The Confidence Interval Coverage (CIC): A Specific Utility Metric for Synthetic Data

Abstract The Confidence Interval Coverage (CIC) metric is proposed to assess the validity—and hence the specific utility—of confidence intervals computed from synthetic datasets. For a population parameter θ, the CIC estimates the coverage of synthetic data confidence intervals, by repeatedly computing confidence intervals for θ and recording the proportion of times they would be expected to enclose the true value. The efficacy of the CIC is demonstrated via both a simulation study and an empirical illustration using a teaching dataset from the England and Wales 2021 Census. The CIC is compared to the Confidence Interval Overlap (CIO) metric (Karr et al. 2006), a well-established metric for assessing specific utility.