DataTags: Sharing Privacy Sensitive Data by Michael Bar-sinai

Latanya Sweeney, Mercè Crosas, David O'Brien, Alexandra Wood, Urs
Gasser, Steve Chong, Micah Altman, Michael Bar-Sinai and others.
 
presented by
Michael Bar-Sinai 
mbarsinai@iq.harvard.edu @michbarsinai
the programs interactively executed by
The language we develop to create them
suggested wording for questions, sub-
e harmonized decision-graphs take a long
we plan to support a special TODO type,
harmonized decision-graphs can be
ing useful views of the harmonized
Below are two views of a harmonized
(based on HTML5) and another static
was automatically generated by our
al information as well as basic wording
researcher may be different).
gulations related to IRBs, consent and
w chart of questions for an interview of a
gal experts review our approach and all
and prudent with respect to data sharing
show parts of the HIPAA harmonized
ed'Decision@Graph' CONCLUSIONS'
The DataTags project will allow researchers to publish their data, without
breaching laws or regulations. Using a simple interview process, the
system and researcher will generate a machine actionable data policy
appropriate for a dataset – its “DataTags”. This policy will later by used by
systems like Dataverse to decide how the data should be made available,
and to whom. The system will also be able to generate a customized DUA
based on these tags – a task that is currently done manually, consuming a
lot of time and resources.
The programming language for Tag Space and Harmonized decision-graph
description, and the tools related to it, will be able to describe general
harmonized decision-graphs, not just in the legal field. While easy to learn,
the language relies on Graph Theory, a robust foundation that will allow
various tools, including model checking and program/harmonized
decision-graph validations.
We believe DataTags will dramatically improve the rate of data sharing
among researchers, while maintaining legal compliance and at no cost to
the researcher or her institution. As a result, we expect more data to be
available for researchers, with fewer barriers of access.
REFERENCES'
[1] Sweeney L. Operationalizing American Jurisprudence for Data Sharing.
White Paper. 2013
[2] http://www.security.harvard.edu/research-data-security-policy
ACKNOWLEDGEMENTS''
Bob Gellman – validating the current harmonized decision-graph we have
is HIPAA compliant.
1.
Person-specific
[PrivacyTagSet ]
2.
Explicit consent
[PrivacyTagSet ]
YES
1.1.
Tags= [GREEN, store=clear, transfer=clear, auth=none, basis=not applicable, identity=not person-specific, harm=negligible]
[PrivacyTagSet (EncryptionType): Clear(AuthenticationType): None(EncryptionType): Clear(DuaAgreementMethod): None]
NO
ng?
YES
NO
nt, effort=___, harm=___]
e(EncryptionType): Clear]
shing and Credit Cards)
deo Cameras)
y)
D)
sts)
rmonized*decision*graph,*compu/ng*
AA*compliance*
r DataTags to be successful. From the data
gging process may be experienced as a
unfamiliar terms, and carrying dire legal
tly. Thus, the interview process and its user
nviting, non-intimidating and user-
legal or technical terms are used, a
ily available.
ocess depends on the answers, existing
display (such as progress bars or a check
to convey the progress made so far in a
engaged in the process is an open research
dy.
er'Interface'
Dataset&
Interview&
Handling1
Access&
Control&
DUAs,&Legal&
Policies&
Data&Tags&
Dataset&
Dataset&
Dataset& Dataset&
Shame&
Civil&
Penal>es&
Criminal&
Penal>es&
Max&
Control&
No&Risk&
Minimal&
Direct&&
Access&
Criminal&
Penal>es&
Privacy&
Preserving&
Access&
Minimal&
Privacy&
Preserving&
Minimal&
Differen>al&
Privacy&
ε=1&
ε=1/10&
ε=1/100&
ε=1/1000&
Custom&
Agreement&
Overview*of*a*dataset*ingest*workflow*in*Dataverse,*showing*the*
role*of*the*DataTags*project*in*the*process.**
Create and maintain a user-friendly system that allows researchers to
share data with confidence, knowing they comply with the laws and
regulations governing shared datasets.
We plan to achieve the above by the following efforts:
1.  Describe the space of possible data policies using orthogonal
dimensions, allowing an efficient and unambiguous description of each
policy.
2.  Harmonize American jurisprudence into a single decision-graph for
making decisions about data sharing policies applicable to a given
dataset.
3.  Create an automated interview for composing data policies, such that
the resulting policy complies with the harmonized laws and regulations
(initially assuming the researcher’s answers correctly described the
dataset).
4.  Create a set of “DataTags” – fully specified data policies (defined in
Describing a Tag Space), that are the only possible results of a tagging
process.
5.  Create a formal language for describing the data policies space and the
harmonized decision-graph, complete with a runtime engine and
inspection tools.
6.  Create an inviting, user-friendly web-based automated interview system
to allow researchers to tag their data sets, as part of the Dataverse
system.
and credible. Funding agencies and publications increasingly require data
sharing too. Sharing data while maintaining protections is usually left to
the social science researcher to do with little or no guidance or assistance.
It is no easy feat. There are about 2187 privacy laws at the state and federal
levels in the United States [1]. Additionally, some data sets are collected or
disseminated under binding contracts, data use agreements, data sharing
restrictions etc. Technologically, there is an ever-growing set of solutions
to protect data – but people outside of the data security community may not
know about them and their applicability to any legal setting is not clear.
The DataTags project aims to help social scientists share their data widely
with necessary protections. This is done by means of interactive
computation, where the researcher and the system traverse a decision
graph, creating a machine-actionable data handling policy as they go. The
system then makes guarantees that releases of the data adhere to the
associated policy.
OBJECTIVES'
executed and reasoned about.
Part of the tooling effort is creating useful views of the harmonized
decision-graph and its sub-parts. Below are two views of a harmonized
decision-graph – one interactive (based on HTML5) and another static
(based on Graphviz). The latter was automatically generated by our
interpreter. Nodes show technical information as well as basic wording
(actual wording presented to the researcher may be different).
We have already harmonized regulations related to IRBs, consent and
HIPAA and made a summary flow chart of questions for an interview of a
researcher. We have also had legal experts review our approach and all
agreed it was sufficient, proper and prudent with respect to data sharing
under HIPAA. The views below show parts of the HIPAA harmonized
decision-graph.
and to whom. The sy
based on these tags –
lot of time and resour
The programming lan
description, and the t
harmonized decision-
the language relies on
various tools, includi
decision-graph valida
We believe DataTags
among researchers, w
the researcher or her
available for research
[1] Sweeney L. Oper
White Paper. 2013
[2] http://www.securi
A
Bob Gellman – valida
is HIPAA compliant.
Harm(Level( DUA(Agreement(
Method(
Authen$ca$on( Transit( Storage(
No(Risk( Implicit( None( Clear( Clear(
Data$is$non)confiden.al$informa.on$that$can$be$stored$and$shared$freely
Minimal( Implicit( Email/OAuth( Clear( Clear(
May$have$individually$iden.fiable$informa.on$but$disclosure$would$not$cause$material$harm$
Shame( Click(Through( Password( Encrypted( Encrypted(
May$have$individually$iden.fiable$informa.on$that$if$disclosed$could$be$expected$to$damage$a$
person’s$reputa.on$or$cause$embarrassment
Civil(Penal$es( Signed( Password( Encrypted( Encrypted(
May$have$individually$iden.fiable$informa.on$that$includes$Social$Security$numbers,$financial$
informa.on,$medical$records,$and$other$individually$iden.fiable$informa.on
Criminal(
Penal$es(
Signed(
(
Two:Factor( Encrypted( Encrypted(
May$have$individually$iden.fiable$informa.on$that$could$cause$significant$harm$to$an$individual$
