R help - May 2008 - computational ontology to research questions, and statistical methods

our group at Duke is currently attempting to connect two data sets fed
through a web-based system, one containing research questions (i.e.,
variables positioned with certain roles) and a database of statistical
methods.  this connection is done informally in a number of R packages
and related software such as Rcmdr, R4calc, and Rkward where a
variable of a certain type is matched to a certain test.  For example,
a logistic regression model requires a dichotomous variable as its
dependent variable.  our goal is to formalize this link through a
computational ontology, which would standardize and hopefully
facilitate package and library development for R.

we have started some parallel efforts regarding both our requirements
for this ontology as well a search for existing ontologies.  So far we
couldn't find any existing ontology, which seems odd given the
widespread application such an ontology has

Question for the list members:  Is anybody aware of any existing
ontologies computational focusing on this topic?

thanks


Ricardo

>  Question for the list members:  Is anybody aware of any existing
>  ontologies computational focusing on this topic?
Have you look at zelig (http://gking.harvard.edu/zelig/) ?  It unifies
a lot of statistical models within a framework that describes variable
types for input and output.

You might also try searching for "statistical expert systems" - e.g.
More Intelligent Statistical Software and Statistical Expert Systems:
Future Directions.  Gerald J. Hahn.  The American Statistician, Vol.
39, No. 1 (Feb., 1985), pp. 1-8.

Hadley


-- 
http://had.co.nz/