Bayesian nonparametric disclosure risk assessment

Any decision about the release of microdata for public use is supported by estimation measures disclosure risk, most popular being number τ1 sample uniques that are also population uniques. In such a context, parametric and nonparametric partition-based models have been shown to have: i) strength leading estimators with desirable features, including ease implementation, computational efficiency scalability massive data; ii) weakness producing underestimates in realistic scenarios, underestimation getting worse as tail behaviour empirical distribution gets heavier. To fix this phenomenon, we propose Bayesian model can be tuned microdata. Our relies on Pitman–Yor process prior, it leads novel estimator all features that, addition, allows reduce tuning “discount” parameter. We show effectiveness our through its application synthetic data real data.