A practical advantage of the analysis of data using an IRT framework is the use of incomplete item administration designs and handling of missing data. In some situations, intentionally incomplete item administration designs can greatly improve the efficiency of data collection. With an IRT approach one can also effectively deal with problems specific to longitudinal research where items differ across waves. When using IRT models in a maximum likelihood or a Bayesian framework, it is easy to include individuals that have missing data on one or more items if the data are missing at random (Little and Rubin 1987). When data are not missing at random, the non-randomness can be modelled within an IRT framework by expanding the model with an IRT model that describes the pattern of the missing data (see, for instance, Moustaki and Knott 2000; Moustaki and O’Muircheartaigh 2000; Holman and Glas 2005). The encompassing framework for handling missing data using IRT offers an important advantage over classical test theory. In classical test theory sum scores are only meaningful if the items are the same in all individuals and at all measurement waves.
