Skip to main navigation
Skip to main content
Nonparametric mixture models have become increasingly popular over the last 20 years. One of the reasons for this is that they provide practitioners, such as biometricians, epidemiologists or ecologists, with a natural framework for dealing with unobserved heterogeneity (eg heterogeneity that cannot be found in the form of covariate information).
Whereas conventional random effect models assume a normal distribution for the random effect, nonparametric mixture models provide a distribution-free alternative to the normal random effects model. Since the nonparametric estimator of the random effect is always discrete nonparametric mixture models provide also a basis for likelihood-based clustering. Recent applications involve the small area clustering of life expectancy growth in North-Rhine Westphalia. Current research focuses on a semiparametric alternative for mixing distribution with applications to meta-analysis.