Assessing the Application of Experimental Design Principles in Optimizing Statistical Power and Data Efficiency

Posted: Jun 13, 2018

• Levi Campbell
• Mateo Nelson
• Riley Baker

Abstract

The exponential growth of data availability has paradoxically created new challenges in experimental design and statistical inference. While massive datasets are increasingly accessible, the principles governing efficient experimental design have not kept pace with computational advancements. This research addresses the critical gap between classical experimental design methodology and contemporary data science practices. Traditional approaches to experimental design, developed primarily for physical and biological sciences, often fail to account for the unique characteristics of computational experiments, including their reproducibility, scalability, and the often complex, high-dimensional parameter spaces involved. Our work introduces a novel framework that systematically applies and extends experimental design principles to optimize both statistical power and data efficiency in computational contexts. The central research question we address is: How can classical experimental design principles be adapted and extended to maximize statistical power while minimizing data requirements in computational experiments? This question is particularly relevant in domains where data collection is expensive, time-consuming, or ethically constrained, such as clinical trials, educational interventions, or environmental monitoring. We challenge the prevailing assumption that larger sample sizes invariably lead to better experimental outcomes. Instead, we demonstrate that strategic experimental design can achieve superior statistical power with significantly less data. This counterintuitive finding has profound implications for research efficiency, resource allocation, and the generalizability of computational findings. Our approach integrates elements from response surface methodology, optimal design