A compilation of my statistical and data analysis projects, including:
Robust Variance Estimation to Model Complex Meta-Analytic Structures in the Immediate Antihypertensive Effects of Exercise
Objective: To determine and compare performance of the classical random effects and Robust Variance Estimation meta-analysis methods among studies with dependencies and moderators. We will apply these findings through the lens of the immediate blood pressure response to exercise termed, postexercise hypotension (PEH) and provide practical solutions for modeling common dependencies. Methods: Through the use of a Monte Carlo simulation, we created several simulation scenarios modeling a variety of common settings in PEH studies, including repeated measurements within the same subject. We compared the estimation and inference results of the Robust Variance Estimation (RVE) method to three variations of classical random effects meta-analysis and meta-regression with a time moderator: average effect across all time points (“Average”), one randomly selected observation (“Random”) across time points, and all observations across all time points (“Naïve). Results: Overall, RVE method performs best in terms of: 1) reducing bias of the overall and moderator effects; 2) correctly estimating coverage probability containing the true effect; 3) having a high power in detecting a significant nonzero overall or moderator effect; and 4) most accurately estimating the true t2 value. Under a few specific simulation settings, the classical methods performed just as well as, or better than the RVE method in some performance metrics when there was no time moderator effect. Conclusions: The RVE method performs best overall across the simulation scenarios considered and is recommended for PEH studies with time moderators and various levels of dependencies of effect sizes, along with other types of studies.
A Two Part Analysis on Model Comparison of Heart Rate across Single Aerobic Session and the Physiological Models between Eight Lagged Biometric Variables
In recent years, personal heart rate monitors have grown in popularity and are used anyone from professional athletes to the average weekend warrior. Athletes in particular have begun to utilize this technology to monitor their training and recovery status. These personal fitness tracking devices can track many variables at any given moment, including biomarkers such as heart rate, heart rate variability, heart rate zone, resting heart rate, as well as training characteristics such as distance and pace, and sleep variables such as time spent in each sleep phase. In addition to activity and biomarker tracking, many of the more sophisticated monitors are able to input the data into a model or algorithm to provide a score that indicates how well the user recovered or the strain of their training. This data can also be utilized to adjust the athletes training in order to allow the athletes to recover and minimize risk of injury. Most of these heart rate monitors are worn on the wrist or around the chest. However, “Convenience and comfort of the wrist-based devices has enabled them to largely replace chest straps that employ electrodes that measure cardiac electrical activity” (Pasadyn, 2019). The heart rate monitor worn on the chest has been set as the gold standard for heart rate monitors and is used to determine the accuracy of wrist worn monitors. Because the chest-worn heart rate monitors lack much of the sophistication and ease that the wrist monitors have, recent studies have attempted to compare the accuracy of the monitors. In this two-part analysis, I will (1) compare the heart rate capabilities of the Polar Ignite and WHOOP 3.0 wrist worn fitness tracker during an aerobic training session using ARIMA models, and (2) determine the physiological relationship between eight daily, lagged variables collected by the WHOOP 3.0 over the course of one month, using the vector autoregression method (VAR).
The goals of this indpendent project were to 1) analyze expenses and revenues of University of Connecticut (UConn) Athletic Department from 2015 to 2020 in light of recent budget cuts, 2)to determine any possible discrepencies in fund allocation between Men’s and Women’s teams, 3) to investigate trends of specific expenses by team and gender, and to determine areas of significant increase, and 4) to determine any statistically significant relationship between revenue and expense categories in an effort to increase revenue.
LinkedIn: https://www.linkedin.com/in/magdalene-mlynek-600695159/
Email: m.mlynek@lse.ac.uk