if$exposed,$including$serious$risk$of$criminal$liability,$psychological$harm,$loss$of$insurability$or$
employability,$or$significant$social$harm
Maximum(
Control(
Signed(
(
Two:Factor( Double(
Encrypted(
Double(
Encrypted(
Defined$as$such,$or$may$be$life)threatening$(e.g.$interviews$with$iden.fiable$gang$members).
Screenshot*of*a*ques/on*screen,*part*of*the*tagging*process.*Note**
the*current*data*tags*on*the*right,*allowing*the*user*to*see*what*
was*achieved*so*far*in*the*tagging*process.*
In order to define the tags and their possible values, we are developing a
formal language, designed to allow legal experts with little or no
programming experience to write interviews. This will enable frequent
updates to the system, a fundamental requirement since laws governing
research data may change. Below is the full tag space needed for HIPAA
compliance, and part of the code used to create it.
Representing the tag space as a graph allows us to reason about it using
Graph Theory. Under these terms, creating DataTags to represent a data
policy translates to selecting a sub-graph from the tag space graph. A single
node n is said to be fully-specified in sub-graph S, if S contains an edge
from n to one of its leafs. A Compound node c is said to be fully-specified
in sub-graph S if all its single and compound child nodes are fully
specified in sub-graph S.
A tagging process has to yield a sub-graph in which the root node (shown
in yellow) is fully-specified.
Describing'a'Tag'Space'
DataType: Standards, Effort, Harm.!
!
Standards: some of HIPAA, FERPA,!
ElectronicWiretapping,!
CommonRule.!
Effort: one of Identified, Identifiable, !
DeIdentified, Anonymous.!
Harm: one of NoRisk, Minimal, Shame, Civil,!
Criminal, MaxControl.!
!
The*tag*space*graph*needed*for*HIPAA*compliance,*and*part*of*the*code*used*to*
describe*it.*Base*graph*for*the*diagram*was*created*by*our*language*
interpreter.*
DataTags
blue
green
orange
red
crimson
None1yr
2yr
5yr
No Restriction
Research
IRB
No Product
None
Email
OAuth
Password
none
Email
Signed
HIPAA
FERPA
ElectronicWiretapping CommonRule
Identified
Reidentifiable
DeIdentified
Anonymous
NoRisk
Minimal Shame
Civil
Criminal
MaxContro
l
Anyone
NotOnline
Organization
Group
NoOne
NoMatching
NoEntities
NoPeople NoProhibition
Contact NoRestriction
Notify
PreApprove
Prohibited
Click
Signed
SignWithId
Clear
Encrypt
DoubleEncrypt
Clear
Encrypt
DoubleEncrypt
code
Handling
DataType
DUA
Storage
Transit
Authentication
Standards
Effort
Harm
TimeLimit
Sharing
Reidentify
Publication
Use
Acceptance
Approval
Compund
Simple
Aggregate
Value
1.
Person-specific
[PrivacyTagSet ]
2.
Explicit consent
[PrivacyTagSet ]
YES
1.1.
NO
2.1.
Did the consent have any restrictions on sharing?
[PrivacyTagSet ]
YES
3.
Medical Records
[PrivacyTagSet ]
NO
2.1.2.
Add DUA terms and set tags from DUA specifics
[PrivacyTagSet ]
YES
2.1.1.
Tags= [GREEN, store=clear, transfer=clear, auth=none, basis=Consent, effort=___, harm=___]
[PrivacyTagSet (EncryptionType): Clear(AuthenticationType): None(EncryptionType): Clear]
NO
YES NO
3.1.
HIPAA
[PrivacyTagSet ]
YES
3.2.
Not HIPAA
[PrivacyTagSet ]
NO
3.1.5.
Covered
[PrivacyTagSet ]
YES
4.
Arrest and Conviction Records
[PrivacyTagSet ]
NO
3.1.5.1.
Tags= [RED, store=encrypt, transfer=encrypt, auth=Approval, basis=HIPAA Business Associate, effort=identifiable, harm=criminal]
[PrivacyTagSet (EncryptionType): Encrypted(AuthenticationType): Password(EncryptionType): Encrypted(DuaAgreementMethod): Sign]
YES NO
5.
Bank and Financial Records
[PrivacyTagSet ]
YESNO
6.
Cable Television
[PrivacyTagSet ]
YESNO
7.
Computer Crime
[PrivacyTagSet ]
YESNO
8.
Credit reporting and Investigations (including ‘Credit Repair,’ ‘Credit Clinics,’ Check-Cashing and Credit Cards)
[PrivacyTagSet ]
YESNO
9.
Criminal Justice Information Systems
[PrivacyTagSet ]
YESNO
10.
Electronic Surveillance (including Wiretapping, Telephone Monitoring, and Video Cameras)
[PrivacyTagSet ]
YESNO
11.
Employment Records
[PrivacyTagSet ]
YESNO
12.
Government Information on Individuals
[PrivacyTagSet ]
YESNO
13.
Identity Theft
[PrivacyTagSet ]
YESNO
14.
Insurance Records (including use of Genetic Information)
[PrivacyTagSet ]
YESNO
15.
Library Records
[PrivacyTagSet ]
YESNO
16.
Mailing Lists (including Video rentals and Spam)
[PrivacyTagSet ]
YESNO
17.
Special Medical Records (including HIV Testing)
[PrivacyTagSet ]
YESNO
18.
Non-Electronic Visual Surveillance (also Breast-Feeding)
[PrivacyTagSet ]
YESNO
19.
Polygraphing in Employment
[PrivacyTagSet ]
YESNO
20.
Privacy Statutes/State Constitutions (including the Right to Publicity)
[PrivacyTagSet ]
YESNO
21.
Privileged Communications
[PrivacyTagSet ]
YESNO
22.
Social Security Numbers
[PrivacyTagSet ]
YESNO
23.
Student Records
[PrivacyTagSet ]
YESNO
24.
Tax Records
[PrivacyTagSet ]
YESNO
25.
Telephone Services (including Telephone Solicitation and Caller ID)
[PrivacyTagSet ]
YESNO
26.
Testing in Employment (including Urinalysis, Genetic and Blood Tests)
[PrivacyTagSet ]
YESNO
27.
Tracking Technologies
[PrivacyTagSet ]
YESNO
28.
Voter Records
[PrivacyTagSet ]
YESNO
YES NO
YES NO
Two*views*of*the*same*harmonized*decision*graph,*compu/ng*
HIPAA*compliance*
Usability is a major challenge for DataTags to be successful. From the data
publisher point of view, a data tagging process may be experienced as a
daunting chore containing many unfamiliar terms, and carrying dire legal
consequences if not done correctly. Thus, the interview process and its user
interface will be designed to be inviting, non-intimidating and user-
friendly. For example, whenever legal or technical terms are used, a
layman explanation will be readily available.
As the length of the interview process depends on the answers, existing
best practices for advancement display (such as progress bars or a check
list) cannot be used. Being able to convey the progress made so far in a
gratifying way, keeping the user engaged in the process is an open research
question which we intend to study.
User'Interface'
In*order*to*make*the*tagging*process*approachable*and*nonE
in/mida/ng,*whenever*a*technical*or*a*legal*term*is*used,*an*
explana/on*is*readily*available.*Shown*here*is*part*of*the*final*
tagging*page,*and*an*explained*technical*term.**
Dataset&
Inter
Hand
Acc
Con
DUAs,
Poli
Data
Overview*of*a*d
role
earchers to handle research data. We extend this to a 6-level scale for
cifying data policies regarding security and privacy of data. The scale is
ed on the level of harm malicious use of the data may cause. The
umns represent some of the dimensions of the data policy space.
the runtime and the researcher. The language we develop to create them
will support tagging statements, suggested wording for questions, sub-
routines and more. As we realize harmonized decision-graphs take a long
time to create and verify legally, we plan to support a special TODO type,
such that partially implemented harmonized decision-graphs can be
executed and reasoned about.
Part of the tooling effort is creating useful views of the harmonized
decision-graph and its sub-parts. Below are two views of a harmonized
decision-graph – one interactive (based on HTML5) and another static
(based on Graphviz). The latter was automatically generated by our
interpreter. Nodes show technical information as well as basic wording
(actual wording presented to the researcher may be different).
We have already harmonized regulations related to IRBs, consent and
HIPAA and made a summary flow chart of questions for an interview of a
researcher. We have also had legal experts review our approach and all
agreed it was sufficient, proper and prudent with respect to data sharing
under HIPAA. The views below show parts of the HIPAA harmonized
decision-graph.
breaching laws or regulations. Using a simple interview process, the
system and researcher will generate a machine actionable data policy
appropriate for a dataset – its “DataTags”. This policy will later by used by
systems like Dataverse to decide how the data should be made available,
and to whom. The system will also be able to generate a customized DUA
based on these tags – a task that is currently done manually, consuming a
lot of time and resources.
The programming language for Tag Space and Harmonized decision-graph
