We have compared an IRT model with a sum score approach with indirectly measured phenotypes. Under a range of conditions, the IRT framework is to be preferred over using sum scores. For example, in longitudinal studies with data missing by design or changing measurement instruments, when some items in a questionnaire change across birth cohorts or across different ages or when item data are missing, a sum score approach may no longer be appropriate, but in many cases the analysis can still be meaningfully carried out in an IRT framework using parameter expansion (see, for instance, Glas 1998).
