How can you ensure privacy in a dyadic data set (e.g., data on romantic couples)?

Question

Some data sets are based on dyadic data, such as data on romantic couples (e.g., measuring daily relationship satisfaction in both partners), perceptions of mentor-mentee dyads, or parent-child dyads.

Such data sets often can be properly anonymized, so that nobody can reidentify a participant except one potential attacker: The partner who also participated in the study. (Record/remember your own answer --> identify your own data set --> get the data of your partner based on the dyad ID which is necessary to link both partners)

How can open data practices and confidentiality be handled with dyadic data?

Answer 1

These are our current lab practices:

We have to distinguish operational privacy (during data collection, when you still have your pseudonym lists), and the "final" privacy when we release a data set.

During data collection, the study PI in our studies always has the assignment list and could in principle re-identify participants. The pseudonym list should be stored in a secure place, separately from the other primary data, ideally password protected.

Upon publication of the data set, we delete the pseudonym list. After that (making sure that we have no critical variables in the data set, or unique combinations, etc.), the data set is anonymous except for one person: If a participant records or remembers his or her own answers, the person can re-identify his or her partner. We consider this risk to be large enough that we release dyadic data sets only as scientific use files (German Wikipedia). This means, access to these data sets is restricted to persons with an academic affiliation, and access is controlled via an independent public institution (i.e., not by the original data provider). Data re-users sign a contract where they agree not to try to re-identify certain persons and only use the data for scientific purposes.

Here is an example from a recent study: https://www.psychdata.de/index.php?main=search&sub=browse&id=zrce16dy99&lang=eng