description, and the tools related to it, will be able to describe general
harmonized decision-graphs, not just in the legal field. While easy to learn,
the language relies on Graph Theory, a robust foundation that will allow
various tools, including model checking and program/harmonized
decision-graph validations.
We believe DataTags will dramatically improve the rate of data sharing
among researchers, while maintaining legal compliance and at no cost to
the researcher or her institution. As a result, we expect more data to be
available for researchers, with fewer barriers of access.
REFERENCES'
[1] Sweeney L. Operationalizing American Jurisprudence for Data Sharing.
White Paper. 2013
[2] http://www.security.harvard.edu/research-data-security-policy
ACKNOWLEDGEMENTS''
Bob Gellman – validating the current harmonized decision-graph we have
is HIPAA compliant.
m(Level( DUA(Agreement(
Method(
Authen$ca$on( Transit( Storage(
isk( Implicit( None( Clear( Clear(
is$non)confiden.al$informa.on$that$can$be$stored$and$shared$freely
mal( Implicit( Email/OAuth( Clear( Clear(
have$individually$iden.fiable$informa.on$but$disclosure$would$not$cause$material$harm$
me( Click(Through( Password( Encrypted( Encrypted(
have$individually$iden.fiable$informa.on$that$if$disclosed$could$be$expected$to$damage$a$
on’s$reputa.on$or$cause$embarrassment
Penal$es( Signed( Password( Encrypted( Encrypted(
have$individually$iden.fiable$informa.on$that$includes$Social$Security$numbers,$financial$
ma.on,$medical$records,$and$other$individually$iden.fiable$informa.on
inal(
l$es(
Signed(
(
Two:Factor( Encrypted( Encrypted(
have$individually$iden.fiable$informa.on$that$could$cause$significant$harm$to$an$individual$
posed,$including$serious$risk$of$criminal$liability,$psychological$harm,$loss$of$insurability$or$
oyability,$or$significant$social$harm
mum(
rol(
Signed(
(
Two:Factor( Double(
Encrypted(
Double(
Encrypted(
ned$as$such,$or$may$be$life)threatening$(e.g.$interviews$with$iden.fiable$gang$members).
order to define the tags and their possible values, we are developing a
mal language, designed to allow legal experts with little or no
gramming experience to write interviews. This will enable frequent
ates to the system, a fundamental requirement since laws governing
earch data may change. Below is the full tag space needed for HIPAA
mpliance, and part of the code used to create it.
presenting the tag space as a graph allows us to reason about it using
ph Theory. Under these terms, creating DataTags to represent a data
icy translates to selecting a sub-graph from the tag space graph. A single
e n is said to be fully-specified in sub-graph S, if S contains an edge
m n to one of its leafs. A Compound node c is said to be fully-specified
ub-graph S if all its single and compound child nodes are fully
cified in sub-graph S.
agging process has to yield a sub-graph in which the root node (shown
yellow) is fully-specified.
Describing'a'Tag'Space'
pe: Standards, Effort, Harm.!
rds: some of HIPAA, FERPA,!
ElectronicWiretapping,!
CommonRule.!
: one of Identified, Identifiable, !
DeIdentified, Anonymous.!
one of NoRisk, Minimal, Shame, Civil,!
Criminal, MaxControl.!
tag*space*graph*needed*for*HIPAA*compliance,*and*part*of*the*code*used*to*
describe*it.*Base*graph*for*the*diagram*was*created*by*our*language*
interpreter.*
DataTags
blue
green
orange
red
crimson
None1yr
2yr
5yr
No Restriction
Research
IRB
No Product
None
Email
OAuth
Password
none
Email
Signed
HIPAA
FERPA
ElectronicWiretapping CommonRule
Identified
Reidentifiable
DeIdentified
Anonymous
NoRisk
Minimal Shame
Civil
Criminal
MaxContro
l
Anyone
NotOnline
Organization
Group
NoOne
NoMatching
NoEntities
NoPeople NoProhibition
Contact NoRestriction
Notify
PreApprove
Prohibited
Click
Signed
SignWithId
Clear
Encrypt
DoubleEncrypt
Clear
Encrypt
DoubleEncrypt
code
Handling
DataType
DUA
Storage
Transit
Authentication
Standards
Effort
Harm
TimeLimit
Sharing
Reidentify
Publication
Use
Acceptance
Approval
Compund
Simple
Aggregate
Value
1.
Person-specific
[PrivacyTagSet ]
2.
Explicit consent
[PrivacyTagSet ]
YES
1.1.
NO
2.1.
Did the consent have any restrictions on sharing?
[PrivacyTagSet ]
YES
3.
Medical Records
[PrivacyTagSet ]
NO
2.1.2.
Add DUA terms and set tags from DUA specifics
[PrivacyTagSet ]
YES
2.1.1.
Tags= [GREEN, store=clear, transfer=clear, auth=none, basis=Consent, effort=___, harm=___]
[PrivacyTagSet (EncryptionType): Clear(AuthenticationType): None(EncryptionType): Clear]
NO
YES NO
3.1.
HIPAA
[PrivacyTagSet ]
YES
3.2.
Not HIPAA
[PrivacyTagSet ]
NO
3.1.5.
Covered
[PrivacyTagSet ]
YES
4.
Arrest and Conviction Records
[PrivacyTagSet ]
NO
3.1.5.1.
Tags= [RED, store=encrypt, transfer=encrypt, auth=Approval, basis=HIPAA Business Associate, effort=identifiable, harm=criminal]
[PrivacyTagSet (EncryptionType): Encrypted(AuthenticationType): Password(EncryptionType): Encrypted(DuaAgreementMethod): Sign]
YES NO
5.
Bank and Financial Records
[PrivacyTagSet ]
YESNO
6.
Cable Television
[PrivacyTagSet ]
YESNO
7.
Computer Crime
[PrivacyTagSet ]
YESNO
8.
Credit reporting and Investigations (including ‘Credit Repair,’ ‘Credit Clinics,’ Check-Cashing and Credit Cards)
[PrivacyTagSet ]
YESNO
9.
Criminal Justice Information Systems
[PrivacyTagSet ]
YESNO
10.
Electronic Surveillance (including Wiretapping, Telephone Monitoring, and Video Cameras)
[PrivacyTagSet ]
YESNO
11.
Employment Records
[PrivacyTagSet ]
YESNO
12.
Government Information on Individuals
[PrivacyTagSet ]
YESNO
13.
Identity Theft
[PrivacyTagSet ]
YESNO
14.
Insurance Records (including use of Genetic Information)
[PrivacyTagSet ]
YESNO
15.
Library Records
[PrivacyTagSet ]
YESNO
16.
Mailing Lists (including Video rentals and Spam)
[PrivacyTagSet ]
YESNO
17.
Special Medical Records (including HIV Testing)
[PrivacyTagSet ]
YESNO
18.
Non-Electronic Visual Surveillance (also Breast-Feeding)
[PrivacyTagSet ]
YESNO
19.
Polygraphing in Employment
[PrivacyTagSet ]
YESNO
20.
Privacy Statutes/State Constitutions (including the Right to Publicity)
[PrivacyTagSet ]
YESNO
21.
Privileged Communications
[PrivacyTagSet ]
YESNO
22.
Social Security Numbers
[PrivacyTagSet ]
YESNO
23.
Student Records
[PrivacyTagSet ]
YESNO
24.
Tax Records
[PrivacyTagSet ]
YESNO
25.
Telephone Services (including Telephone Solicitation and Caller ID)
[PrivacyTagSet ]
YESNO
26.
Testing in Employment (including Urinalysis, Genetic and Blood Tests)
[PrivacyTagSet ]
YESNO
27.
Tracking Technologies
[PrivacyTagSet ]
YESNO
28.
Voter Records
[PrivacyTagSet ]
YESNO
YES NO
YES NO
Two*views*of*the*same*harmonized*decision*graph,*compu/ng*
HIPAA*compliance*
Usability is a major challenge for DataTags to be successful. From the data
publisher point of view, a data tagging process may be experienced as a
daunting chore containing many unfamiliar terms, and carrying dire legal
consequences if not done correctly. Thus, the interview process and its user
interface will be designed to be inviting, non-intimidating and user-
friendly. For example, whenever legal or technical terms are used, a
layman explanation will be readily available.
As the length of the interview process depends on the answers, existing
best practices for advancement display (such as progress bars or a check
list) cannot be used. Being able to convey the progress made so far in a
gratifying way, keeping the user engaged in the process is an open research
question which we intend to study.
User'Interface'
In*order*to*make*the*tagging*process*approachable*and*nonE
in/mida/ng,*whenever*a*technical*or*a*legal*term*is*used,*an*
explana/on*is*readily*available.*Shown*here*is*part*of*the*final*
tagging*page,*and*an*explained*technical*term.**
Dataset&
Interview&
Handling1
Access&
Control&
DUAs,&Legal&
Policies&
Data&Tags&
Dataset&
Dataset&
Dataset& Dataset&
Shame&
Civil&
Penal>es&
Criminal&
Penal>es&
Max&
Control&
No&Risk&
Minimal&
Direct&&
Access&
Criminal&
Penal>es&
Privacy&
Preserving&
Access&
Minimal&
Privacy&
Preserving&
Minimal&
Differen>al&
Privacy&
ε=1&
ε=1/10&
ε=1/100&
ε=1/1000&
Custom&
Agreement&
Overview*of*a*dataset*ingest*workflow*in*Dataverse,*showing*the*
role*of*the*DataTags*project*in*the*process.**
Ben-Gurion University
of the Negev

