Longitudinal modelling with longitudinal households

Publication type

Conference Paper


Paul Clarke

Publication date


The longitudinal analysis of household surveys is complicated by changes
in the composition of the sampled households over time: so-called
‘longitudinal households’. If one chooses household as the unit of
analysis then questions arise as to how to define the household as new
members arrive, old ones leave, or the household breaks up completely.
These questions can be avoided by taking individual as the level of
analysis, but this approach treats individuals as being in their own
separate worlds and ignores the influences of other household members.
In this workshop, we will review the methodology that has been developed
to address this problem, and give guidance for practice. This workshop
will comprise three papers:
1. About longitudinal households: We will discuss household change in
the British Household Panel Survey (BHPS) and Understanding Society, the
forces behind these changes, and the implications of these changes for
the definition of the longitudinal household.
2. Longitudinal modelling with longitudinal households: Standard
multilevel models include random effects for households but are
unrealistic for longitudinal households. Analyses can be divided into
those with individual-level and household-level outcomes. We will
review the use of multilevel models for this purpose, including
multiple-membership models, and illustrate with BHPS examples on health
for individual-level outcomes, and residential mobility for
household-level outcomes.
3. Handling drop-out: Techniques for handling drop-out (and other
missing data) using weights and model-based techniques will be reviewed.


Statistical Analysis and Surveys