptivo de su comportamiento, el testeo de su sensibilidad ante presencia de ruido (distintos grados de contaminación de diferentes patrones homogéneos de textura), la evaluación de su algoritmo paralelo y su capacidad en la clasificación supervisada de diferentes texturas presentes en una imagen satelital concreta. La segunda parte aborda la problemática de la medida de la calidad de una imagen en el contexto de fusión. En el Capítulo 4 se presentan las características de las medidas de calidad basadas en el Índice de similitud estructural, como así también algunas propiedades de interés de este índice. Luego, en el Capítulo 5 se propone una nueva medida de calidad de fusión de imágenes que tiene como objetivo ser ciega (sin imagen de referencia), de cálculo significativamente simple y de fácil implementación. Para validar esta propuesta se desarrolla tres experimentos que evalúan: su propósito general, la estabilidad y capacidad discriminativa, como así también, la correlación con la opinión subjetiva. Por úlltimo, se presenta una aplicación de la medida propuesta en la obtención de imágenes de pinturas de alta calidad, a través de un proceso de exclusión que selecciona con éxito el mejor subconjunto de imágenes de entrada en términos de la imagen fusionada. / This work consists of two parts, framed in the area of statistical image processing. The first corresponds to the statistical modeling of an image and the second to the measurement of the quality of an image in the fusion context. In Chapter 1, the theoretical framework concerning the stochastic modeling of a texture image using the Gibbs-Markov Random Fields is described. Furthermore, sufficient conditions are established to obtain the decomposition of a Gibbs potential in con figurations based on the Besag coding scheme. In Chapter 2, we propose a new estimation method based on the Besag coding scheme, called the Coding Conditional Least Square estimator (CCLS), implemented by a parallel algorithm. In order to evaluate its performance, we have conducted a Monte Carlo simulation study in specifi c situations of a Gibbs-Markov texture model, the Auto-binomial Model. The study consisted of a descriptive analysis of its behavior, the sensitivity testing before the presence of noise (different contamination degrees in different homogeneous texture patterns), the evaluation of its parallel algorithm, and its capacity in the supervised classi cation of different textures on a specifi c satellite image. The second part addresses the problem of measuring the quality of an image in the fusion context. Chapter 4 presents the quality measures characteristics of based on the structural similarity index, as well as some interest properties of this index. ubsequently, in Chapter 5, a new image fusion quality measure is proposed. This measure aims to be blind (without a reference image)and signifi cantly simple to calculate. It is also, easily to implemented. In order to validate this proposal, we have developed three experiments to asses the following aspects: its general purpose, stability and discriminative capacity, as well as the correlation with subjective opinion. Finally, we present an application of the proposed measure in obtaining images of high quality paintings, by means of an exclusion process (leave one out) that successfully selects the best subset of input images in terms of the fused image.