D-optimal Designs for Quadratic Logistic Regression Models
2007 (English)In: International Journal of Statistical Sciences, ISSN 1683-5603, Vol. 6, no Special issue, 269-303 p.Article in journal (Refereed) Published
D-optimal designs are derived for certain quadratic logistic regression models. The performance of the D-optimal designs regarding maximum likelihood estimation of model parameters and estimation of the optimum of the response function is studied for different sample sizes. Comparisons are made with a couple of non-optimal designs. There were found to be disagreements between the asymptotic distribution and the small sample distribution of the maximum likelihood estimator. The designs are also evaluated as to what extent they suffer from the problem of non-existence of the maximum likelihood estimator. The probability that the maximum likelihood estimate exists is compared for the various designs. Non-existence proved to be a substantial problem for these quadratic logistic models.
Place, publisher, year, edition, pages
2007. Vol. 6, no Special issue, 269-303 p.
D-optimal design; Logit model; Response Surface Methodology; Maximum likelihood estimator; Separation
IdentifiersURN: urn:nbn:se:su:diva-14371OAI: oai:DiVA.org:su-14371DiVA: diva2:180891