Lydia Pan / Mathematics / Faculty Mentor: Xinlei (Sherry) Wang
Many clinical studies report summary statistics such as the median, quartiles, and range in boxplots, rather than providing sample means and standard deviations, which are crucial for meta-analyses. To bridge this gap, we propose a novel Bayesian approach that utilizes the joint distribution of order statistics and weakly informative priors to estimate the mean and standard deviation while also quantifying uncertainty. Our method achieves lower Relative Mean Squared Error compared to existing approaches, producing nearly unbiased estimates for normally distributed data. It remains robust to small sample sizes, outliers, and deviations from normality. Furthermore, its uncertainty quantification provides insights into study quality and supports informed decisions regarding study inclusion in meta-analyses. Ultimately, our method enhances the reliability of quantitative synthesis in clinical research.
Leave a Reply