Jeffrey Bardzell, Program Head/Chair Justin Silverman, Thesis Advisor/Co-Advisor Sharon Xiaolei Huang, Committee Member
Keywords:
Covairation Analysis Uncertainity Quantification Sequence Count Data Bioinformatics
Abstract:
People may be curious about how the false assumption of scale will affect the correlation estimates. A good way to explore this impact is to separate the covariation equation into compositional and scale parts, and see how the error in scale parts caused by erroneous assumption propagates to the final estimand. This thesis explores the impact of erroneous modeling assumptions regarding scale on the fidelity of
covariation analysis estimates using high-throughput genomic data. We develop a mathematical representation of covariation analysis in terms of composition and scale and then use this foundation to study how errors in unmeasured scale parts propagate into correlation estimates. This study allows us to create new tools which allow applied researchers to characterize the sensitivity of correlation estimates to model misspecification and thereby identify when putative associations are likely false positives secondary to erroneous modeling assumptions.
