#### Abstract

In stratified sampling with k different variables and H strata it is often of interest to minimize the survey cost with respect to variance restrictions on each of the k variables. This problem has previously been solved using compromise solutions or using a linear approximation to this nonlinear problem. In this paper a nonlinear optimization routine is tested on this problem. The formulation of the problem in its original form proved problematic. For the test cases run, the transformation t_{h} = l/n_{h}, where n_{h} is the number of samples in stratum h, performed best when k and H are less than 7. As the number of strata and variables increase, the transformation t_{h } = n_{h} ^{2} performs better. In addition, simple modifications to the routine used can improve the convergence.

#### Creative Commons License

This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

#### Recommended Citation

Williams, M. and Schreuder, H. T.
(1993).
"A COMPARISON OF ALGORITHMS FOR SELECTING AN OPTIMUM SAMPLE FROM H STRATA USING k VARIABLES,"
*Annual Conference on Applied Statistics in Agriculture*.
http://newprairiepress.org/agstatconference/1993/proceedings/17

A COMPARISON OF ALGORITHMS FOR SELECTING AN OPTIMUM SAMPLE FROM H STRATA USING k VARIABLES

In stratified sampling with k different variables and H strata it is often of interest to minimize the survey cost with respect to variance restrictions on each of the k variables. This problem has previously been solved using compromise solutions or using a linear approximation to this nonlinear problem. In this paper a nonlinear optimization routine is tested on this problem. The formulation of the problem in its original form proved problematic. For the test cases run, the transformation t_{h} = l/n_{h}, where n_{h} is the number of samples in stratum h, performed best when k and H are less than 7. As the number of strata and variables increase, the transformation t_{h } = n_{h} ^{2} performs better. In addition, simple modifications to the routine used can improve the convergence.