Replication Package for "A Practical Method to Reduce Privacy Loss when Disclosing Statistics Based on Small Samples"

Authors: Raj Chetty and John Friedman

The article "A Practical Method to Reduce Privacy Loss when Disclosing Statistics Based on Small Samples" appears in Journal of Privacy and Confidentality 2019 Vol 9 Issue 2 (doi.org/10.29012/jpc.716).

This copy of the code was forked from https://github.com/Opportunitylab/Differential-Privacy. No modifications were made to the code.

Figures Replication

This folder contains the Stata code to reproduce figures 1 to 3 in “[A Practical Method to Reduce Privacy Loss when Disclosing Statistics Based on Small Samples]](https://doi.org/10.29012/jpc.716)” by Chetty and Friedman (2019).

To run the code, please set the file paths in in the do-file all_figures_paper.do to the files contained in this replication package (tract_covariates.dta and tract_outcomes_early_dta_dp.dta).

Implementation of Noise-Infusion Algorithm: Example Code

The sub-folder example_code_implementation_guide contains two examples of how to implement the noise-infusion algorithm outlined in the Appendix A of “A Practical Method to Reduce Privacy Loss when Disclosing Statistics Based on Small Samples” by Chetty and Friedman (2019). The three files in the sub-folder contain the following:

Example 1 – Simple regression coefficient as statistic of interest

Implementation_guide_simple_reg.do

Stata do-file showing a step-by-step example of how to apply the noise-infusion algorithm to publicly release the estimated coefficients of a simple regression estimate of child income rank on parent income rank in each of the cells of a simulated dataset.

Example 2 – Predicted value of Y at a certain value of X as statistic of interest

Implementation_guide_p_25_prediction.do

Stata do-file showing a step-by-step example of how to apply the noise-infusion algorithm to publicly release the predicted value of child income rank at the 25th percentile of the parental income distribution in each of the cells of a simulated dataset.

Dataset

private_data_by_cells.dta

Simulated dataset in Stata format containing information on child income rank and parent income rank of 10,000 fictitious individuals grouped in 111 cells. Variable names: parent_rank, kid_rank, cell.