Finite mixture modeling of censored regression models
2014 (English)In: Statistical papers, ISSN 0932-5026, E-ISSN 1613-9798, Vol. 55, no 3, 627-642 p.Article in journal (Refereed) Published
A finite mixture of Tobit models is suggested for estimation of regression models with a censored response variable. A mixture of models is not primarily adapted due to a true component structure in the population; the flexibility of the mixture is suggested as a way of avoiding non-robust parametrically specified models. The new estimator has several interesting features. One is its potential to yield valid estimates in cases with a high degree of censoring. The estimator is in a Monte Carlo simulation compared with earlier suggestions of estimators based on semi-parametric censored regression models. Simulation results are partly in favor of the proposed estimator and indicate potentials for further improvements.
Place, publisher, year, edition, pages
2014. Vol. 55, no 3, 627-642 p.
finite mixture models, censoring, Tobit, EM-algorithm
Probability Theory and Statistics
Research subject Statistics
IdentifiersURN: urn:nbn:se:umu:diva-67461DOI: 10.1007/s00362-013-0509-yISI: 000339339100004OAI: oai:DiVA.org:umu-67461DiVA: diva2:612016