Institut Français des Sciences et Technologies des Transports, de l'Aménagement et des Réseaux

Type de document

RAPPORT DE STAGE

Langue

francais

Auteur

RANDRIANASOLO

Résumé / Abstract

This report investigates the nonresponse correction when we are confronted with mechanisms of non-ignorable nonresponse e.g. when nonresponse is not totally explained by auxiliary information or when the probability of the response only depends on the variable of interest. In the presence of unit nonresponse form, the probability of response estimation is computed using linear regression, data analysis and likelihood resolution method. This probability is necessary to have an estimation of the variable of interest with weights equaled the inverse of the inclusion probability, times the computed probability of the response, using the estimator (also known as Horvitz-Thompson's estimator). Thanks to the auxiliary information, we can attach calibrated weights (re-weighting method) in order to obtain more efficient estimators : calibration and regression estimators. Then, the variance estimation for these estimators using linearization and residual techniques allows us to build more precise condent intervals. An application of the correction method is the case of the motorcycles and bikes trips in the French National Travel Survey of 2007 - 2008. An application of this variance estimation is also the case of Kish individuals in the same travel survey.

Editeur

Univ Strasbourg, INRETS