But by adding random data, differential privacy destroys the ability to apply the usual statistical analysis techniques to the dataset.