Balancing Inferential Integrity and Disclosure Risk Via Model Targeted Masking and Multiple Imputation
Bibliographic record
Abstract
There is a growing expectation that data collected by government-funded studies should be openly available to ensure research reproducibility, and so is the concern on data-privacy. A strategy to protect individuals' identity is to release multiply imputed (MI) synthetic datasets with masked sensitivity values (Rubin, 1993). However, information loss or incorrectly specified imputation models can weaken or invalidate the inferences obtained from the MI-datasets. Studying a restricted-use Canadian Scleroderma Research Group (CSRG) dataset, the authors investigate the use of a new masking framework with a data-augmentation (DA) component and a tuning mechanism that balances between protecting identity-disclosure and preserving data-utility. They found, respectively in a work-disability and an interstitial lung disease study, using this DA-MI strategy reached 0% identity disclosure-risk, preserved all inferential conclusions, and on average produced 98.5% and 95.5% confidence intervals (CI) overlaps when compared to the 95% CIs constructed using the generic CSGR dataset; the lowest CI-overlap value is 91%. In contrast, the same is not true for the currently used methods; with the CI-overlap values ranging from 73.9% to 91.8% and the lowest value being 28.1%. These findings indicate that the DA-MI masking framework facilitates sharing of useful research data while protecting participants' identities.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.147 | 0.364 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.002 | 0.003 |
| Science and technology studies | 0.002 | 0.005 |
| Scholarly communication | 0.006 | 0.010 |
| Open science | 0.005 | 0.012 |
| Research integrity | 0.003 | 0.005 |
| Insufficient payload (model declined to judge) | 0.001 | 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 source (direct Gemma or distilled Codex), 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".