Optimal design for an inverse Gaussian regression model (Q1087280)

Let \(Y_ 1,Y_ 2,...,Y_ n\) be n independent random variables following inverse Gaussian distribution such that \(Y_ i\sim IG(\theta_ i,\lambda)\), where \(\theta_ i^{-1}=\beta 'x_ i\). Let \(\psi =(\beta ',\lambda)'\) and \(\Sigma\) be the dispersion matrix of the maximum likelihood estimator of \(\psi\). With \(x\in [0,1]\), a D-optimal design corresponds to an allocation of the n design points \(x_ 1,x_ 2,...,x_ n\) which maximizes a specified approximation of the determinant of \(\Sigma\) and is shown to be a two point design with \(n_ 0\) trials at \(x=0\) and \(n-n_ 0\) trials at \(x=1\), where \(n_ 0=n/2\) if n is even, and \(n_ 0=(n-1)/2\) or \((n+1)/2\) if n is odd.