Economics, Econometrics and Finance
4 years ago

Estimating Marginal Likelihood: Complete-data Bias Leads to Wrong Model Selection

2 views
RePEc: Research Papers in Economics
Pitfalls of Estimating the Marginal Likelihood Using the Modified Harmonic Mean
Joshua C. C. Chan, Angelia L. Grant

Paper Summary

The modified harmonic mean method is commonly used to estimate the marginal likelihood in statistics. Two versions of this method were tested using US and UK inflation data. The version based on observed data worked well, while the one based on complete data had a bias and often chose the wrong model.

Related papers

Theme background image
Mathematics
8 years ago

New method predicts ice-hockey goals with precision using relevance weighting.

Multicollinearity
Statistical Computing
Missing Data
Theme background image
Mathematics
4 years ago

New estimator reduces bias in complex models, simplifying statistical analysis.

Regularization Methods
Bayesian Monte Carlo
Risk Modeling
Theme background image
Mathematics
4 years ago

New method reveals true impact of World Trade Organization on environment

Causal Inference
Regularization Methods
Quality of Life