Search Filter
- Publications
- Abstract of Theses and Dissertations
- Database
- Optimum Stratification by Non-linear Programming
Optimum Stratification by Non-linear Programming
Thesis Abstract:
Optimum stratification in the case of multivariate surveys was obtained by minimizing sampling cost given fixed error levels. A mathematical model, in which every equation involved was expressed only in terms of the total sample size, n, and the number of strata, L, was set up and solved by a non-linear programming procedure called Box algorithm. The transformations of variance constraint equations in terms of n and L were based in Neyman's allocation which was the most efficient in the case of univariate surveys and Dalenius cum (squareroot of f) method which yielded optimum strata, boundaries given a fixed number of strata L.
The result of these transformations was an equal allocation of sample size to strata, nh = nL, h = 1,2,...L, where both n and L were determined using a FORTRAN program of Box algorithm.
Post-optimality analysis was also performed to determine the effect on the optimal solution by changes on certain constants and parameters. This was done by making the necessary changes in the mathematical formulation of the problem and the running the FORTRAN program again.