Methods for Multiple Systems Estimation
Methods for multiple systems estimation using
Poisson log-linear models, with particular attention to sparse capture
data, model selection and its associated uncertainty, and Bayesian
thresholding of interaction effects. The package implements and extends methodology
from Silverman (2020) ,
Chan, Silverman and Vincent (2021)
, and Silverman, Chan and Vincent
(2024) .
MultipleSystemsEstimation
Population size estimation from multiple incomplete lists, with
particular attention to sparse data and model-selection uncertainty.
The MultipleSystemsEstimation R package implements Poisson log-linear
methods for multiple systems estimation of population size. Working from
a table of counts of cases falling on a number of lists and their
possible overlaps, the aim is to estimate the size of the dark figure
— individuals appearing on none of the lists.
Among other developments, the package addresses two important
challenges: sparse capture data, including combinations of lists with no
observed overlap, and the proper assessment of the uncertainty arising
from the selection of a model from a large range of possibilities.
Key features
- A range of alternative estimation methods: BIC-based model
selection, stepwise selection, Bayesian thresholding and fixed-model
estimation.
- Model-selection uncertainty: Bootstrap confidence intervals that
account for model selection, rather than conditioning on the selected
model.
- Sparse data: Methods for dealing properly with zero capture counts
and for testing the existence and identifiability of estimates.
- Hierarchical models: Tools for generating, exploring and comparing
hierarchical log-linear models. The implementation goes beyond the
previously published papers in extending all methods to handle
higher-order interactions.
- Consistent interface: A common structure for estimates,
uncertainty intervals, fitted models and diagnostic information.
Documentation
The User
Guide
is the recommended starting point. It explains the estimation methods,
their application and the interpretation of results.
Five companion vignettes provide further details of the methods and
their implementation.
References
- Chan, L., Silverman, B. W. and Vincent, K. (2021). Multiple systems
estimation for sparse capture data: inferential challenges When there
are nonoverlapping lists. Journal of the American Statistical
Association, 116(535), 1297–1306.
https://doi.org/10.1080/01621459.2019.1708748
- Silverman, B. W. (2020). Multiple-systems analysis for the
quantification of modern slavery: classical and Bayesian approaches.
Journal of the Royal Statistical Society: Series A (Statistics in
Society), 183(3), 691–736.
doi:10.1111/rssa.12505..
- Silverman, B. W., Chan, L. and Vincent, K. (2024). Bootstrapping
multiple systems estimates to account for model selection. Statistics
and Computing, 34, article 44.
https://doi.org/10.1007/s11222-023-10346-9