Skip to contents

Linked Dataset (n = 1000) for a Simple Illustration

Source: R/LD1000.R
LD1000.Rd

The LD1000 data set contains a simulated cohort of 1,000 individuals designed to replicate a secondary analysis scenario where records from disparate data sources must be linked prior to performing regression analyses.

Usage

data(LD1000)

Format

A data frame with 1,000 rows and 10 variables.

Details

To establish this cohort, demographic covariates were generated in one file and a binary clinical outcome was generated in a separate file. The two files were linked probabilistically using the Fellegi-Sunter model without unique identifiers, and 1:1 matching was enforced via a linear sum assignment problem (LSAP). This record linkage results in a mismatch rate of approximately 20%.

LD1000 provides variables of primary interest for secondary analysis alongside examples of linkage paradata.

  • Disease_Status: Binary clinical outcome variable (0 or 1).

  • BMI: Body Mass Index of the individual.

  • Age: Age of the individual.

  • Treatment: Binary indicator of treatment assignment (0 or 1).

  • fl_prob: The posterior match probability assigned to the linked record pair by the probabilistic linkage algorithm (fastLink).

  • safe_match: A logical indicator (TRUE / FALSE) denoting whether the linked record pair had perfect string agreement on both first and last names and exact agreement on birth year.

  • bm: Birth month (1-12) used as a blocking variable during the probabilistic record linkage process and treated as blocks for the clerical audit.

  • block_mrate: Estimated block-level mismatch rate. This variable was generated by simulating a clerical review (audit) of 20 randomly sampled records per birth-month block. Rates were adjusted as in Chambers (2009) to prevent values of exactly 1 or 0.

  • True_Disease_Status: The unobservable true clinical outcome variable (0 or 1) that corresponds to the individual's actual record.

  • is_match: The latent true match status (TRUE = correct match, FALSE = mismatch). Note: This variable is provided for benchmarking and is unobserved in real-world secondary analysis settings.

References

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

Sariyar, M., & Borg, A. (2026). RecordLinkage: Record Linkage Functions for Linking and Deduplicating Data Sets. R package version 0.4-12.6.

Enamorado, T., Fifield, B., & Imai, K. (2023). fastLink: Fast Probabilistic Record Linkage with Missing Data. R package version 0.6.1.