UNIVERSITY OF HERTFORDSHIRE STATISTICS GROUP SEMINAR SERIES

Sponsored by Department of Statistics, Accounting and
Management Systems, Business School, University of Hertfordshire.

Wednesday 22nd March 2000, time 2:30pm, Hatfield Campus, Room A209
Tony Fielding (University of Birmingham &
                    Institute of Education, University of London)

Title:
"Generalised models for ordered responses in multilevel and
 other complex data structures: examples in educational progress"

Abstract:

"Social science data in the form of ordered grades in research and
 official publications is often converted to points scores before
 analysis. This is a prevalent practice, for instance in dealing
 with educational achievement and progress data. In particular,
 much statistical modelling of hierarchically structured data
 through multilevel models with ordered grade dependent responses
 proceeds in this way. Statisticians and others have for some time
 pointed out the statistical and substantive drawbacks of imposing
 arbitrary continuous measurement scales on such lower level
 measures. These are particularly evident when the responses are
 very broadly grouped.

 Set against these inappropriate measurement assumptions is the
 familiarity and undoubted ease through widely available software
 of conventional linear models (multilevel or otherwise). Models
 and methods directly treating ordered grade responses have been
 available for simple single level structures for some time
 through generalised linear modelling (GLMs). These treat ordered
 data at no higher a level of measurement than their construction
 allows. There have been further theoretical and practical
 advances in random effects modelling for more hierarchical data
 structures through generalised linear mixed models (GLMMs). So
 far these have not seen widespread application and a general
 reliance has been on scores.

 This contribution  also recognises the need for the development
 and promotion of such modelling procedures for non-hierarchical
 but quite complex data structures, through cross classified and
 structures involving weighted random effects. For instance one of
 the applications to be discussed necessitates the disentangling
 of student, group, and teacher effects in a complex structure.
 Without attention to appropriate modelling and data collection
 such effects are often confounded.

 The contribution will introduce estimation procedures that have
 been developed for such ordered responses and complex situations
 through a macro ORDCAT. This can be implemented within the
 framework of the widely used, flexible, and well-supported
 programme multilevel modelling programme MLwiN. An opportunity
 will be afforded to contrast the application of these methods
 with traditional points scores from the viewpoints of
 appropriateness, practical feasibility, and estimator
 properties."


Wednesday 3rd May 2000, time 2:30pm, Hertford Campus, Room M1
Roger Jones (University of Hertfordshire)
Title and Abstract to follow.

Wednesday 7th June 2000, time 2:30pm, Hatfield Campus, Room A209
Michael Healy (Visiting Professor, University of Hertfordshire)
Title and Abstract to follow.


For further information regarding location, parking, etc.,
please contact Neil Spencer (details below)

****************************************
Dr Neil H Spencer,
Senior Lecturer in Statistics,
Department of Statistics, Accounting and Management Systems,
University of Hertfordshire,
Hertford Campus, Mangrove Road,
Hertford, SG13 8QF,
ENGLAND.
Telephone: +44 1707 285529
Fax: +44 1707 285489
E-mail: [log in to unmask]
****************************************



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%