Dienstag, 3. November 2009, 11:30-12:00 Uhr - Raum: W9-109

Genetic Validity of Psychological Constructs: Modeling of Multimethod Twin Data
Dipl.-Psych. Christian Kandler
Universität Bielefeld

Constructs can be considered ‘valid’ to the degree to which different methods converge (Campbell & Fiske, 1959). Previous research on the accuracy of psychometric measurement has already shown that an increase of quantity and quality of construct-relevant information yield an increase of correlations between different methods measuring one construct (e.g., Blackman & Funder, 1998; Letzring, Wells, & Funder, 2006). This supports the concept of convergent validity. However, further studies have shown that method-specific components may provide incremental validity, in particular if multiple rater perspectives were used as different methods (e.g., Kraemer, Measelle, Ablow, Essex, Boyce, et al., 2003; Vazire & Mehl, 2008). Multiple-rater twin studies offer additional insight into the sources of both common variance in different rater perspectives and variance in rater-specific perspectives. Referring to a specific application, the identification and the size of genetic influences on both common variance and specific variance in self-reports and peer reports on adult personality can have important implications for how to interpret convergent valid and method variance and thus for the research on accuracy in personality judgments.

  

Dienstag, 1. Dezember 2009, 11-12 Uhr - Raum: W9-109

Forecasting Stationary Processes Under Asymmetric Loss
JProf Dr. Matei Demetrescu
Johann Wolfgang Goethe-Universität Frankfurt am Main

 

Dienstag, 15. Dezember 2009, 11-12 Uhr - Raum: W9-109

Expectile smoothing
Sabine Schnabel
Biometris, Wageningen University and Research Center

As a least squares alternative to quantile regression Newey and Powell proposed in 1987 expectile estimation using asymmetric least squares. This is a weighted generalization of classical least squares and results in curves at different locations in the data not only the mean trend. We combine this estimation approach with P-splines for a flexible smooth functional form. Several models have been developed in this context. The simple least asymmetrically weighted squares (LAWS) model is the basic model. It estimates single expectile curves for different levels of asymmetry p, 0 <p<1. In order to overcome the possible problem of crossing expectile curves for different levels of p we propose the expectile bundle model. This is a location-scale model fitting all expectiles curves simultaneously. It results in a set of smooth non-crossing expectile curves. These two models are complemented by the so-called expectile sheets as a third approach in expectile estimation. This technique includes the asymmetry parameter p as an additional dimension. It is based on a tensor product of two B-spline bases and implies smoothness both over x and p. Similar to quantiles expectiles give us information about the distributing underlying the data. Therefore we can use the expectiles curves to estimate the density of the data. The presentation includes an overview over possible areas of application and includes examples from demography, public health and other areas of research.

  

Dienstag, 19. Januar 2010, 11-12 Uhr - Raum: W9-109

Analysis of the salary trajectories in Luxembourg : a finite mixture model approach
Professor Jang Schiltz
Université du Luxembourg

We present the finite mixture model of Daniel Nagin. In  a second part of the talk, we analyze the salaries of about 700.000 employees who worked in Luxembourg between 1940 and 2006 with the aim of detecting groups of typical salary trajectories with respect to some covariants like sex, workstatus, residentship and nationality. We use the proc traj SAS procedure from Bobby L. Jones to classify the workers by Nagin’s finite mixture model and descriptive statistical methods like the CHAID procedure to get a caracterization with respect to the covariants of the different groups. In a third part of the talk we use these results to analyse the long term sustainability of the Luxembourg pension system. Finally, we compare the current pure repartition model to a possible mix of repartition and capitalization. More precisely, we propose a probabilistic economic model to decide for each of the groups found above which part of the pension should be financed by repartition and which part by capitalization. It is based on a portfolio type risk management approach to compare the demographic risk inherent to the repartion strategy to the financial risk inherent to the capitalization strategy.