starling-package: starling: Probabilistic Record Linkage for Public Health...

In starling: Record Linkage for Public Health Surveillance

starling: Probabilistic Record Linkage for Public Health Surveillance

Description

**Bird note**: Starlings form murmurations — thousands of individual birds moving as one coordinated, fluid shape with no central conductor, each responding only to its nearest neighbours until a single coherent pattern emerges from the flock. That is the visual the package is named for: murmuration() takes two separate, uncoordinated sets of records and lets pairwise local comparisons resolve into one coherent linked dataset. The supporting functions — flock(), preflight(), check_medicare(), and murmuration_plot() — are the pre-flight preparation and post-flight review that make the murmuration trustworthy.

Details

## Core workflow

1. preflight() — Run a battery of pre-linkage data quality checks: completeness, duplicates, date plausibility, Medicare validity, name quality, and factor-level consistency across both datasets.

2. check_medicare() — Validate Medicare numbers using the Services Australia Modulus 10 weighted checksum (can also be called standalone or from preflight()).

3. flock() — Generate blocking variables (single-field, phonetic, or composite) to partition records before linkage.

4. murmuration() — Perform probabilistic record linkage using the Fellegi-Sunter EM algorithm (reclin2) across four linkage types: case-to-hospitalisation ("c2h"), vaccination-to-case ("v2c"), vaccination-to-hospitalisation ("v2h"), and vaccination-to-event ("v2e").

5. murmuration_plot() — Plot the linkage weight distribution to confirm the threshold is placed at the natural valley between match and non-match clusters before finalising.

## Threshold guidance

See murmuration_plot for a discussion of threshold selection. The default of 12 in murmuration() is a conservative starting point; 17 is a reasonable default for Australian linkage with Medicare + 2 names + DOB. Always plot the distribution first.

Author(s)

Maintainer: Nicolas Smoll [e-mail] (ORCID)

Authors:

• Nicolas Smoll [e-mail] (ORCID)

See Also

Useful links:

• https://github.com/nrsmoll/starling

• Report bugs at https://github.com/nrsmoll/starling/issues

starling documentation built on July 10, 2026, 9:07 a.m.