Comparing Randomised and Systematic Designs for Optimal Input Mapping in Business, Marketing, and IT Education Field Trials

Abstract

Randomisation is a fundamental principle for en- suring unbiased treatment effects in experimental designs across various domains, including business, marketing, and information technology (IT) education. However, the choice between ran- domised and systematic designs must align with the experiment’s goals, particularly in large-scale field trials. This study investi- gates the suitability of these designs when mapping optimal input levels across a structured grid for educational interventions. A simulation study employing Bayesian hierarchical models and geographically weighted regression (GWR) revealed that, for extensive trials, randomised and systematic designs produce comparable results when fitting linear models or ignoring spatial variation. Conversely, for quadratic models, especially when spatial variation is significant, systematic designs outperform randomised designs in terms of achieving lower true mean squared errors (MSE) for coefficient estimation. These findings suggest systematic designs may offer enhanced robustness and reliability for designing and analyzing large-scale interventions in business, marketing, and IT education contexts, where precise spatial mapping and optimization are critical.