PSYCHOMETRICS

Author: FERNÁNDEZ ALONSO RUBEN.
Year: 2004.
University: OVIEDO [www.uniovi.es].
Place of defense: FACULTAD DE PSICOLOGÍA.
Place of preparation: FACULTAD DE PEDAGOGÍA.

Author: HERRERA ROJAS AURA NIDIA.
Year: 2005.
University: BARCELONA [www.ub.es].

Summary: In recent decades there has developed a variety of methods for the detection of Functioning Differential of Items (DIF, by its initials in English) and is reported multiple investigations that seek to identify the conditions under which may be more or less adequate, or identify the factors that affect their power and error type I. It is well known that the methods, including the Mantel (MH) from Holland & Thayer (1988), the logistic regression (RL) Swaminathan & Rogers (1990) and Ji square Lord (1980) is leave affected by the sample size of the focus groups and reference but has not been studied so far, what is the effect of the reason for sample sizes of the groups on their power and error type I. This aspect is particularly important when you want to analyze the possible bias of an item or a test against a minority group significantly. In this paper were conducted three experiments Monte Carlo to analyze the effect of sample size of the reference group (500 and 1500), the reason for sample sizes (2, 2.5, 3, 4 and 5) and their interaction; on the power and type I error of MH, RL and Ji Square. In the latter case was also considered as the effect of the length of trial (500 and 100 items). The data were generated using a three-parameter logistic model, in all cases, the factorial design was completely crossed and the dependent variables were the detection rate as a measure of the statistical power and the rate of false positive as a measure of their error, calculated from 100 aftershocks. The results identify significant factors, compare and generate statistical practice on the use made of it by the psychologists involved in the development and analysis of psychological tests.

Author: HERRERA ROJAS AURA NIDIA.
Year: 2005.
University: BARCELONA [www.ub.es].

Summary: In recent decades there has developed a variety of methods for the detection of Functioning Differential items (DIF, by its initials in Spanish), and is reported multiple investigations that seek to identify the conditions under which may be more or less appropriate , or identify the factors affecting its power and error type I. It is well known that the methods, including the Mantel (MH) from Holland & Thayer (1988), the logistic regression (RL) Swaminathan & Rogers (1990) and Ji square Lord (1980) are allowed to affect the sample size of the focus groups and reference but has not been studied so far, what is the effect of the reason for sample sizes of the groups on their power and error type I. This aspect is particularly important when you want to analyze the possible bias of an item or a test against a minority group significantly. In this paper were conducted three experiments Monte Carlo to analyze the effect of sample size of the reference group (500 and 1500), the reason for sample sizes (2, 2.5, 3, 4 and 5) and their interaction; on the power and Type I error MH, RL and Ji square. In the latter case also examined the effect of the length of trial (500 and 100 items). The data were generated using a three-parameter logistic model, in all cases, the factorial design was completely crossed and the dependent variables were the detection rate as a measure of the statistical power and the rate of false positives as a measure of their error, calculated from 100 aftershocks. The results identify significant factors, compare and generate statistical practice on the use made of it by psychologists involved in the development and analysis of psychological tests. KEYWORDS: DIF Mantel, logistic regression, size sample.

Author: JIMENEZ TORRES SILVIA.
Year: 2005.
University: SALAMANCA [www.usal.es].
Place of defense: FACULTAD DE PSICOLOGIA.
Place of preparation: FACULTAD DE PSICOLOGIA.

Summary: The instruments developed for the diagnosis and assessment of the intensity of social anxiety have generally been developed since the Classical Test Theory. The objective of this research was to analyze psicométricamente versions of the questionnaires in Spanish Social Anxiety and Distress Scale, Liebowitz Social Anxiety Scale and Sheehan Disability Inventory for the assessment of social anxiety adapted by Bobes and collaborators in 1999, both from the Classical Theory Tests and from the Item Response Theory, as well as to compare the performance of both models. The sample consisted of 1218 subjects at the University of Salamanca. The results of the analysis support the classical goodness of these instruments and put evidenced by its adapters (Bobes et al. 1999). The analysis of data according to the Item Response Theory resulted in a proper adjustment overall both items as subjects in the questionnaires. It found a lack of items to assess the entire continuum of social anxiety and overlap in index intensity of some of them, which would render the assessment of this personality trait. The Item Response Theory seems not offer more information than the classical theory when the implementation of the questionnaires is collective, autoaplicada and paper and pencil. However offers significant advantages to construct tests tailored to the subject.

Author: GARCÍA FORERO CARLOS.
Year: 2006.
University: BARCELONA [www.ub.es].
Place of defense: FACULTAD DE PSICOLOGÍA.
Place of preparation: UNIVERSITAT DE BARCELONA.