Why Share Data?
• Transparency
• Collaboration
• Research acceleration
• Reproducibility
• Data citation
• Compliance with requirements from
sponsors and publishers

Sharing Data is Nontrivial
• Law is complex
• Thousands of privacy laws in the US alone, at federal,
state and local level, usually context-specific
• 2187 [Sweeney, 2013]
• Technology is complex
• e.g. Encryption standards change constantly, as new
vulnerabilities are found
• Specific dataset details (may be) complex

User-friendly, interactive algorithmic system
for
generating proper dataset handling policies

deterministic, does not require
 anyone except the user
(at runtime)
algorithm may prompt the user for information while running
non-intimidating, does
not assume user is a
legal or a technological
expert
how to transfer, store and disseminate a
 dataset, license and DUAs…
Human readable, machine actionable
legally and
technologically
up-to-date.
User-friendly, interactive algorithmic system
for
generating proper dataset handling policies

Describing a
Dataset Handling Policy

A 2-D Example
Double'Encrypt/
Encrypt/
Clear/
No'
Restriction/
Notify/ Pre'Approve/ Prohibited/
Publication+
Storage+

DataTag Definition
DataTags: code, basis, Handling, DataType, DUA, IP, identity, FERPA, CIPSEA.
TODO: IP.
code: one of
blue (Non-confidential information that can be stored and shared freely.),
green (Potentially identifiable but not…),
yellow (Potentially harmful personal information…),
orange (May include sensitive, identifiable personal information…),
red (Very sensitive identifiable personal information…),
crimson (Requires explicit permission for each transaction…)
.
Handling: storage, transit, authentication, auth.
Storage: one of clear, encrypt, doubleEncrypt.
Authentication: some of None, Email, OAuth, Password
- - - continued

single value
set of values
compound value
future value
DataTag Definition
DataTags: code, basis, Handling, DataType, DUA, IP, identity, FERPA, CIPSEA.
TODO: IP.
code: one of
blue (Non-confidential information that can be stored and shared freely.),
green (Potentially identifiable but not…),
yellow (Potentially harmful personal information…),
orange (May include sensitive, identifiable personal information…),
red (Very sensitive identifiable personal information…),
crimson (Requires explicit permission for each transaction…)
.
Handling: storage, transit, authentication, auth.
Storage: one of clear, encrypt, doubleEncrypt.
Authentication: some of None, Email, OAuth, Password
- - - continued

Current DataTag Definition
• 10 Compound 
DataTags, Legal, Assertions…
• 9 Sets 
Mostly referring to specific parts of a law
• 19 Single values 
TimeLimit, Sharing, Acceptance, Approval…

DataTags Legal
Code
Assertions
Handling
MedicalRecords
EducationRecords
ContractOrPolicy
GovernmentRecords
Part2
HIPAA
deidentified
veteransMedicalData
consent
scientificResearch
waiver
authorization
safeHarborDeidentified
expertDetermination
limitedDataset
businessAssociateContract
FERPA
PPRA
deidentified
directoryOptOut
directoryInfo
schoolOfficial
study
consent
audit
protected
protectedDeidentified
consent
optOutProvided
marketing
no
yes
DPPA
CIPSEA
PrivacyAct
Census
ESRA
highlyRestricted
required
stateConsentLimited
stateConsentBroad
requesterConsentLimited
requesterConsentBroad
research
exception
deidentified
identifiable
deidentified
identifiable
CensusPublished
restricted
public
blue
green
yellow
orange
red
crimson
IP (todo)
DataType
Identity
Harm
Effort
noRisk
minimal
shame
civil
criminal
maxControl
anonymous
deidentified
identifiable
identified
noPersonData
notPersonSpecific
personSpecific
Authentication
Transit
Storage
Approval
auth
Acceptance
DUA
None
Email
OAuth
Password
clear
encrypt
doubleEncrypt
clear
encrypt
doubleEncrypt
None
Email
Signed
approval
none
Click
Signed
SignWithID
Use
Publication
Reidentify
TimeLimit
Sharing
Auditing
NoRestriction
Research
IRB
NoProduct
NoRestriction
Notify
PreApprove
Prohibited
NoMatching
NoEntities
NoPeople
NoProhibition
Reidentify
Contact
none
_1yr
_2yr
_5yr
_50yr
Anyone
NotOnline
Organization
Group
NoOne
NotNeeded
Yearly
Monthly

DataTags Legal
Code
Assertions
Handling
MedicalRecords
EducationRecords
ContractOrPolicy
GovernmentRecords
Part2
HIPAA
deidentified
veteransMedicalData
consent
scientificResearch
waiver
authorization
safeHarborDeidentified
expertDetermination
limitedDataset
businessAssociateContract
FERPA
PPRA
deidentified
directoryOptOut
directoryInfo
schoolOfficial
study
consent
audit
protected
protectedDeidentified
consent
optOutProvided
marketing
no
yes
DPPA
CIPSEA
PrivacyAct
Census
ESRA
highlyRestricted
required
stateConsentLimited
stateConsentBroad
requesterConsentLimited
requesterConsentBroad
research
exception
deidentified
identifiable
deidentified
identifiable
CensusPublished
restricted
public
blue
green
yellow
orange
red
crimson
IP (todo)
DataType
Identity
Harm
Effort
noRisk
minimal
shame
civil
criminal
maxControl
anonymous
deidentified
identifiable
identified
noPersonData
notPersonSpecific
personSpecific
Authentication
Transit
Storage
Approval
auth
Acceptance
DUA
None
Email
OAuth
Password
clear
encrypt
doubleEncrypt
clear
encrypt
doubleEncrypt
None
Email
Signed
approval
none
Click
Signed
SignWithID
Use
Publication
Reidentify
TimeLimit
Sharing
Auditing
NoRestriction
Research
IRB
NoProduct
NoRestriction
Notify
PreApprove
Prohibited
NoMatching
NoEntities
NoPeople
NoProhibition
Reidentify
Contact
none
_1yr
_2yr
_5yr
_50yr
Anyone
NotOnline
Organization
Group
NoOne
NotNeeded
Yearly
Monthly
co
DUA
Reidentify
TimeLimit
Sharing
Auditing
NoRestriction
Prohibited
NoMatching
NoEntities
NoPeople
NoProhibition
Reidentify
Contact
none
_1yr
_2yr
_5yr
_50yr
Anyone
NotOnline
Organization
Group
NoOne
NotNeeded
Yearly
Monthly

