MLwiN uses both maximum likelihood estimation and Markov Chain Monte Carlo (MCMC) methods. MLwiN is based on an earlier package, MLn, but with a graphical user interface (as well as other additional features). MLwiN represents multilevel models using mathematical notation including Greek letters and multiple subscripts, so the user needs to be (or become) familiar with such notation.
Features:
Statistical procedures
MCMC estimation has been enhanced considerably (see the new manual downloadable from the web site). The following features have now been added (Note: some of these were also available in the development version 1.20):
- • The DIC diagnostic is now available (Spiegelhalter et al., 2002. Bayesian measures of model complexity and fit (with discussion). Journal of the Royal Statistical Society, B, (64): 583-640).
- • Multilevel factor analysis models can be fitted with multiple (correlated or uncorrelated) factors at each level.
- • Multicategory ordered and unordered response models can be fitted.
- • Multivariate mixtures of continuous and binary responses can be fitted.
- • Complex level 1 models can be fitted.
- • Multivariate models, including those with missing responses, can be fitted.
- • Adjustments for measurement error in predictors can be fitted.
- Cross classified models can be fitted.
- • Multiple membership models can be fitted.
- • Autoregressive structures at level 1 can be fitted.
- • Spatial data models can be fitted.
- • An interface with the WINBUGS software package is available.
Single level models can be specified using standard notation.
Interface design
Changes have been made to improve ease of use for the following menus:
- • The multivariate window has been removed and multivariate models are now set up using the responses button on the equations window.
- • There is now an add term button on the equations window for adding continuous variables, categorical variables and interactions to a model.
- • A notation button has been added to the equations window that allows switching between different notational representations.
- • A separate MCMC menu has been added.
- • Much improved interface for the specification of ordered and unordered categorical response models.
The default worksheet is now 1500 variables. The previously reserved columns have now had their column numbers increased by 1000.
