1. The ‘Five Safes’ framework for data access management
Felix Ritchie
University of the West of England, Bristol
Administrative Data Service
Presentation by
Felix Ritchie
Professor of Applied Economics
Bristol Centre for Economics and Finance
6th September 2017

2. What is the most important factor in data access?
Overlapping problems:
‘Five Safes’: framework for structuring decision-making
anonymisation IT
users
allowed use
publication model

3. ‘Five safes’ approach The Five Safes = safe use Five components:
safe projects
safe people
safe settings
safe data
safe outputs
= safe use
Five components:
Is this an appropriate use of the data?
How trustworthy are the researchers?
Does the environment prevent misuse?
Is the data detail appropriate?
Is there any confidentiality risk from publication?
Consider jointly and severally

4. Joint but several Example: safe projects Several: Joint:
“can this data be used for research, assuming the users and environment can be certified to any standard we like?”
Joint:
“does our strategy for accrediting researchers meet the standard we assumed when looking at projects?”

5. Scales, not targets Balance is the key:
Similarities to multi-criteria decision analysis

6. Types of use Description and evaluation Design Training
Statistical authorities, data archives, HMRC, Cancer Research UK, Health Foundation, NIST, Scottish Health Informatics Project, American Academy of Sciences, RCUK, OECD, Eurostat…
Design
ADRN, Australian Bureau of Social Services, Greek Statistical Authority…
Training
consistent with DEA, GDPR, other modern laws

7. Use in design/evaluation
Where do you start?
No necessary priority for any component
Emerging best practice: focus on managerial choices
Project
People
Setting
Managerial controls
Data
Output
Statistical controls

8. Use in design/evaluation
Identify user needs
Identify broad shape
projects, people, setting
system view
Identify specific structures
projects, people, setting
managerial controls
residual statistical control
evidence-based
relevant to context
Address statistical risks
data, outputs

9. Summary Around for 15 years Well established, esp in UK, Australia, NZ
Part of a suite of evidence-based data access work
Framework for making decisions, not the decisions
attitudes
assessment of benefits
ethics etc
next set of presentations