Current Legal Scope
Privacy of medical records
• The Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule, 45 C.F.R. Part 160 and Subparts A and E of Part
164, which regulates the use and disclosure of protected health information held by covered entities such as health care
providers
• The substance abuse confidentiality regulations, 42 C.F.R. Part 2, which protect the confidentiality of the medical records of
patients seeking treatment for alcohol or drug abuse
Privacy of education records
• The Family Educational Rights and Privacy Act (FERPA), 20 U.S.C. § 1232g, which safeguards student records maintained by an
educational agency or institution
• The Protection of Pupil Rights Amendment (PPRA), 20 U.S.C. § 1232h, which establishes privacy-related procedures for certain
surveys, analyses, and evaluations funded by the US Department of Education
• The Education Sciences Reform Act of 2002, 20 U.S.C. § 9573, which restricts the collection, use, and dissemination of
education data in research conducted by the Institute of Education Sciences
Privacy of government records
• The Privacy Act of 1974, 5 U.S.C. § 552a, which establishes fair information practices for protecting personally identifiable
records maintained by federal agencies
• The Confidential Information Protect and Statistical Efficiency Act (CIPSEA), 44 U.S.C. § 3501 note, which protects confidential
data collected by U.S. statistical agencies
• Title 13 of the U.S. Code, which protects the confidentiality of Census Bureau data
• The Driver’s Privacy Protection Act (DPPA), 18 U.S.C. §§ 2721-2725, which restricts the disclosure of personal information from
state department of motor vehicle records Warning - we are still in beta!

Selecting the DataTags
• Many possibilities
• Active effort using DataLog/Prolog
• Supervised by Prof. Steve Chong, Harvard SEAS
• Procedural Questionnaire

Questionnaire Code
(>ppraCompliance< ask:
(text: Do the data contain any information collected from students...?)
(terms:
(Elementary or secondary school: Refers to...))
(no: (end))
)
(>ppra2< ask:
(text: Do the data contain information ... )
(terms:
(Survey or assessment: Refers to any academic or non-academic ...))
(yes:
(>ppra2a< ask:
(text: Do the data contain any personally identifiable information ...)
(terms:
(Deidentified information: Information that..)
(Biometric record: Refers to a record of one ...)
(Targeted request: Refers to a request by a ...)
)
(yes:
(set: PPRA=protected, Code=orange, Harm=civil,
Transit=encrypt, Storage=encrypt, Effort=identifiable))
(no:
(set: PPRA=protectedDeidentified, Code=green, Harm=minimal,
Transit=clear, Storage=clear, Effort=deidentified))
) continued...

ppra2a
ask
Do the data from the survey or
assessment contain any personally
identifiable information about a
student? Personally identifiable
informati...
Set
Legal=[ EducationRecords:[ PPRA:[ protectedDeidentified ] ] ]
Code=green
Assertions=[ DataType:[ Harm:minimal Effort:deidentified ] ]
Handling=[ Transit:clear Storage:clear ]
no
Set
Legal=[ EducationRecords:[ PPRA:[ protected ] ] ]
Code=orange
Assertions=[ DataType:[ Harm:civil Effort:identifiable ] ]
Handling=[ Transit:encrypt Storage:encrypt ]
yes
ppraCompliance
ask
Do the data contain any information
collected from students in an elementary
or secondary school?
ppra2
ask
Do the data contain information
collected from students in a survey or
assessment concerning any of the
following eight categories? ...
yes no
yes no
Current Questionnaire

