A Kiefer-Wolfowitz NPMLE procedure for estimation of a Gaussian model with independent mean and variance prior components with weighted longitudinal data. This version iterates back and forth from Gamma and Gaussian forms of the likelihood.

1 |

`y` |
A vector of observations |

`id` |
A strata indicator vector indicating grouping of y |

`w` |
A vector of weights corresponding to y |

`u` |
A vector of bin boundaries for the mean effects |

`v` |
A vector of bin boundaries for the variance effects |

`eps` |
Convergence tolerance for iterations |

`maxit` |
A limit on the number of allowed iterations |

`...` |
optional parameters to be passed to KWDual to control optimization |

A list consisting of the following components:

`u` |
midpoints of the mean bin boundaries |

`fu` |
the function values of the mixing density of the means |

`v` |
midpoints of the variance bin boundaries |

`fv` |
the function values of the mixing density of the variances. |

`logLik` |
vector of log likelihood values for each iteration |

`du` |
Bayes rule estimate of the mixing density means. |

`dv` |
Bayes rule estimate of the mixing density variances. |

`status` |
Mosek convergence status for each iteration |

J. Gu and R. Koenker

Gu, J. and R. Koenker (2015) Empirical Bayesball Remixed, preprint

WGLVmix for a more general bivariate mixing distribution version and WTLVmix for an alternative estimator exploiting a Student/Gamma decomposition

Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.

All documentation is copyright its authors; we didn't write any of that.