Prediction variance properties of bridge designs Online publication date: Tue, 03-Nov-2015
by Kathryn Kennedy; Rachel T. Silvestrini; Douglas C. Montgomery; Bradley Jones
International Journal of Experimental Design and Process Optimisation (IJEDPO), Vol. 4, No. 3/4, 2015
Abstract: Bridge designs are compromise designs between D-optimal designs and Latin hypercube designs. Bridge designs are a good choice of experiments for computer simulation models with only small variance in the output response. This can occur in a wide range of computer experiments because variance reduction techniques are widely used in these studies. We expect that people will use results from bridge designs to fitting low order polynomial regression models. Bridge designs do not contain replicates, either in the original design or in projection, so they are also potentially good choices for other modelling strategies such as Gaussian process models. We compare bridge designs to D-optimal, I-optimal and Latin hypercube designs in terms of their theoretical prediction variance for experiments with 2-5 factors and polynomial models of order 2-5. The empirical results are also provided using test functions for Gaussian process models. We find that bridge designs are good choices for computer experiments when the underlying model is hypothesised to be a second or third-order polynomial, or a fourth-order polynomial of up to four factors. They also perform well in fitting Gaussian process models.
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Experimental Design and Process Optimisation (IJEDPO):
Login with your Inderscience username and password:
Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.
If you still need assistance, please email subs@inderscience.com