Events and Meetings of Italian Statistical Society, Advances in Latent Variables - Methods, Models and Applications

Font Size: 
Smooth Nonparametric Heterogeneity Estimation with an Application to Meta-Analysis
Dankmar Boehning

Last modified: 2013-06-14

Abstract


Frequently, the question of heterogeneity for a parametric distributional family, such as the normal family, arises. it has been common to approach this question by either modelling the mixing distribution induced by heterogeneity through a parametric mixing distribution such as the normal or through a discrete, nonparametric mixing distribution. Both approaches have been discussed as alternatives in the sense ’either or’. We suggest to follow a combined smooth nonparametric approach in which the mixing distribution follow a mixture of parametric distributions itself. The approach is illustrated with a meta-analytic application on the effect of smoking on liver cancer. The smooth nonparametric approach will allow combining the target of a latent, continuous concept (which is thought traditionally only possible via a parametric approach) with the flexibility of the excellent fit of the nonparametric, discrete modelling approach. It will allow also investigating these questions by allowing the formulation of appropriate hypothesis.

Full Text: PDF