Latent class multivariate probit and latent trait models for evaluating test accuracy without a gold standard: A simulation study

Enzo Cerullo, Sean Pinkney, Alex J. Sutton, Tim Lucas, Nicola J. Cooper, Hayley E. Jones · 2025

Read Latent class multivariate probit and latent trait models for evaluating test accuracy without a gold standard: A simulation study by Enzo Cerullo, Sean Pinkney, Alex J. Sutton, Tim Lucas, Nicola J. Cooper, Hayley E. Jones (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-18489.txt.

Introduction

When no perfect reference test exists, judging new medical tests gets tricky. Enzo Cerullo and five co-authors compare two statistical models for that job in this 2025 simulation study. Both models assume a hidden true state and let test results depend on each other. One, the latent trait model, forces a simple link structure. The other allows a full web of links.

The team simulated data under five patterns, from no links to rich, varied ones. They also tried varied prior settings, including new priors that keep correlation guesses readable. Their software fits these slow models at graphics-chip-like speed. The full-web model won overall. The simple trait model did fine on its own kind of data but failed badly on rich patterns, sometimes worse than assuming no links. High test sensitivity could mask poor link recovery. The authors advise the flexible model for real use.

Catalog record

CREATOREnzo Cerullo, Sean Pinkney, Alex J. Sutton, Tim Lucas, Nicola J. Cooper, Hayley E. Jones
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE174 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-18489

Browse the full library · About the Tower of Babel