| Verfügbarkeit |
Autoren | Rohwer, Götz |
Titel | Using sampling weigths for model estimation? |
URL | https://www.neps-data.de/Portals/0/Working Papers/WP_IV.pdf |
Erscheinungsjahr | 2011 |
Seitenzahl | 20 S. |
Verlag | Bamberg: Otto-Friedrich-Universität |
Reihe | NEPS Working Paper. Band 4 |
Dokumenttyp | Graue Literatur; Monographie; online |
Beigaben | Literaturangaben |
Sprache | englisch |
Forschungsschwerpunkt | Bildungspanel (NEPS) |
Schlagwörter | Bildungsforschung; Sampling; Statistisches Modell; Statistische Methode; Modellierung; Variable |
Abstract | This paper discusses the question whether one should use (design-based) sampling weights when estimating statistical models. It is argued that the answer depends, in particular, on the kind of model to be estimated. The paper distinguishes three kinds. (1) Descriptive models that intend to provide simplified descriptions of the distribution of variables defined for a target population. It is argued that, except for some special situations, sampling weights should be taken into account when estimating such models. (2) Probabilistic data models which start from the idea that the data in a given sample can be viewed as realizations of random variables. It is argued that thinking about the usage of sampling weights in the estimation of such models depends on the understanding of the relationship between the model and the random variables serving to represent the given data. (3) Probabilistic functional models which intend to formulate rules for a generic unit defined without reference to any particular target population. It is argued that using sampling weights in the estimation of such models is required only if the selection probabilities used in the sampling procedure depend on endogenous variables of the model. (DIPF/Orig.) |
Förderkennzeichen | 01GJ0888 |