Protecting microdata (k-anonymity)
In brief
Under "Edit data" → "Protect microdata", Statable checks how often each combination of key variables (for example state, age, gender) occurs. If a combination occurs fewer than k times, the person is easier to recognise. The result is a new version of the data; "Undo" brings back the original.
How it works
- Choose key variables. The order is the importance: what is at the top is kept longest when suppressing.
- Choose k (2, 3 or 5; default 3). The figures appear right away: how many cases are below k and, if a case weight applies, how many re-identifications in the population are expected (an approximation).
- Steps as you wish: remove direct identifiers (name, address), cap outliers at the top or bottom (percentile or fixed limit), suppress locally (for cases below k, the least important entry is cleared until every combination occurs at least k times).
- "Check" shows before and after and the log: for each step the variable and number of cases. The log can be copied and belongs in the report or documentation.
- "Create the protected version" replaces the data in the project. The steps run in this order: cap, suppress, remove columns.
Important
- k-anonymity does not protect against everything. Before release, the responsible body checks the data.
- Which variables are key variables depends on the data and on what a possible attacker knows; the organisation decides, not Statable.
- Missing entries match every category (as in sdcMicro). With a case weight in the project, the figures count with the weights (Fk).
For specialists
Calculated as in R sdcMicro: freqCalc (alpha = 1) for fk and Fk, localSuppression with importance, percentiles as quantile(type = 7).