Jornal article

Hobza T., Pardo L., Vajda I.

Journal of Statistical Planning and Inference

2008

12 (2008)

3822-3840

0378-3758

Ing. Tomáš Hobza, Ph.D. (ÚTIA - Oddělení stochastické informatiky)

Logistic regression, Median, Robustness, Consistency and asy

A new method of smoothing of discrete data by means of a continuos random statistically sufficient transformation is proposed. This provides median of the trasformed data sensitive to the regrassion parameters of the original discrete model, thus allowing to apply their robust median estimation. Another new idea is the independent repeating of the smoothing to the same discrete data, called stochastic enhancing, which improves the performance of the estimation. Extensive simulations are used to demonstrate that this robust estimation outperforms the former methods known from the literature. Asymptotic theory of the estimation is established as well.