Abstract:
|
Non-probability sampling is prevailing in practice, but ignoring the underlying response mechanism leads to erroneous inferences. Incorporating auxiliary information from an independent probability sample, we propose a unified semi-parametric method to estimate the sampling weights for a non-probability sample by balancing functions of auxiliaries in a reproducing kernel Hilbert space. The consistency and the central limit theorem of the proposed estimator are established under general rejective sampling designs, and the corresponding variance estimator is also investigated. Compared with existing works, the proposed method is more robust since no parametric assumption on the response mechanism associated with the non-probability sample is required. Numerical results show that the proposed estimator outperforms its competitors, especially when the model is mis-specified and the coverage rate is satisfactory.
|