Unpublished,
Assessing quality of income data in a survey vs an administrative source
(2014)
Abstract
Research assesses the quality of labour income reported in a social
survey against indivuduals' records in the administrative income
tax file. Income data are known to be misreported in both sources,
for different reasons, giving rise to substantial descrepancies at
micro and macro levels of analysis. Income reportage in surveys is
also plagued by high rates of item non-response. We focus on three
novel issues: analyzing the distribution of discrepancies; investigating
factors associated with positive and negative discrepancies, separately
for the wage-earners and the self-employed individuals; utilizing
the survey software Blaise Auditrail feature for pinpointing the
effect of gros- and net-income items design on the quality of response.
We find a clear regression to the mean: overstatement of labor income
in the survey is more common among low-income respondents whereas
understatement is more widespread among high-income respondents.
We conclude that income mis-reportage by a wage-earner is determined
by individual's employment characteristics and fringe benefits. Misreporing
by the self-employed, however, remains merely unexplained, supposedly
due to the conceptual abiguity of income definition and measurement
in the survey. The Auditrail data appear to make an invaluable contribution
to the analysis of a respondent's cognitive behavior, monitoring
and improvement of data quality.
