Exploring the Impact of Finite Sample Size and Missing Data on Statistical Modeling Decisions

Statistical models play a critical role in data applications for explanatory and predictive purposes. The model-building process involves use of various statistical tools, some of which make certain assumptions to yield good statistical properties like consistency. Such properties enable researchers to make reliable statistical inferences. However, when samples are small to moderate in size, issues arise when applying common model-building tools due to deviations from underlying assumptions. Such deviations can lead to unreliable parameter estimates, reduced statistical power, thereby affecting data-driven decisions. The process of model-building and conducting statistical tests is further impeded when data are incomplete due to missing values.

One arm of this research project investigates the impact of finite sample size on data-driven decisions. By understanding this impact, new modeling tools can be developed to improve data-driven decisions in cases where sample sizes are small to moderate. The second arm of this research project studies the impact of missing data on statistical inference, which is followed by developing methods to mitigate this issue. In data applications, traditional methods for dealing with missing data can be biased, inefficient, (mathematically) incongruent, and computationally intensive. This arm of research will develop new statistical tests and modeling methods for data with missing values. Methods developed in this research project will be applied in real health-related datasets.

Fields of research

• Statistical Modeling
• Asymptotic Theory
• Statistical Consistency/Efficiency
• Missing Data