This repo contains the code used to produce the data, figures, and tables for the paper "Data Anonymization for Open Science: A Case Study", by Paul Francis, Gregor Jurak, Bojan Leskosek, Thierry Meurers, Karen Otte, and Fabian Praßer.
The paper takes the original data from a base study, anonymized it using three different anonymization tools (ARX 2-anonymity, SDV CTGAN, and SynDiffix), and then determines whether the scientific conclusions of the original paper still hold when the anonymized data is used.
The base paper for this study is "Associations of mode and distance of commuting to school with cardiorespiratory fitness in Slovenian school children: a nationwide cross-sectional study" by Gregor Jurak, Maroje Soric, Vedrana Sember, Sasa Djuric, Gregor Starc, Marjeta Kovac, and Bojan Leskosek.
A copy of the base study is at papers/commute.pdf. It can be found online at:
https://bmcpublichealth.biomedcentral.com/articles/10.1186/s12889-021-10326-6
The original data is not publicly available due to privacy reasons. Please contact Bojan Leskosek at Bojan.Leskosek@fsp.uni-lj.si if you are interested in the original data. (Note this is not a promise to give you the data.)
CommDataSyn.csv is a synthetic version of the original data. It can be used for testing purposes, but is not used in any evaluation.
The true original data should be placed in CommDataOrig.csv.
The directories ARX, SDV, and SynDiffix contain the code to generate the synthetic data for the respective tools. The README files in those directories contain more information.
The synthetic data for ARX and SDV can be found in the following locations respectively:
ARX/datasets/syn_dataset.csv and syn_dataset.parquet
SDV/datasets/syn_dataset.csv and syn_dataset.parquet
SynDiffix is unique in that it produces multiple datasets when it anonymizes. The datasets are written into a zip file known as a SynDiffix blob. The blob can be found at:
synDiffix/datasets/commute.sdxblob.zip
The different datasets contain different columns, and the basic idea is that the dataset with only the columns needed for any given analytic purpose are used. This dataset will have the best accuracy for the analytic task.
The R script CommCode.R does the linear regression on the data. It uses two of the SynDiffix synthetic data files (one for each direction of commute), located at:
synDiffix\datasets\sdx_toHome_target_VO2max.csv
synDiffix\datasets\sdx_toSchool_target_VO2max.csv
Because there is no R extension to read a SynDiffix blob, these two files are extracted using the script:
synDiffix/files_from_blob.py
Execute the R script CommCode.R:
Rscript CommCode.R if vscode terminal.
This generates eight files. They are named:
results/r_xxx_plot.png
results/r_xxx.json
where xxx can be one of orig, arx, sdv, and sdx. The plots are displayed in the paper.
Execute:
python getPaperValues.py
This reads in the json files produced by CommCode.R as will as the original and synthetic data files. It generates the following files:
results/paper_values.csv
results/paper_values.json
(These two files actually contain the same data, but in the two different formats.)
Execute:
python figs_and_tabs.py
This reads in paper_values.json as well as the synthetic and original data, and produces all of the remaining figures and tables used in the paper (plus some additional plots and tables not used in the paper). The output if placed in results/tables/.
The figures are placed in results/tables/figs. The tables are placed in:
results/tables/table1.tex
results/tables/table2.tex
results/tables/table3a.tex
results/tables/table3b.tex
(Note that these table numbers map to those of the original study paper.)
figs_and_tabs.py also produces the file figs_and_tabs.tex. This is a main latex file that can be compiled to produce figs_and_tabs.pdf as a convenient way of viewing all of the figures and tables.