In this paper the Total Survey Error (TSE) paradigm is combined
with detailed data quality indicators to develop a framework for investigating
micro data quality. The TSE framework is widely used
in the survey methodology literature to identify different components
of error that arise in the survey process. Consequently, it provides a
very useful typology for researchers to understand which data quality
issues are relevant in applied work based on these surveys. In order to
demonstrate how the framework sheds light on micro data quality, two
labour market household surveys conducted by Statistics South Africa
are reviewed, spanning a time-frame from 1995-2007. It is argued that
efforts to improve data quality should involve a virtuous interaction
between producers and consumers of micro data and should be considered
an evolving process. For producers of data, the preparation
and publication of detailed data quality frameworks is recommended,
and two examples of these frameworks are reviewed. For consumers of
data, judicious analyses of the univariate, bivariate and multivariate
relationships in public-use versions of the datasets can help shed light
on different components of survey error, and should be communicated
back to survey organisations. Ultimately, improving data quality is
about being more explicit about the limitations of data production at
each stage of the process, which does not stop at initial public release.