Fig. 8

The effect of the sample size on utility loss and disclosure risk of non-stochastic anonymised data. A The dataset-specific utility loss as measured by the summary statistic U of propensity scores. B The variables-specific utility loss as measured by the Euclidean distance-based metric. C The analysis-specific information loss as measured by the standardised difference of regression model coefficients. D The robust Mahalanobis distance-based disclosure risks. Each point and error bar in the four panels indicates the mean plus minus one standard deviation of the metrics across 100 simulations with constant k = 5