questionnaire.flow-c1
$1
ferpaCompliance
ppraCompliance
dua
DPPA1
medicalRecordsCompliance
Gov1
govRecsCompliance
start
ask
Do the data concern living persons?
questionnaire.flow-c1/ppraCompliance
questionnaire.flow-c1/ferpaCompliance
Set
Code=green
yes
Set
Code=blue
Assertions=[ Identity:noPersonData ]
no
questionnaire.flow-c1/medicalRecordsCompliance
questionnaire.flow-c1/govRecsCompliance
todo
Arrest and Conviction Records, Bank and
Financial Records, Cable Television,
Computer Crime, Credit reporting and
Investigations [including 'Credit
Repair', 'Credit Clinics', Check-Cashing
and Credit Cards], Criminal Justice
Information Systems, Electronic
Surveillance [including Wiretapping,
Telephone Monitoring, and Video
Cameras], Employment Records, Government
Information on Individuals, Identity
Theft, Insurance Records [including use
of Genetic Information], Library
Records, Mailing Lists [including Video
rentals and Spam], Special Medical
Records [including HIV Testing],
Non-Electronic Visual Surveillance.
Breast-Feeding, Polygraphing in
Employment, Privacy Statutes/State
Constitutions [including the Right to
Publicity], Privileged Communications,
Social Security Numbers, Student
Records, Tax Records, Telephone Services
[including Telephone Solicitation and
Caller ID], Testing in Employment
[including Urinalysis, Genetic and Blood
Tests], Tracking Technologies, Voter
Records
questionnaire.flow-c1/dua
ferpa15
ask
Are the data being disclosed for the
purpose of conducting an audit or
evaluation of a federal- or
state-supported education program?
Set
Legal=[ EducationRecords:[ FERPA:[ audit ] ] ]
Code=yellow
Assertions=[ DataType:[ Harm:shame Effort:identifiable ] ]
Handling=[ Transit:encrypt ]
yes
REJECT
An educational agency or institution is
likely breaching its FERPA duties
because it is disclosing, or a third
party is re-disclosing, non-directory
PII without parental consent where no
obvious FERPA exception applies.
no
Set
Legal=[ EducationRecords:[ FERPA:[ study ] ] ]
Code=yellow
todo
DUA subroutine
Set
Legal=[ EducationRecords:[ FERPA:[ directoryInfo ] ] ]
Code=green
Assertions=[ DataType:[ Harm:minimal Effort:identifiable ] ]
todo
DUA subroutine
ferpa10
ask
Do the data include personally
information about a student includes,
but i...
Set
Legal=[ EducationRecords:[ FERPA:[ deidentified ] ] ]
Code=green
no
ferpa11
ask
Did the educational agency or
institution designate all of the
personally identifiable information in
the data as directory information?
The...
yes
Set
Legal=[ EducationRecords:[ FERPA:[ consent ] ] ]
Code=yellow
todo
Consent sub-routine
ferpa13
ask
Are the data being disclosed to a school
official or contractor with a legitimate
educational interest?
ferpa14
ask
Are the data being disclosed for the
purpose of conducting a study for, or on
behalf of, the educational agency or
institution in order to d...
no
Set
Legal=[ EducationRecords:[ FERPA:[ schoolOfficial ] ] ]
Code=yellow
yes
ferpa12
ask
Did the parents or guardians of the
students, or the students themselves if
they were adults or emancipated minors
at the time of the data c...yes
no
Set
Legal=[ EducationRecords:[ FERPA:[ directoryOptOut ] ] ]
Code=yellow
ferpa9
ask
derived from institutional records
directly related to a student? This
refers to records maint...
yes no
ferpaCompliance
ask
derived from records maintained by an
educational agency or institution, or by
a person or entity acting...
yes no
ferpa11a
ask
Did any of the students in the data, or
their parents, if the students are under
18, request to opt out of the release of
their information ...
no
yes
no
yes
no
yes
ppra6
ask
Does the school in which the information
was collected receive any funds from the
US Department of Education?
ppra7
ask
Were parents provided with notice and an
opportunity to opt out of the collection
and disclosure of the information?
yes no
ppra2a
ask
Do the data from the survey or
assessment contain any personally
informati...
Set
Legal=[ EducationRecords:[ PPRA:[ protectedDeidentified ] ] ]
Code=green
no
Set
Legal=[ EducationRecords:[ PPRA:[ protected ] ] ]
Code=orange
yes
REJECT
Please ask the party that collected the
data whether written consent from the
parents was obtained. Dataset cannot be
accepted without clarifying this issue.
Set
Legal=[ EducationRecords:[ PPRA:[ optOutProvided ] ] ]
yes
REJECT
Possible PPRA violation due to school’s
failure to provide parents with the
opportunity to opt out of the data
collection
no
Set
Legal=[ EducationRecords:[ PPRA:[ marketing ] ] ]
ppra3
ask
Was the survey or assessment funded, in
whole or in part, by the US Department
of Education?
ppra5
ask
Was any of the personal information in
the data collected from students for the
purpose of marketing or sale?
no
ppra4
ask
Was written consent obtained from the
parents of the students who participated
in the survey or assessment, or from the
students themselves ...
yes
ppraCompliance
ask
collected from students in an elementary
or secondary school?
ppra2
ask
Do the data contain information
collected from students in a survey or
assessment concerning any of the
following eight categories? ...
yes no
REJECT
Possible PPRA violation due to failure
to obtain written consent for collection
of information protected under the PPRA
yes no
Set
Legal=[ EducationRecords:[ PPRA:[ consent ] ] ]
no
yes
Not Sure
no yes
Set
Handling=[ DUA:[ TimeLimit:_1yr ] ]
duaAdditional
ask
Did the data have any restrictions on
sharing, e.g. stated in an agreement or
policy statement?
dua
todo
Data use agreements
duaTimeLimit
ask
Is there any reason why we cannot store
the data indefinitely? Limiting the time
a dataset could be held interferes with
good scienc...
ask
For how long should we keep the data?
yes
Set
Handling=[ DUA:[ TimeLimit:none ] ]
no
Set
Legal=[ ContractOrPolicy:yes ]
Set
Set
Handling=[ DUA:[ TimeLimit:none ] ]
1 year 5 yearsindefinitely
Set
50 years
Set
Legal=[ ContractOrPolicy:no ]
yes no
DPPA2
ask
Do the department of motor vehicles
records contain an individual's
photograph or image, Social Security
number, or medical or disability in...
DPPA2a
ask
Has express consent been obtained from
each person whose photograph, Social
Security number, or medical or
disability information appears in...
yes
Set
Code=yellow
no
DPPA6
ask
Was consent obtained from the
individuals whose information is
contained in the data by the requester
of the records?
DPPA7
ask
Were the data obtained for use in
research activities?
no
DPPA6a
ask
Was the obtained consent limited in any
way, such as purpose, duration, parties
that may receive the data?
yes
DPPA3
ask
Is disclosure of the personal
information required by the Driver's
Privacy Protection Act? Disclosure is
required, among other reasons, in c...
Set
Legal=[ GovernmentRecords:[ DPPA:[ required ] ] ]
yes
DPPA4
ask
Is the disclosure of these records made
in accordance with the driver's license
record disclosure law of the state from
which the records we...
no
Set
Legal=[ GovernmentRecords:[ DPPA:[ highlyRestricted ] ] ]
Code=orange
Set
Legal=[ GovernmentRecords:[ DPPA:[ research ] ] ]
DPPA5
ask
Was consent obtained from the person
