Mixed Stratification

Created by: Georgi Borros Date: 5 July 2026

Overview

This repository contains the code and data used to replicate a comparison of stratification methods: medoid-based stratification with genetic algorithm (GA) sample allocation, medoid-based stratification with proportional allocation, and the SamplingStrata method.

---

Code Replication

Folder/File	Description
medoids_mixed_gasampsi/	Code used to run the medoids for stratification with the GA allocation method, for each dataset
Proportional/	Code used to run the medoids for stratification with proportional allocation
SamplingStrata/	Code used to run the SamplingStrata method for the GHS application

---

Functions

core_functions/

Key functions used to perform the methods:

• function_calc_variance — Calculates measures of variation: the design effect and the coefficient of variation.
• function_sample_size — Calculates proportional allocation.
• medmix_gasampsi — The genetic algorithm function used by medoids_mixed_gasampsi.

---

Data

• multivar_datasets/ — The datasets used for replication.
• dataset_generation2026.R — Script used to generate the simulated datasets.

---

Running the Code

HPC

All code was run on the UCT HPC cluster (Ada). Simulated data results were parallelised across 40 cores to get a sense of variability across 40 runs. Application results were run over 10 runs. To reproduce locally, this code can be minimally edited by adjusting the number of cores and parallel runs (iter).

local_testing/

For a quicker, non-parallelised sanity check, local_testing/ contains scripts that let a user run each method for a single specified seed and inspect the outcome directly, without needing HPC access. This is useful for debugging a method or checking output structure before committing to a full 40-core HPC run.

---

SamplingStrata Folder

SamplingStrata/ contains the scripts used to run the SamplingStrata method for the GHS application, across both the continuous and atomic variants, using different CV inputs. Naming follows ss_ghs_<method>_<CV input>[_fpc], plus a set of Bethel allocation scripts that use MED-GA-derived stratifications and CVs.

Script	Description
ss_ghs_cont_1	Continuous SamplingStrata method, with manually specified 0.05 CVs as input
ss_ghs_atomic_1	Atomic SamplingStrata method, with manually specified 0.05 CVs as input
ss_ghs_cont_2	Continuous SamplingStrata method, with MED-GA CVs as input
ss_ghs_cont_2_fpc	Continuous SamplingStrata method, with MED-GA (FPC-adjusted) CVs as input
ss_ghs_atomic_2	Atomic SamplingStrata method, with MED-GA CVs as input
ss_ghs_atomic_2_fpc	Atomic SamplingStrata method, with MED-GA (FPC-adjusted) CVs as input
bethel_ghs_cont	Bethel allocation method, using MED-GA stratification and CVs (MED-GA inputs derived from ss_ghs_cont_1 inputs)
bethel_ghs_atomic	Same as bethel_ghs_cont, but using atomic-derived inputs
bethel_ghs_cont_fpc	Bethel allocation method, using MED-GA stratification and FPC-adjusted CVs (MED-GA inputs derived from ss_ghs_cont_1 inputs)
bethel_ghs_atomic_fpc	Same as bethel_ghs_cont_fpc, but using atomic-derived inputs

---

medoids_mixed_gasampsi/ Folder

Contains the scripts used to run the MED-GA method for the GHS application, using strata numbers and/or sample size inputs sourced from the SamplingStrata and Bethel runs above.

Script	Description
medmix_ghs_atomic	MED-GA method, using ss_ghs_atomic_1 inputs
medmix_ghs_cont	MED-GA method, using ss_ghs_cont_1 inputs
medmix_ghs_bethel	MED-GA method, using Bethel sample size inputs and ss_ghs_cont_1 strata number inputs
medmix_ghs_bethel_fpc	MED-GA method, using Bethel sample size inputs (where Bethel was supplied with FPC-adjusted CVs) and ss_ghs_cont_1 strata number inputs
medmix_ghs_bethel_atomic	MED-GA method, using Bethel sample size inputs and ss_ghs_atomic_1 strata number inputs
medmix_ghs_bethel_fpc_atomic	MED-GA method, using Bethel sample size inputs (where Bethel was supplied with FPC-adjusted CVs) and ss_ghs_atomic_1 strata number inputs
run_medmix_gasampsiMIX1 – run_medmix_gasampsiMIX6	Scripts to run the MED-GA method on the simulated datasets

---