OECD Conference: Assessing the Feasibility of Micro-Data Access – The Feasibility Study

OECD Conference:
Assessing the Feasibility of Micro-Data Access
Discussion Session – The Feasibility Study
Luxembourg, 26-27 October 2006
Nadim Ahmad, Statistics Directorate, OECD
Options (I and II) - Indicators
Is it feasible to produce new suites of indicators? In
which areas?
If an average can be shown is it true that other
descriptive statistics related to it can be shown, for
example Var?
What about Co-variances and Regression
How can we proceed?
Options (III) - Bilateral Projects
How feasible are these? The beauty of international
access is the facilitation of international comparisons.
Are the resource/travel/logistic costs to great to
render this approach impractical.
Options (IV) – Remote Access
E-mail based systems versus Automatic. Are the
latter possible. Can algorithms be designed to make
them work (couple with creation of trusted and
educated user network and penalties)?
Are different approaches needed for different
What are the costs involved (for NSOs) in creating
common datasets?
Options (V) – Transmission of Micro-Data
How many countries will be able to provide
anonymised or confidentialised (Public-Use files)?
And in what areas? Social vs Economic
Will it be possible to harmonise formats, meta-data
and confidentiality rules?
Is the option of using privately sourced data more
practical? Despite the risks to NSOs. Are, and how,
these sources used in your office?
Developing Governance Rules
– No attempt made to identify individual responses
– Access restricted to vetted, trained and trusted OECD Users,
who are able to demonstrate a need for micro-data access
citing the potential impact on policy and benefits to microdata access providers.
– Recognition of potential for access-removal if breach occurs
– Dedicated OECD research centre team responsible for
secondary vetting of (trusted) users outputs.
– No access possible outside list of vetted OECD users.
– Widespread dissemination of results
Developing Confidentiality Rules
No statistic with <3? 5? 10? etc observations
Majority rules? What % constitutes disclosure?
Maximum number of variables?
Historic data