Online Program

Towards Necessary and Sufficient Conditions for Transparency in Establishment Surveys (308202)

Keywords: metadata, standards, conformance, interoperability, usability, quality

In the past several years, transparency has become an important topic in the statistical community. The Promise of Evidence-Based Policy-Making report and the subsequent Evidence-Based Policy-Making Act in the US are important drivers. Reading the report, one realizes transparency is a precondition for evidence-based policy-making.

This coincides with the recent push to consider alternative sources for data in establishment surveys and the need for combining data from multiple sources. In this new world, transparency will make it clear what data sources are used to create some table or time series, how those sources are combined, statements about why the particular sources were suitable for use, etc. In the traditional sample survey world, transparency is just as important, for similar reasons.

How, then do you ensure transparency exists for some statistical program? First, we know transparency is situational. Different users have different needs. These needs translate into metadata requirements - providing enough information for the user to locate, understand, or use some statistical artifact for some specific purpose.

But, metadata requirements and specifications of those lead to questions: 1) How does one comply with a specification? 2) How good (what is the quality) of the metadata provided? 3) Is the system connecting the user with the instance of the metadata specification usable? 4) Is the information communicated understandable?

These questions lead to necessary criteria for transparency of statistical artifacts. Are these criteria also sufficient? In this paper, we identify 4 specific criteria for transparency and explore whether they are sufficient. Examples from the US Bureau of Labor Statistics are provided.