Characterizing Standard Variable Importance Measures and Growth Modeling of Bangledeshi Children over Two Years of Life
Cook, Heather, Statistics - Graduate School of Arts and Sciences, University of Virginia
Keenan, Daniel, AS-Statistics, University of Virginia
Creating interventions to avoid adverse events is an ongoing topic in numerous settings and thus it is often important to answer questions such as which treatments can be applied to avoid outcomes such as death or stunted growth. One may hope to answer these queries through the use of variable importance measures and through modeling the growth and development of individuals. Variable importance is an up and coming aspect of statistics that ranks variables in terms of some measure of importance which is often applied without the notion of either the exact meaning or how it can be compared with other regression or classification methods. Thus, characterizing standard variable importance measures could go a long way in the applicability and practicality of these ideas. In addition, confidence intervals and a lower threshold of importance was created and explored in order to advance the understanding and interpretability of such measures and methods. Simulations were conducted to show the behavior of such metrics with theoretical results stemming from a simple setting. In a specific population of Bangladeshi children from the PROVIDE study, growth models were explored where previous models have not correctly described these children's heights over the first two years of time, especially considering the plethora of covariates (900+). Developmental outcomes from 2 to 5 years of age were additionally modeled and explored. Throughout this research, the variable importance is described and explored in diverse manners while the children's heights and development is explained through various inclusive models.
PHD (Doctor of Philosophy)
statistics, growth modeling, variable importance
All rights reserved (no additional license for public reuse)