Summary: What is the most appropriate method for estimating models questionnaires graduated response? The argument that the estimation methods based on the Theory of Response Item (TRI) have better properties in endless samples and using much information has been relegated to the methods of analysis of Factorial Items (AEF). However, what is the most appropriate method in finite sample is a matter that still has no clear answer. METHODS Using data categorical modeled along the lines of Samejima, we compare the performance of the estimator TRI Maximum Verosimilitud Marginal with the EM algorithm (MVM-EM) with the estimator AFI not weighted least squares (ungweighted least squares, ULS). The latter was compared using two parametrizations in the estimation: TRI metrics and standardized metrics. We carried out an extensive study of Monte Carlo simulation to compare the performance of these estimators based on: 1-Number of points (200, 500 and 2000) 2-Number of items of the test (9, 21 and 42) 3-Number of categories in item (2 and 5) 4-Asymmetry of the responses observed (centered, asymmetrical and high asymmetry). 5-Load Factor of the following items: 0.4, 0.6 and 0.8. 6-number of variables latent (1 and 3). For each condition were carried out 1000 replications. The criteria for performance of the estimators were the number of replications converged bias on the parameters and their errors typical coverage parameters and typical errors and the average time for each replication. RENDERING The main interest lies in the study compare the performance of an estimator with good properties in samples finite (ULS) for the properties asintóticas optimal estimator (MVM-EM), the latter much more widespread. Three are the main contributions. First, it is the first time in which they compare the errors typical of the estimator ULS in the case dichotomous and politómico. Given the length of questionnaires scale, the behavior of the typical errors is of great significance for assessing the structure of the models. Secondly, are completed and are integrating the results of published literature simultaneously considers all factors traditionally studied. In addition, while in the literature were analyzed results mainly through inspection tables, we use a statistical model for analyzing results. This allows to know the trends in terms of the manipulated variables. Finally, ULS and MVM-EM are compared fairly, while using the same metrics for both in the estimation of the magnitude of the bias is comparable. RESULTS AND DISCUSSION The results suggest that the performance of MVM-EM and ULS is very similar in all conditions studied. Both estimators failed in the same conditions, the most restrictive for the estimation of the model: small samples with only a few items dichotomous far from normal. In cases less restrictive, ULS earned a performance comparable or even better than MVM-EM. These results, together with the flexibility for modeling ULS and its speed ULS should continue to be an alternative to consider more often in practice with data scale. KEYWORDS Maximum Verosimilitud, Methods of least squares, Model graduated response, ULS, Model Samejima, factor analysis of items, ordinal variables, variables Likert, Model factorial and sample size and No-normalidad in the indicators.

Author: BARRA SALAZAR ANA MARIA.
Year: 2006.
University: POLITÉCNICA DE CATALUÑA [www.upc.edu].
Place of defense: Aula 28.8 (ETSEIB).
Place of preparation: Edifici C5 DESPATX: 003, PLANTA 0 ND.

Summary: The goal of this thesis is to analyze the effects of the Organizational Learning Process via the 4I model, over the capabilities of organizations to generate competitive advantages. We analyze how the 4I model, developed by Crossan, Lane, and White (1999), helps organizations to develop advantages over other organizations. We define global variables that influence both the organizational learning capabilities of organizations, and the development of competitive advantages, and analyze them statistically. We perform surveys over an important sample of organizations from Chile that exports salmon to the European Union. These surveys are divided into the following three important aspects: (a) The organization. (b) Organizational learning. c) Competitive advantages. The surveys allows us to measure the correlation of global variables, and they reliability. Also, we do some field work in important Chilean organizations to obtain more information regarding this research. We apply a deductive and inductive research approach to prove the initial proposicion of the influences of the 4I model in the capabilities of enterprises to generate competitive advantages. Moreover, we establish the scope of the organizational learning concept, such as the organizational knowledge, knowledge management, and learning organization. We also review the important concepts of core competence, and competitive advantages, to explain the relationship between them and the organizational learning process.

Author: GRAUPERA SANZ JOSE LUIS.
Year: 2006.
University: CASTILLA-LA MANCHA [www.uclm.es].
Place of defense: FACULTAD DE CIENCIAS DEL DEPORTE.
Place of preparation: FACULTAD DE CIENCIAS DEL DEPORTE.

Summary: The theoretical framework of the thesis is built from the analysis of the concept of learning styles and major taxonomies attempting to organize the different theoretical perspectives from which to construct such deals. It develops a revised and updated classification of learning styles, which is based on previous scientific tradition and supports research thesis. They are presented and critically all models of learning styles that have had some connection with the investigation in the area of physical activity and sport, with particular emphasis on derivatives of the experiential learning theory (Kolb, 1984). There will always be a comprehensive bibliographic scrutiny, prepared with a rigorous search strategy and clearly defined in scope, boundaries and procedure. We present a study bibliométrico of publications and compiled an outline detailed documentation obtained, depending on the topography of learning styles, the theoretical models of reference, tools used, samples used and methodological characteristics. It performs a critical analysis of each and every one of the works collected in the literature, thus establishing the state of affairs, ie scientific knowledge about learning styles in physical activity and sport. The empirical research of the thesis was directed toward the study of style experiential learning students Activity Science and Sports, athletes and sports coaches. First used the model of learning styles Kolb, whichever Learning Styles Inventory (IEA) in its Spanish version (Gonzalez Tirados, 1983) representative samples of the three people mentioned above. In a second research applies Questionnaire Honey-Alonso Learning Styles (CHAEA), which also comes from a model experiential (Alonso, Galician and Honey, 1994). Design investigations are within the ambit paradigm associative or differential, using as a group selection variables: gender, age, sport experience, expertise sports, job sports practiced and specialty university extended. The main conclusions it leads the analysis of the results are as follows: students Activity Science and Sports and athletes have a way of learning more rich and balanced that the university population generally, the sports coaches are characterized accused by a predominance of analytical dimensions or assimilative style, the style of learning is different in men and women's sports, with age and experience will produce slight variations in style, the relationship between the style and skill sports is inconclusive, athletes Task open and closed tasks show a slightly different learning style. The two instruments used in investigations show considerable gaps psychometric reliability low and unsatisfactory (on the edge of acceptability to study differential) and deficiencies in terms of construct validity.