about whom the information pertains by
the State DMV in response to a request
for an individual record?...
no
DPPA5a
ask
Was the obtained consent limited in any
way, such as purpose, duration, parties
that may receive the data?
yes
DPPA1
ask
Do the data contain personal information
obtained from a state department of
motor vehicles?yes no
yes
REJECT
Possible violation of the DPPA due to
disclosure of highly restricted personal
information without the consent of the
individuals in the data
no
yes
DPPA8
ask
Was the personal information obtained
under one of the other fourteen
permissible uses for the disclosure of
drivers' records, including: ...
no
Set
Legal=[ GovernmentRecords:[ DPPA:[ exception ] ] ]
Set
Legal=[ GovernmentRecords:[ DPPA:[ stateConsentLimited ] ] ]
REJECT
Possible violation of state law on
disclosure of drivers' records
Set
Legal=[ GovernmentRecords:[ DPPA:[ requesterConsentLimited ] ] ]
yes
Set
Legal=[ GovernmentRecords:[ DPPA:[ requesterConsentBroad ] ] ]
no
REJECT
possible violation of the DPPA because
the depositor has not cited one of the
required or permissible uses for
disclosure.
yes
no
yes
Set
Legal=[ GovernmentRecords:[ DPPA:[ stateConsentBroad ] ] ]
noyes
no
MR2a
ask
Were the data obtained from a federally
assisted drug abuse program after March
20, 1972, or a federally assisted
alcohol abuse program afte...
MR7
ask
Do the data contain information from a
covered entity or business associate of
a covered entity?
no
MR3
ask
Do any of the substance abuse records
contain patient identifying information,
or information that would identify a
patient as an alcohol or...
yes
MR2
ask
Do the data contain information related
to substance abuse diagnosis, referral,
or treatment?
yes no
MR8a
ask
Has an expert in statistical or
scientific principles and methods for
deidentification certified that the data
have been deidentified?
MR9
ask
Do the data constitute a limited data
set under the HIPAA Privacy Rule?
no
Set
Legal=[ MedicalRecords:[ HIPAA:[ expertDetermination ] ] ]
Code=green
yes
medicalRecordsCompliance
ask
Do the data contain health or medical
information?
yes no
Set
Legal=[ MedicalRecords:[ HIPAA:[ limitedDataset ] ] ]
Code=green
Set
Legal=[ MedicalRecords:[ HIPAA:[ businessAssociateContract ] ] ]
Code=orange
REJECT
For additional guidance on determining
whether an individual or organization is
a covered entity, please review the
charts provided by the Centers for
Medicare & Medicaid Services at
http://www.cms.gov/Regulations-and-Guidance/HIPAA-Administrative-Simplification/HIPAAGenInfo/Downloads/CoveredEntitycharts.pdf
http://www.cms.gov/Regulations-and-Guidance/HIPAA-Administrative-Simplification/HIPAAGenInfo/Downloads/CoveredEntitycharts.pdf
MR4a
ask
Were the data maintained by the
Department of Veterans Affairs either
shared with the patients’ consent for
the purposes of scientific resea...
Set
Legal=[ MedicalRecords:[ Part2:veteransMedicalData ] ]
Code=orange
yes
REJECT
reject data and flag for review;
possible violation of 38 U.S.C. 4132 due
to the release of VA medical records
without a valid exception.
no
Not Sure
no
MR8
ask
Have all direct identifiers been removed
from the data?
yes
Set
Legal=[ MedicalRecords:[ Part2:scientificResearch ] ]
Code=orange
MR4
ask
Were the data maintained in connection
with the US Department of Veterans
Affairs or US Armed Services?
yes
MR5
ask
Did the individuals provide written
consent for the disclosure of their
information?
no
MR10
ask
Do the data contain protected health
information from medical records? In
other words, do the data contain any
health information that can b...
MR11
ask
Do the data contain health information
from patients who have provided written
authorization for their information to
be disclosed to or use...
yes no
MR12
ask
Has an oversight body such as an
Institutional Review Board [IRB] or
Privacy Board reviewed how the data will
be used or disclosed and appro...
Set
Legal=[ MedicalRecords:[ HIPAA:[ waiver ] ] ]
Code=orange
yes
MR13
ask
Were the data disclosed pursuant to a
HIPAA business associate contract?
no
MR6
ask
Were the data shared by the substance
abuse program for scientific research
purposes?
no yes
yes no
no
Set
Legal=[ MedicalRecords:[ HIPAA:[ authorization ] ] ]
Code=orange
yes
no
Set
Legal=[ MedicalRecords:[ Part2:consent ] ]
Code=orange
yes
yes
Set
Legal=[ MedicalRecords:[ Part2:deidentified ] ]
Code=green
Assertions=[ DataType:[ Effort:deidentified ] ]
no
no
Set
Legal=[ MedicalRecords:[ HIPAA:[ safeHarborDeidentified ] ] ]
Code=green
yes
yes
no
todo
Privacy Act DUA
PA2
ask
Do the data contain personally
identifiable information?
Set
Legal=[ GovernmentRecords:[ PrivacyAct:[ identifiable ] ] ]
yes
Set
Legal=[ GovernmentRecords:[ PrivacyAct:[ deidentified ] ] ]
no
CIPSEA3
ask
Are the data being shared in
identifiable form?
Set
Legal=[ GovernmentRecords:[ CIPSEA:[ deidentified ] ] ]
no
Set
Legal=[ GovernmentRecords:[ CIPSEA:[ identifiable ] ] ]
yes
Census3
ask
Has a Census Bureau Disclosure Review
Board issued an approval memorandum
granting formal permission for the data
to be published?
Set
Legal=[ GovernmentRecords:[ Census:[ CensusPublished ] ] ]
Code=green
yes
REJECT
Possible violation of Title 13 due to
failure to obtain permission from the
Census Bureau to publish the Census
data.
no
ESRA1
ask
Do the data contain information from an
education research project that receives
funding from the Institute of Education
Sciences? Examples ...
REJECT
Possible violation of ESRA due to
failure to obtain authorization to
disclose the data from the Institute of
Education Sciences
CIPSEA1
ask
Were the data collected directly from
respondents for exclusively statistical
purposes under a pledge of
confidentiality?
PA1
ask
Are the data maintained by the agency in
a system of records?
no
CIPSEA2
ask
Is the agency that collected the data a
statistical agency or unit from this
list: USDA Economic Research Service,
National Agricultural Sta...
yes
Gov2
ask
Are you an employee or an agent, such as
a contractor, consultant, researcher, or
employee of a private organization,
operating under a cont...
Census1
ask
Were the data collected by the US Census
Bureau?
yes
no
no
Census2
ask
Were the data published by the Census
Bureau?
yes
Set
Legal=[ GovernmentRecords:[ Census:[ CensusPublished ] ] ]
Code=green
no
ESRA2
ask
Have the data been published previously,
or did you submit a copy of the data to
the Institute of Education Sciences
[IES] Data Security Off...
yes
Set
Legal=[ GovernmentRecords:[ ESRA:[ public ] ] ]
Code=green
Assertions=[ DataType:[ Harm:minimal ] ]
no
ESRA3
ask
Has the IES Data Security Office limited
access to the data to individuals
authorized by an IES license agreement
or individuals who have ex...
yes
yes no
no
Set
Legal=[ GovernmentRecords:[ ESRA:[ restricted ] ] ]
Code=yellow
Assertions=[ DataType:[ Harm:shame ] ]
yes
todo
CIPSEA DUA
Gov1
ask
Were the data collected by a federal
agency or a researcher under contract
with or with funds from a federal
agency?
no
yes
yes no
no yes
govRecsCompliance
questionnaire.flow-c1/Gov1
questionnaire.flow-c1/DPPA1
Current Questionnaire

