README - GLM design assessment algorithms

D C Woods
Southampton Statistical Sciences Research Institute
University of Southampton
UK

Email: D.C.Woods@maths.soton.ac.uk
WWW: http://www.maths.soton.ac.uk/staff/woods

[This is a temporary readme - more detailed documentation will be available shortly]

Two algorithms are available to carry out design assessments as described in Woods, Lewis,
Eccleston and Russell (2005).

1. coefficients.cpp - compares a number of designs across a specified parameter space 
through simulation. A set number of parameter vectors are drawn at random from the 
parameter space and a locally optimal design found via simulated annealing search for each. 
The efficiency of the designs under consideration is them compared to each of these locally 
optimal designs to approximate an efficiency distribution. See Woods et al (2005) Section 5 
for examples.

2. perturbation.cpp - used in Woods et al (2005), Sections 6 and 7, to evaluate the 
effectiveness for different parameter values of a design method for which compromises 
across GLMs with different link functions and/or linear predictors. Sets of parameter 
values are generated from specified normal distributions for the "largest" model (with most 
terms in the linear predictor). These are then perturbed via the addition of normal random 
errors (with variance proportional to the size of the original coefficient). For each 
generated set of initial parameters, a set of fully specified models is then obtained and 
compromise and locally optimal designs are then found and compared across this set. Hence, 
the effectivness of the design method can be assessed across a variety of parameter 
assumptions.

Compilation

Both algorithms require the Gnu Scientific Library (GSL) (http://www.gnu.org/software/gsl/).
Typical compilation commands when the GSL is installed as standard would be

g++ -o coeff.run coefficients.cpp -lm -lgsl -lgslcblas

g++ -o pert.run perturbations.cpp -lm -lgsl -lgslcblas

Input instructions and example input files can be found within the assessment.zip archived
file.

References

Woods, D.C., Lewis, S.M., Eccleston, J.A. and Russell, K.G. (2005). Designs for generalized 
linear models with several variables and model uncertainty. Technometrics, in press.