EINLADUNG
zum
Betriebswirtschaftlichen Forschungsseminar
des Instituts fuer Betriebswirtschaftslehre
der Universitaet Wien
Bruenner Strasse 72, 1210 Wien
Freitag, 14.06.1996; 15.30 Uhr; HS 8 des BWZ
DR. HERIBERT REISINGER (Universitaet Wien)
"DER EINFLUSS DES FORSCHUNGSDESIGNS AUF DIE HOEHE VON
BESTIMMTHEITSMASSEN IN LINEAREN REGRESSIONSMODELLEN"
Abstract:
The classical linear regression model is the standard procedure for
analyzing dependencies between variables that are measured on a
metric scale. In the course of model estimation it is common practice to
assess the appropriateness of a single descriptive model for the
problem under study with the help of coefficients of determination (R^2
and ADJ. R^2 ). When considering the advantages of calculating these
measures in empirical studies the question arises whether it makes
sense to evaluate a model by means of a single descriptive measure at
all. For example, from a statistical point of view the analyzed data set
is irrelevant when deciding on the appropriateness of the model under
consideration. However, a market researcher clearly distinguishes
whether he studies time series or cross sectional data. A well known
fact says that on the average one may expect larger coefficients of
determination for time series data than for cross sectional data.
Starting from this known phenomenon it is tried to identify various
impacts on R^2 and ADJ. R^2 that originate in the research designs of
empirical studies rather than in the research subjects within the
framework of a meta-analysis. One important result claims a strong
negative correlation between the sample size and the values of R^2
and ADJ. R^2 .
=========================================================================