Evaluating tools for capture-recapture model selection to estimate the size of hidden populations: it works in practice, but does it work in theory?

Wesson, Paul S.; Jewell, Nicholas P.; McFarland, Willi; Glymour, M. Maria

doi:10.1016/j.annepidem.2022.10.010

Cited by 4 publications

(4 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The R package shinyrecap ( 24 ) was used to implement the Bayesian latent class model, which seeks to meet the identifiability constraint of list independence by conditioning on latent classes based on observed capture histories. Both the decomposable graph approach and Bayesian latent class models have shown less bias than conventional log-linear models in previous simulation studies ( 10 , 11 ).…”

Section: Methodsmentioning

confidence: 80%

“…The poor performance from the log-linear models was consistent with long-standing criticism that this approach cannot effectively model complex list dependencies, especially in the presence of sparce cells ( 10 – 12 ). More unexpected was the underperformance of the Bayesian latent class and decomposable graph approach models, both of which performed favorably in recent simulation studies ( 10 , 11 ).…”

Section: Discussionmentioning

confidence: 96%

“…The R package Rcapture ( 21 ) was used to fit the log-linear regression models, adjusting for potential list dependencies. The R package SparseMSE ( 22 ) implements SparseMSE, a model designed to correct the bias resulting from small or no overlap between lists ( 11 , 13 ). The R package DGA ( 23 ) was used to implement the decomposable graph approach.…”

Section: Methodsmentioning

confidence: 99%

“…Furthermore, model selection based on the information criterion can be extremely sensitive to cell sparsity (i.e., few people observed in cells corresponding to different capture patterns). Because coefficients in the model are jointly determined, searching the extreme regions of the parameter space to maximize the likelihood of probabilities for the sparse cells also impacts predictions for other cells ( 11 , 13 ). This sometimes results in implausibly large size estimates and a deceptively small information criterion, making it the (statistically) preferred model.…”

Section: Abbreviationsmentioning

confidence: 99%

See 3 more Smart Citations

Evaluating a Targeted Minimum Loss-Based Estimator for Capture-Recapture Analysis: An Application to HIV Surveillance in San Francisco, California

Wesson,

Das,

Chen

et al. 2023

American Journal of Epidemiology

Self Cite

View full text Add to dashboard Cite

Capture-recapture is a common tool in epidemiology to estimate the size of “hidden” populations and correct the under-ascertainment of cases, based on incomplete and overlapping lists of the target population. Log-linear models are often used to estimate the population size yet may produce implausible and unreliable estimates due to model misspecification and small cell sizes. A novel Targeted Minimum Loss-based Estimation (TMLE) model developed for capture-recapture makes several notable improvements to conventional modeling: “targeting” the parameter of interest, flexibly fitting the data to alternative functional forms, and limiting bias from small cell sizes. Using simulations and empirical data from the San Francisco Department of Public Health’s HIV surveillance registry, we evaluated the performance of the TMLE model and compared results to other common models. Based on 2,584 people observed on three lists reporting to the surveillance registry, the TMLE model estimated the number of San Francisco residents living with HIV as of 12/31/2019 to be 13,523 (95% CI: 12,222 – 14,824). This estimate, compared to a “ground truth” of 12,507, was the most accurate and precise of all models examined. The TMLE model is a significant advancement in capture-recapture studies, leveraging modern statistical methods to improve the estimation of hidden populations.

show abstract

Section: Methodsmentioning

confidence: 80%