|
Activity Number:
|
384
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #309298 |
|
Title:
|
Process Modeling To Explain Ordinal Categorical Data
|
|
Author(s):
|
Simone Gray*+ and Alan E. Gelfand
|
|
Companies:
|
Duke University and Duke University
|
|
Address:
|
Institute of Statistics Decision Scien, Durham, NC, 27708,
|
|
Keywords:
|
Ordinal ; Categorical ; Data
|
|
Abstract:
|
Let us suppose that all classifying variables in a multi-way contingency table are ordinal. Then expected cell counts in the table can be viewed as driven by an intensity surface over R^m. We propose to model this surface as a realization of an m-dimensional spatial process. We can envision two versions - an ordinal classification associated with known scale and known categorical intervals, and an ordinal classification associated with a conceptual latent scale and unknown intervals. Moreover, we may have some of each, as well as nominal categories. Benefits of this approach include flexible modeling for the joint probabilities in the table and interpolation to other ordinal classifications for the variables. We detail modeling for such settings and the associated computation to fit these models. We illustrate with both simulated data and real data from the North Carolina birth records.
|
