Constrained parameters? Use Metropolis-Hastings

Constrained parameters? Use Metropolis-Hastings

by Ben Lambert

Formal & Physical Sciences Mathematics PBTProbability and statisticsPBTBBayesian inference
🎙 Ben Lambert 👥 148K 📅 May 15, 2018 ⏱ 13 min 👁 12K 📄 tutorial 🧭 2026-08-17
Available in: English (current) Français

Keywords

Metropolis-Hastingsconstrained parametersMCMCBayesian inferenceproposal distribution

Summary

This video, part of a lecture course on Bayesian statistics, addresses the challenge of sampling from a posterior distribution when parameters are constrained (e.g., a standard deviation must be positive). The presenter, Ben Lambert, first demonstrates two naive approaches: using a symmetric normal proposal in the standard Metropolis algorithm, which leads to many rejections and inefficiency, and using rejection sampling to ensure positive proposals, which introduces bias due to asymmetric proposal probabilities. He then introduces the Metropolis-Hastings algorithm, which allows asymmetric proposal distributions and corrects for the asymmetry in the acceptance ratio, thus providing an unbiased and efficient sampler. He illustrates these points with simulations in Mathematica, showing the superior performance of Metropolis-Hastings. Finally, he mentions an alternative approach: transforming the parameter (e.g., using log) to make it unbounded, while accounting for the Jacobian, allowing the use of standard Metropolis.

140 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides a clear and valuable explanation of a common issue in MCMC sampling. The argumentation is logically structured: it identifies the problem, demonstrates two flawed solutions, and then presents the correct approach. The use of simulations to visually confirm the theoretical points strengthens the argument. The presenter’s expertise is evident, and the content is accurate and well-presented.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is high: the content is based on established statistical theory and the presenter is an academic. The video does not cite specific sources, but the description links to the author’s course materials and textbook, which are credible. The title accurately reflects the content. No comments were provided for analysis.

126 words

Title / Content Match

The title accurately reflects the content: the video explains why standard Metropolis fails for constrained parameters and demonstrates the Metropolis-Hastings solution.

Quality & Reliability

8/10

The video is a clear, pedagogically sound tutorial on a specific Bayesian computational technique. The presenter is an academic (author of a textbook on Bayesian statistics) and the content aligns with established statistical theory. The simulations illustrate the concepts effectively. No sources are cited in the video itself, but the description links to the author's course materials and textbook, which are credible.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Concurring Sources

Contribution & Novelties

The video provides a clear pedagogical explanation of a specific MCMC technique, filling a gap for learners who encounter constrained parameters. It contrasts naive approaches with the correct Metropolis-Hastings method, and also mentions the transformation approach. The simulations are helpful for understanding.

Pour aller plus loin :

86 words

Radar Profile

The radar profile shows high scores in information quantity, quality, and technical level, with a slightly lower but still strong reliability score. This indicates a well-produced, informative tutorial with solid scientific grounding.

Reliability 8/10