Creating the Questionnaire
Legal team at the Berkman Center for Internet and Society
• Identifying regulatory and contractual
requirements most relevant to research data sharing
• Clustering legal requirements and approaches into
general categories
• Drafting annotated DataTags questionnaires
• Supporting implementation process

The Big Picture
• DataTags is part of the Privacy Tools for Sharing Research Data project 
(http://privacytools.seas.harvard.edu)
• (near!) Future versions of Dataverse will be able to consult a DataTags tagging
server during the data deposit process, and carry out the resultant dataset
handling policy
• Resultant tags will be used to create, e.g. DUAs, licenses, parameters for
privacy-preserving access
Risk%Assessment%
and-
De/
Identiﬁcation-
Diﬀerential%
Privacy-
Customized%&%
Machine/
Actionable-
Terms%of%Use-
Data%Tag%
Generator-
Data%
Set-
Query%
Access-
Restricted%
Access-
Consent%
from%
subjects-
Open%
Access%to%
Sanitized%
Data%Set-
IRB%
proposal%
&%review-
Policy%
Proposals%
and%Best%
Practices-
Database%of%
Privacy%Laws%
&%
Regulations-
Deposit%in%repository-
IntegraFon'of'Tools'

Thank you!
I Data
http://datatags.org
http://datascience.iq.harvard.edu/about-datatags

DataTags: Sharing Privacy Sensitive Data by Michael Bar-sinai

More Related Content

What's hot (20)

Viewers also liked (11)

Similar to DataTags: Sharing Privacy Sensitive Data by Michael Bar-sinai (20)

More from datascienceiqss (11)

Recently uploaded (20)

DataTags: Sharing Privacy Sensitive Data by Michael Bar-sinai