postlink

The postlink package is dedicated to providing a unified suite of statistical tools designed to rigorously account for record linkage errors in post-linkage data analysis.

Record linkage is often error-prone, particularly when identifiers used for matching records are noisy or non-unique. Mismatches (false matches) act as a contaminant in the linked data, typically leading to attenuated estimates in downstream analysis. The postlink R package currently supports three statistical frameworks to account for potential mismatch errors during downstream regression modeling:

• Weighting: Solves adjusted estimating equations using block-specific proportions of correct links. It assumes an Exchangeable Linkage Error (ELE) model (Chambers, 2009; Chambers et al., 2023; Vo et al., 2024).
• Mixture Modeling: Uses an Expectation-Maximization (EM) algorithm and infers the latent correct match status using a two-component mixture model (Slawski et al., 2025).
• Bayesian Mixture Modeling: Performs posterior inference via data augmentation, alternating between imputing the latent match status and updating model parameters to propagate linkage uncertainty (Gutman et al., 2016).

The postlink package currently focuses on methods for secondary analysis, where the individual files that were linked are not accessible. For the primary analysis setting, when individual files are accessible, methods that perform record linkage and analysis jointly with direct propagation of uncertainty would be more suitable.

The long-term goal of postlink is to extend support for a wide array of linkage and post-linkage analysis scenarios.

Package Design

The package is built on a modular, object-oriented S3 architecture that decouples the specification of linkage error from the substantive statistical modeling. This provides a familiar, standard formula based modeling interface.

Phase 1: Adjustment Specification

First, we define the linked data and the chosen adjustment methodology using a constructor function. These constructors validate the data and return a lightweight S3 adjustment object.

• adjELE(): Specifies the Exchangeable Linkage Error (ELE) model, using known or audited mismatch rates.
• adjMixture(): Specifies a frequentist mixture model approach that treats match status as a latent variable, estimating linkage error rates directly from data (e.g., using an overall mismatch rate, safe matches, predictors of match status, or predicted correct match probabilities). If no record linkage information is available, a constant mismatch rate is assumed.
• adjMixBayes(): Specifies a Bayesian mixture model approach, with default or informative priors, enabling parameter estimation and multiple imputation of latent match statuses using Stan.

Phase 2: Estimation & Inference

The adjustment object is subsequently passed to a standard modeling wrapper, integrating the linkage error correction into the familiar R modeling syntax:

• plglm() for generalized linear models (linear, logistic, Poisson, Gamma).
• plcoxph() for Cox proportional hazards regression.
• plsurvreg() for parametric survival models.
• plctable() for contingency table analysis.

Estimation and inference supported for each type of adjustment object vary. Please refer to the adj* or pl* documentation for models currently supported.

Standard R workflows (e.g., summary(), predict(), vcov(), and confint()) can be used for display and processing of results. These methods specially are derived to consider the additional steps introduced by the linkage error adjustment.

Note: While the two-phase workflow is recommended for standard analyses, the package’s architecture isolates the core logic of each method into independent internal routines. If preferred, the underlying computational functions can be used directly by supplying pre-computed design matrices and response vectors (e.g., coxphELE(), glmMixture(), survregMixBayes())

Installation

The development version of postlink can be installed from GitHub or locally:

# Using devtools:
# install.packages("devtools")
devtools::install_github("postlink-group/postlink")

# Or, using pak:
# install.packages("pak")
pak::pkg_install("postlink-group/postlink")

System Requirements

Because postlink includes Bayesian mixture models powered by rstan, installing the development version of the package from source requires a working C++ toolchain to compile the underlying models.

Depending on your operating system, please ensure you have the following installed:

• Windows: Install the version of Rtools that matches your current R version.
• macOS: Install the Xcode Command Line Tools by opening your terminal and running: xcode-select --install
• Linux: Ensure you have the standard C++ compiler and R development tools installed (e.g., sudo apt-get install r-base-dev on Ubuntu/Debian).

(Note: Once postlink is officially released on CRAN, Windows and macOS users will be able to download pre-compiled binaries, bypassing this requirement).

Quick Start

Below is a brief example illustrating the typical workflow using postlink.

We analyze the relationship between age at death and year of birth linked using historical records from the LIFE-M project. The linked dataset contains a mix of hand-linked records (assumed correct) and purely machine-linked records subject to an approximate 5% mismatch rate. Instead of fitting the standard glm model ignoring the mismatch errors, we use the postlink to adjust for potential mismatches using the entire linked dataset.

library(postlink)

# Load the LIFE-M demo dataset
data(lifem)

# Phase 1: Adjustment Specification
# We model the correct match indicator via logistic regression using 
# name commonness scores (commf, comml) and a 5% expected mismatch rate.
adj_object <- adjMixture(
  linked.data = lifem,
  m.formula = ~ commf + comml,
  m.rate = 0.05,
  safe.matches = hndlnk
)

# Check specified adjustment
print(adj_object)

# Phase 2: Estimation & Inference
# Fit a Gaussian regression model utilizing a cubic polynomial for year of birth.
fit <- plglm(
  age_at_death ~ poly(unit_yob, 3, raw = TRUE),
  family = "gaussian",
  adjustment = adj_object
)

# View model results
summary(fit)
confint(fit)

References

Chambers, R. (2009). Regression analysis of probability-linked data. , 4, 1-15.

Chambers, R. L., Fabrizi, E., Ranalli, M. G., Salvati, N., & Wang, S. (2023). Robust regression using probabilistically linked data. , 15(2), e1596.

Gutman, R., Sammartino, C., Green, T., & Montague, B. (2016). Error adjustments for file linking methods using encrypted unique client identifier (eUCI) with application to recently released prisoners who are HIV+. , 35(1), 115–129.

Slawski, M., West, B. T., Bukke, P., Wang, Z., Diao, G., & Ben-David, E. (2025). A general framework for regression with mismatched data based on mixture modelling. , 188(3), 896-919.

Vo, T. H., Garès, V., Zhang, L. C., Happe, A., Oger, E., Paquelet, S., & Chauvet, G. (2024). Cox regression with linked data. , 43(2), 296-314.