Dienstag, 3. Mai 2011

Berufungsvorträge

 

Dienstag, 17. Mai 2011, 11-12 Uhr - Raum: W9-109

Dipl.-Kfm. Joachim Schnurbus
Universität Bielefeld

Black box bandwidths — how hat matrix analysis illuminates nonparametric mixed kernel regression

Nonparametric mixed kernel regression allows for flexible modeling of the functional form for a setting of mixed discrete and continuous covariates, based on some a priori decisions on the nonparametric configuration such as choosing bandwidths. These smoothing parameters influence the subsequent analysis, but the interpretation of their effect is -especially in a multivariate setting- difficult. We show how the hat matrix of nonparametric mixed kernel regression can help to choose among different configurations. Our hat matrix analysis does not depend on the covariate dimension and enables a detailed analysis of the importance of each observation for the nonparametric estimation. The hat matrix is analyzed for simulated artificial covariates and also for different nonparametric specifications using a real data set.

 

Dienstag, 31. Mai 2011, 11-12 Uhr - Raum: W9-109

Prof. Dr. Mark Trede
Westfälische Wilhelms-Universität Münster

Estimating Continuous-Time Income Models

A fundamental component of inter-temporal consumption-saving and portfolio allocation models is a statistical model of the income process. While income processes are commonly unobservable income flows which evolve in continuous time, observable income data are usually discrete, having been aggregated over time. We consider continuous-time earning processes, specifically non-linearly transformed Ornstein-Uhlenbeck processes, and the associated integrated, i.e. time aggregated process. Both processes are characterized, and we show that time ggregation alters important statistical properties. The parameters of the earning process are estimable by GMM, and the finite sample properties of the estimator are investigated. Our methods are applied to annual earnings data for the US. It is demonstrated that the model replicates well important features of the earnings distribution.

 

Dienstag, 14. Juni 2011, 11-12 Uhr - Raum: W9-109

Dr. Otto B. Walter
Universität Bielefeld

Methodische und praktische Probleme des adaptiven Testens für gesundheitsbezogene Konstrukte

Die Mehrzahl der vorliegenden methodischen und praktischen Beiträge zu adaptiven Tests konzentieren sich auf die Leistungsdiagnostik. Dabei lassen sich Vorzüge von adaptiven Testverfahren auch für die Erfassung von klinischen oder gesundheitsbezogenen Konstrukten nutzen. Der Vortrag geht auf Besonderheiten und jüngere Entwicklungen in diesem neuen Anwendungsgebiet ein.

 

Dienstag, 21. Juni 2011, 10:30-12:00 Uhr - Raum: W9-109

Dipl.-Kfm. Jan Henckens
Universität Bielefeld

A Fuel Economy Model for the German Car Fleet in 2010 - Estimates of manufacturers’ relative ability to obtain fuel economy

Emissions from private transport account for 12% of all carbon dioxide emissions within the EU making it one of the most important sources of greenhouse gases. As has been shown that downsizing strategies do not seem to be effective, improvement of fuel economy provides an interesting approach in reducing this share. Based on specially collected data with predictor variables which describe vehicle features, such as type of transmission, and vehicle line specific measurements (e.g. compression ratio) we present different methods of variable selection and propose a multiple linear regression model which is used to investigate the relative efficiencies of manufactures present on the German automotive market in 2010. Furthermore, an application of quantile regression is shown in this context.

 

Dipl.-Soz. Daniel Seddig
Westfälische Wilhelms-Universität Münster

The analysis of count data in criminology - Statistical assumptions and challenges

The analysis of the developmental aspects of deviance and delinquency over time is a main object of life-course and developmental criminology. The aim is to distinguish between developmental patterns that represent the dynamics of outcomes within distinct groups or classes of offenders. The most common method in several studies is the application of growth mixture models (Muthén & Shedden 1999, Nagin 2005) to longitudinal crime-related data (e.g. D’Unger et al. 1998, Nagin 1999, Bushway et al. 2003, Kreuter & Muthén 2008). However, since the application of common normal theory maximum likelihood estimation procedures seems inappropriate to count data, a common solution to the distributional specifics of crime-related measures is the use of the (zero-inflated) poisson growth mixture model. Still, there remains doubt if crime related count variables effectively meet the requirements of the poisson distribution, particularly the assumption of equidispersion. It has been demonstrated for count regression models that the use of the poisson distribution may lead to biased parameter estimates and standard errors if the assumption is violated, i.e. data is overdispersed (Gardner et al. 1995, Osgood 2000, Berk & MacDonald 2008). An alternative approach is the negative binomial model that adds a dispersion parameter to the probability model (Long 1997, Cameron & Trivedi 1998, Hilbe 2008). The perfomrance of normal theory maximum likelihood, poission and negative binomial estimation in first and second order polynomial growth curve models under varying levels of dispersion in ex-ante generated count data is examined in a monte carlo simulation study. Results suggest that the poisson based parameter estimates and standard errors are severly biased in the conditions of moderate and high overdispersion. In contrast, estimates and standrad errors based on the negative binomial model are not prone to bias. Since the growth curve model (Bollen & Curran 2006) can be seen as a special case of the growth mixture model (i.e. growth mixture model with a single class), the results raise implications for delvelopmental research in criminology.

 

Dienstag, 12. Juli 2011, 11-12 Uhr - Raum: W9-109

Prof. Dr. Jörg-Peter Schräpler
Ruhr-Universität Bochum

Die Nutzung von georeferenzierten Daten in der sozialwissenschaftlichen Forschung am Beispiel der Entwicklung eines Schulstandorttyps

In dem Vortrag wird eine Analyse-Methode vorgestellt, mit deren Hilfe es möglich ist, aus Punktdaten bzw. geographischen Koordinaten (wie etwa Adressdaten von SGB II-Empfängern) Häufigkeitsdichten und Dichteflächen zu erzeugen, die unabhängig von vorgegebenen Gemeinde- oder Kreisabgrenzungen sind. Die Kernel-Density-Methode stammt ursprünglich aus der ökologischen Statistik und fand inzwischen Eingang in die kriminologische sowie der epidemiologischen Forschung und wird nun im Rahmen der Sozialberichterstattung angewandt. Die SGB II-Dichten lassen sich mit amtlichen Schuldaten, wie dem Anteil der Schülerinnen und Schüler mit Migrationshintergrund, sinnvoll zu einem Standort-Index verknüpfen. In der vorgestellten Studie wird dieser Index mit einem Referenzindex des Instituts für Schulentwicklungsforschung (IFS) sowie mit den bisherigen Standorttypen des Ministeriums für Schule und Weiterbildung (MSW) in NRW evaluiert. In dem Vortrag wird zum Schluss auch auf das Problem des Datenzugangs für Wissenschaftler eingangen.