The ever-increasing complexity of Rasch datasets, combined with a desire to use standard statistical software for estimation has motivated advances in log-linear Rasch models. The relationship between log-linear Rasch models and logit-linear (or exponential) Rasch models is shown at www.rasch.org/rmt/rmt113r.htm
Log-linear models simplify estimation by eliminating nuisance parameters (usually those of the subjects), but often add complexity through the need for design matrices. They are also awkward to implement when data-points are missing.
Hatzinger & Katzenbeisser (2008) derive dichotomous and partial-credit log-linear Rasch models incorporating multiple time-points. This can be estimated using conditional maximum-likelihood estimation (CMLE) with standard statistical software, such as R. It models dependency across time-points, and also allows different subjects to be observed at different numbers of time-points. The example dataset has dichotomous data, 3 items, 45 subjects observed at up to 11 time-points.
The estimation of the parameters of multidimensional polytomous Rasch models presents an even greater technical challenge. Anderson et al. (2007) achieve it by formulating the Rasch model as a log-linear-by-association (LLLA) model. The multidimensional structure renders conventional CMLE impossible in general, so a pseudo-likelihood technique is employed. The overall likelihood of the data is decomposed into a set of parallel regression models which are maximized simultaneously. This is implemented in the plRasch package for R, and the SAS plgRasch macro. Example datasets have up to 30 items, 1000 subjects, 3 response categories and 2 dimensions.
Anderson, C.J., Li, Z., & Vermunt, J.K. (2007). Estimation of models in the Rasch family for polytomous items and multiple latent variables. Journal of Statistical Software, 20:4. www.jstatsoft.org/v20/i06
Hatzinger R. & Katzenbeisser W. (2008). Log-linear Rasch-type Models for Repeated Categorical Data with a Psychobiological Application. Department of Statistics and Mathematics, Wirtschaftsuniversitaet Wien. Research Report 69.
http://epub.wu-wien.ac.at/dyn/virlib/wp/eng/showentry?ID=epub-wu-01_dab
Advances in Log-Linear Rasch Models … Rasch Measurement Transactions, 2008, 22:2 p. 1167
The URL of this page is www.rasch.org/rmt/rmt222f.htm
Website: www.rasch.org/rmt/contents.htm