DEPARTMENT OF MEDICAL STATISTICS
DE MONTFORT UNIVERSITY, LEICESTER
Seminar Announcement
Time: Wednesday, 30th June 1999, 1 p.m.
Place: James Went Building, DMU Leicester
Room JW 6.06
Speaker: Professor J.K. Lindsey
De Montfort University, Leicester
Title: Data objects and model formulae for the future: some proposals.
Abstract:
Rectangular data matrices and linear formula specification are two of
the major impediments to developing and implementing advanced
statistical models.
A set of three data objects is proposed to replace the rectangular
data matrix used in statistical software packages. Although developed
for repeated measurements data, these have much wider application,
such as to independent data and to time series. The three objects
contain, respectively, the response variable, with all of the
information required to specify its probability distribution,
the inter-unit covariates, and the intra-unit covariates, necessary for
a set of models of interest.
As an extension to the Wilkinson and Rogers (1973) notation widely
used to specify the linear part of regression models, two general
procedures for describing nonlinear models are proposed. The first
simply uses the builtin function construction facilities of R or S,
whereas the second is more closely related to the Wilkinson and Rogers
notation, but allows unknown parameters to have individual names.
Then, such a formula can be automatically transformed into a
suitable function, with the software distinguishing between knowns and
unknowns.
Both the data objects and the model formulation have been implemented
in R. A wide collection of model-fitting functions for nonlinear
regression, including repeated measurements, based on them is available.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|