Heaping and its Consequences for Duration Analysis

This paper analyses the consequences of heaping in duration models.
Heaping is a specific form of response error typical to
retrospectively collected labor force status data. Respondents
round-off the spell length, when duration data is collected by
episode-based questionnaires. Calendar-based questionnaires instead
may lead to abnormal concentrations of the start and/or end of
spells at specific calendar months. The investigation concentrates
on this latter type of heaping, which Kraus and Steiner [1995]
identified for the unemployment spell data from the German
Socio-Economic Panel (GSOEP). In the special case of an exponential
model heaping with a symmetric zero-mean measurement error does not
bias the parameter estimate. In the Weibull model with duration
dependence, however, it is proven that even such a symmetric
heaping would lead to inconsistent estimation. We discuss the bias
for general heaping patterns and derive from this a proposal for
bias correction. In a number of simulation studies we check the
theoretical results. The Monte Carlo simulations also show that an
amount of heaping, that characterizes the GSOEP-West does not lead
to considerably biased parameter estimates of a Weibull model.
However, it clearly leads to spurious seasonal effects. Finally,
some directions of future work are indicated.