Justin Zhu is a senior with a joint concentration in Statistics and CS. Justin works under Professor Susan Murphy at the Statistical Reinforcement Learning Lab on Zero-inflated Poisson Mixed Models for Mobile Health.
Student's perspective:
In my project, I focus on the bias and the coverage probability of Wald confidence intervals of the estimators within mobile health (mHealth). Preliminary simulation results show that the glmmTMB package is able to output consistent estimators with nominal coverage probability.
Mobile health (mHealth) aims to use smartphones and wearable sensors to deliver interventions to promote healthier behavior. Micro-randomized trial (MRT) is an experimental design to provide data to optimize mHealth interventions. In an MRT, each individual is repeatedly randomized among treatment options (e.g., receiving or not receiving a push notification) at hundreds or thousands of deci- sion points.
After each decision, a near-term, proximal outcome is measured. In this project, we focus on HeartSteps, a mHealth study where the intervention is a push notification for encouraging physical activity, and the proximal outcome is the minute-level step count of the individual for 30 minutes after the decision point. In order to understand how the mHealth intervention impacts an individ- ual’s physical activity level, it is important to model the treatment effect of the push notification on the 30-minute step count curve. To account for the excess zeros in the step count data, as well as the within person correlation in step count across de- cision points, we propose to use Zero-Inflated Poisson Mixed Models (ZIPMMs) for modeling the treatment effect on the minute-level step count. To effectively model ZIPMMs, we have worked on constructing generative models and conduct- ing simulation studies to evaluate the performance of the glmmTMB package in the R programming language for fitting ZIPMMs.