https://github.com/computorg/published-202311-delattre-fim

Computing an empirical Fisher information matrix estimate in latent variable models through stochastic approximation
https://github.com/computorg/published-202311-delattre-fim

Last synced: about 10 hours ago
JSON representation

Computing an empirical Fisher information matrix estimate in latent variable models through stochastic approximation

• Host: GitHub
• URL: https://github.com/computorg/published-202311-delattre-fim
• Owner: computorg
• License: cc-by-4.0
• Created: 2023-11-16T09:30:58.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2025-09-12T13:39:41.000Z (10 months ago)
• Last Synced: 2025-09-12T16:04:18.436Z (10 months ago)
• Language: R
• Homepage: http://computo-journal.org/published-202311-delattre-fim/
• Size: 5.87 MB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Computing an empirical Fisher information matrix estimate in latent variable models through stochastic approximation

Maud Delattre, Estelle Kuhn

2023-11-21

### Citation

Maud Delattre and Estelle Kuhn (November 2023). Computing an empirical Fisher information matrix estimate in latent variable models through stochastic approximation. Computo.

### Badges

[![build and

publish](https://github.com/computorg/published-202311-delattre-fim/actions/workflows/build.yml/badge.svg)](https://github.com/computorg/published-202311-delattre-fim/actions/workflows/build.yml)

[![reviews](https://img.shields.io/badge/review-report-blue)](https://github.com/computorg/published-202311-delattre-fim/issues?q=is%3Aopen+is%3Aissue+label%3Areview)

[![SWH](https://archive.softwareheritage.org/badge/origin/https://github.com/computorg/published-202311-delattre-fim)](https://archive.softwareheritage.org/browse/origin/?origin_url=https://github.com/computorg/published-202311-delattre-fim)

[![DOI:10.57750/r5gx-jk62](https://img.shields.io/badge/DOI-10.57750%2Fr5gx--jk62-034E79.svg)](https://doi.org/10.57750/r5gx-jk62)

[![Creative Commons

License](https://i.creativecommons.org/l/by/4.0/80x15.png)](http://creativecommons.org/licenses/by/4.0/)

### Authors’ affiliations

- Maud Delattre (Université Paris-Saclay, INRAE, MaIAGE, 78350, Jouy-en-Josas, France)

- Estelle Kuhn (Université Paris-Saclay, INRAE, MaIAGE, 78350, Jouy-en-Josas, France)

### Abstract

The Fisher information matrix (FIM) is a key quantity in statistics.

However its exact computation is often not trivial. In particular in

many latent variable models, it is intricated due to the presence of

unobserved variables. Several methods have been proposed to approximate

the FIM when it can not be evaluated analytically. Different estimates

have been considered, in particular moment estimates. However some of

them require to compute second derivatives of the complete data

log-likelihood which leads to some disadvantages. In this paper, we

focus on the empirical Fisher information matrix defined as an empirical

estimate of the covariance matrix of the score, which only requires to

compute the first derivatives of the log-likelihood. Our contribution

consists in presenting a new numerical method to evaluate this empirical

Fisher information matrix in latent variable model when the proposed

estimate can not be directly analytically evaluated. We propose a

stochastic approximation estimation algorithm to compute this estimate

as a by-product of the parameter estimate. We evaluate the finite sample

size properties of the proposed estimate and the convergence properties

of the estimation algorithm through simulation studies.