Does anyone know of software to fit a mixed model to nonlinear ordinal
regression data. In particular, I have confidence rating from memory
experiments which are often modelled by a set of thresholds on normal latent
variables (this is often called ROC analysis).
I want to estimate, using a mixed model (or similar, prehaps Baysian) approach,
the distributions over subjects of the threshold and mean and variance
parameters of the normal latent variables. The distribution of parameters (or
suitable transformation of the parameters) over subjects can be assumed normal.
Often the experiments have several within subjects conditions (e.g., the same
subjects might memorise common and rare words, each with each word type having
a corresponding latent variable). A mixed model approach is necessary both
becasue I want to do inference (e.g., do common and rare words have different
means and/or variances), and because shrinkage will be useful to control wild
parameter estiamtes that sometimes happen when fitting invidual data. Note that
when both the mean and variance of the normal latent variables are allowed to
vary the ordinal regression problem is nonlinear, so I cant use standard
packages (at least standard packages I know of).
Please let me know if anything is unclear in this description, I have tried to
avoid using jargon common to my applicaiton (analysis of recognition memory
data) but am not sure if I have the terminology right.
Cheers
Associate Professor ANDREW HEATHCOTE
Deputy Head, School of Behavioural Sciences
Aviation Building, University Avenue, Callaghan
University of Newcastle Fax: (61-2) 49216980
NSW, 2308, AUSTRALIA Office: 49215952
www.newcastle.edu.au/school/behav-sci/ncl
|