A consistent estimation approach that does not rely on confidential information specified by auxiliary weighting variables Seminar

Unfortunately, there are numerous situations when the users do not have access to these variables and probabilities, because of confidentiality issues. We proposed a new concept: the matrix of generalised weights, which is constructed from the auxiliary variables. This matrix would need to be computed by the data provider (e.g. National Statistical Agencies) and made available to the users. This matrix protects the confidentiality, because the values of auxiliary variables and the inclusion probabilities cannot be retrieved from this matrix. We show how this matrix can be used for inference (point estimation and confidence intervals). The effect of the auxiliary variables is taken into account through the matrix of generalised weights. We propose two approaches: an empirical likelihood approach and an approach based on a linearised variance estimator. We focus on the empirical likelihood approach because its consistency relies on weak conditions and our simulations shows that this approach provides more robust confidence intervals with skewed data.

http://eprints.soton.ac.uk/389379/