A Robbins–Monro-based Sequential Procedure (Q4365076)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 1082044
Language Label Description Also known as
default for all languages
No label defined
    English
    A Robbins–Monro-based Sequential Procedure
    scientific article; zbMATH DE number 1082044

      Statements

      A Robbins–Monro-based Sequential Procedure (English)
      0 references
      0 references
      0 references
      14 December 1998
      0 references
      Fortran 77
      0 references
      sequential regression design
      0 references
      Robbins-Monro process
      0 references
      subroutine
      0 references
      \textit{B. K. Moser} and \textit{W. C. Fei} [Metrika 38, No. 6, 321-333 (1991; Zbl 0743.62076)] developed a sequential regression design based on the Robbins-Monro process. The design provides estimates of a regression function and its roots which are strongly convergent and asymptotically normal with minimum variance. The algorithm presented here implements the Moser-Fei procedure when \(E(Y| x)= M(x, {\mathbf L})\) is an increasing regression function of \(x\) and \(K\) unknown parameters \({\mathbf L}= (L_{p_1},\dots, L_{p_K})'\). The function \(M(x,{\mathbf L})\) is parameterized such that \(L_{p_i}\) is the root of the equation \(M(x,{\mathbf L})= p_i\) for \(i= 1,\dots, K\), i.e. the value of \(x\) such that \(M(x,{\mathbf L})= p_i\). The algorithm is presented as a subroutine which provides one update of the sequential process. After each update the program informs the user where to observe the next set of \(K\) design points. The \(K\) design points at the \((n+1)\)th update of the process are the values of \(x_{n+1,1},\dots, x_{n+1,K}\) given by \[ x_{n+1,i}= x_{n,i}- (y_{n,i}- p_i)/n \widehat{M}' (x,{\mathbf L}) \] for \(i= 1,\dots, K\), where \(x_{n,i}\) is the \(i\)th design point from the \(n\)th update, \(y_{n,i}\) is the corresponding observed value at \(x_{n,i}\), \(p_i\) is a specified constant in the range space of \(y_{n,i}\) and \(\widehat{M}' (x,{\mathbf L})\) is the derivative of \(M(x,{\mathbf L})\) with respect to \(x\) evaluated at \(x= L_{p_i}\) and \({\mathbf L}= (x_{n,1},\dots, x_{n,K})'\). The algorithm also produces estimates of any root of \(M(x,{\mathbf L})\), estimates of any function of the \(K\times 1\) vector \({\mathbf L}\) and estimates of \(M(x,{\mathbf L})\) for specified values of \(x\). Estimated asymptotic variances and covariances of these estimators are then generated by the subroutine.
      0 references

      Identifiers