Faculty: Qixuan Chen
Qixuan Chen, PhD
Associate Professor of Biostatistics
Director for MS Programs
My current research focuses on data integration, leveraging multiple data sources to provide more robust and efficient inference than using any single data source alone. My projects fall into three areas. The first focuses on combining probability and non-probability samples with population data sources. In a recent project, we utilized auxiliary information in the population to improve inference of population quantities from nonrandom samples in data-rich settings using Bayesian machine learning (Liu, Gelman, and Chen 2022). We applied the proposed method to improve generalizability, with the first application combining a telephone survey with administrative records and the second application combining data from a COVID-19 epidemiologic study with electronic health records. The second area centers on combining summary measures across multiple data sources. In another recent project, we developed a bivariate hierarchical Bayesian model for combining summary measures and their uncertainties from multiple sources, with applications in brain imaging, meta-analysis, and small area estimation (Yao, Ogden, Zeng, and Chen, 2022). Our study showed that the commonly used methods for meta-analysis can lead to biased inference when summary measures are correlated with their uncertainty measures, and that our proposed model can be used to correct the bias. The third area involves the integration of data prone to error with calibration data to correct for measurement error. An ongoing project is examining the association between exposure to indoor allergens and pediatric asthma morbidity. We utilize standards data in the immunoassay plates for measuring allergen concentrations to correct for measurement error in indoor allergens using a Bayesian general location model approach. We have also developed a constrained iterative conditional imputation approach with bootstrap inference for multiple imputation to correct for measurement error in a regression setting. This method allows leveraging two or more databases with disjoint measures in some variables due to measurement or classification error and with measurement error in exposure mixtures. We apply it in two ongoing projects; a marijuana use and homicide association study to correct for misclassification in self-reported marijuana use; and an HIV study to correct for measurement error in HIV viral load using the error-prone dry blood method.
As Director for all MS programs and Director of the Theory and Methods track, my mission is to strive for the success of our MS programs in preparing students to meet the challenges of real-world problem-solving. Our students are required to complete a research practicum to gain practical work and research experience. I am currently leading an effort to improve the practicum search, approval, and advising systems and thus enhance the quality of practicum experiences for our Master students.
References
Liu, Y., Gelman, A., Chen, Q. (2022). Inference from nonrandom samples using Bayesian machine learning, Journal of Survey Statistics and Methodology, https://doi.org/10.1093/jssam/smab049, to appear.
Yao, Y., Ogden, R.T., Zeng, Q., Chen, Q. (2022). Bivariate hierarchical Bayesian model for combining summary measures and their uncertainties from multiple sources, Annals of Applied Statistics, to appear.