Evaluating PPRL Vs Clear Text Linkage with Real-World Data
Bibliographic record
Abstract
IntroductionPrivacy-preserving Record Linkage (PPRL) is a record linkage technique that can increase the security of personal information. PPRL uses techniques of either hashing identifiers (where exact matches are required) or Blooming identifiers (where partial matches are of interest before they are provided for linkage.
 Objectives and ApproachWe use LinXmart software to evaluate performance of PPRL linkage compared to linkage using clear text identifiers. The test linkage dataset is one that is routinely linked (N=2,672,257) at our linkage centre. The population spine (N=8,440,442) includes a record for every person who has resided in British Columbia, Canada over the past 30 years. Weights were determined using LinXmart’s implementation of the Expectation Maximization (EM) algorithm. For both linkages, accepted links were the highest-weighted candidate link with a weight above the threshold suggested by EM estimation. We compare linkage rates and quality and differences in weight and threshold estimations between clear-text and PPRL linkages
 ResultsClear-text and PPRL methods resulted in 97% and 90% linkage rates, respectively. Approximately 67% of records in the linked datasets contained a nominally unique ID. Records with a unique ID linked at higher rates (>99% for both clear-text and PPRL) while the linkage rate for records missing the ID differed substantially (92% /70% for clear-text/PPRL). Comparing PPRL linkage to the clear-text linkage, we obtain F-measures of 0.99 and 0.80 for records with and without the unique ID, respectively.
 Conclusion / ImplicationsLinkage performance may be attributable to differences in comparison operators between the two methods. Bloomed fields compared with Dice coefficient allow for partial matching but may not be as sensitive as clear-text string comparisons. Numerical comparisons in PPRL are exact matches while clear-text comparisons allow for more sophisticated matching. Further refinements in PPRL are being explored to improve these results.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.015 | 0.013 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.003 | 0.010 |
| Open science | 0.014 | 0.004 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".