Research
Experimental Design
Experimental design is a cornerstone of statistical analysis, crucial for establishing causal relationships and ensuring the reliability of study outcomes. By systematically planning how data is to be collected and analyzed, experimental design allows researchers to control for confounding variables, minimize bias, and increase the validity of their results. This structured approach enables statisticians to draw meaningful conclusions with greater confidence. Whether in clinical trials, social sciences, or engineering, a well-designed experiment maximizes the efficiency of data collection, ensuring that valuable resources are not wasted.
Multi-Stratum Factorial Designs: a unified theory akin to minimum aberration [Chang and Cheng (2018); Chang (2022)]
Effect Aliasing in Gaussian Processes [Chang, Cheng, and Cheng (2019)]
De-Aliasing using Conditional Models via a Bayesian Perspective [Chang (2019)]
Design Theory for Conditional Models [Mukerjee, Wu, and Chang (2017); Chang (2023)]
Subdata Selection, Subsampling
Subsampling from massive datasets is an essential statistical technique that addresses several critical challenges inherent in big data analysis. As data volumes exponentially increase, full dataset analysis often becomes computationally expensive or even infeasible. Subsampling offers a practical solution by enabling the extraction of a representative, smaller subset of data that maintains the statistical properties of the entire dataset. This approach significantly reduces computational load and processing time, making data analysis more manageable and cost-effective.
Subdata Selection for Large Scale Deterministic Computer Models [Chang (2023)]
Supervised Stratified Subsampling (SSS): a model-independent subsampling method with model-robust predictive performance for regression problems [Chang (2024)]
