Disclosure Risk from Interactions and Saturated Models in Remote Access

Remote access as well as remote analysis solve many problems arising from granting researchers access to sensitive data. Both allow to run analyses without actually seeing the data. Therefore none of them demand either substantively altering the data or strictly restricting the access to it. Still remote access and remote analysis bear the risk to disclose sensitive information though the actual data is not directly available. An intruder has nothing to do but to apply standard procedures in a sophisticated way to exploit certain features enabling disclosure. Even usual and unsuspicious multivariate analyses bear great potential for data snoopers. We will illustrate how an intruder could employ commonly used factor analysis to disclose sensitive variables in a data set. We will derive the approach and evaluate it using the IAB Establishment Panel. There is theoretical and empirical evidence for the high risk for violation of confidentiality from all variants of factor analysis.