Non-parametric maximum likelihood estimation of mixture models
Researchers at the Frisch Centre have developed a computer program designed for non-parametric maximum likelihood estimation of multivariate mixture models; i.e. statistical models where the subjects under analysis are characterized by vectors of unobserved heterogeneity with an unknown statistical distribution. The program has been used in a number of scientific publications. Due to space limitations, it has not been possible to provide a detailed description of the program in each of these publications. Below, we provide a link to a file containing a detailed documentation of the workings of the program and its optimization algorithms. We also provide a link to the list of scientific publications where the program has been used. For some of the publications, this list contains a link to additional information that is particular for the publication in question.
The program is tailored to handle the combination of a large dataset (with millions of observations) and a flexible model specification (with thousands of parameters to estimate), and it runs on a cluster of computers. It is particularly suitable for life course / event history / hazard rate analyses, based on large-scale administrative register data.
The program has been written by Simen Gaure in FORTRAN.
We emphasize that the program is under construction, and that new features may be added in the future, both to enhance the generality and usefulness of the program and to improve its computational efficiency. The documentation file will be updated accordingly.
We hope that interested researchers may find the detailed documentation of our program useful in their own research. We are also thankful for comments and ideas regarding the workings of the program as well as ideas for its future development.
Updated 12.01.2015, email@example.com