Over the past decades there has been an increasing use of panel surveys at the household
or individual level, instead of using independent cross-sections. Panel data have important
advantages, but there are also two potential drawbacks: attrition bias and panel conditioning
eÆects. Attrition bias can arise if respondents drop out of the panel non-randomly, i.e., when
attrition is correlated to a variable of interest. Panel conditioning arises if responses in one
wave are in∞uenced by participation in the previous wave(s). The experience of the previous
interview(s) may aÆect the answers of respondents in a next interview on the same topic, such
that their answers diÆer systematically from the answers of individuals who are interviewed
for the Ørst time. The literature has mainly focused on estimating attrition bias; less is known
on panel conditioning eÆects.
In this study we discuss how to disentangle the total bias in panel surveys due to attrition
and panel conditioning into a panel conditioning and an attrition eÆect, and develop a test for
panel conditioning allowing for non-random attrition. First, we consider a fully nonparametric
approach without any assumptions other than those on the sample design, leading to interval
identiØcation of the measures for the attrition and panel conditioning eÆect. Second, we
analyze the proposed measures under additional assumptions concerning the attrition process,
making it possible to obtain point estimates and standard errors for both the attrition bias
and the panel conditioning eÆect.
We illustrate our method on a variety of questions from two-wave surveys conducted in a
Dutch household panel. We found a signiØcant bias due to panel conditioning in knowledge
questions, but not in other types of questions. The examples show that the bounds can be
informative if the attrition rate is not too high. Point estimates of the panel conditioning
eÆect do not vary a lot with the diÆerent assumptions on the attrition process.