Probability and Quantile Estimation from Individually Micro-Aggregated Data

Micro-aggregation is a frequently used strategy to anonymize data before they are released to the scientific public. A sample of a continuous random variable is individually micro-aggregated by first sorting and grouping the data into groups of equal size and then replacing the values of the variable in each group by their group mean. In a similar way, data with more than one variable can be anonymized by individual micro-aggregation. Data thus distorted may still be used for statistical analysis. We show that if probabilities and quantiles are estimated in the usual way by computing relative frequencies and sample quantiles, respectively, these estimates are consistent and asymptotically normal under mild conditions.