Content area
Full Text
This is my third review of the ML series of packages from the Institute of Education's Multilevel Modelling Project for this journal. The theme of my first review (in 1996) was that multilevel modelling was important and the package (MLn) could do more than any other program and therefore was necessary for many researchers. It was statistics state-of-the-art in an area where applied researchers would begin to use multilevel modelling more frequently. It was a bit cumbersome, but there were not many alternatives.
Thanks to my first review, multilevel modelling did become popular! The multilevel discussion list ([email protected]) became larger and the contributors were less often statisticians. Several other specialist packages were developed (some freeware), functions for S-Plus were written (lme, nlme, glmmpql), and mainstream packages like SAS, SPSS, and SYSTAT began developing procedures (see http://multilevel.ioe.ac.uk/softrev/index.html for reviews). The multilevel project needed something to distinguish it from the competitors. With MLwiN 1.0, the team produced a great GUI where you could actually watch the equations solve, watch bootstrap estimates converge, etc. For people like me (and many of you), this was exciting. We showed our non-statistics colleagues this, and to our amazement they were less excited than us. With 1.0, many of the procedures became simpler and the state-of-the-art element continued. Through their research papers, their website, their newsletters, and help to people on the discussion list (particularly from Jon Rasbash and Min Yang), the project remained the intellectual hub of multilevel modelling. Of course, being the intellectual hub does not mean people should necessarily buy their software.
MLwiN 2.0 has to compete with specialist packages like HLM, WinBUGS, and the Mixed-up collection, as well as...