Review of revision of
Confidentiality and Differential Privacy in the Dissemination of Frequency Tables
Yosef Rinott, Christine M. O’Keefe, Natalie Shlomo and Chris Skinner

SUMMARY: The authors have addressed most of my concerns. I have some very minor comments/typo corrections etc.


* (p2) "This paper focuses on frequency tables, which is the most common form" --> "are the most common form

* (p2) "individuals in $D$, see O’Keefe and Chipperfield (2013)." --> "individuals in $D$ (O’Keefe and Chipperfield, 2013)."  (use \citep)

* (p2) "contestable assumptions about an intruders prior knowledge" --> "an intruder's prior knowledge"

* (p3) "see for example the recent book, Dwork and Roth (2014), and its references" --> "see for example the recent monograph by Dwork and Roth (2014) and its references"

* (p3) Ungrammatical sentence: "The idea was introduced in a mathematically rigorous framework with the potential for wide application and, by employing a ‘worst case’ approach, avoids strong assumptions about which variables are sensitive to disclosure, and intruders’ prior knowledge and attack scenarios, leading to a well-defined quantification of the confidentiality protection guarantee."

* (p3) "This worst case approach may be deemed overprotective of confidentiality, for example" --> "confidentiality; for example"

* (p3) In the discussion on pre-tabular vs. post-tabular methods it might be worth mentioning that this distinction is described in much of the DP literature as input vs. output perturbation.

* (p4) "methods could also be used, see Drechsler (2011)" --> "methods could also be used (Drechsler , 2011)"

* (p7) The authors should decide if they wish to use "differential privacy" or "Differential Privacy"

* (p8) "worst case" --> "worst-case"

* (p10) "The likelihood ratio could alternatively be viewed as a posterior odds ratio, or Bayes factor, from a Bayesian perspective." --> perhaps a citation?

* (p11) "and inference on whether and where the target is in $D$ is impossible." This is incorrect. Small $\epsilon$ simply means the inference is more difficult. See Wasserman and Zhou (2010).

* (p12) Some comments on how small $\delta$ can "practically" be would be useful.

* (p12) Is the proof needed?

* It might be nice to have a summary of the implications of the different structural constraints in Section 4 at the end before moving to Section 5.

* (p30) sentence before Section 7, some extra space around the comma.

* (p31) `spending'